Download to read offline
![Investigating the empirical evidence for
understanding vulnerability and the associations
between poverty, HIV infection and AIDS impact
Stuart Gillespiea, Robert Greenerb, Alan Whitesidec and
James Whitworthd
AIDS 2007, 21 (suppl 7):S1–S4
It is just over 25 years since the first cases of AIDS were were dead, killed in the First World War. It is only in the
reported. Over this quarter-century, AIDS has become past decade that the last of these spinsters has died. The
one of most highly studied diseases in history. There impacts of AIDS will take even longer to work through
have been significant medical advances in understanding the population.
the consequences of HIV infection and treating AIDS, as
is well documented in many journals, including AIDS. Second, HIV is diverse in its spread. Early fears that the
The complex and place-specific social, economic, virus would spread rapidly outside Africa have not
behavioural and psychological drivers of the spread of materialized. For example, the UNAIDS 2006 ‘Report
HIV remain less well delineated. The consequences of on the global AIDS epidemic’ estimated that there were
increased illness and death in poor countries and commu- 5.7 million people living with HIV in India. In July 2007,
nities are still unfolding. this was revised downward to 2.5 million, reflecting much
less spread of the infection than had been feared [2].
In 2000, HIV was placed firmly on the global development Similar downward revisions of estimates have been made
agenda by UN Security Council Resolution 1308, which in China. In a recent book, James Chin [3] argued that
stated: ‘the spread of HIV can have a uniquely devastating there are many populations in which heterosexual
impact on all sectors and levels of society’. A year later, in epidemics will not occur in the general population and
July 2001, there was a UN General Assembly Special the epidemic will remain confined to specific risk groups.
Session on HIV/AIDS. Since then our understanding of Chin’s examples of where the potential for HIVepidemics
the epidemic and its potential impacts has deepened. This has been overstated are primarily from Asia, and in
supplement, written by social scientists, looks at how particular China and the Philippines. This is not to
socioeconomic determinants drive HIV spread and how understate the individual tragedy of each infection, but
AIDS illness and mortality is impacting on communities. rather to recognize that there are countries where AIDS
will have a considerable impact and others where its
It is helpful to locate the contents of this supplement in importance can be downgraded.
the context of the history of the epidemic. There are three
overarching points to be made in introduction. First, the It is not just globally that there is wide variation. In
epidemic is complex both in terms of what is driving it mainland sub-Saharan Africa HIV prevalence in adults
and the effects it has. It has been described as a ‘long wave ranges from 0.7% in Mauritania to 33.4 % in Swaziland.
event’. It takes years for the epidemic to spread through The hardest-hit countries are all in southern Africa; these
society and generations for the full impact to be felt. A are shown in Fig. 1, the so-called ‘red’ countries. Adult
recent book highlights the nature of such long wave HIV prevalence exceeds 20% in four of these countries:
events [1]. ‘Singled out: how two million women Swaziland, Lesotho, Botswana and Zimbabwe. South
survived without men after the First World War’ describes Africa, Namibia, Zambia, Mozambique, and Malawi all
how in the United Kingdom a generation of women were have adult prevalence rates in the range of 10–20% [2].
unable to marry, as the men they would have partnered These countries are the focus of this supplement.
From the aInternational Food Policy Research Institute, Geneva, Switzerland, the bJoint United Nations Programme on HIV/AIDS,
Geneva, Switzerland, the cHealth Economics and HIV/AIDS Research Division, University of KwaZulu-Natal, South Africa, and
the dWellcome Trust, London, United Kingdom
Correspondence to Alan Whiteside, Health Economics and HIV/AIDS Research Division, University of KwaZulu-Natal, Block
J418 Westville, University Road Westville, Private Bag XS4001, Durban, 4000, South Africa.
Fax: +27 (31) 260 25 87; e-mail: whitesid@ukzn.ac.za
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S1](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-6-2048.jpg)
![S2 AIDS 2007, Vol 21 (suppl 7)
deficiency virus (HIV) was identified as the cause. The
number of cases rose rapidly across the United States and
was quickly identified in Europe, Australia, New Zealand
and Latin America. In central Africa, health workers were
observing new illnesses such as Kaposi’s sarcoma (a cancer)
in Zambia, cryptococcosis (an unusual fungal infection) in
Kinshasa, and there were reports of ‘slim disease’ and
unexpectedly high rates of death in Lake Victoria fishing
villages in Uganda [6–8]. These illnesses were occurring in
heterosexual adults, not just gay men, individuals with
haemophilia, blood transfusion recipients, and intravenous
drug users, who formed the main groups at risk in
developed countries. By 1982, cases were being seen in the
partners and infants of those infected [8,9].
The initial response of public health specialists, epide-
miologists and scientists was to try to identify what was
causing the disease and to understand how it was
spreading. This would inform prevention strategies and
Fig. 1. Map of adult HIV prevalence in Africa. 20–34%; medical interventions. Early responses were therefore
10–< 20%; 5–< 10%; 1–< 5%; < 1%. predominantly scientific and technical in nature.
Third, social science faces problems in addressing the It soon became apparent, however, that this was not
phenomenon of HIVand its consequences. The epidemic enough, and attention shifted to understanding why
is only 25 years old, which means that it, and its effects, are people were being exposed. This led to early knowledge
still unfolding. Social science relies on assessing what has attitude and practice surveys, which sought to understand
happened. This is done through surveys and panel data, high-risk behaviours [3] p.73. This emphasis on
and sometimes the picture is at odds with what we expect. prevention gained momentum because medical scientists
For example in the 1980s it was suggested, on the basis of had not yet discovered drugs that could cure, or even slow,
models, that AIDS would cause economies to grow more the progress of the disease. Initial optimism for developing
slowly than otherwise would be the case. In 2007, at the an effective vaccine soon faded and is now seen to be
individual country level, this does not seen to have many years, if not decades, away.
occurred. Uganda had the worst epidemic in the world
during the early 1990s yet managed consistent economic Internationally, the World Health Organization (WHO)
growth estimated at 6.5% per annum from 1991 to 2002. took the lead in response to HIV in 1986; teams visited
Botswana’s growth rate over the same period was 5.6%. most developing countries to establish short and
South Africa has seen steady growth since 1999. Yet it is medium-term AIDS programmes, which then evolved
only through longitudinal and cross-sectional studies that into national AIDS programmes [10]. International
we can hope to understand the impact of the disease. responses to HIV were, however, limited and character-
Longitudinal panel data give a picture of what has ized by denial, underestimation, and oversimplification.
happened in a population over the period for which the HIV was not placed high on the agenda of any other
data are collected. An alternative is to gather cross- United Nations agency. Although life expectancy was
sectional data: if we can understand what has happened in plummeting in certain African countries, for example,
Uganda will it help predict what might happen in the United Nations Development Programme waited
Lesotho? The one thing we have not been good at is until 1997 to take this into account in calculating its
predicting the future, although UNAIDS made a brave human development index [11].
attempt at this through its ‘AIDS in Africa: three scenarios
to 2025’ report launched in March 2005 [4]. By the 1990s there was a new perspective developing, as
interest in the individual, social, and economic milieux
that lead to vulnerability to HIV infection began to grow.
Academics and programme officers increasingly recog-
A brief history of 25 years of response nized that social justice, poverty and equity issues were
driving the uneven spread of the virus within and
1981–1996 between communities and societies [12–15].
The AIDS epidemic was recognized in 1981, initally
among gay men in New York and San Francisco [5]. It was 1996–2007
officially named ‘acquired immune deficiency syndrome’ In 1996, there were major changes in response to HIV,
(AIDS) in July 1982, and in 1983 the human immuno- reflecting and reflected in the scholarship of the time. In](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-7-2048.jpg)
![Introduction Whiteside et al. S3
the 1994 book ‘AIDS in Africa’ of 33 chapters only three inequity, long-term concurrent partnerships, the lack of
were on preventive strategies and four on socioeconomic male circumcision, and the prevalence of co-infections
impact, the rest were scientific or epidemiological [16]. are factors that have been identified and need further
By 1996, when the second edition of ‘AIDS in the world’ examination. There are no easy solutions to curbing the
was published, of 41 chapters only approximately 18 were spread of the epidemic. There are countries, outside
pure science [17]. southern Africa, where the epidemic appears to be under
control: Uganda brought early hope to Africa by showing
In 1996, the new UN agency charged with coordinating how high levels of political commitment and com-
the response to the epidemic, UNAIDS, began operations munity-led responses can work to stabilize HIV
in Geneva. This was significant as it acknowledged that prevalence. In other locations, such as Tanzania, infection
the international health body the WHO was not able to rates peaked at a lower level than those currently seen in
respond to the epidemic in all its facets, and there needed most of southern Africa.
to be international coordination for an exceptional
disease. At the XIth International AIDS Conference in The focus of this supplement is on bringing together and
Vancouver, the arrival of new drugs in developed understanding the data on the socioeconomic dimensions
countries to treat AIDS was announced, and mortality of the epidemic. It came out of a meeting sponsored by
among those being treated plummeted. UNAIDS and hosted by the Health Economics and
HIV/AIDS Research Division of the University of
At the XIIIth International AIDS Conference in KwaZulu-Natal held in Durban from 16 to 18 October
Durban, South Africa, in July 2000, Nelson Mandela, 2006. The aim of the symposium was to bring together
closed the conference with a call for drugs to be made people, especially those involved in field research, to share
accessible to all. Since then, the response to AIDS has knowledge and experience and to address gaps in our
been dominated by new initiatives for making treatment understanding of the spread of HIV and impact of AIDS.
accessible, especially in developing countries. The price In particular, we were looking for community-
of drugs has fallen dramatically with the manufacture of based longitudinal studies currently being carried out
generic drugs.1 In 2001, United Nation’s Secretary in Africa.
General, Kofi Annan, called for spending on AIDS to be
increased 10-fold in developing countries, and the The outputs of this meeting were to be a review of the
Global Fund for AIDS, TB and Malaria was established. main longitudinal socioeconomic data collections in
The same year, President George W. Bush announced Africa with a bearing on HIV, the publication of the
the Presidential Emergency Plan for AIDS Relief participants’ best papers, and an opportunity to network
(PEPFAR) targeting 15 developing countries. In 2003, and share ideas.
the WHO and UNAIDS proclaimed the ‘3 by 5’ plan, to
treat 3 million people in poor countries by the end The meeting was a qualified success in that papers were
of 2005. presented and we have this interesting and thought-
provoking supplement. There are, however, a number of
Over the decade from 1996 to 2006, more financial caveats, and these cut to the heart of the issues we are
resources than ever before were made available for the dealing with. South African research and papers
response to AIDS, with emphasis increasingly on making dominate. Of the 11 papers we publish, eight are from
treatment available in developing countries. In 1996, South Africa, two compare data from across the continent
there was approximately US$300 million for HIV/AIDS and one is from Zimbabwe. This is also true of the
in low and middle-income countries; by 2006, this authors, the vast majority are either South African or
increased to US$8.3 billion. It is noteworthy that this based in the developed world. Clearly, there are real issues
response, largely a result of treatment becoming with developing capacity in African countries. The global
available and affordable, led to a ‘remedicalization’ of emphasis is on delivery not research, but, as this
HIV/AIDS. supplement shows, quality data and good science are
essential.
It is not clear why southern Africa has been so hard hit by
HIV. Socioeconomic variables, cultural factors and sexual Of the ten papers we publish, seven are from South Africa
behaviour all play a role. Poverty, income inequality, sex two compare data from across the continent and one is
from Zimbabwe. This is a good spread. What do the
papers tell us? Put simply, the causes and consequences of
1
Presentation by Peter Graaf of the HIV/AIDS Department of the the epidemic are complex and policy needs to take this
WHO to an ‘Informal technical consultation on the relevance and into account.
modalities of implementation of an observatory for HIV commodities
in Africa’ organized by Health Economics and HIV/AIDS Research
Division (HEARD), University of KwaZulu Natal, the World Health Although poor individuals and households are likely to be
Organization, and Swedish/Norwegian HIV/AIDS Team on 25 June hit harder by the downstream impacts of AIDS than their
2007. less poor counterparts, their chances of being exposed to](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-8-2048.jpg)
![Is poverty or wealth driving HIV transmission?
Stuart Gillespiea, Suneetha Kadiyalab and Robert Greenerc
Evidence of associations between socioeconomic status and the spread of HIV in
different settings and at various stages of the epidemic is still rudimentary. Few existing
studies are able to track incidence and to control effectively for potentially confounding
factors. This paper reviews the findings of recent studies, including several included in
this volume, in an attempt to uncover the degree to which, and the pathways through
which, wealth or poverty is driving transmission in sub-Saharan Africa. We investigate
the question of whether the epidemic is transitioning from an early phase in which
wealth was a primary driver, to one in which poverty is increasingly implicated. The
paper concludes by demonstrating the complexity and context-specificity of associ-
ations and the critical influence of certain contextual factors such as location, sex and
age asymmetries, the mobility of individuals, and the social ecology of HIV trans-
mission. Whereas it is true that poor individuals and households are likely to be hit
harder by the downstream impacts of AIDS, their chances of being exposed to HIV in the
first place are not necessarily greater than wealthier individuals or households. What is
clear is that approaches to HIV prevention need to cut across all socioeconomic strata of
society and they need to be tailored to the specific drivers of transmission within
different groups, with particular attention to the vulnerabilities faced by youth and
women, and to the dynamic and contextual nature of the relationship between socio-
economic status and HIV. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 7):S5–S16
Keywords: socioeconomic status, poverty, inequality, HIV, gender, prevention
Introduction to have better access to reproductive healthcare, condom
use is generally low in Africa and other parts of the
Evidence of the association between HIV transmission developing world. Pre-existing sexual behaviour patterns
and socioeconomic status is mixed [1–3]. Although early (from ‘pre-HIV’ times) therefore make the richer and the
studies tended to find positive correlations between better educated more vulnerable to HIV infection,
economic resources, education and HIV infection [4,5], especially in the early stages of the epidemic, when
as the epidemic has progressed, it has increasingly been information about the virus and how to protect oneself is
assumed that this relationship is changing. Evidence of the usually low [6,8]. At a later stage, however, it has been
degree, type and dynamics of the influence of socio- argued that individuals with higher socioeconomic status
economic factors on rates of HIV transmission in different tend to adopt safer sexual practices, once the effects of
settings and at various stages of the AIDS epidemic is, AIDS-related morbidity and mortality become more
however, still rudimentary. This paper seeks to bring apparent, adding greater credibility to HIV prevention
together what is known on this, drawing especially on the messages [9,10].
findings of some recent studies, including several in
this supplement. Another currently postulated dynamic is that poverty
(possibly itself fuelled by AIDS) is increasingly placing
In most countries, relatively rich and better educated men individuals from poor households at greater risk of
and women have higher rates of partner change because exposure to HIV via the economically driven adoption of
they have greater personal autonomy and spatial mobility risky behaviours. Poverty and food insecurity are thought
[4,6,7]. Although the richer and better educated are likely to increase sexual risk taking, particularly among women
From the aInternational Food Policy Research Institute, Geneva, Switzerland, the bInternational Food Policy Research Institute,
Washington, DC, USA, and the cJoint United Nations Programme on HIV/AIDS, Geneva, Switzerland.
Correspondence and requests for reprints to Stuart Gillespie, International Food Policy Research Institute, c/o UNAIDS, 20 Avenue
Appia, CH-1211 Geneva 27, Switzerland.
E-mail: s.gillespie@cgiar.org
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S5](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-10-2048.jpg)
![S6 AIDS 2007, Vol 21 (suppl 7)
who may engage in transactional sex to procure food Does poverty increase exposure to HIV?
for themselves and their children. Women’s economic
dependence on their partners may also make it difficult for At the country level there is a weak positive relationship
them to insist on safer sex (e.g. condom use). In addition, between national wealth and HIV prevalence across
poor people are more likely to be food insecure and countries in sub-Saharan Africa, where higher prevalence
malnourished. Malnutrition is known to weaken the is seen in the wealthier countries of southern Africa
immune system, which in turn may lead to a greater risk of (Fig. 1). Strong urban–rural economic linkages, good
HIV transmission in any unprotected sexual encounter transport links and high professional mobility may
(although this remains under-researched). This strand of translate into both higher incomes and higher HIV
literature on HIV transmission in Africa stresses the reversal incidence. National poverty rates, on the other hand, do
in the distribution of the epidemic across population not show a strong association with HIV prevalence
subgroups as the epidemic advances within countries, with (Fig. 2). There is, however, a clear and significant pattern
those of lower socioeconomic status experiencing a higher of association between income inequality and HIV
subsequent rate of HIV transmission. prevalence across countries; countries with greater
inequality have higher HIV prevalence, especially in
We aim to present an overview of the findings of key sub-Saharan Africa but also to a lesser extent in Asia and
recent African studies (primarily 2004–2007) examining Latin America (Fig. 3).
the relationship between economic resources/status and
the risk of HIV infection (see Table 1). The starting point Household level evidence that poverty is a major driver of
was the evidence presented in this supplement on this the epidemic is rather mixed. It is important, however, to
relationship, but our search then expanded to draw upon note that most studies focus on relative poverty in the
other recent literature from sub-Saharan Africa where the context of generalized chronic poverty. In most cases, it is
epidemic is most severe. only the highest one or two quintiles (or possibly three in
middle-income southern African countries) that can be
First, PUBMED and ECONLIT searches (2004–2007) thought of as representing the non-poor, using the
were used to identify all studies addressing the link standard poverty line definitions, or the US$1 or US$2
between socioeconomic status (poverty and education in per day measures adopted for the purpose of global
particular) and the risk of HIV. Searches were limited to comparison. Comparisons are thus between ‘wealthier’
English language and Africa. Keywords pertaining to the and ‘poorer’ groups.
explanatory variables were ‘poverty’, ‘wealth’, ‘socio-
economic status’, ‘socioeconomic’, ‘education’ and Studies adopting ethnographic methodologies suggest
‘education level’. Keywords pertaining to the outcome that material poverty increases the risks of contracting
variable of interest were ‘HIV risk’, ‘HIV transmission’, HIV mainly through the channel of high-risk behaviour
‘sexual behaviour’ and ‘HIV prevalence’. Studies on adoption. The respondents of an ethnographic study in
special groups of populations such as truck drivers and the southern province of Zambia [26] identified frequent
uniformed services have been excluded. Conceptual/ droughts and limited wage labour opportunities, after the
theoretical papers have not been included in the review of post-economic liberalization closure of companies, as the
the association between socioeconomic status, poverty, ‘push’ factors behind the increasing resort of women to
education and the risk of heterosexual HIV transmission, transactional sex. In a qualitative study in Malawi [27]
although such studies have been used from a reference certain social groups were found to continue to engage in
perspective. Quantitative studies with only descriptive high-risk behaviours despite knowing the risks. They did
statistics have been excluded. Sixteen of the 49 retrieved so, the authors contend, to affirm their social identity and
articles were thus excluded. In addition, a Dissertation to deny that ‘anything they do makes a difference to what
Abstracts Online search and a Google Scholar search were they perceive as a life of powerlessness and despair’ (p. 17).
also conducted to identify pertinent recent grey literature. The ‘culture of poverty’, as documented by Lewis [28] in
Whenever possible, the authors of such papers that met Latin America, may thus be as significant as material
the above criteria were contacted for the latest drafts and poverty in motivating risky behaviours.
updates on the status of their articles.
The findings from several recent quantitative surveys that
As such, this overview is intended to complement earlier investigated the relationship between economic depri-
reviews examining this relationship [23,24]. It then seeks vation and the adoption of high-risk behaviours are
to delve deeper into the pathways and interactions that generally consistent with much of the qualitative research
contextualize the link between wealth/poverty and [29–31], although there are important differences
heterosexual HIV transmission risk. We stress at the between behaviours and regarding the influence of
outset that we are not reviewing evidence of the gender in different contexts [12,14,32].
downstream impacts of AIDS on poverty, a subject that
has been comprehensively covered recently elsewhere Employing the Cape Area Panel Study, which surveys
[23–25]. individual youths aged 14–22 years in Cape Town, South](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-11-2048.jpg)
![Table 1. Recent quantitative studies examining the relationship between HIV and socioeconomic status.
Study Objective Study design and statistical analyses Key findings
Dinkelman et al. [11] Estimate if sexual debut between 2002 Cape Area Panel Study data that surveyed Household income negatively associated with sexual
and 2005, number of recent partners 4752 boys and girls, 14–22 years of age debut, and economic shocks positively associated
and lack of condom use at last sex in Cape Town, South Africa (2002–2005). with multiple partnerships among girls. Community
in 2005 is affected by household Multivariate probit models poverty rates predict earlier sexual debut and
income constraints and income shocks. higher rates of unprotected recent sex for boys.
Schooling positively associated with a significant
condom use, but negatively associated with
multiple partners for both boys and girls.
Weiser et al. [12] Studies the association between food Cross-sectional population-based survey of Food insufficiency associated with inconsistent
insufficiency (not having enough food 1255 adults in Botswana and 796 adults condom use with a non-primary partner, sex
to eat over the previous 12 months) in Swaziland. exchange, intergenerational sexual relationships, and
and inconsistent condom use, sex Multivariable logistic regression analyses, lack of control in sexual relationships. For men,
exchange, and other measures of risky sex. clustered by country, and stratified by sex. food insufficiency was associated with increase in
the odds of unprotected sex only. Higher
educated women, but not men, were less likely to
report high-risk behaviours.
Johnson and Way [13] Investigates the association between Cross-sectional, 2003 Kenya Demographic Wealth was positively related to HIV-positive
demographic, social, behavioural, and Health Survey. serostatus for both men and women. Women
and biological variables and HIV Multivariate logistic regression model with primary education were nearly twice as likely
serostatus in Kenya. stratified by sex. to be HIV positive as those with no education.
Sexual behaviour factors were not significantly
associated with HIV serostatus.
Nii-Amoo Dodoo et al. [14] Examines the relationship between Quantitative data are drawn from the Although poverty was significantly associated with
HIV-related sexual activity outcomes, Demographic & Health Surveys (DHS) the examined sexual outcomes in all settings, the
specifically age at first sex and multiple and qualitative data from the Sexual urban poor are significantly more likely than their
sexual partnerships, and socioeconomic Networking and Associated Reproductive rural counterparts to have an early sexual debut
deprivation amenities index, (based on and Social Health Concerns study. and a greater incidence of multiple sexual partnerships.
asset index and amenities index) in rural Multivariate Cox regressions. The disadvantage of the urban poor is accentuated
and urban Kenya. for married women; those in Nairobi’s slums are at
least three times as likely to have multiple sexual
Poverty, wealth, HIV transmission Gillespie et al.
partners as their rural counterparts.
Lopman et al. [15] Studies the association between wealth Manicaland, Zimbabwe HIV/STD Prevention The greatest decrease in HIV prevalence occurred in
index (based on household asset ownership) Project’s population-based open cohort the highest wealth index tercile in both men and
and HIV incidence, HIV mortality, sexual (baseline between 1998 and 2001 and women. In men (but not women), HIV incidence
risk behaviour, and sexual mixing patterns. follow-up between 2001 and 2003). was lowest in the top wealth index tercile. Mortality
Multivariate logistics and Poisson regression rates were significantly lower in both men and women
models. of higher wealth index. Men of higher wealth
index reported more sexual partners, but were also
more likely to use condoms, controlling for age and
site type. Better-off women reported fewer partners
and were less likely to engage in transactional sex.
Hargreaves et al. [16] To assess the evidence that HIV incidence Prospective cohort of 1967 individuals Among men, there was little evidence that HIV
rates and sexual behaviour patterns differed (14–35 years of age) in Limpopo province, seroconversion was associated with any
by wealth, education and migration. South Africa (2001 and 2004). socioeconomic factor. Among women, HIV
Multivariate logistic regression models, seroconversion was negatively associated with
stratified by sex. education, but not wealth or migration. Migrant
men more often reported multiple partners. Migrant
and more educated individuals of both sexes, and
women from wealthier households, reported
higher levels of condom use.
Mishra et al. [17] Examines the association between wealth Cross-sectional nationally representative In all eight countries, adults in the wealthiest quintiles
(index based on household ownership surveys from eight sub-Saharan African have higher prevalence of HIV than those in the
of consumer durables) and HIV serostatus countries conducted during 2003–2005. poorer quintiles, but the positive association
of 15–49-year-old individuals. Multivariate logistic regression models, between wealth and HIV status was statistically
stratified by sex. insignificant in multivariate models.
(continued overleaf )
S7](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-12-2048.jpg)
![S8
AIDS
Table 1. (continued )
Study Objective Study design and statistical analyses Key findings
2007, Vol 21 (suppl 7)
Barnighausen et al. [18]
¨ Investigates the effect of educational Longitudinal data (2003–2005) on 3325 adults Belonging to a household in the middle
attainment, household wealth categories from Africa Centre Demographic Information wealth category increased the risk of
(based on a ranking of households on an System in KwaZulu-Natal, South Africa. HIV seroconversion. One additional grade
assets index scale) and total household Semiparametric and parametric survival models. of educational attainment reduced the
expenditure, on HIV incidence. hazard of HIV seroconversion by
approximately 7%. Urban residence was
associated with a 65% increase in the
hazard of HIV seroconversion.
Chapoto and Jayne [19] To determine the ex-ante socioeconomic Nationally representative panel data set of Relatively non-poor men (ranked by
characteristics of individuals who died 18 821 individuals from 5420 households assets levels) were 43% more likely
in their prime age (15–59 years) surveyed between 2001 and 2004. to die than poor men. Poor and non-poor
in Zambia. Multivariate probit models, stratified by sex women were equally likely to die. No clear
and assets. relationship observed between education
attainment and probability of prime-age
mortality. Poor women with business
income were 15% less likely, and non-poor
women with business income 7% more
likely, to die than those without business income.
Kirimi and Jayne [20] Estimates the potentially changing Nationwide data set of 5755 individuals Over time, the probability of disease-related
relationship over time between from 1500 Kenyan rural households death declined for both men and women.
household and individual-level collected in 1997, 2000, 2002 and 2004. A reversal in the effect of education on death
indicators of poverty and subsequent Multivariate probit models, stratified by sex. was observed, with more educated women
death of prime-age adults in Kenya. and men, and particularly younger ones,
being at greater risk of death. Although weak,
there is also a delayed but significant
negative effect of landholding size and asset
value on male mortality.
Glynn et al. [9] Investigates the associations between Cross-sectional population-based survey No association between schooling and HIV
schooling and both HIV and herpes conducted in 1997–1998 in four African infection and a significant negative association
simplex 2 infection and risky behaviours cities including approximately with herpes simplex 2 in women observed in
in Cotonou (Benin), Yaounde (Cameroon), 2000 adults in each city. Kisumu or Ndola,. In Yaounde, women with
Kisumu (Kenya) and Ndola (Zambia). Multivariate models, stratified by sex. more schooling were less likely to be HIV
positive. Similar association observed among
men in Cotonou for herpes simplex 2. In all
cities, those with more education tended to
report less risky sexual behaviours.
De Walque et al. [10] Investigates the association between Population-based cohort followed between In 1989/90, there was no significant relationship
changing HIV prevalence, condom 1989/1990 and 1999/2000. between education and HIV prevalence.
use and education in rural south-west Multivariate and bivariate (condom versus In 1999–2000 women aged 18–29 years
Uganda. education) analyses. with post-primary education were at
significantly lower risk of HIV-1 infection
than women with no education. Condom
use increased during the study period and
this increase has been concentrated among
more educated individuals.
Luke [21] To study the trade-off between transfers Cross-sectional survey of Luo men aged Men’s income was not significantly associated
and condom use at last sexual intercourse 21–45 years in Kisumu, Kenya. with condom use. Having an adolescent
in non-commercial, non-marital sexual Multivariate models including male fixed female partner does not have a significant
relationships in Kenya. effects models. effect on condom use. For every Ksh500,
approximately the mean amount given in
transfers per partnership, the probability
of condom use decreased by approximately 8%.
Trade-off between transfers and condom use
does not vary between adolescents and
adult women.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-13-2048.jpg)
![Poverty, wealth, HIV transmission Gillespie et al. S9
level of gender inequality, age is protective. Similarly, the
not always significant. Conditional on gender inequality,
Africa (2002–2005), Dinkelman et al. [11] show that for
effect of gender inequality for women decreased with
the share of young women who live in poverty in the
was associated with a 1% increase in the probability
girls, sexual debut appears to be earlier in poor
increasing household assets, although this effect was
A one standard deviation increase in gender inequality
of being HIV positive for young women. For a given
households, especially those who have experienced an
in inherited land, the total amount of transfers
community did not increase the probability of
Economic status was positively and significantly
economic shock (a death, illness or job loss). A recent
increases by Ksh10 on average. Wealth was
associated with both the giving of transfers
and the amount. For every additional acre
additional year of education increased the
cross-sectional study in Kenya found asset poverty to be
not correlated with condom use. Each
significantly related to risky sexual outcomes, such as
early sexual debut, multiple sexual partnerships, in all
three residential settings studied [14]. In a study in
probability of condom use by
Botswana and Swaziland [12], although protective in
individual HIV infection.
unadjusted analyses, controlling for other variables,
approximately 3.4%.
income was not associated with intergenerational sex
and a lack of control in sexual relationships among
women. Wealthier men reported having more sex
exchange [adjusted odds ratios (aOR) 1.94, 95%
confidence interval (CI) 1.59–2.37] but were also more
likely to report condom use (aOR 0.78, 95% CI 0.72–
0.84).
Another recent cross-sectional study of Luo men aged
21–45 years of age in urban Kisumu, Kenya, found male
economic status, controlling for age and education, to
Cross-sectional survey of Luo men aged
and Housing Census, Kenya Poverty
be positively associated with transactional sex and the
Three sources of cross-sectional data:
value of transfers [22]. For every Ksh1000 in male
Health Survey, 1999 Population
21–45 years in Kisumu, Kenya.
2003 Kenya Demographic and
income, the probability of giving a transfer in the past
month increases approximately 1%, and the total amount
Multivariate probit models
of transfers increases Ksh29 (US$0.40). Wealth (income
Multivariate models.
and inherited land) was not, however, correlated with
condom use, suggesting that larger transfers are not being
Map (2003).
given by wealthier men as an incentive for condom-free
(riskier) sex.
Two prospective cohort studies examining the relation-
ship between economic resources and high-risk sexual
behaviours are presented in this volume. In a 3-year
women and adult men within an individual’s
follow-up study (baseline between 1998 and 2001 and
Examines the relationship between HIV status
women’s poverty status on individual HIV
inherited land), transfers, and non-marital
follow-up between 2001 and 2003) in Manicaland,
Empirical investigation of the connection
between economic status (income and
and gender inequality between young
non-commercial, sexual relationships
Zimbabwe, Lopman et al. [15], found wealthier men
community and to examine young
reporting more sexual partners, but also more frequent
use of condoms, controlling for age and site type. This
relationship became insignificant, however, after con-
trolling for education level, in addition to age and site
type, suggesting that the effect of wealth is at least partly
status in Kenya.
the result of differences in education across wealth levels.
Better-off women reported fewer partners and were less
in Kenya.
likely to engage in transactional sex, adjusting for age,
education level and site type. Hargreaves et al. [16] in
Limpopo, South Africa (2001–2004) found women, but
not men, from wealthier households reporting higher
levels of condom use (aOR comparing household ‘doing
OK’ with ‘very poor’ 2.03, 95% CI 1.29–3.20).
Beegle and Ozler (unpublished)
Using Demographic and Health Survey (DHS) data from
eight countries, Mishra et al. [17] found a positive
association between an asset-based wealth index and HIV
status. This relationship was stronger for women, and it
was clear that HIV prevalence was generally lower among
Luke [22]
the poorest individuals in these countries. This is partly
accounted for by an association of wealth with other](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-14-2048.jpg)
![S10 AIDS 2007, Vol 21 (suppl 7)
35%
Swaziland
30%
Botswana
25%
Lesotho
HIV prevalence 20% Zimbabwe Namibia
Southern Africa Zambia
R squared = 0.2952 Mozambique South Africa
15% not significant
Malawi
Central African Republic
10%
Gabon
Côte d'Ivoire
Tanzania Kenya
E&W Africa Uganda
5%
R squared = 0.0000 Angola
not significant Sierra leone
Ethiopia
0%
US$100 US$1 000 US$10000
GDP per capita (PPP, logarithmic scale)
Fig. 1. HIV and per-capita gross domestic product in Africa. Sources: Economic data from UNDP Human Development Report
2006; HIV prevalence data from UNAIDS Epi Update, May 2006.
underlying factors. Wealthier individuals tend to live in likely than the poorest women to be HIV positive [13].
urban areas where HIV is more prevalent, they tend to be Similar findings were reported in Tanzania [33] and in
more mobile, more likely to have multiple partners, more Burkina Faso [34].
likely to engage in sex with non-regular partners, and
they live longer; all factors that may present greater Studies of cross-sectional associations between HIV
lifetime HIV risks. On the other hand, however, they serostatus and socioeconomic status (such as those above
tend to be better educated, with better knowledge of HIV and the cross-sectional studies featured in another
prevention methods, and are more likely to use condoms; comprehensive review [1]) suffer from important
factors that reduce their risk compared with poorer limitations: They are unable to distinguish between the
individuals. Controlling for these associations, however, effect of economic status on HIV infection and the effect
does not reverse the conclusion: there is no apparent of HIV infection on economic status, and they are unable
association between low wealth status and HIV. to control for the fact that individuals from richer
households may survive longer with HIV, and are thus
Using data from the cross-sectional, population-based more likely to be present in the population to be tested,
2003 Kenya Demographic and Health Survey, a recent thereby increasing HIV prevalence rates.
study found increased wealth to be positively related to
HIV infection, with the effect being stronger for women In a cross-sectional study, it is thus conceivable to find a
than men; the wealthiest women being 2.6 times more positive association between economic status and HIV
25% Botswana
Lesotho
Zimbabwe
20% Namibia
South Africa
Southern Africa
R squared = 0.0996 Zambia
Mozambique
not significant
HIV prevalence
15% Malawi
Central African Republic
10%
E&W Africa
Côte d'Ivoire Uganda
Tanzania R squared = 0.0307
Kenya not significant
5% Cameroon
Nigeria
Rwanda Burundi
Ghana
Ethiopia Gambia Mali
Burkina Faso Niger
Senegal
Mauritania Sierra Leone Madagascar
0%
0 10 20 30 40 50 60 70 80
Percentage below US$1 per day
Fig. 2. HIV and poverty in Africa. Sources: Economic data from UNDP Human Development Report 2006; HIV prevalence data
from UNAIDS Epi Update, May 2006.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-15-2048.jpg)
![Poverty, wealth, HIV transmission Gillespie et al. S11
35%
Swaziland
30%
2
R = 0.4881
P = 0.005%
25% Botswana
Lesotho
HIV prevalence
Zimbabwe Namibia
20%
South Africa
Zambia
Mozambique
15% Malawi
Central African Republic
10%
Tanzania Uganda Côte d'Ivoire
Kenya Cameroon
5% Rwanda Nigeria
Burundi
Ghana Mali
Ethiopia
Senegal Niger
0%
0.25 0.35 0.45 0.55 0.65 0.75
GINI coefficient
Fig. 3. HIV and income inequality in Africa. Sources: Economic data from UNDP Human Development Report 2006;
HIV prevalence data from UNAIDS Epi Update, May 2006.
infection, even if higher economic status protects transmission. Those studies are nationally representative
individuals from acquiring HIV. Both the above-stated rural household panel surveys, unlike the studies reviewed
limitations can be overcome by using prospective cohorts above (in which the national level surveys are cross-
to track HIV incidence. This volume presents three such sectional and longitudinal cohorts are limited to
studies with differing results: (i) Lopman et al. [15] in provinces). Although they do not directly measure
Manicaland, Zimbabwe, reported a significantly lower HIV prevalence or incidence, they do employ innovative
male HIV incidence (between baseline in 1998/2000 and methodologies to infer the extent of HIV-related prime
follow-up in 2001/2003) in the wealthiest asset tercile age adult mortality.
(15.4/1000 person-years) compared with the lowest
tercile (27.4/1000 person-years), controlling for age and A nationally representative rural panel data survey (2001–
site of residence. This trend was even more marked in 2004) in Zambia [19], sought to determine the ex ante
young men under 17–24 years of age. No such socioeconomic characteristics of individuals who died in
association between wealth and HIV seroconvesion was their prime age (15–59 years). When ranked by asset
observed among women. Mortality rates were signifi- levels, relatively wealthier men were 43% more likely to
cantly lower in both men and women of higher wealth die of disease-related causes than men in poor households,
groups. They also found a decrease in HIV prevalence with no clear association among women.
across all asset wealth groups during the study period,
with the largest decrease in the wealthiest tercile for both In contrast, a nationwide rural panel survey (1997–2004)
men at 25% and women at 21%. (ii) Controlling for place in Kenya [20], performing similar analyses to the above,
of residence, migration status, partnership status, sex and reported men and women from relatively asset-poor
age, a study in rural KwaZulu Natal by Barnighausen et al.
¨ households to be more likely to die than those from
[18] found that individuals from households in the middle wealthier households. The authors also found a shift in
asset wealth tercile had a significantly higher hazard of the relationship between landholding size and prime-age
HIV seroconversion (1.7 times that of the poorest tercile), mortality in which no significant association was observed
whereas there was no significant difference between the between 1997 and 2000, but in both the 2000–2002 and
wealthiest and poorest terciles. Per capita household 2000–2004 periods, access to more land was associated
expenditures on the other hand did not significantly with reduced male mortality.
influence the hazard of HIV seroconversion. (iii) In a
study of HIV incidence in Limpopo Province of South
Africa between 2001 and 2004, Hargreaves et al. [16] did
not find a statistically significant association between HIV Does education reduce exposure to HIV?
seroconversion and economic status (assessed through
participatory wealth ranking methods) in either men Education is one of the most studied socioeconomic
or women. factors in the context of AIDS epidemics. Although
education and economic resources are often jointly
A few other longitudinal studies have added to our determined, empirical evidence has shown that education
understanding of socioeconomic differentials in HIV predicts health independently of income [35].](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-16-2048.jpg)
![S12 AIDS 2007, Vol 21 (suppl 7)
A systematic review in 2002 of 27 studies [7], mostly especially among women [16,18]. Hargreaves et al. [16]
cross-sectional, with data predominantly collected before found that among women (but not men) HIV
1996, found that increased schooling was either not seroconversion was negatively associated with education
associated with HIV infection or was associated with an (aOR comparing attended secondary school versus
increased risk of HIV infection among men and women none/primary 0.49, 95% CI 0.28–0.85; comparing
from both rural and urban communities in Africa. As the those completing secondary school versus none/primary
epidemic within countries has advanced, the evidence 0.25, 95% CI 0.12–0.53). Barnighausen et al. [18]
¨
suggests a shift towards a reduced relative risk of HIV reported that one additional grade of educational
infection among adults, especially younger women, who attainment reduced the hazard of HIV seroconversion
have a secondary education [9,10,36]. by approximately 7%.
The hypothesis that the ability to process and access In sum, a relatively clear picture emerges for education, the
information is one channel through which education majority of studies suggest that education is increasingly
affects health outcomes has been examined in a study in associated with less risky behaviours. Sustained efforts to
Uganda [10], in which changes in association between improve education levels as well as targeted and tailored
schooling levels, HIV prevalence, and condom use were messages on HIV prevention efforts can yield positive
estimated among a population-based rural cohort in results.
Masaka District between 1989/1990 and 1999/2000.
During the early years of the epidemic in 1990, there was
no robust relationship between HIV and years of
education for either sex for all individuals older than Poverty and HIV: pathways and
17 years of age. By 2000, however, each additional year of interactions
education was found to lower the risk of being HIV
positive significantly among 18–29-year-old women Links between socioeconomic conditions, such as wealth
(aOR 0.863, 95% CI 0.77–0.96). Condom use was found and education, and HIV risk and vulnerability are clearly
to be positively associated (using bivariate analysis only) complex, perhaps too complex for a single explanation. A
with schooling levels between 1995 and 2000, with the major analytical challenge is to define the causal pathways
gradient between higher educational achievement and operating from distal socioeconomic factors to proximal
greater condom use being steeper for women than men individual behaviours and ultimately physiological
(chi-square for trend of odds in 1996/1997, 69.10 for factors. Different socioeconomic factors may affect health
men and 82.13 for women, and in 1999/2000, 103.01 for at different times in the life course [40,41], operating at
men and 164.18 for women). different levels (e.g. individual, household and neigh-
bourhoods) [42,43] and through different causal pathways
One study in Cote d’Ivoire [37] found more highly [44,45].
educated people to be more likely to engage in multiple
sexual partnerships, although they were also more likely The sections below highlight some of the more important
to use condoms, thus offsetting some of the risk of factors and processes that condition the relationship
exposure to HIV. Similar observations of a higher between poverty, wealth and HIV. Here we focus on the
probability of condom use among the more educated key issues of gender inequality, mobility and social
have been reported elsewhere [38,39]. ecology. Malnutrition is another potentially important
conditioning factor affecting the risk of HIV infection.
A cross-sectional study in Botswana and Swaziland found Given space limitations, the reader is referred to other
that higher educated women were less likely to report a reviews and ongoing work in this area [24].
lack of control in sexual relationships (aOR 0.36, 95% CI
0.36–0.37), were less likely to report inconsistent
condom use (aOR 0.72, 95% CI 0.57–0.91) and Gender and economic asymmetries
intergenerational sex (aOR 0.68, 95% CI 0.53–0.86). The issue of gender is front and central to any discussion
No association between risk behaviours and education of HIV and poverty. Women’s dependence on men’s
among men was observed [12]. Studies employing economic support throughout much of the developing
longitudinal rural panel datasets from Zambia, Kenya world means that women’s personal resources, including
and Ethiopia have shown a pattern of negative association their sexuality, has economic potential. Economic
between educational attainment and disease-related asymmetries within a couple are reinforced by various
mortality [19] (A. Chapoto et al., unpublished). contextual factors, such as family and peer pressures,
social and economic institutions and pervasive and deeply
As with economic status, few studies have prospectively entrenched sex-based inequalities. Social norms in many
investigated the relationship between education and HIV sub-Saharan African contexts, for example, permit (and
incidence. Two such studies presented in this volume even encourage) men to engage in sex with multiple
found a significantly protective effect of education, partners, with much younger partners, and to dominate](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-17-2048.jpg)
![Poverty, wealth, HIV transmission Gillespie et al. S13
sexual decision-making. In a study of four communities in HIV prevalence becomes even stronger. Finally, they
a southern province of Zambia [26] respondents blamed show how the relationship between inequality and HIV is
women. Women were perceived to move around and stronger when inequality is generated more by higher
‘give love for money’; women who some believe could proportions of richer men than poorer women.
otherwise work hard and do not need to have sex for
money. The fact that men, often much older than girls/ Using a combination of data sources on HIV status at the
women, pay for sex was rarely mentioned as a cause of individual level and poverty and inequality measures at
the problem. the community level, a study in Kenya (K. Beegle, B.
Ozler, unpublished data) found, conditional on a set of
Pre and extramarital sex may involve the male to female individual and community characteristics, gender
transfer of material resources, such as money and gifts. inequality between young women and adult men to be
Such exchanges may take the form of commercial sex or significantly correlated with the individual’s HIV-positive
more informal transactional sex, which is common in status. This effect is stronger for young women, especially
high HIV contexts [26,46,47]. In a study of young Luo in western Kenya where HIV prevalence is highest, and is
men in Kenya [21], male to female transfers were given in robust to various definitions of economic inequality
three-quarters of recent non-marital partnerships, and between young women and older men.
transfers were substantial. Men on average provided
approximately US$8.50 (Ksh600) to each non-marital In Botswana and Swaziland, food insufficiency among
partner in the past month, equivalent to 9% of a male women was found to be significantly associated with
mean monthly income. The author also reported a inconsistent condom use with a non-primary partner
negative and significant relationship between the value of (aOR 1.73, 95% CI 1.27–2.36), sex exchange (aOR
transfer and reported consistent condom use [21]. For 1.84, 95% CI 1.74–1.93), intergenerational sexual
every transfer of (monetary or non-monetary) Ksh500, relationships (aOR 1.46, 95% CI 1.03–2.08), and lack
approximately the mean amount given in transfers per of control in sexual relationships (aOR 1.68, 95% CI
non-marital partnership, the probability of condom use 1.24–2.28). For men, food insufficiency was associated
decreased approximately 8%. with only a 14% increase in the odds of reporting
unprotected sex, and was not associated with other risky
Evidence points to significant positive associations between sexual behaviours [12]. Although food insufficiency is
larger age differences between partners, the value of certainly influenced by income, it is a distinct entity with
economic transactions and unsafe sexual behaviours [46– different causes and consequences; there are many steps
48]. In South Africa, low socioeconomic status has been between an aggregated household income variable and
found not only to increase female odds of exchanging sex the ability of an individual woman to access, control and
for money or goods, but also to raise female chances of use income to buy food. A specific focus on protecting
experiencing coerced sex, and male and female odds of and promoting access to food may thus decrease exposure
having multiple sexual partners. It also lowers female to HIV, especially among women.
chances of abstinence, female and male age at sexual debut,
condom use at last sex, and communication with most Mobility
recent sexual partner about sensitive topics. Low socio- The link between mobility and the spread of HIV is
economic status has more consistent negative effects on determined by the structure of the migration process, the
female than on male sexual behaviours; it also raises the conditions under which it occurs, including poverty,
female risk of early pregnancy [48]. exploitation, separation from families and partners, and
separation from the sociocultural norms that guide
A few interesting recent studies have suggested that behaviours within communities [49]. Mobility can
increased economic inequality between men and women increase vulnerability to high-risk sexual behaviour as
leads to partnerships that are riskier in terms of HIV migrants’ multilocal social networks create opportunities
exposure. In one (B. Penman, B. Ozler, K. Beegle, S. Baird, for sexual networking. Mobility also makes individuals
unpublished data), a basic model of HIV epidemiology more difficult to reach for preventive, care or treatment
was combined with population demographic processes, services.
factoring in the marital and economic status of sexually
active heterosexual individuals. Using a few simple There is convincing empirical evidence of a link between
assumptions regarding partnership patterns, the data human mobility and the risk of HIV transmission. In sub-
generate a clear correlation between gender inequality Saharan Africa, the risk of HIV infection has been found
(defined by economic inequality between young women to be higher near roads, and among individuals who
and older men) and HIV prevalence in a completely either have personal migration experience or have sexual
susceptible population after 20–25 years. As expected, if partners who are migrants [18,49–53].
rich men or poor women contribute a higher share of
their respective populations to the high sexual activity In eastern and southern Africa, mining, plantations and
group, then the relationship between sex inequality and related agricultural industries (typically producing tea,](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-18-2048.jpg)
![S14 AIDS 2007, Vol 21 (suppl 7)
coffee, tobacco, sugar cane, and rice) are often associated rates in the Cape Area Panel Study by Dinkelman et al. [11]
with situations of significant risk. Risks may be enhanced significantly predicted earlier sexual debut for girls and
by regularized single-sex migration as in the case of boys, and higher rates of unprotected recent sex for boys.
southern African mines [53]; high and seasonal demands
for agriculture labour on estates; workers moving on their The structural context of labour arrangements also
own, sometimes from considerable distances and lodged contributes to the demand for transactional sex. Although
in single-sex dormitories; long and often irregular pay considered very arduous and physically demanding, cane
intervals; and a dependent population of occasional or cutting jobs, for example, command higher monthly
commercial sex workers from nearby villages or further wages than most permanent positions. In a Zambian
afield [26,54,55]. Ownership structures, the national study [26], in the two worker compounds that make up
policy environment, and the economics of the industries the study area in Mazabuka, there was widespread
are all important drivers of HIV transmission risk. awareness that married women sleep with cane cutters to
access resources they either need or want.
The social ecology of HIV
The socio-ecological systems perspective of disease Social cohesion and social capital are other important
transmission fosters a deliberate analysis of the dynamics conditioning factors. A higher HIV risk has been found to
of population patterns of health and wellbeing at each be significantly associated with structural factors related to
level of biological, ecological and social organization [45]. the community in a study in Limpopo, South Africa [57].
Although most attention has been paid to the socio- Such factors included easier access to a trading centre,
economic conditions of individuals and their households, higher proportions of short-term residents, and lower
relatively little attention has been paid to the socio- levels of social capital (particularly significant among
ecological conditions that shape norms, behaviours and men); the latter being an index based on social network
access to various resources. membership and responses to questions on levels of trust,
reciprocity, solidarity in a time of crisis, collective action
For example, in a Tanzanian study [50], community (positive) and local serious and violent crime rate
characteristics, such as the type of economic activity, ratio (negative). In other words, HIV prevalence was higher
of bar girls to men, share of migrants, and distance to big in settings in which the social order had broken down, or
cities have all been found to correlate positively with HIV had never been established in the first place. Among men,
seroprevalence; traits that are usually associated with higher HIV prevalence was also seen among communities
higher income. with easier access to a local mine, a higher density and
activity of local bars, a higher numbers of sex workers per
Slum populations may be particularly vulnerable. In village, and lower proportions of outmigrants. More
Kenya, for example, slum residence has been found to be research is needed on social cohesion and HIV risk.
unique in its adverse impact on sexual outcomes,
presumably because monetary currency is central to
existence in cities where difficult economic circumstances
coerce women to use sex as a means of survival [56]. Conclusions
Using two separate indicators of deprivation, a Kenyan
study has shown that although poverty is significantly In conclusion, this paper has drawn up recent studies to
associated with the examined sexual outcomes in all examine what is known about the degree to which, and
settings, the urban poor were significantly more likely the ways in which, socioeconomic status is associated
than their rural counterparts to have an early sexual with HIV transmission. The notion that poverty is the
debut and multiple sexual partnerships, even among main driver of HIV transmission is too simplistic. Relative
married women [14]. Complementing their quantitative wealth appears to have a mixed influence on HIV risk
evidence with qualitative research, the authors posited depending on an array of contextual factors. Gender
that, beyond purely economic factors, other social inequality appears to be particularly important. In the
conditions contributed to higher levels of sexual activity most comprehensive multicountry, cross-sectional study
in the very poor slum communities. Young children were to date [17], the residual effect of wealth was found to be
socialized into sexual behaviour because of: a lack of statistically insignificant after controlling for variables
alternate recreational opportunities; residential con- such as urban residence, age, education and differences in
ditions that precluded privacy for adult sexual activity; sexual behaviour. There are very few cohort studies that
and role modeling by adults who either transacted sex for are able to relate wealth or poverty to the incidence of
money or were more generally involved in casual sexual HIV, and they tend to be rurally based. One such study,
activity. reported here, shows the highest risk of infection in the
middle wealth group.
Consistent with these findings, Barnighausen et al. [18]
¨
showed that living in urban and peri-urban areas increases Education in general appears to be protective with regard
the hazard of HIV seroconversion. Community poverty to HIV risk, and the interaction effects between](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-19-2048.jpg)
![Poverty, wealth, HIV transmission Gillespie et al. S15
education and wealth could be very positive; when and women, and to the dynamic and contextual nature
individuals have resources and the ability to use those of the relationship between socioeconomic status and
resources, they can act on safeguarding their sexual health. HIV.
In investigating the relationship between poverty, wealth
and HIV it is important to state the following Sponsorship: Financial support was provided by the
qualifications: (i) Many of the studies presented in this Joint United Nations Programme on HIV/AIDS
volume and elsewhere suffer from important limitations (UNAIDS) and the RENEWAL programme of the
such as: low statistical power (especially when measuring International Food Policy Research Institute.
HIV incidence); high attrition rates (especially of
Conflicts of interest: None.
educated, mobile men); difficulty in tracking certain
individuals (e.g. mobile commercial sex workers, truck
drivers); a paucity of longitudinal nationally representa- References
tive panels tracking HIV incidence; and a lack of a
1. Wojciki JM. Socioeconomic status as a risk factor for HIV
comprehensive measure of economic status. (ii) On the infection in women in East, Central and Southern Africa: a
latter, different aspects of economic status are likely to systematic review. J Biosoc Sci 2005; 37:1–36.
behave differently; e.g. assets, income, expenditures, cash 2. Nyindo M. Complementary factors contributing to the rapid
spread of HIV-1 in sub-Saharan Africa: a review. East Afr Med J
flows, sources and control of income, and their 2005; 82:40–46.
intrahousehold and sex differentials. Most studies here 3. Buve A, Lagarde E, Carael M, Rutenberg N, Ferry B, Glynn JR,
use an asset-based index of household wealth. (iii) Most et al. Study Group on Heterogeneity of HIV Epidemics in
African Cities. Interpreting sexual behavior data: validity issues
studies also examine relative not absolute wealth. Many of in the multicentre study on factors determining the differential
the wealthiest groups in affected communities may spread of HIV in four African cities. AIDS 2001; 15 (Suppl. 4):
actually fall below an absolute poverty line. (iv) Patterns S117–S126.
4. Ainsworth M, Semali I. Who is most likely to die of AIDS?
are not uniform across, or even within, countries, and a Socioeconomic correlates of adult deaths in Kagera Region,
variety of socioeconomic factors and processes are likely Tanzania. In: Confronting AIDS: evidence for developing world.
Edited by Ainsworth M, Fransen L, Over M. Brussels: European
to be at play in complex and interlinked ways. (v) The Commission; 1998.
relationship between wealth and HIV is dynamic and may 5. Gregson S, Waddell H, Chandiwana S. School education and
change over time. Most studies are cross-sectional in HIV control in sub-Saharan Africa: from discord to harmony?
J Int Dev 2001; 13:467–485.
nature, not longitudinal, and thus ‘snapshots’ at a single 6. Gregson S, Garnett GP, Nyamukapa CA, Hallett TB, Lewis JJ,
point in time. It is important therefore to track the rates of Mason PR, et al. HIV decline associated with behaviour change
new infections. If incidence is higher in groups moving in eastern Zimbabwe. Science 2006; 311:664–666.
7. Hargreaves JR, Glynn JR. Educational attainment and HIV-1
out of poverty, this implies a greater need to ‘HIV-proof ’ infection in developing countries: a systematic review. Trop
poverty reduction so as to make the options for increasing Med Int Health 2002; 7:489–498.
wealth less risky. (vi) The role and influence of social 8. Anderson RM, May RM, Boily MC, Garnett GP, Rowley JT. The
spread of HIV-1 in Africa – sexual contact patterns and the
capital, social cohesion and community-level structural predicted demographic-impact of AIDS. Nature 1991; 352:581–
factors is under-researched, little understood but poten- 589.
tially very important; especially given the known ´
9. Glynn JR, Carael M, Buve A, Anagonou S, Zekeng L, Kahindo M,
et al. Does increased general schooling protect against HIV
association between different forms of inequality and infection? A study in four African cities. Trop Med Int Health
risk and vulnerability. (vii) Likewise, the literature is 2004; 9:4–14.
somewhat biased towards explaining the relationship 10. De Walque D, Nakiyingi-Miiro JS, Busingye J, Whitworth JA.
2005. Changing association between schooling levels and HIV-
between socioeconomic conditions and HIV through 1 infection over 11 years in a rural population cohort in south-
high-risk behaviour adoption pathways, with less west Uganda. Trop Med Int Health 2005; 10:993–1001.
attention being paid to the ways in which pre-existing 11. Dinkelman T, Lam D, Leibbrandt M. Household and commu-
nity income, economic shocks and risky sexual behavior of
health and nutritional status may have compromised the young adults: evidence from the Cape Area Panel Study 2002
immune status of individuals [58]. and 2005. AIDS 2007; 21 (Suppl. 7):S49–S56.
12. Weiser SD, Leiter K, Bangsberg DR, Butler ML, Percy-de Korte F,
Hlanze Z, et al. Food insufficiency is associated with high risk
In summary, when examining the interplay between sexual behavior among women in Botswana and Swaziland.
wealth or poverty and HIV transmission, there is no PLoS Medicine 2007; 4:e260.
simple explanation, no magic bullet. AIDS cannot 13. Johnson K, Way A. Risk factors for HIV infection in a national
adult population: evidence from 2003 Kenya Demographic and
accurately be termed a ‘disease of poverty’. Although Health Survey. J Acquir Immune Defic Syndr 2006; 42:627–636.
it is true that poor individuals and households are likely 14. Nii-Amoo Dodoo F, Zulu EM, Ezeh AC. Urban–rural differ-
ences in the socioeconomic deprivation – sexual behavior link
to be hit harder by the downstream impacts of AIDS, in Kenya. Soc Sci Med 2007; 64:1019–1031.
their chances of being exposed to HIV in the first place 15. Lopman B, Lewis J, Nyamukapa C, Mushati P, Chandiwana S,
are not necessarily greater than wealthier individuals Gregson S. HIV incidence and poverty in Manicaland,
Zimbabwe: is HIV becoming a disease of the poor? AIDS
or households. What is clear is that approaches to HIV 2007; 21 (Suppl. 7):S57–S66.
prevention need to cut across all socioeconomic strata 16. Hargreaves JR, Bonell CP, Morison LA, Kim JC, Phetla G, Porter
of society, and they need to be tailored to the specific JDH, et al. Explaining continued high HIV prevalence in South
Africa: socioeconomic factors, HIV incidence and sexual
drivers of transmission within different groups, with behaviour change among a rural cohort, 2001–2004. AIDS
particular attention to the vulnerabilities faced by youth 2007; 21 (Suppl. 7):S39–S48.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-20-2048.jpg)
![HIV infection does not disproportionately affect the
poorer in sub-Saharan Africa
Vinod Mishraa, Simona Bignami-Van Asscheb, Robert Greenerc,
Martin Vaessena, Rathavuth Honga, Peter D. Ghysc,
J. Ties Boermad, Ari Van Asschee, Shane Khana
and Shea Rutsteina
Background: Wealthier populations do better than poorer ones on most measures of
health status, including nutrition, morbidity and mortality, and healthcare utilization.
Objectives: This study examines the association between household wealth status and
HIV serostatus to identify what characteristics and behaviours are associated with HIV
infection, and the role of confounding factors such as place of residence and other risk
factors.
Methods: Data are from eight national surveys in sub-Saharan Africa (Kenya, Ghana,
Burkina Faso, Cameroon, Tanzania, Lesotho, Malawi, and Uganda) conducted during
2003–2005. Dried blood spot samples were collected and tested for HIV, following
internationally accepted ethical standards and laboratory procedures. The association
between household wealth (measured by an index based on household ownership of
durable assets and other amenities) and HIV serostatus is examined using both
descriptive and multivariate statistical methods.
Results: In all eight countries, adults in the wealthiest quintiles have a higher
prevalence of HIV than those in the poorer quintiles. Prevalence increases mono-
tonically with wealth in most cases. Similarly for cohabiting couples, the likelihood that
one or both partners is HIV infected increases with wealth. The positive association
between wealth and HIV prevalence is only partly explained by an association of wealth
with other underlying factors, such as place of residence and education, and by
differences in sexual behaviour, such as multiple sex partners, condom use, and male
circumcision.
Conclusion: In sub-Saharan Africa, HIV prevalence does not exhibit the same pattern of
association with poverty as most other diseases. HIV programmes should also focus on
the wealthier segments of the population.
ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 7):S17–S28
Keywords: Africa, AIDS, HIV, poverty, sexual behaviour, wealth
Introduction health status, including nutrition, morbidity and
mortality, and of healthcare utilization [1–3]. Consistent
The relationship between socioeconomic status and with these findings, there is evidence of an inverse
health is well documented. There is ample evidence that relationship between socioeconomic status and the risk
wealthier populations do better on most measures of of sexually transmitted infections (STI), such as herpes,
From the aMacro International Inc., Calverton, Maryland, USA, the bUniversity of Montreal, Montreal, Canada, the cJoint United
Nations Programme on HIV/AIDS, Geneva, Switzerland, the dWorld Health Organization, Geneva, Switzerland, and the eHEC
Montreal, Montreal, Canada.
Correspondence to Vinod Mishra, DHR Division, Macro International Inc., 11785 Beltsville Drive, Calverton, MD 20705, USA.
Tel: +1 301 572 0220; fax: +1 301 572 0999; e-mail: vinod.mishra@macrointernational.com
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S17](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-22-2048.jpg)
![S18 AIDS 2007, Vol 21 (suppl 7)
chlamydia, gonorrhoea, syphilis, and bacterial vaginosis
SOCIAL AND EPIDEMIOLOGICAL CONTEXT
[4–13]. Although much of this evidence is from Social attitudes and practices, level of economic development, stage of HIV/AIDS epidemic, HIV
prevalence, availability and access to prevention and treatment methods
developed countries, it is reasonable to expect that
poverty increases vulnerability to HIV in the same
UNDERLYING PROXIMATE
manner in low and middle-income countries. It is indeed FACTORS FACTORS
OUTCOME
often argued that poverty is the root cause of the spread
Sexual behaviour
of HIV [14]. A recent article in the Lancet argued
Abstinence
that ‘[s]ince poverty plays a role in creating an Sexual debut
Multiple sexual partners
environment in which individuals are particularly Concurrent partners
susceptible and vulnerable to HIV/AIDS, poverty Wealth
Non-regular partners
Commercial sex
reduction will undoubtedly be at the core of a sustainable Partner faithfulness
Type of sexual activity
Age
solution to HIV/AIDS’ [15]. Analogous views have been Sex/gender (MSM, anal, oral, etc.)
Ethnicity
expressed in numerous public statements and publi- Religion Transmission cofactors
Urban/rural residence
cations, and guide HIV prevention efforts in several Geographical region Other STI
countries. Education
Male circumcision
HIV status
Nutritional status
Occupation
Exposure to media
At the global level, there is evidence of a positive Marital status Treatment & care
Duration in union
correlation between countries’ HIV prevalence and Mobility Treatment & care
Alcohol use ART
poverty, as measured by per capita income, income HIV knowledge &
attitudes
inequality, or absolute poverty [16]. The HIVepidemic in Other risk factors
sub-Saharan Africa represents a notable exception to this
Condom use
general pattern. On the one hand, at the macro level Injecting drug use
Medical injections
African nations with high HIV prevalence, such as South Blood transfusion
Skin cutting/tattooing
Africa and Botswana, tend to be the wealthier countries in
the region [17,18]. On the other hand, at the individual
level, wealth has been found to be positively associated
with HIV serostatus [19–21]. Reviews of the existing
literature about the association between socioeconomic Fig. 1. Association between wealth status and HIV preva-
status and HIV infection indicate that only a few studies lence: a conceptual framework.
have found a negative association, whereas most have
found a positive or no association [22,23]. To account for and epidemiological context. For wealth to have an effect
this finding, it has been argued that a greater prevalence of on HIV incidence and prevalence, the underlying factors
risky sexual behaviours among the wealthier may increase must affect one or more of the proximate factors, which
their vulnerability to HIV infection, whereas better in turn affect either the rate of infection or the duration of
nutritional status, greater access to healthcare, and greater infectivity with HIV [24] (a detailed discussion of this
use of antiretroviral drugs may improve their survival if conceptual framework and possible pathways between
infected [21]. wealth and HIV status is provided elsewhere [25]).
In this study, using data from eight recent population- We analysed data from six Demographic and Health
based, nationally representative surveys with HIV Surveys (DHS; Kenya, Malawi, Lesotho, Cameroon,
testing in sub-Saharan Africa, we conducted an in- Ghana, and Burkina Faso) and two AIDS Indicator
depth analysis of the association between household Surveys (AIS; Tanzania and Uganda) with linked HIV test
wealth status and HIV prevalence in sub-Saharan Africa. results. These surveys, conducted during the period
Our aim is to identify what specific characteristics and 2003–2005, collected sociodemographic and behavioural
behaviours of the wealthier are associated with HIV data as well as blood samples for HIV testing from
infection, and to what extent confounding factors such as nationally representative samples of adult women and
place of residence and other risk factors mediate this men. Table 1 gives basic information about the eight
association. surveys. The sampling design and survey implementation
procedures for each country are described in detail in the
individual country survey reports [26–33]. Although the
age ranges of adults included in the surveys varied across
Methods countries (see Table 1), for consistency the present
analysis is limited to men and women aged 15–49 years in
We conceptualized the association between wealth and each country.
HIV as being influenced by a host of underlying
background factors and mediated by several proximate In all surveys, HIV testing was carried out using dried
factors (Fig. 1). We viewed this relationship as transitional blood spot samples collected on a special filter paper using
in nature, operating within and depending on the social capillary blood from a finger prick, except in Uganda](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-23-2048.jpg)
![Wealth and HIV in sub-Saharan Africa Mishra et al. S19
Table 1. Number of men and women interviewed and tested for HIV, Demographic and Health Surveys/AIDS Indicator Surveys with linked
HIV testing.
Country (year) No. eligible for No. Interview No. eligible for No. tested HIV response No. ever
Sex (age group, years) interview interviewed response rate (%) HIV testing for HIV rate (%) had sex
Kenya 2003
Male (15–54) 4183 3578 86 4183 2941 70 3038
Female (15–49) 8717 8195 94 4303 3285 76 6784
Tanzania 2003/04
Male (15–49) 6194 5659 91 6194 4774 77 4690
Female (15–49) 7154 6863 96 7154 5973 83 5963
Uganda 2004/05
Male (15–59) 9905 8830 89 9905 8298 84 7390
Female (15–59) 11 454 10 826 95 11 454 10 227 89 9483
Malawi 2004
Male (15–54) 3797 3261 86 3797 2404 63 2863
Female (15–49) 12 229 11 698 96 4071 2864 70 10 397
Lesotho 2004/05
Male (15–59) 3305 2797 85 3305 2246 68 2291
Female (15–49) 7522 7095 94 3758 3032 81 5917
Cameroon 2004
Male (15–59) 5676 5280 93 5676 5098 90 4424
Female (15–49) 11 304 10 656 94 5738 5287 92 9280
Ghana 2003
Male (15–59) 5345 5015 94 5345 4274 80 3861
Female (15–49) 5949 5691 96 5949 5311 89 4807
Burkina Faso 2003
Male (15–59) 3984 3605 90 3984 3418 86 2769
Female (15–49) 12 952 12 477 96 4575 4223 92 10 911
where venous blood was collected. Participation in HIV by means of an index based on household ownership of
testing was voluntary and, before collecting blood samples consumer durables (such as a television and a bicycle;
for HIV testing, each selected participant was asked to materials used for housing construction; and the avail-
provide informed consent to the testing [34]. Informed ability of amenities such as electricity, source of drinking
consent was obtained separately for the questionnaire water, and type of toilet facility) that tend to be correlated
interview. In each country, HIV testing was conducted in with household economic status. The index, constructed
a central laboratory by following a standard testing using principal components analysis, is a composite
algorithm designed to maximize the sensitivity and measure of the cumulative living standard of a household,
specificity of HIV test results, and an approved quality which places individual households on a continuous scale
assurance and quality control plan [35]. The testing of relative wealth [36,37]. The wealth index is divided
algorithm used two HIV enzyme immunosorbent assays, into population quintiles, with the lowest quintile
based on different antigens. All discordant samples that representing the poorest 20% and the highest quintile
were positive on the first test and negative on the second representing the wealthiest 20% of households within
test were retested with the same enzyme immunosorbent each country. The wealth index defined in this manner
assays, and if still discordant, were resolved by Western captures well the relative economic status within each
blot testing. These steps were also repeated for 5–10% of country, and it correlates strongly with the health and
randomly selected samples that tested negative on the first wellbeing of people [37].
test. For external quality assessment, a subset of dried
blood spot samples (usually approximately 5%) was Using the conceptual framework illustrated in Fig. 1, we
retested at an outside reference laboratory using the systematically examined the association between wealth
same algorithm. and HIV infection. For each country, we first descrip-
tively evaluated whether household wealth was associated
In order to ensure confidentiality, the HIV test results with key risk behaviours and protective factors, including
were anonymously linked to individual and household those that may increase the risk of HIV exposure [age at
questionnaire information through bar codes, after first sexual intercourse, age at first cohabitation, number
scrambling the household and cluster identifiers [35]. of times married, duration in current union, polygamy,
All HIV testing procedures were reviewed by the ethical partner living elsewhere (for women only), number of
review boards of Macro International Inc. (a US-based lifetime and recent sexual partners, sexual intercourse
company that provides technical assistance to DHS/AIS with a non-regular (non-marital, non-cohabiting) part-
surveys around the world), and the host country. ner], and those that may be associated with an increased
risk of transmission per exposure [condom use with the
As DHS/AIS surveys do not include direct questions on last non-regular partner and consistent condom use,
income or expenditure, we measured household wealth alcohol use at last sexual intercourse, reported STI or STI](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-24-2048.jpg)
![S20 AIDS 2007, Vol 21 (suppl 7)
symptoms, circumcision (for men only)]. We also the exception of men in Ghana and Lesotho where HIV
examined the association with knowledge of the prevalence in the highest wealth quintile is slightly lower
respondent’s HIV status, and knowledge of HIV than in the lowest quintile. In most cases, HIV prevalence
prevention methods. increases monotonically with household wealth status,
with the notable exception of Ghana where there is an
We used multivariate logistic regression to measure the inverted U-shaped relationship between the two.
independent relationship between household wealth and Similarly, for cohabiting couples the likelihood that
HIV status after controlling for underlying and mediating one or both partners is HIV infected increases
proximate factors. In particular, for women and men aged with household wealth, with the wealthiest 20% of
15–49 years who reported ever having sex, we estimated couples being two to seven times more likely to have
five alternative logistic regression models. The first model HIV than the poorest 20%. The only exception is
estimated unadjusted effects of household wealth on HIV Lesotho, where this ratio is smaller (1.2) but in the same
prevalence. The second model added controls for several direction.
underlying background factors, including age, ethnicity,
religion, urban/rural residence, and geographical region Wealthier men and women tend to be more educated,
of residence. The third model additionally controlled for more mobile, and more likely to live in urban areas,
education, occupation, media exposure, marital status, where HIV is more prevalent. Wealthier men and
duration in union, number of years at current place of women are also more likely to be older, regularly
residence, alcohol use at last sex in the previous 12 exposed to mass media, to be working in professional or
months, knowledge of HIV prevention methods, and service jobs than poorer individuals (not shown). In
knowledge of own HIV status. The fourth model added a addition, consistently across countries, wealthier indi-
set of proximate factors that were likely to mediate the viduals tend to start cohabiting at an older age than
relationship between the underlying background factors poorer individuals, with an average age difference
and HIV prevalence (as indicated in Fig. 1). These between the highest and lowest wealth quintile of 2–
included age at first sexual intercourse, the number of 4 years in most cases (Tables 3 and 4). Wealthier men
lifetime sexual partners (replaced with whether the and women are less likely to be in a polygamous union
respondent had two or more partners in the previous than poorer men and women. Knowledge of HIV
12 months in Kenya, Ghana, Burkina Faso, and prevention (being faithful to one’s regular partner and
Malawi, where information on lifetime partners was using condoms) increases with household wealth for
not available), reported STI or STI symptoms in the both men and women in all countries, except for men in
previous 12 months, circumcision (for men only), and Tanzania, Malawi, and Cameroon and for women
consistent condom use in the previous 12 months. Finally, in Lesotho, where there is little difference in such
the fifth model added a control for community-level knowledge by wealth status. Wealthier men are more
wealth, computed by averaging the household wealth likely to report having had two or more sexual partners
scores in each cluster. We also fitted a similar set of models in the past 12 months than poorer men, with the notable
to the data for cohabiting couples to examine the exception of Tanzania where the pattern is reversed.
association between household wealth and the likelihood Wealthier men also tend to have more lifetime sexual
that one or both partners was HIV positive. Results from partners and are more likely to have sex with non-regular
only the first model (unadjusted) and the fifth model partners than poorer men. In all countries, wealthier
(adjusted for all potential confounders and mediating men and women are more likely to use condoms
factors) are presented. The analysis accounts for the than poorer individuals. Also, wealthier men are more
complex survey design of the DHS/AIS to estimate likely than poorer men to be circumcised, except in
efficient regression coefficients and robust standard errors Lesotho.
adjusting for intracluster correlation and by using
sampling weights. In Table 5 we present unadjusted and adjusted odds of
HIV infection by wealth quintile. Unadjusted odds
indicate that, in all countries except Lesotho and Ghana,
men belonging to the highest wealth quintile are more
Results likely to be HIV infected than those belonging to the
lowest wealth quintile. In Lesotho and Ghana, there is
The overall HIV prevalence in the eight countries an inverted U-shaped relationship between household
considered ranges from 1.8% in Burkina Faso to 23.5% in wealth and HIV prevalence among men; in other
Lesotho, with women having a higher HIV prevalence terms, the odds of HIV infection peak in the middle
than men in all countries except Burkina Faso (Table 2). wealth quintile. Higher HIV prevalence with increasing
In all countries, HIV prevalence tends to be much higher household wealth is also observed for women; the odds of
among adults belonging to the wealthiest 20% of HIV infection are two to five times greater in the highest
households than among those from the poorest 20%. wealth quintile than in the lowest wealth quintile
This pattern holds for men and women separately, with (statistically significant in all countries). This suggests a](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-25-2048.jpg)
![S26 AIDS 2007, Vol 21 (suppl 7)
more likely to be circumcised, which may reduce their risk to underreport and men tend to exaggerate their
of HIV infection. Also, wealthier adults may live longer premarital and extramarital sexual activity [39]. Epide-
with HIV than poorer individuals as a result of their better miological studies in Africa have also observed weak
health and nutritional status. associations between self-reported risky sexual behaviour
and HIV status [40]. The findings of our study may be
Women are less likely than men to report having multiple biased to the extent that men and women misreport their
partners and non-regular partners. We found that the number of sexual partners, sex with non-regular partners,
positive association between wealth status and HIV condom use, and other related behaviours, or to the
prevalence tends to be stronger for women than for men extent that the degree of misreporting is different across
in most countries, suggesting disproportionately greater the wealth quintiles.
vulnerability of women in the wealthier groups.
A fourth limitation is that the surveys included in the
There are several limitations of this study that should be analysis did not collect data on concurrent partnerships
kept in mind when interpreting our findings. One and sexual networks. We were thus unable to examine the
important limitation is that DHS/AIS surveys do not extent to which wealthier individuals are more likely to
collect data on household income or expenditure, which engage in such complex patterns of sexual relations,
would traditionally be included in an assessment of wealth which may increase the risk of HIV infection in Africa by
status. The assets-based wealth index used here is only a allowing the virus to spread rapidly to others [41–45].
proxy indicator for household economic status [36]. In
addition, wealth index scores cannot be compared across Moreover, because of the cross-sectional nature of the
countries both because the level and distribution of data used in this study, endogeneity might bias our results
wealth differs from one country to another and because at several levels. First, when considering the effect of
the choice of assets included in the construction of the wealth status on HIV prevalence we do not allow for the
index varies somewhat from country to country. In spite opposite, detrimental effect of HIV infection on wealth
of these issues, in developing-country settings, the wealth status, which is well established [46,47]. Excluding HIV-
index has been shown to produce superior results and positive individuals who reported being seriously ill for
equal or greater distinctions in health outcomes than three or more months in the previous 12 months (in
household expenditure-based measures [37]. Moreover, Tanzania, Uganda, Cameroon, and Malawi, where such
as income and expenditure measures can be volatile and information was collected) had virtually no effect on the
temporary, wealth status (which results from the observed associations between wealth and HIV status
accumulation of income) is a preferred measure to relate (data not shown). Second, if HIV-positive adults were
to HIV prevalence (which results from an accumulation aware of their serostatus, they might have adjusted their
of incidence). sexual and reproductive behaviour. We tested this by
excluding HIV-positive individuals who were previously
A second limitation is differential non-response in the tested and received the result, which made little difference
surveys considered. Non-response rates for HIV testing to the observed associations between wealth and HIV
tend to be higher among the wealthier, urban, and more status (data not shown). Third, when infected with HIV,
educated adults, who also tend to have higher HIV wealthier individuals are likely to survive longer than
prevalence. Previous research has, however, indicated that poorer individuals because of better nutrition and access
in these surveys differential non-response has small and to healthcare. Cross-sectional data used in this study did
insignificant effects on the observed HIV prevalence, so not allow taking into account such selective survival of
any bias caused by differential non-response by wealth wealthier respondents. A lack of information on the
status should be small [38]. In any case, if there were no availability and access to treatment and care (antiretroviral
differential non-response by wealth status, the positive drugs in particular) further limited the possibility of
association between wealth status and HIV prevalence disentangling this effect. Antiretroviral therapy coverage
would be even stronger. In addition, the surveys was, however, still very low at the time of the survey data
considered exclude population groups that are difficult collection in most countries.
to locate or interview, most notably the homeless. The
observed positive association between wealth status and Finally, for many HIV-positive adults, the infection may
HIV prevalence may be overestimated to the extent that have preceded their sexual and other behaviours recorded
the homeless are poorer and have higher HIV prevalence in the survey, which may have biased some of the
than those included in the survey. Given that the associations. Moreover, the strength and direction of
proportion of the homeless in the total population tends the relationship between wealth status and HIV
to be small, any effect of excluding this group on the prevalence and the roles of risk behaviours and protective
observed associations is likely to be small. factors are likely to change over time, depending on the
stage and spread of the epidemic [48]. Cross-sectional data
Another limitation is that our analysis is based on self- used in our study do not allow the examination of causal
reported behaviours. There is evidence that women tend effects and these transitional phenomena.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-31-2048.jpg)
![Wealth and HIV in sub-Saharan Africa Mishra et al. S27
In conclusion, this study found a positive association 2. Adler NE, Newman K. Socioeconomic disparities in health:
pathways and policies. Health Aff (Millwood) 2002; 21:60–
between household economic status and HIV prevalence 76.
among adult men and women in sub-Saharan Africa. 3. Fotso JC, Kuate-Defo B. Socioeconomic inequalities in early
Accounting for various underlying factors and proxi- childhood malnutrition and morbidity: modification of the
household-level effects by the community SES. Health Place
mate determinants explains much of this positive 2005; 11:205–225.
association, but in most cases wealthier adults remain 4. Ellen JM, Kohn RP, Bolan GA, Shiboski S, Krieger N. Socio-
at least as likely as poorer individuals to be HIV infected. economic differences in sexually transmitted disease rates
among black and white adolescents, San Francisco, 1990 to
We found that HIV prevalence does not follow the same 1992. Am J Public Health 1995; 85:1546–1548.
pattern of association with poverty within countries in 5. Fleming DT, McQuillan GM, Johnson RE, Nahmias AJ, Aral SO,
sub-Saharan Africa as most other diseases. Although Lee FK, et al. Herpes simplex virus type 2 in the United States,
1976 to 1994. N Engl J Med 1997; 337:1105–1111.
poverty reduction is an essential strategy to improve 6. Lacey CJN, Merrick DW, Bensley DC, Fairley I. Analysis of the
health and combat the HIV epidemic, our analysis sociodemography of gonorrhoea in Leeds, 1989-93. BMJ 1997;
314:1715–1718.
suggests that HIV prevention, care, and treatment 7. Holtgrave DR, Crosby RA. Social capital, poverty, and income
programmes should also be focused on the better-off inequality as predictors of gonorrhoea, syphilis, chlamydia and
segments of the population. Focusing on the most AIDS cases rates in the United States. Sex Transm Infect 2003;
79:62–64.
important modes of exposure will probably be more 8. Kyriakis KP, Hadjivassiliou M, Paparizos VA, Flemetakis A,
effective than focusing broadly on poverty reduction Stavrianeas N, Katsambas A. Incidence determinants of gonor-
[49,50]. It will also be important to extend programmes rhea, chlamydial genital infection, syphilis and chancroid in
attendees at a sexually transmitted disease clinic in Athens,
to the rural areas where a majority of the population in Greece. Int J Dermatol 2003; 42:876–881.
sub-Saharan Africa resides. 9. Miller GC, McDermott R, McCulloch B, Fairley CK, Muller R.
Predictors of the prevalence of bacterial STI among young
disadvantaged Indigenous people in north Queensland,
It is important to stress that our findings do not imply that Australia. Sex Transm Infect 2003; 79:332–335.
the poor are not disproportionately affected by HIV when 10. Wald A. Synergistic interactions between herpes simplex virus
they do get infected. Poverty reduction is an extremely type-2 and human immunodeficiency virus epidemics. Herpes
2004; 11:70–76.
important goal in itself for many reasons, and it will 11. Chawla R, Bhalla P, Garg S, Meghachandra Singh M, Bhalla K,
certainly help combat the HIV epidemic in the long run Sodhani P, et al. Community based study on sero-prevalence of
and deal with its many adverse consequences. syphilis in New Delhi (India). J Commun Dis 2004; 36:205–
211.
12. Uuskula A, Nygard JF, Kibur-Nygard M. Syphilis as a social
disease: experience from the post-communist transition period
in Estonia. Int J STD AIDS 2004; 15:662–668.
13. Bukusi EA, Cohen CR, Meier AS, Waiyaki PG, Nguti R, Njeri JN,
Acknowledgements Holmes KK. Bacterial vaginosis: risk factors among Kenyan
women and their male partners. Sex Transm Dis 2006; 33:361–
367.
The authors wish to thank the numerous people who 14. Fitzgerald DW, Behets F, Caliendo A, Roberfroid D, Lucet C,
provided comments on the two presentations and on Fitzgerald JW, Kuykens L. Economic hardship and sexually
earlier drafts of this paper. transmitted diseases in Haiti’s rural Artibonite Valley. Am J
Trop Med Hyg 2000; 62:496–501.
15. Fenton L. Preventing HIV/AIDS through poverty reduction:
Sponsorship: Funding for this research was provided the only sustainable solution? Lancet 2004; 364:1186–1187.
by the Joint United Nations Programme on HIV/AIDS 16. Bloom DE, Sevilla J. Health, wealth, AIDS, and poverty. In:
(UNAIDS) (no. HQ/05/423963). Additional support Report of the Asia-Pacific Ministerial Meeting. 9–10 October
2001. Melbourne, Australia: The Australian Government Over-
was provided by the United States Agency for seas Aid Programme.
International Development (USAID) through the MEA- 17. Whiteside A. Poverty and HIV/AIDS in Africa. Third World Q
SURE DHS project (no. GPO-C-00-03-00002-00). 2002; 23:313–332.
18. UNAIDS. Report on the global AIDS epidemic. Geneva:
The preliminary findings of this research were pre- UNAIDS; 2006.
19. Menon R, Wawer MJ, Konde-Lule JK, Sewankambo NK, Li C.
sented at the US President’s Emergency Plan for AIDS The economic impact of adult mortality on households in Rakai
Relief (PEPFAR) Annual Meeting in Durban, South district, Uganda. In: Ainsworth M, Fransen L, Over M, editors.
Africa, on 12–15 June 2006, and at the International Confronting AIDS: evidence from the developing world. Brus-
AIDS Economic Network Meeting in Toronto, Canada, sels: European Commission; and Washington, DC: The World
Bank; 1998. pp. 321–335.
on 11–12 August 2006. 20. Kirunga CT, Ntozi JP. Socio-economic determinants of HIV
serostatus: a study of Rakai District, Uganda. Health Transit Rev
The views expressed are those of the authors and do 1997; 7 (Suppl.):175–188.
not necessarily reflect the views of UNAIDS, USAID, 21. Shelton JD, Cassell MM, Adetunji J. Is poverty or wealth at the
root of HIV? Lancet 2005; 366:1057–1058.
the United States Government, or the organizations 22. Ainsworth M, Semali I. Who is most likely to die of AIDS?
with which the authors are affiliated. Socioeconomic correlates of Adult Deaths in Kagera Region,
Tanzania. In: Ainsworth M, Fransen L, Over M, editors. Con-
fronting AIDS: evidence from the developing world. Brussels:
References European Commission; and Washington, DC: The World Bank;
1998. pp. 95–109.
1. Kuate-Defo B. Effects of socioeconomic disadvantage and wo- 23. Wojciki JM. Socioeconomic status as a risk factor for HIV
men’s status on women’s health in Cameroon. Soc Sci Med infection in women in East, Central and Southern Africa: a
1997; 44:1023–1042. systematic review. J Biosoc Sci 2005; 37:1–36.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-32-2048.jpg)
![S28 AIDS 2007, Vol 21 (suppl 7)
24. Boerma JT, Weir SS. Integrating demographic and epidemio- 37. Rutstein SO, Johnson K. The DHS wealth asset index. DHS
logical approaches to research on HIV/AIDS: the proximate- comparative report no. 6. Calverton, Maryland, USA: ORC
determinants framework. J Infect Dis 2005; 191 (Suppl. 1):S61– Macro; 2004.
S67. 38. Mishra V, Vaessen M, Boerma JT, Arnold F, Way A, Barrere B,
25. Mishra V, Bignami S, Greener R, Vaessen M, Hong R, Ghys P, et al. HIV testing in national population-based surveys: experi-
et al. A study of the association of HIV infection with wealth in ence from the Demographic and Health Surveys. Bull WHO
sub-Saharan Africa. DHS working papers no. 31. Calverton, 2006; 84:537–545.
Maryland, USA: Macro International Inc.; 2007. 39. Zaba B, Pisani E, Slaymaker E, Boerma JT. Age at first sex:
26. Central Bureau of Statistics (CBS), Ministry of Health (MOH), understanding recent trends in African demographic surveys.
and ORC Macro. Kenya demographic and health survey 2003. Sex Transm Infect 2004; 80 (Suppl. 2):ii28–ii35.
Calverton, Maryland, USA: CBS, MOH, and ORC Macro; 2004. ´
40. Buve A, Lagarde E, Carael M, Rutenberg N, Ferry B, Glynn JR,
¨
27. Ghana Statistical Service (GSS), Noguchi Memorial Institute for et al. Study Group on Heterogeneity of HIV Epidemics in
Medical Research (NMIMR), and ORC Macro. Ghana demo- African Cities. Interpreting sexual behaviour data: validity
graphic and health survey 2003. Calverton, Maryland, USA: issues in the multicentre study on factors determining the
GSS, NMIMR, and ORC Macro; 2004. differential spread of HIV in four African cities. AIDS 2001;
´
28. Institut National de la Statistique et de la Demographie (INSD), 15 (Suppl. 4):S117–S126.
and ORC Macro. Burkina Faso demographic and health survey 41. Hudson CP. AIDS in rural Africa: a paradigm for HIV-1 pre-
2003 [in French]. Calverton, Maryland, USA: INSD and ORC vention. Int J STD AIDS 1996; 7:236–243.
Macro; 2004. 42. Morris M, Kretzschmar M. Concurrent partnerships and the
29. Tanzania Commission for AIDS (TACAIDS), National Bureau of spread of HIV. AIDS 1997; 11:641–648.
Statistics (NBS), and ORC Macro. Tanzania HIV/AIDS indicator 43. Lagarde E, Auvert B, Chege J, Sukwa T, Glynn JR, Weiss HA,
survey 2003–04. Calverton, Maryland, USA: TACAIDS, NBS, et al. Condom use and its association with HIV/sexually trans-
and ORC Macro; 2005. mitted diseases in four urban communities of sub-Saharan
30. Institut National de la Statistique (INS) and ORC Macro. Africa. AIDS 2001; 15 (Suppl. 4):S71–S78.
Cameroun demographic and health survey 2004 [in French]. 44. Halperin DT, Epstein H. Concurrent sexual partnerships help to
Calverton, Maryland, USA: INS and ORC Macro; 2005. explain Africa’s high HIV prevalence: implications for preven-
31. National Statistical Office (NSO), and ORC Macro. Malawi tion. Lancet 2004; 364:4–6.
demographic and health survey 2004. Calverton, Maryland, 45. Helleringer S, Kohler HP. Social networks, perceptions of risk,
USA: NSO and ORC Macro; 2005. and changing attitudes towards HIV/AIDS: new evidence from
32. Ministry of Health and Social Welfare (MOHSW), Bureau of a longitudinal study using fixed-effects analysis. Popul Stud
Statistics (BOS), and ORC Macro. Lesotho demographic and 2005; 59:265–282.
health survey 2004. Calverton, Maryland, USA: MOHSW, BOS, 46. Piot P, Bartos M, Ghys PD, Walker N, Schwartlander B. The
and ORC Macro; 2005. global impact of HIV/AIDS. Nature 2001; 410:968–973.
33. Ministry of Health (MOH) and ORC Macro. Uganda HIV/AIDS 47. UNAIDS. Report on the global HIV/AIDS epidemic. Geneva:
sero-behavioural survey 2004–2005. Calverton, Maryland, UNAIDS; 2000.
USA: MOH and ORC Macro; 2006. 48. Hargreaves J, Boler T. Girl power: the impact of girls’ education
34. ORC Macro. Anemia and HIV testing field manual: demo- on HIV and sexual behaviour. Johannesburg, South Africa:
graphic and health surveys. Calverton, Maryland, USA: ORC ActionAid International; 2006.
Macro; 2005. 49. Pisani E, Garnett GP, Grassly NC, Brown T, Stover J, Hankins C,
35. ORC Macro. HIV testing laboratory manual: demographic and et al. Back to basics in HIV prevention: focus on exposure. BMJ
health surveys. Calverton, Maryland, USA: ORC Macro; 2005. 2003; 326:1384–1387.
36. Filmer D, Pritchett L. Estimating wealth effects without 50. Gouws E, White PJ, Stover J, Brown T. Short term estimates of
expenditure data – or tears: an application to educational adult HIV incidence by mode of transmission: Kenya and
enrollments in states of India. Demography 2001; 38:115– Thailand as examples. Sex Transm Infect 2006; 82 (Suppl. 3):
132. iii51–iii55.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-33-2048.jpg)
![The socioeconomic determinants of HIV incidence:
evidence from a longitudinal, population-based study
in rural South Africa
Till Barnighausena,b, Victoria Hosegooda,c, Ian M. Timaeusa,c and
¨
Marie-Louise Newella,d
Background: Knowledge of the effect of socioeconomic status on HIV infection in
Africa stems largely from cross-sectional studies. Cross-sectional studies suffer from
two important limitations: two-way causality between socioeconomic status and
HIV serostatus and simultaneous effects of socioeconomic status on HIV incidence
and HIV-positive survival time. Both problems are avoided in longitudinal cohort
studies.
Methods: We used data from a longitudinal HIV surveillance and a linked demo-
graphic surveillance in a poor rural community in KwaZulu-Natal, South Africa, to
investigate the effect of three measures of socioeconomic status on HIV incidence:
educational attainment, household wealth categories (based on a ranking of households
on an assets index scale) and per capita household expenditure. Our sample comprised
of 3325 individuals who tested HIV-negative at baseline and either HIV-negative or
-positive on a second test (on average 1.3 years later).
Results: In multivariable survival analysis, one additional year of education reduced
the hazard of acquiring HIV by 7% (P ¼ 0.017) net of sex, age, wealth, household
expenditure, rural vs. urban/periurban residence, migration status and partnership
status. Holding other factors equal, members of households that fell into the middle
40% of relative wealth had a 72% higher hazard of HIV acquisition than members of the
40% poorest households (P ¼ 0.012). Per capita household expenditure did not sig-
nificantly affect HIV incidence (P ¼ 0.669).
Conclusion: Although poverty reduction is important for obvious reasons, it may not be
as effective as anticipated in reducing the spread of HIV in rural South Africa. In
contrast, our results suggest that increasing educational attainment in the general
population may lower HIV incidence.
ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 7):S29–S38
Keywords: Africa, economics, epidemiology, HIV incidence, longitudinal study,
socioeconomic factors, surveillance
Introduction 5.5 million adults and children living with HIV and an
estimated 320 000 deaths as a result of HIV/AIDS in 2005
The HIV epidemic remains one of the greatest health and [1]. From the beginning of the HIVepidemic, researchers
development challenges facing sub-Saharan Africa. South have tried to understand its relationship with socio-
Africa bears a substantial brunt of the HIVepidemic, with economic status. Reviews of studies in Africa show a wide
From the aAfrica Centre for Health and Population Studies, University of KwaZulu-Natal, Mtubatuba, South Africa, the
b
Department of Population and International Health, Harvard School of Public Health, Boston, Massachusetts, USA, the cCentre
for Population Studies, London School of Hygiene and Tropical Medicine, London, UK, and the dCentre for Paediatric
Epidemiology and Biostatistics, Institute of Child Health, University College London, London, UK.
Correspondence to Till Barnighausen, Africa Centre for Health and Population Studies, University of KwaZulu-Natal, PO Box 198,
¨
Mtubatuba 3935, South Africa.
E-mail: tbarnighausen@africacentre.ac.za
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S29](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-34-2048.jpg)
![S30 AIDS 2007, Vol 21 (suppl 7)
variety of relationships between HIV and socioeconomic has been published, and there have only been a few
status [2–5]. Most of the studies, however, examine longitudinal studies reporting on the effect of socio-
cross-sectional associations between HIV serostatus and economic status on the hazard of HIV seroconversion in
socioeconomic status and thus suffer from two other African countries [17–21]. With one exception,
important limitations. these longitudinal studies were conducted in cohorts of
sexually active young urban women attending antenatal
First, cross-sectional studies are usually unable to care or family planning services [17,18,20], or in cohorts
distinguish between the effect of socioeconomic status of urban factory workers [19,21]. Relationships between
on HIV infection and the effect of HIV infection on socioeconomic status and HIV incidence in these
socioeconomic status [6–8]. Several pathways have been subpopulations may differ from those in the general
suggested through which a decrease in socioeconomic population, or may not be detectable because of limited
status may increase the risk of HIV infection. Malnu- variation in socioeconomic status [22].
trition and micronutrient deficiencies, which are more
common among the poor than in the rest of society can No clear pattern of the relationships between socio-
disrupt the integrity of the vaginal epithelium, increasing economic status and HIV seroconversion risk emerges
its permeability to HIV [9]. Ulcerative genital diseases, from these studies. In urban women attending antenatal
which are associated with low socioeconomic status, care or family planning services, the risk of HIV
increase the transmission probability of HIV [7]. Women seroconversion has been found not to be associated with
of low socioeconomic status may be economically education [18], to decrease with the educational
dependent on male partners, limiting their ability to attainment of women’s partners [20], and to decrease
negotiate condom use in relationships, or forcing them to with women’s incomes [17]. In studies in urban factory
sell sex for money [2,10]. On the other hand, an increase workers, the risk of HIV seroconversion was positively
in socioeconomic status may increase the risk of HIV associated with occupational status [19] and educational
infection because wealthy or educated people have more attainment [21]. The only published study using
resources with which to attract and maintain multiple longitudinal, population-based data to examine the
partners [11]. association between socioeconomic status and HIV
incidence in Africa found that in rural Uganda
HIV status may also be an important determinant of educational attainment is not a significant predictor of
socioeconomic status. HIV-related diseases can limit HIV incidence [23].
people’s ability to work, which may decrease socio-
economic status, especially among individuals who work Wojcicki [4] concluded from a review of studies on
in economic sectors in which income is closely linked to socioeconomic status and HIV infection in sub-Saharan
productivity associated such as agriculture or the informal Africa that ‘‘further studies should use multiple measures of
sector [12]. In addition, HIV and its associated diseases socioeconomic status’’ because different dimensions of
may lead to substantial expenditure for health care, socioeconomic status influence HIV infection risk in
decreasing household wealth [12–14]. different ways. We report findings about the effect of three
different measures of socioeconomic status (educational
A second problem of cross-sectional studies is that they attainment, household wealth categories and household
usually cannot distinguish between the effect of socio- expenditure) on HIV incidence from a longitudinal HIV
economic status on HIV incidence and the effect of surveillance of the general population in a rural community
socioeconomic status on survival with HIV. Previous in KwaZulu-Natal, South Africa.
studies have shown that HIV survival time increases with
socioeconomic status [15], possibly because diet or access
to antiretroviral therapy improve with socioeconomic
status [4,16]. In a cross-sectional study, it is thus possible to Methods
find a positive association between socioeconomic status
and HIV infection, even if higher socioeconomic status Study area
protects individuals from acquiring HIV. We used data from the longitudinal population-based
HIV surveillance conducted by the Africa Centre for
Both problems – two-way causality between socio- Health and Population Studies, University of KwaZulu-
economic status and HIV serostatus and the simultaneous Natal, and from the Africa Centre Demographic
effects of socioeconomic status on HIV incidence and on Information System (ACDIS) to investigate the relation-
HIV-positive survival time – are avoided when a cohort ship between socioeconomic status and the hazard of HIV
of HIV-negative individuals is followed over time and the seroconversion. ACDIS has been collecting demographic
hazards of HIV seroconversion are compared across data since January 2000 and socioeconomic data since
individuals of different socioeconomic status in survival February 2001 [24]. The ACDIS demographic surveil-
analyses. To date, no such longitudinal study on lance area (DSA) is located in the rural district of
socioeconomic status and HIV incidence in South Africa Umkhanyakude in northern KwaZulu-Natal, South](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-35-2048.jpg)
![The socioeconomic determinants of HIV incidence Barnighausen et al.
¨ S31
Africa. It covers 435 square kilometres and a total resident were contacted and refused to consent to an HIV test
population of approximately 86 000 Zulu-speaking (n ¼ 2551)]. In comparison with the individuals included
people (June 2003). Although the study took place in in the sample, individuals for whom information on the
an overall rural community, the area includes an urban independent variables used in this study was available, but
township and periurban areas (informal settlements with a a second HIV test was not, were not significantly different
population density of more than 400 people per square at the 5% confidence level with regard to per capita
kilometer). In 2001, 61% of households in the ACDIS household expenditure, rural vs. urban/periurban place
area had a toilet and only 38% had access to piped water. of residence or the probability of having a partner at
The unemployment rate in the same year was 30% [25]. baseline. They were, however, slightly younger (25.4 vs.
26.4 average years of age, P < 0.001), slightly more
Data collection educated (8.2 vs. 7.7 average educational grade attained,
Teams of two trained fieldworkers visited each eligible P < 0.001), wealthier (a value of À0.073 vs. À0.352 on
individual in his or her household. Fieldworkers revisited the household assets index scale, P < 0.001), more likely
households up to four times to contact absentees. If a to be male (49% vs. 40%, P < 0.001), more likely to
subject no longer lived at the household, the field worker have migrated out of the DSA between the two
handed the case to a specially trained tracking team that surveillance rounds (14% vs. 7%, P < 0.001) and less
made up to 10 attempts to find the individual in his or her likely to be married (9% vs. 11%, P ¼ 0.001) than the
new residence. After written informed consent, the field individuals in the sample. Table 1 shows the character-
workers collected blood by finger stick and prepared istics of the 3325 individuals included in the sample.
dried blood spots for HIV testing according to the Joint Even though the study took place in an overall rural
United Nations Programme on HIV/AIDS (UNAIDS) community, a substantial proportion of participants (28%)
and World Health Organization (WHO) Guidelines for lived in either an urban or a periurban area (Table 1).
Using HIV Testing Technologies in Surveillance [26].
HIV status was determined by antibody testing with a Although in comparison with South Africa as whole, the
broad-based HIV-1/HIV-2 enzyme-linked immunosor- average indicators of socioeconomic status in this
bent assay (ELISA; Vironostika, Organon Teknika, community are low [25], their dispersion within the
Boxtel, the Netherlands) followed by a confirmatory community is quite wide (see Table 1). For example, in
ELISA (GAC-ELISA; Abbott, Abbott Park, Illinois, our sample the 10th percentile of educational attainment
USA). All covariates used in this study were collected by was 2nd grade, whereas the 90th percentile was 12th
the ACDIS demographic surveillance system conducted grade, and the 10th percentile of daily total per capita
between January 2003 and September 2004 [25]. household expenditure was 1.2 South African Rand
(ZAR), whereas the 90th percentile was 7.2 ZAR,
Sample suggesting that there is sufficient variation in these two
Our sample includes all individuals who met the measures of socioeconomic status to warrant investigating
following criteria: they were age-eligible for inclusion their effects on HIV incidence.
in the HIV surveillance (women between 15 and 49 years
of age and men between 15 and 54 years of age) both Table 1. Sample characteristics (N U 3325).
during the first surveillance round (from June 2003 to N (%)a
December 2004) and during the second round (from
January to December 2005); they were residents in the Sex
ACDIS DSA during the first round of the HIV Male 1317 (40%)
Female 2008 (60%)
surveillance and either residents in the DSA or non- Mean (SD) age (years) 26.4 (11.0)
resident household members during the second round; Mean (SD) grade of educational attainment 7.7 (3.4)
they tested HIV-negative during the first round and tested Wealth category
either HIV-negative or HIV-positive during the second Poorest 40% 1330 (40%)
Middle 40% 1330 (40%)
round; and data on all independent variables used in this Wealthiest 20% 665 (20%)
analysis were available at the time of the HIV test in the Mean (SD) daily total per capita 5.37 (30.50)
first round. On average the time between the first and the household expenditures (ZAR)
Place of residence
second HIV test was 1.3 years. Rural 2396 (72%)
Urban/periurban 929 (28%)
Of the 8952 participants in the population-based HIV Migration status
surveillance who were HIV-negative during the first Migrant 231 (7%)
Non-migrant 3094 (93%)
round of testing and still eligible to be tested during the Partnership status
second round, information on at least one of the Not married, without partner 1687 (51%)
independent variables used in this study was missing for Married 374 (11%)
1597 individuals, and information on HIV status during Not married, with partner 1264 (38%)
the second round of testing was missing for 4030 [either a
Or mean (SD) where indicated. SD, standard deviation; ZAR, South
because individuals could not be contacted (n ¼ 1479) or African rand.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-36-2048.jpg)
![S32 AIDS 2007, Vol 21 (suppl 7)
Independent variables spending for household members (on shopping, rent,
The focus of our analyses is the socioeconomic clothes, water, fuel, electricity, health, transport, religious
determinants of HIV incidence. We investigated the activities, telephones, cell phones, payments for goods
relationship of HIV infection with three measures of bought by hire-purchase or lay-bye, funerals, life
socioeconomic status: educational attainment, household insurance, and school), as well as spending for individuals
wealth and expenditure. Increased educational attainment outside the household (money, goods and food). We used
(the highest education grade that an individual has daily total household expenditure and divided it by the
completed within a country’s educational system) has number of members in each household to adjust for
been hypothesized to lead to a lower risk of HIV infection, household size [34]. The resulting variable, i.e. daily total
because it improves the ability to understand and act on per capita household expenditure, is henceforth referred to
health promotion messages and because it is associated with as household expenditure. We logarithmically transformed
increased exposure to school-based HIV prevention the household expenditure variable to reduce skewness.
programmes or increased access to health services [27].
As expected, educational attainment, the household assets
Wealth and expenditure are likely to capture different index, and household expenditure were positively
financial aspects of social status. Households generate correlated. However, the three measures of socioeconomic
wealth through saving of income after spending money status were not very highly correlated (with Pearson’s
on consumption. There is commonly greater variation in correlation coefficient at 0.256 between educational
wealth than in expenditure because wealth is accumulated attainment and the household assets index; at 0.174
and because some expenditure on basic items such as food between educational attainment and the log transform of
and clothing are indispensable for human survival. Wealth household expenditure; and at 0.380 between the house-
may thus be a more sensitive measure of the long-term hold assets index and the log transform of household
social position of a household and may capture influences expenditure), reinforcing the theoretical conrideration
of social status on the risk of HIV infection better than that each captures a different aspect of socioeconomic status
expenditure. and cannot be used in place of one of the other two in
multiple regression analysis.
We used a household assets index as a measure of wealth.
As shown by Morris and colleagues [28], household assets In addition to the three measures of socioeconomic status,
indices are valid proxies for wealth in health surveys in we controlled for a number of variables that have been
rural Africa. Following Filmer and Pritchett [29], we used found to be associated with HIV infection in
the first principal component obtained in a principal cross-sectional surveys in South Africa (sex [5,35–37],
component analysis of information on house ownership, age [38–41], rural vs. urban/periurban residence [36],
water source, energy, toilet type, electricity and 27 migration status [5,39,41] and partnership status [5,37,39,
household assets as an assets index. The assets included 41–43]). All independent variables were measured at
items that can be used for consumption, production or baseline and assumed to be time-invariant.
both, such as beds, bicycles, tables, telephones, television
sets, sewing machines, block makers, wheelbarrows, Statistical analysis
tractors, cattle, and other livestock. We categorized In order to investigate the effects of explanatory variables
households as either belonging to the poorest 40%, the on the time to HIV seroconversion, we used semipara-
middle 40% or the wealthiest 20% on the assets index metric and parametric survival models in the following
scale. We chose these three categories of relative wealth, proportional hazards specification [44].
because they have been found to capture wealth effects
hðt; Xi Þ ¼ h0 ðtÞ Á expðXi bÞ (1)
well in a number of studies in poor provinces of South
Africa [30,31], including studies investigating the effect of
wealth on health [32,33]. where h0(t) is the baseline hazard function that is obtained
when all explanatory variables are equal to 0. A unit
Household expenditure captures the short-term financial change of an independent variable in this model leads to a
liquidity of the members of a household and should thus constant parallel shift of the baseline hazard function. If
be a better measure of the influences of current the baseline hazard function is left unspecified, the model
consumption of costly services on HIV infection than is the semiparametric Cox proportional hazards model
wealth. For example, in a cash-strapped period, an (CPHM).
individual may not have sufficient funds to seek treatment
for sexually transmitted diseases (STD) whose presence Although parametric models that specify the functional
increases the risk of HIV transmission, or may not form of the baseline hazard function have the disadvan-
be able to travel to a place where condoms are available. tage that they may lead to inconsistent estimates if the
We measured total household expenditure by summing baseline function is misspecified, they will be more
spending across all categories on which expenditure efficient than the semiparametric model if the base-
information is available in the ACDIS, including line hazard is correctly specified. We thus estimated](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-37-2048.jpg)
![The socioeconomic determinants of HIV incidence Barnighausen et al.
¨ S33
parametric models in addition to the CPHM. First, we Results
used the exponential model
During 4352 person-years at risk, 131 of the 3325
hðt; Xi Þ ¼ expðaÞ Á expðXi bÞ (2)
individuals in our sample became seropositive. The
overall incidence of HIV infection was 3.0 per 100
which assumes that the hazard function is constant over person-years (95% confidence interval 2.5–3.6). Table 2
time. Next, we estimated the Weibull model shows the unadjusted hazard ratios when we examined
hðt; Xi Þ ¼ p Á tpÀ1 Á expðaÞ Á expðXi bÞ (3) separately the effects on time to seroconversion of sex;
age, age2 and age3; educational attainment; wealth
categories; household expenditure; place of residence;
which allows the hazard function to increase (p > 1) or
migration status; and partnership status in CPHM. In
decrease (p < 1) monotonically over time. The Weibull
these separate regressions, female sex, age, belonging to
model includes the exponential model as a special case
the middle wealth category, urban/periurban place of
(p ¼ 1).
residence, having migrated out of the DSA between the
first and the second round of the HIV surveillance, and
Finally, we estimated random effects generalizations to the
not being married but having a partner were positively
proportional hazards models, frailty models, which allow
associated with the hazard of HIV seroconversion (all
for variability in unobserved individual-level factors that
P < 0.010). Educational attainment and household
is unaccounted for by the independent variables included
expenditure were not significantly associated with the
in the models [45]. The unobservable individual effect
hazard of HIV seroconversion (both P ! 0.556).
(frailty), v, is considered a random variable over the
population that multiplicatively enters the hazard func-
By experimentation we found that a third-order poly-
tion in the above proportional hazards specification, i.e.
nomial age specification provided a good fit for the
hðt; Xi jvi Þ ¼ vi Á hðt; Xi Þ ¼ vi Á ho ðtÞ Á expðXi bÞ (4) relationship between age and time to HIV seroconversion.
In order to reduce multicollinearity we expressed age as its
The random variable v is assumed to be positive, to be deviation from its mean [46]. When we adjusted for sex and
distributed independently of t and X, and to follow a age (in the third-order polynomial specification) two of the
gamma distribution with unit mean (required for relationships from the individual regressions that do not
identification) and finite variance (u). If u is not significantly adjust for any other factor changed significantly (Table 2).
different from zero, individual heterogeneity is not First, the hazard ratio of educational attainment changed
important and it is appropriate to estimate the non-frailty from 0.99 (P ¼ 0.785) to 0.93 (P ¼ 0.022). Holding age
models. and sex constant, each additional year of educational
Table 2. Unadjusted hazard ratios of HIV seroconversion and hazard ratios adjusted for sex and age.
HR (s.e.) P value aHR (s.e.) P value
Sex
Male 1 – –
Female 1.8314 (0.3616) 0.002 – –
Age (years) 0.9954 (0.0162) 0.778 – –
Age2 0.9915 (0.0020) < 0.001 – –
Age3 1.0003 (0.0001) 0.002 – –
Educational attainment (years) 0.9931 (0.0251) 0.785 0.9338 (0.0279) 0.022
Wealth category
Poorest 40% 1 1
Middle 40% 1.8525 (0.3716) 0.002 1.8688 (0.3750) 0.002
Wealthiest 20% 1.0247 (0.2789) 0.928 1.0427 (0.2844) 0.878
Daily total per capita household expenditures (ZAR, ln) 0.9156 (0.1370) 0.556 0.9591 (0.1415) 0.777
Place of residence
Rural 1 1
Urban/periurban 1.5852 (0.2836) 0.010 1.6272 (0.2923) 0.007
Migration status
Migrant 1 1
Non-migrant 0.4387 (0.1067) 0.001 0.4764 (0.1172) 0.003
Partnership status
Not married, without partner 1 1
Married 0.8437 (0.3065) 0.640 0.8224 (0.3505) 0.646
Not married, with partner 2.1285 (0.3946) < 0.001 1.5992 (0.3680) 0.041
HR, hazard ratio; aHR, hazard ratio adjusted for sex and age; s.e., standard error; ZAR, South African rand; Ln, natural Logarithm.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-38-2048.jpg)
![S34 AIDS 2007, Vol 21 (suppl 7)
attainment reduced the hazard of HIV seroconversion by significant determinant of HIV seroconversion in any
7%. Second, the hazard ratio for the group of unmarried of the models. Urban residence was associated with a 65%
individuals with a partner changed from 2.13 (P < 0.001) increase in the hazard of HIV seroconversion (P ¼ 0.012;
to 1.60 (P ¼ 0.041). Table 3, IIB). Individuals who remained residents in the
ACDIS DSA between the two rounds of the HIV
Table 3 shows estimation results of the semiparametric surveillance faced approximately half the hazard of HIV
CPHM and the parametric exponential and Weibull seroconversion of those who migrated out of the DSA
regression models in their proportional hazard specifica- between the two rounds (P ¼ 0.006). Once all other
tion (expressions 1, 2 and 3, respectively). We tested the variables were controlled for, the hazard ratio of
proportional hazards assumption for all variables jointly unmarried individuals with a partner at baseline remained
and for each variable individually using the tests proposed borderline significant (P ¼ 0.074) and the partnership
by Grambsch and Therneau [47]. The null hypothesis variables jointly increased the model fit with borderline
that the hazard rates are proportional could not be significance (P ¼ 0.099).
rejected at the 10% confidence level in any of the tests. A
unit change in one of the independent variables leads to a In order to test whether the effects of education and wealth
proportional shift of the hazard rate. on HIV incidence are modified by sex, we added in turn
education-sex and wealth category-sex interaction terms
The sizes and significance levels of the adjusted hazard to all the models reported in Table 3. None of the
ratios were very similar in all three estimations. Whereas interaction terms was significant at the 5% confidence
the CPHM leads to consistent estimates and is more level. Furthermore, we added the square of the educational
flexible than the parametric models, it is less efficient than attainment variable to the regression models IB, IIB and
the appropriate parametric model. In the Weibull IIIB in order to investigate whether the effect of education
regression, the null hypothesis that P ¼ 1 could not be on HIV incidence is non-linear. In none of the models was
rejected (Table 3), i.e. the hazard function is neither the added term significant; we thus did not include it in the
increasing nor decreasing over time. The exponential final regression equations. Similarly, we replaced the log
estimation is thus preferred over the Weibull estimation. transform of the household expenditure variable in models
In order to check for unobserved individual heterogen- IB, IIB and IIIB with alternative functional forms (linear,
eity, we estimated the exponential model (Table 3, IIB) linear and square). In none of the alternative regression
and the Weibull model (Table 3, IIIB) including an equations was any of the household expenditure terms
individual random effect, or frailty, y (see equation 4). significant at the 5% confidence level.
The null hypothesis that the individual random effect is
equal to zero was not rejected at the 10% significance
level. The exponential estimation without frailty is thus
the preferred parametric model and was used for the Discussion
description of results below.
We show that educational attainment significantly
In multivariable survival analysis, belonging to the middle reduces the hazard of becoming infected with HIV in
40% of households as ranked by the assets index increased a poor rural community in South Africa when controlling
the hazard of HIV seroconversion by a factor of for sex, age, wealth, household expenditure, place of
approximately 2 (P ¼ 0.001; Table 3, IIA). Controlling residence, migration status and partnership status. The
for place of residence, migration status and partnership protective effect of education shown in this study differs
status in addition to sex and age reduced the size of the from the findings of previous studies that suggest that
hazard ratio (to 1.72) but the effect remained significant educational attainment is not significantly associated, or
(P ¼ 0.012; Table 3, IIB). To test whether our finding that positively associated, with the risk of HIV infection [27].
belonging to a household in the middle wealth category
increases the risk of HIV incidence is robust to a change in The differences between our results and the findings of
the choice of household wealth categories, we repeated previous studies may be caused by methodological issues
the regressions in Table 3 with households categorized such as how they control for confounding. We find that
into wealth tertiles on the assets index scale. The educational attainment is not correlated with the risk of
alternative categorization changed the sizes and signifi- HIV seroconversion in univariable analysis but that its
cant levels of all coefficients only slightly. In particular, protective effect against HIV seroconversion becomes
when we replaced the wealth variables in model IIB with apparent once sex and age are controlled for. In the case of
variables capturing wealth tertiles, the coefficient of the South Africa, the relationship between educational
middle wealth category became 1.62 (P ¼ 0.037). attainment and time to HIV seroconversion is likely to
be confounded by sex and age. For example, HIV
One additional grade of educational attainment reduced seroconversion risk decreases with age above certain peak
the hazard of HIV seroconversion by approximately 7% ages in women and men [48], whereas average
(Table 3, IIB). Household expenditure was not a educational attainment decreases with age in older age](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-39-2048.jpg)
![S36 AIDS 2007, Vol 21 (suppl 7)
groups. The age-specific pattern of education reflects Finally, we find that other covariates (sex, age, place of
secular changes in South African education policy, such as residence, migration status and partnership status)
the South African Schools Act of 1996 that abolished influence the hazard of HIV seroconversion as expected
racial segregation in schools [49]. For example, the matric based on previous studies [35–38,40,41]. Studies of risky
pass rate (i.e. attainment of grade 12) increased from 40% sexual behaviour in Africa have shown striking differ-
in the late 1990s to 68% in 2005 [50]. ences between women and men [51–54]. In as far as
education and wealth effects on HIV incidence are
Alternatively, education effects may differ by the stage of conveyed by sexual behaviour we expected to find, but
the HIVepidemic. Most of the published studies that have did not, that the effects of education and wealth on HIV
examined the relationship between education and HIV incidence are modified by sex. It is possible that pathways
infection were conducted in early stages of the epidemic from education and wealth to HIVacquisition that are not
[27], when educational attainment may have been sex-specific (e.g. malnutrition) are relatively more
positively associated with HIV infection, for example important in explaining our findings than sex-specific
because the more educated had more partners in any pathways, or that after controlling for sex and other
given period of time than the less educated. In contrast, as factors different pathways in women and men have similar
the epidemic matured, the more educated may have effects on HIV incidence. Given that our sample includes
adopted HIV risk-reducing behaviours more quickly fewer men than women, however, it is also possible that
than the less educated because they were more exposed to our study lacks the power to detect a sex differential in the
health promotion messages or more empowered to effects of education and wealth on HIV incidence.
negotiate protective behaviours with sexual partners [27].
Another possible limitation of our study are uncontrolled
We also show that in this overall poor community it is not selection effects because of selection into the baseline
the members of the asset-poorest households who are at sample, because of missing information on independent
highest risk of HIV acquisition but people who live in variables, or because of attrition between the first and the
households belonging to the middle category of relative second round of the HIV surveillance. Whereas selection
wealth. Recent analyses of cross-sectional surveys of HIV on observed factors that are associated with HIV
serostatus in Africa have shown that the poor do not have seroconversion will bias estimates of HIV incidence (unless
the highest HIV prevalence [11,22]. Our longitudinal different selection biases balance each other out),
study provides evidence about the causal effect of relative coefficient estimates in multiple regression will be
wealth on the risk of HIV acquisition. First, our results consistent if the observed factors that determine selection
obtain when important other determinants of acquisition are included as independent variables in the regression
of HIV that may be correlated with wealth are controlled equation [55,56]. Our regression equations thus control for
for, particularly urban residence and migration status. selection on sex, age, education, wealth, household
Second, unlike analyses of cross-sectional surveys, we can expenditure, rural versus urban/periurban place of
rule out the possibility that the positive HIV status of residence, migration status and partnership status. As these
study participants caused his or her household to fall into characteristics are among the most commonly observed
poverty because of the loss of employment or increased correlates of HIV infection in South Africa [57], it is
expenses related to disease. Third, unlike results from possible that our model adequately controls for selection
cross-sectional surveys, our findings cannot have been effects. We cannot completely rule out that selection on
caused by a wealth gradient in the survival with HIV. unobserved characteristics that are associated with the risk
of contracting HIV affect our findings. One possibility to
Our third main finding is that household expenditure adjust for selection on unobservable factors are Heckman-
does not seem to influence the hazard of HIV type selection models, which are not as well developed for
seroconversion in this population. In the ACDIS DSA, survival analysis as they are, for example, for ordinary least
government clinics distribute condoms for free and squares regression or probit regression, and whose
provide basic health services, including treatment for performance commonly depends on the existence of a
sexually transmitted diseases, for free. One plausible valid and relevant exclusion restriction, i.e. a variable that is
explanation of our result is thus that access to services that a significant predictor of selection into the sample, but not
can help prevent HIV transmission does not depend on independently associated with the time to HIV serocon-
households’ short-term ability to pay. Alternatively, it is version [58–60]. Future studies are needed to investigate
possible that financial liquidity does improve access (for the effect of selection on unobservable factors in analysis
example, because transport to government clinics is costly of the socioeconomic determinants of HIV incidence in
or because private health care providers offer some this community.
services that are effective in reducing HIV transmission
that are not available at government clinics), but that In sum, our results provide little support for the assertion
access does not translate into actual utilization of such that ‘‘reducing poverty will be at the core of a long-term,
services (for example, because individuals who could sustainable solution to reducing HIV/AIDS’’ [61].
access them do not believe that they are effective). Although poverty reduction is important for obvious](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-41-2048.jpg)
![The socioeconomic determinants of HIV incidence Barnighausen et al.
¨ S37
reasons, it may not be as effective as anticipated in reducing 14. Booysen FL, Mafereka R. The impact of HIV/AIDS on house-
the spread of HIV in rural South Africa. In contrast, hold savings in two Free State communities. In: Conference on
Reducing Poverty and Inequality: how can Africa be included?
increasing educational attainment in the general popu- Oxford: Oxford University, Centre for Study of African Econo-
lation, whatever the precise pathways of the effect, may mies; 2006.
lower HIV incidence. 15. Braitstein P, Brinkhof MW, Dabis F, Schechter M, Boulle A,
Miotti P, et al. Mortality of HIV-1-infected patients in the first
year of antiretroviral therapy: comparison between low-
income and high-income countries. Lancet 2006; 367:817–
824.
16. Wood E, Montaner JS, Chan K, Tyndall MW, Schechter MT,
Acknowledgement Bangsberg D, et al. Socioeconomic status, access to triple
therapy, and survival from HIV-disease since 1996. AIDS
The authors wish to thank the fieldworkers and 2002; 16:2065–2072.
17. Bulterys M, Chao A, Habimana P, Dushimimana A, Nawrocki P,
supervisors at the Africa Centre for Health and Saah A. Incident HIV-1 infection in a cohort of young women in
Population Studies for their excellent work in the HIV Butare, Rwanda. AIDS 1994; 8:1585–1591.
surveillance and the Demographic Information System. 18. Kapiga SH, Lyamuya EF, Lwihula GK, Hunter DJ. The incidence
of HIV infection among women using family planning methods
They would also like to thank the Africa Centre in Dar es Salaam, Tanzania. AIDS 1998; 12:75–84.
community for their participation in the surveys. 19. Mayala M, Minlangu M, Nzila N, Mama A, Jingu M, Mundele L,
et al. Prevalence and incidence of HIV-1 infection among
employees of a large textile business and their wives in
Sponsorship: The research reported in this paper was Kinshasa, 1991–1996 [in French]. Rev Epidemiol Sante Pub-
supported by the Wellcome Trust through grant lique 2001; 49:117–124.
GR065377/Z/01/B for the HIV surveillance and grant 20. Mbizvo MT, Kasule J, Mahomed K, Nathoo K. HIV-1 serocon-
GR065377/Z/01/H for the Africa Centre Demographic version incidence following pregnancy and delivery among
women seronegative at recruitment in Harare, Zimbabwe.
Information System. Some of the methods used in this Cent Afr J Med 2001; 47:115–118.
analysis were devised and tested at workshops run by 21. Senkoro KP, Boerma JT, Klokke AH, Ng’weshemi JZ, Muro AS,
the Wellcome Trust-funded ALPHA network (grant Gabone R, Borgdorff MW. HIV incidence and HIV-associated
GR075886/Z/04/Z). mortality in a cohort of factory workers and their spouses in
Tanzania, 1991 through 1996. J Acquir Immune Defic Syndr
2000; 23:194–202.
Conflicts of interest: None. 22. de Walque D. Who gets AIDS and how? The determinants of
HIV infection and sexual behaviors in Burkina Faso, Cameroon,
Ghana, Kenya, and Tanzania. Policy research working paper
References series 3844. The World Bank: Washington, DC; 2006.
23. Wawer MJ, Serwadda D, Gray RH, Sewankambo NK, Li C,
1. UNAIDS. 2006 Report on the global AIDS epidemic, annex 2: Nalugoda F, et al. Trends in HIV-1 prevalence may not reflect
HIV and AIDS estimates and data. Geneva: Joint United Nations trends in incidence in mature epidemics: data from the Rakai
Programme on HIV/AIDS; 2006. population-based cohort, Uganda. AIDS 1997; 11:1023–1030.
2. Buve A, Bishikwabo-Nsarhaza K, Mutangadura G. The spread 24. Hosegood V, Timaeus IM. Household composition and dy-
and effect of HIV-1 infection in sub-Saharan Africa. Lancet namics in KwaZulu-Natal, South Africa: mirroring social reality
2002; 359:2011–2017. in longitudinal data collection. In: van der Walle, E., Sharpe,
3. Nyindo M. Complementary factors contributing to the rapid M.E., editors. African households: censuses and surveys.
spread of HIV-I in sub-Saharan Africa: a review. East Afr Med J Champaign, IL: University of Illinois Press; 2006. pp. 58–77.
2005; 82:40–46. 25. Case A, Ardington C. Socioeconomic factors. Africa Centre
4. Wojcicki JM. Socioeconomic status as a risk factor for HIV Demographic Information System: sociodemographic and
infection in women in East, Central and Southern Africa: a health events. Africa Centre Population Studies Group. Mtuba-
systematic review. J Biosoc Sci 2005; 37:1–36. tuba: Africa Centre for Health and Population Studies; 2004.
5. Johnson L, Budlender D. HIV risk factors: a review of demo- pp. 68–78.
graphic, socioeconomic, biomedical and behavioural determi- 26. UNAIDS/WHO. Guidelines for using HIV testing technologies in
nants of HIV prevalence in South Africa. CARE monograph no. 8. surveillance: selection, evaluation, and implementation.
Cape Town: Centre for Actuarial Research, University of Cape Geneva: UNAIDS/WHO; 2001.
Town; 2002. 27. Hargreaves JR, Glynn JR. Educational attainment and HIV-1
6. Braveman PA, Cubbin C, Egerter S, Chideya S, Marchi KS, infection in developing countries: a systematic review. Trop
Metzler M, Posner S. Socioeconomic status in health research: Med Int Health 2002; 7:489–498.
one size does not fit all. JAMA 2005; 294:2879–2888. 28. Morris SS, Carletto C, Hoddinott J, Christiaensen LJ. Validity of
7. Mosley A. Does HIV or poverty cause AIDS? Biomedical and rapid estimates of household wealth and income for health
epidemiological perspectives. Theor Med Bioeth 2004; 25:399– surveys in rural Africa. J Epidemiol Commun Health 2000;
421. 54:381–387.
8. Myer L, Ehrlich RI, Susser ES. Social epidemiology in South 29. Filmer D, Pritchett LH. Estimating wealth effects without
Africa. Epidemiol Rev 2004; 26:112–123. expenditure data – or tears: an application to educational
9. Stillwaggon E. HIV/AIDS in Africa: fertile terrain. J Dev Studies enrollments in States of India. Demography 2001; 38:115–132.
2002; 38:1–22. 30. van der Berg S, Louw M. Changing patterns of South African
10. Booysen FL, Summerton J. Poverty, risky sexual behaviour, and income distribution: towards time series estimates of distribu-
vulnerability to HIV infection: evidence from South Africa. tion and poverty. S Afr J Econ 2004; 72:546–572.
J Health Popul Nutr 2002; 20:285–288. 31. Gyekye A, Akinboade O. A profile of poverty in the Limpopo
11. Shelton JD, Cassell MM, Adetunji J. Is poverty or wealth at the province of South Africa. East Afr Soc Sci Res Rev 2003; 19:89–
root of HIV? Lancet 2005; 366:1057–1058. 109.
12. Greener R. The impact of HIV/AIDS on poverty and inequality. 32. Blaauw D, Penn-Kekana L. Socio-economic inequalities and
In: Haacker, M., editor. The macroeconomics of HIV/AIDS. maternal health in South Africa. Johannesburg: Health Systems
Washington, DC: International Monetary Fund; 2004. pp. 167– Development Programme. Centre for Health Policy, University
181. of Witswatersrand; 2002.
13. Bachmann MO, Booysen FL. Health and economic impact of 33. Booysen FL. Using demographic and health surveys to measure
HIV/AIDS on South African households: a cohort study. BMC poverty – an application to South Africa. J Stud Econ Econo-
Public Health 2003; 3:14. metr 2002; 26:53–70.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-42-2048.jpg)
![S38 AIDS 2007, Vol 21 (suppl 7)
34. World Bank. Poverty in Africa – standardized welfare indicators. 47. Grambsch PM, Therneau TM. Proportional hazards tests and
2007. Available at: http://www4.worldbank.org/afr/poverty/ diagnostics based on weighted residuals. Biometrika 1994;
measuring/Indicators/definitions_en.htm. Accessed: 12 June 81:515–526.
2006. 48. Barnighausen T, Tanser F, Mbizana C, Gqwede Z, Herbst K,
¨
35. Auvert B, Ballard R, Campbell C, Carael M, Carton M, Fehler G, Newell ML. HIV incidence in a rural community in KwaZulu-
et al. HIV infection among youth in a South African mining Natal, South Africa. In: 3rd South African AIDS Conference.
town is associated with herpes simplex virus-2 seropositivity Durban, South Africa, 5–8 June, 2007 [Abstract 737].
and sexual behaviour. AIDS 2001; 15:885–898. 49. Maithufi IP. The South African Schools Act of 1996: a break
36. Connolly C, Shisana O, Colvin M, Stoker D. Epidemiology of with the past. In: Robinson, J.A., editor. The law of children and
HIV in South Africa–results of a national, community-based minors in South Africa. South Africa: Butterworth; 1997.
survey. S Afr Med J 2004; 94:776–781. 50. SouthAfrica.info. Education in South Africa. 2006. Available at:
37. Pettifor AE, Rees HV, Kleinschmidt I, Steffenson AE, MacPhail C, http://www.southafrica.info/ess_info/sa_glance/education/edu-
Hlongwa-Madikizela L, et al. Young people’s sexual health in cation.htm. Accessed: 15 January 2007.
South Africa: HIV prevalence and sexual behaviors from a 51. Babalola S. Gender differences in the factors influencing con-
nationally representative household survey. AIDS 2005; sistent condom use among young people in Tanzania. Int J
19:1525–1534. Adolesc Med Health 2006; 18:287–298.
38. Dunkle KL, Beksinska ME, Rees VH, Ballard RC, Htun Y, Wilson 52. Boer H, Mashamba MT. Gender power imbalance and differ-
ML. Risk factors for HIV infection among sex workers in ential psychosocial correlates of intended condom use among
Johannesburg, South Africa. Int J STD AIDS 2005; 16:256– male and female adolescents from Venda, South Africa. Br J
261. Health Psychol 2007; 12:51–63.
39. Lurie MN, Williams BG, Zuma K, Mkaya-Mwamburi D, Garnett 53. Dunkle KL, Jewkes RK, Nduna M, Levin J, Jama N, Khuzwayo N,
G, Sturm AW, et al. The impact of migration on HIV-1 et al. Perpetration of partner violence and HIV risk behaviour
transmission in South Africa: a study of migrant and nonmi- among young men in the rural Eastern Cape, South Africa. AIDS
grant men and their partners. Sex Transm Dis 2003; 30:149– 2006; 20:2107–2114.
156. 54. Jewkes R, Dunkle K, Nduna M, Levin J, Jama N, Khuzwayo N,
40. Shisana O, Hall EJ, Maluleke R, Chauveau J, Schwabe C. HIV/ et al. Factors associated with HIV sero-status in young rural
AIDS prevalence among South African health workers. S Afr South African women: connections between intimate partner
Med J 2004; 94:846–850. violence and HIV. Int J Epidemiol 2006; 35:1461–1468.
41. Zuma K, Gouws E, Williams B, Lurie M. Risk factors for HIV 55. Wooldridge JM. Econometric analysis of cross section and panel
infection among women in Carletonville, South Africa: migra- data. Cambridge, MA: MIT Press; 2001.
tion, demography and sexually transmitted diseases. Int J STD 56. Vella F. Estimating models with sample selection bias: a survey.
AIDS 2003; 14:814–817. J Hum Resources 1998; 33:127–169.
42. Ahmed S, Lutalo T, Wawer M, Serwadda D, Sewankambo NK, 57. Shisana O, Rehle T, Simbayi LC, Parker W, Zuma K, Bhana A,
Nalugoda F, et al. HIV incidence and sexually transmitted et al. South African National HIV prevalence, HIV incidence,
disease prevalence associated with condom use: a population behaviour and communication survey. Cape Town: HSRC Press;
study in Rakai, Uganda. AIDS 2001; 15:2171–2179. 2005.
43. Shisana O, Zungu-Dirwayi N, Toefy Y, Simbayi LC, Malik S, 58. Greene WH. Econometric analysis.. Upper Saddle River, NJ:
Zuma K. Marital status and risk of HIV infection in South Prentice Hall; 2002.
Africa. S Afr Med J 2004; 94:537–543. 59. Heckman JJ. Sample selection bias as a specification error.
44. Cox DR. Regression models and life tables. J Roy Stat Soc 1972; Econometrica 1979; 47:153–161.
34:187–220. 60. Ahn HT, Powell JL. Semiparametric estimation of censored
45. Gutierrez RG. Parametric frailty and shared frailty survival selection models with nonparametric selection mechanism.
models. Stata J 2002; 2:22–44. J Econometr 1993; 58:3–29.
46. Kutner MH, Nachtsheim CJ, Neter J, Li W. Applied linear 61. Fenton L. Preventing HIV/AIDS through poverty reduction: the
statistical models. 5th ed. Boston: McGraw-Hill; 2004. only sustainable solution? Lancet 2004; 364:1186–1187.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-43-2048.jpg)
![Explaining continued high HIV prevalence in
South Africa: socioeconomic factors,
HIV incidence and sexual behaviour change
among a rural cohort, 2001–2004
James R. Hargreavesa,b, Christopher P. Bonella, Linda A. Morisona,
Julia C. Kima,b, Godfrey Phetlab, John D.H. Portera,
Charlotte Wattsa and Paul M. Pronyka,b
Objectives: To estimate HIV incidence and explore evidence for changing sexual
behaviour over time among men and women belonging to different socioeconomic
groups in rural South Africa.
Design and methods: A cohort study conducted between 2001 and 2004; 3881
individuals aged 14–35 years enumerated in eight villages were eligible. At least three
household visits were made to contact each eligible respondent at both timepoints.
Sexual behaviour data were collected in structured, respondent-focused interviews.
HIV serostatus was assessed using an oral fluid enzyme-linked immunosorbent assay at
each timepoint.
Results: Data on sexual behaviour were available from 1967 individuals at both
timepoints. A total of 1286 HIV-negative individuals at baseline contributed to the
analysis of incidence. HIV incidence was 2.2/100 person-years among men and 4.9/
100 person-years in women, among whom it was highest in the least educated group.
Median age at first sex was lower among later birth cohorts. A higher number of
previously sexually active individuals reported having multiple partners in the past year
in 2004 than 2001. Condom use with non-spousal partners increased from 2001 to
2004. Migrant men more often reported multiple partners. Migrant and more educated
individuals of both sexes and women from wealthier households reported higher levels
of condom use.
Discussion: HIV incidence is high in rural South Africa, particularly among women of
low education. Some risky sexual behaviours (early sexual debut, having multiple
sexual partners) are becoming more common over time. Condom use is increasing.
Existing HIV prevention strategies have only been partly effective in generating
population-level behavioural change.
ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 7):S39–S48
Keywords: Education, HIV infection, migration, poverty, sexual behaviour,
South Africa
Introduction behaviour [4–6] from a number of sub-Saharan African
countries. In contrast, antenatal surveillance data and
In recent years, there have been reports of decreases in repeated national HIV prevalence surveys from South
HIV prevalence [1,2], HIV incidence [3] and sexual risk Africa suggest a continued rise in HIV prevalence despite
From the aLondon School of Hygiene and Tropical Medicine, London, UK, and the bRural AIDS and Development Action Research
Programme, Acornhoek, South Africa.
Correspondence to James R. Hargreaves, Infectious Disease Epidemiology Unit, London School of Hygiene and Tropical
Medicine, Keppel Street, London WC1E 7HT, UK.
Tel: +44 020 7927 2955; fax: +44 020 7637 4314; e-mail: james.hargreaves@lshtm.ac.uk
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S39](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-44-2048.jpg)
![S40 AIDS 2007, Vol 21 (suppl 7)
extensive efforts to reduce sexual risk behaviour [7–10]. reporting. Household wealth was assessed through a
HIV prevalence data are, however, an uncertain guide to participatory wealth-ranking technique based on com-
incidence because prevalence is affected both by HIV munity informant rankings of each household’s wealth
incidence and changing mortality patterns. Despite this, repeated three times [13].
few studies have reported direct measures of HIV
incidence in any South African population. There is an Oral fluid samples were collected using the OraSure
urgent need to understand better the pattern of new HIV collection device (UCB Group, Belgium) and analysed
infections in South Africa and whether this is associated with the Vironostika HIV Uni-Form II assay (bioMerieux,
with changes in sexual behaviour. France). HIV data from one interviewer at baseline raised
quality concerns and were excluded from the analysis
We report data from a cohort study conducted in rural (n ¼ 168). Samples testing negative at baseline were
South Africa between 2001 and 2004, as part of a cluster- included in the analysis of HIV incidence. Of these,
randomized trial of a microfinance and training inter- 34% were stored for slightly longer than recommended by
vention reported elsewhere [11]. The current paper has the manufacturer before analysis, but were included in the
three objectives: to estimate HIV incidence among a rural analysis after checking that their inclusion did not bias the
South African cohort; to explore evidence for changing measure of incidence.
sexual behaviour between 2001 and 2004 in this study
population; and to assess the evidence that HIV incidence Statistical analysis
rates and sexual behaviour patterns differed across Data were entered into an Access database (Microsoft,
socioeconomic groups identified on the basis of wealth, California, USA) with statistical analysis conducted using
education and temporary migrancy. Stata version 9 (Stata Corp., College Station, Texas,
USA). The key exposure variable was date of interview
(baseline predominantly in 2001 and follow-up pre-
Methods dominantly in 2004). The exposure period was recorded
as the duration between the first and last interview, or half
Setting of this for those who HIV seroconverted. Household
The study was conducted in Limpopo province in South wealth was assessed at baseline and identified households
Africa’s north east. Poverty remains widespread in the study as ‘very poor’, ‘poor, but a bit better off ’ or ‘doing OK’.
area [12,13] and unemployment rates exceed 40% [14]. Temporary migrancy status was assessed on the basis of
There are high levels of labour migration, with 60% of whether an individual was sleeping in the home at the
adult men and 25% of women residing away from home for time of interview at both timepoints and was coded as a
more than 6 months per year [15]. Few households have binary measure. Educational attainment was coded into
land or livestock sufficient to support livelihoods. three categories (no or primary education only; attended
but did not complete secondary education; completed
Data collection secondary or postsecondary study). This measure used
Ethical approval for the study was granted by committees data collected at follow-up rather than baseline because
at the London School of Hygiene and Tropical Medicine education had not changed for most except the very
and the University of Witswatersrand. young in whom the later data were considered more
relevant to our outcomes.
Two hundred dwellings were randomly sampled in each
of eight study villages in 2001. A household roster was Data from baseline and follow-up surveys were compiled
assembled including all individuals identified as house- to analyse age of sexual debut by a survival approach with
hold members by the household head, regardless of censoring at the current age for those not yet sexually
whether they were currently sleeping at the dwelling, in active because there were many individuals in this
order to account for high levels of temporary labour category. The number of sexual partnerships during the
migration in South Africa. Individuals of both sexes aged past 12 months (including spousal and non-spousal
14–35 years were eligible for inclusion in the cohort. partnerships) was explored via a binary variable (> 1
sexual partner; 1 or no partner). Condom use at last sex
Data were collected by trained female fieldworkers with a non-spousal partner was analysed as a binary
through face-to-face structured interviews conducted in characteristic of sexual partnerships, as opposed to
the local language (Sepedi). Witnessed verbal consent was individuals, and was recorded for the three most recent
obtained from all subjects. Attempts were made to partnerships from the past 12 months for each respondent.
maximize response and follow-up rates by instituting a
full-time field office and making repeated efforts to trace HIV incidence rates by age at baseline, sex and
migrants. Effort was also made to ensure accurate socioeconomic status were calculated among those
reporting through the use of limited recall times, a who were HIV negative at baseline. Logistic regression
respondent-oriented interview and stressing confidenti- was used to assess whether socioeconomic status variables
ality, anonymity and the importance of honesty in were associated with HIV seroconversion. In order to](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-45-2048.jpg)
![Explaining continued high HIV prevalence in South Africa Hargreaves et al. S41
assess whether there was evidence for any change in age of (73.5%) were succesfully interviewed at baseline and 1967
sexual debut over time, we used Cox regression to assess (68.8%) of those interviewed at baseline had data available
the evidence that the rate of sexual debut differed at follow-up. Inability to contact the interviewee was
between those aged 14 and 19 years at baseline (many of the primary reason for lack of interview (16% at
whose sexual debut occurred after 2001) and those aged baseline, 19% at follow-up), refusal being rare (3%, 3%)
20–35 years (whose sexual debut mostly occurred before and missing data accounting for most other exclusions. In
2001). After this, the influence of household wealth on addition, 44 individuals died during the follow-up
age of first sex was investigated. We did not investigate the period, and 371 migrated and could not be traced.
association between current educational or mobility Men were less likely to have complete follow-up, as were
status and age at first sex because these socioeconomic men and women who were older, married, sleeping away
factors were likely to have changed since the time at from the home or had more education (Table 1). The
which first sex occurred. average time between baseline and follow-up was 3.1
years. Among those succesfully followed up, valid data on
The proportion reporting more than one sexual partner HIV serostatus was collected at both timepoints on 1396
during the past 12 months was calculated at both 2001 individuals, of whom 1286 were HIV negative at baseline,
and 2004. To limit selection biases and residual these contributing to the analysis of HIV incidence.
confounding as a result of cohort ageing, we restricted
the analysis to individuals who had data available at both There were 34 seroconversions among men and 108
timepoints and had been previously sexually active, among women. HIV incidence was 2.2/100 person-years
because we expected age to be a strong determinant of [95% confidence interval (CI) 1.5–3.0] for men and 4.9/
first sex, but less strongly associated with the number of 100 person-years (95% CI 4.0–5.9) for women. HIV
partners in the past year among those who had already incidence in the age group 15–24 years at baseline was 2.0
started sexual activity. To explore the influence of time (1.3–3.0) for men and 4.7 (3.7–6.0) for women.
and socioeconomic factors, a dataset was constructed Incidence was lowest in the youngest age group among
containing a record for each individual at each timepoint both sexes and higher among women than men at all ages
with the temporary migration variable being allowed to (Table 2). Among men, there was little evidence that HIV
vary with time. A logistic regression model, specifying seroconversion was associated with any socioeconomic
individual-level clustering via population-averaged gen- factor. Among women, HIV seroconversion was signifi-
eral estimating equations was constructed for each sex cantly less common among those with higher levels of
separately, with multiple partners during the previous 12 education [adjusted odds ratio (aOR) comparing
months as the outcome variable. attended secondary school versus none/primary 0.49,
95% CI 0.28–0.85; comparing those completing
Analysis of condom use at last sex included data on all non- secondary school versus none/primary 0.25, 95% CI
spousal partnerships reported at both baseline and follow- 0.12–0.53]. There was less evidence for differing HIV
up and was thus limited to individuals reporting a non- incidence by marital status, trial arm, household wealth or
spousal partner at each time point. Logistic regression, temporary migrancy.
employing general estimating equations and specifying
individual clustering, was used to estimate the effects of Among men, the median age at first sex was 16 years for
date of interview and socioeconomic status variables. those aged 14–19 years at baseline compared with 17
years for those aged over 20 years at baseline (Fig. 1a;
All analyses were stratified by sex. Variables considered as hazard ratio 0.60, 95% CI 0.50–0.72). Among women,
potential confounders of the effect of time or socio- earlier first sex was also signficantly more often reported
economic status variables on outcome characteristics by those aged 14–19 years at baseline (median 16 years)
were: age; marital status; village-pair; trial arm; and than the older group (17 years, hazard ratio 0.77, 95% CI
(among women for sexual behaviour outcomes) ever 0.67–0.89). Household wealth was not significantly
having had a child. For the partnership-level analysis, associated with age at first sex among either sex.
models were also adjusted for a measure of frequency of
sex during the previous 12 months (more than five times, Men were more likely to report multiple partners in the
five or fewer times). When confounders varied over time past year than women at both timepoints (aOR 5.14 95%,
this was accounted for in the model. For each analysis CI 4.06–6.53; Fig. 1b and Table 3). Among men, having
interaction terms were fitted between the time period had multiple partners during the previous year tended to
variable and each of the socioeconomic status variables. be most common among those aged 20–25 years. The
number of previously sexually active men reporting
multiple partners in the past year increased between the
Results baseline and follow-up interviews (Fig. 1b; aOR 1.34,
95% CI 1.02–1.77). Ever having been married, trial arm,
Some 1482 households were succesfully enumerated, household wealth and educational attainment were not
identifying 3881 eligible 14–35 year olds. Of these, 2858 associated with having had multiple partners in the past](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-46-2048.jpg)
![Explaining continued high HIV prevalence in South Africa Hargreaves et al. S45
Table 3. Multiple partnerships among those previously sexually active reported by men and women in 2001 and 2004 in a rural South African
cohort, by socioeconomic factors.
Men reporting multiple partners in previous year Women reporting multiple partners in previous year
2001 2004 2001 2004
n/N (%) n/N (%) aOR n/N (%) n/N (%) aOR
Interview date
Baseline (2001) 129/495 (26.1) – 1 54/943 (5.7) – 1
Follow-up (2004) – 158/493 (32.1) 1.34 (1.02–1.77) – 82/943 (8.7) 2.09 (1.39–3.17)
Age at baseline (years)
14–19 44/180 (24.4) 20/62 (32.3) 1 23/278 (8.3) 19/73 (26.0) 1
20–25 52/167 (311) 84/214 (39.3) 1.40 (0.96–2.06) 17/323 (5.3) 31/390 (8.0) 0.44 (0.27–0.73)
26þ 33/148 (22.3) 54/217 (24.9) 0.78 (0.51–1.18) 14/342 (4.1) 32/480 (6.7) 0.41 (0.23–0.71)
Marital status
Never married 114/444 (25.7) 136/401 (33.9) 1 48/680 (7.1) 59/622 (9.0) 1
Ever married 15/51 (29.4) 22/92 (23.9) 1.01 (0.62–1.64) 6/263 (2.3) 23/321 (7.2) 0.71 (0.45–1.10)
Trial arm
Control 66/247 (26.7) 85/245 (34.7) 1 23/478 (4.8) 58/478 (12.1) 1
Intervention 63/248 (25.4) 73/248 (29.4) 0.80 (0.60–1.08) 31/467 (6.7) 24/465 (5.2) 0.66 (0.46–0.93)
Ever had a child
No – – – 22/282 (7.8) 20/159 (12.6) 1
Yes – – – 32/652 (4.9) 62/780 (8.0) 0.87 (0.57–1.34)
Household wealth
Very poor 38/141 (27.0) 47/142 (33.1) 1 12/276 (4.4) 22/276 (8.0) 1
Poor, but a bit better off 73/278 (26.3) 84/277 (30.3) 0.91 (0.65–1.27) 35/513 (6.8) 43/513 (8.4) 1.23 (0.81–1.87)
Doing OK 17/75 (22.7) 27/73 (37.0) 0.95 (0.58–1.54) 7/143 (4.9) 13/143 (9.1) 1.13 (0.62–2.05)
Migrancy status
Non-migrant 108/424 (25.5) 118/383 (30.8) 1 50/865 (5.8) 61/761 (8.0) 1
Migrant 21/71 (29.6) 31/71 (43.7) 1.51 (1.03–2.20) 4/78 (5.1) 6/73 (8.2) 1.05 (0.53–2.07)
Educational attainment at follow-up
None/primary only 22/73 (30.1) 18/73 (24.7) 1 11/138 (8.0) 9/137 (6.6) 1
Attended secondary 72/286 (25.2) 93/285 (32.6) 1.02 (0.65–1.60) 35/602 (5.8) 61/602 (10.1) 0.97 (0.58–1.64)
Completed secondary 35/136 (25.7) 47/135 (34.8) 1.12 (0.68–1.84) 8/203 (3.9) 12/204 (5.9) 0.69 (0.36–1.30)
aOR, Adjusted odds ratio comparing socioeconomic groupings across both timepoints, adjusted for interview date, age, marital status, village pair
and trial arm, and, for women only, ever had a child.
more commonly reported in 2004 than in 2001. Condom represented in the final sample, and these individuals were
use at last sex with non-spousal partners was, however, more likely to be migrants and well educated. It is possible
more commonly reported in 2004. that their sexual behaviour or risk of HIV infection
differed from those included in the study; if so, our
Regarding socioeconomic patterns among these out- estimates may have been biased. As the cohort was ageing,
comes, HIV incidence among men was not associated with there may have been residual confounding by age for
socioeconomic factors, but among women infections outcomes in which age was an important determinant.
occurred fastest among the least educated. Sexually active Many authors have also pointed to the difficulties
migrant men more often reported multiple sexual partners, inherent in capturing accurate sexual behaviour infor-
but migrant and more educated men also reported more mation in one-off interviews, and it is likely that some
common condom use with non-spousal partners. Among misreporting occurred [16,17]. If such misreporting
sexually active women, having had multiple sexual partners varied between individuals from different socioeconomic
in the past year was not associated with socioeconomic groups or at different timepoints, this may also have
factors, but women who were migrants, from wealthier produced some bias, although this is difficult to assess.
households and with higher levels of education were more Misreporting of age of sexual debut might have differed
likely to report condom use at last sex with a non-spousal with respect to age at baseline because of the likely greater
partner. There was little evidence that the strength of time intervals involved in recall for older participants.
association between socioeconomic variables and sexual Therefore, the finding of lower age at first sex among later
behaviours had changed over time. age cohorts should be treated with some caution.
The strengths of the study included explicit attempts to Another limitation of the study was relatively low
maximize follow-up rates and ensure accurate reporting. statistical power, particularly with respect to HIV
Furthermore, important potential confounders such as incidence analyses and interaction tests. Finally, although
age, childbirth and partnership characterstics were our educational exposure may have been relatively simple
adjusted for in the analysis. Nevertheless, the study had to record, our assessment of migrancy did not identify
limitations. A proportion of eligible individuals were not migrations at times other than when surveys were](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-50-2048.jpg)
![S46 AIDS 2007, Vol 21 (suppl 7)
Table 4. Condom use at last sex within a non-spousal sexual partner reported by men and women in 2001 and 2004 in a rural South African
cohort, by socioeconomic factors.
Men reporting condom use at last sex with non-spousal Women reporting condom use at last sex with
partner non-spousal partner
2001 2004 2001 2004
n/N (%) n/N (%) aOR n/N (%) n/N (%) aOR
Interview date
Baseline (2001) 327/1002 (32.6) – 1 253/915 (27.7) – 1
Follow-up (2004) – 272/684 (39.8%) 1.43 (1.07–1.92) – 237/723 (32.8) 1.46 (1.14–1.87)
Age (years)
14–19 101/321 (31.5) 26/66 (39.4) 1 117/340 (34.4) 36/82 (43.9) 1
20–25 139/411 (33.8) 114/261 (43.7) 1.26 (0.86–1.85) 87/361 (24.1) 87/298 (29.2) 0.60 (0.43–0.84)
26þ 87/270 (32.2) 132/357 (37.0) 0.91(0.59–1.40) 40/214 (22.9) 114/343 (33.2) 0.56 (0.37–0.86)
Marital status
Never married 312/964 (32.4) 242/629 (38.5) 1 240/880 (27.3) 214/658 (32.5) 1
Ever married 15/38 (39.5) 30/55 (54.6) 1.50 (0.73–3.09) 13/35 (37.1) 23/65 (35.4) 1.81 (1.04–3.14)
Trial arm
Control 173/473 (36/6) 142/347 (40.9) 1 110/422 (26.1) 107/373 (28.7) 1
Intervention 154/529 (29.1) 130/337 (38.6) 0.64 (0.47–0.88) 143/443 (29.0) 130/371 (37.1) 1.25 (0.94–1.66)
Ever had a child
No – – – 122/365 (33.4) 79/18 (42.0) 1
Yes – – – 127/543 (23.4) 158/532 (29.7) 0.72 (0.53–0.99)
Household wealth
Very poor 100/291 (34.4) 79/203 (38.9) 1 67/274 (24.5) 65/223 (29.2) 1
Poor, but a bit better off 177/588 (30.1) 143/380 (37.6) 0.87 (0.61–1.24) 137/508 (27.4) 131/395 (33.2) 1.26 (0.91–1.75)
Doing OK 49/119 (41.2) 50/100 (50.0) 1.20 (0.75–1.92) 45/125 (36.0) 37/94(39.4) 2.03 (1.29–3.20)
Migrancy status
Non-migrant 251/807 (31.1) 202/531 (38.0) 1 215/794 (27.1) 178/583 (30.5) 1
Migrant 76/195 (39.0) 58/119 (48.7) 1.46 (0.98–2.19) 38/120 (31.7) 32/84 (38.1) 1.48 (0.98–2.23)
Educational attainment at follow-up
None/primary only 39/143 (23.9) 24/76 (27.9) 1 31/134 (23.1) 26/90 (28.9) 1
Attended secondary 178/563 (31.6) 160/418 (38.3) 2.06 (1.73–4.90) 150/583 (25.7) 152/477 (31.9) 1.31 (0.83–2.05)
Completed secondary 110/276 (39.9) 88/180 (48.9) 2.91 (1.73–4.90) 72/197 (36.6) 57/156 (37.8) 2.25 (1.34–3.78)
Frequency of sexual intercourse in past year
More than 5 time 194/639 (30.4) 179/522 (34.3) 1 149/632 (23.6) 172/554 (31.1) 1
Five or fewer times 133/363 (36.6) 93/162 (57.4) 2.20 (1.63–2.97) 104/283 (36.8) 65/169 (38.5) 1.45 (1.11–1.89)
aOR, Adjusted odds ratio comparing socioeconomic groupings across both timepoints, adjusted for interview date, age, marital status, village pair,
trial arm, frequency of sexual intercourse and, for women only, ever had a child.
conducted. This is likely to have resulted in an These figures shed light on why HIV prevalence is not,
underestimation of associations involving migrancy. uniquely among sub-Saharan African countries, decreasing
Furthermore, the assessment of household wealth in in South Africa. Our estimates of incidence do not suggest
developing countries is complex [18]. Our participatory a declining epidemic, being higher, for example, than
approach had high internal consistency [13], but a low annual HIV incidence measures among adult men and
level of correlation with an indicator based on multiple women in Uganda in the mid-1990s (1.72% per annum,
assets (J. Hargreaves, L. Morison, J. Gear, J.D.H. Porter, 95% CI 1.38–2.16 and 1.69% per annum, 95% CI 1.38–
M.B. Makhubele, J.C. Kim, et al., in preparation). 2.08, respectively) [19,20]. This is particularly worrying
given that previous studies have suggested wide inter-
This study provides direct measures of annual HIV provincial variation in adult HIV incidence within South
incidence from a South African population, among men Africa (0.5–4.2%), with Limpopo, the province under
(2.2%, 95% CI 1.5–3.0) and women (4.9%, 95% CI 4.0– study here, lying only at the midpoint of this range [9].
5.9) aged 14–35 years at baseline. National estimates from
cross-sectional research employing a detuned enzyme- Furthermore, our research suggests that although
linked immunosorbent assay that detects infections in the condom use has increased over time, young people, if
past 180 days have previously estimated HIV incidence anything, may be initiating sex earlier and the proportion
among 15–24 year olds at 0.8% per annum for men reporting multiple partners has, if anything, increased.
(compared with 2.0% for this age group in this study) and These data confirm findings from recent cross-sectional
6.5% for women (compared with 6.0%), with the overall studies in South Africa [9,10,21], and stand in contrast
estimate for Limpopo province among 15–49 year olds at to the experience of Uganda [20], Kenya [22] and
2.4% per annum [9]. Although not always directly Zimbabwe [23,24], where reductions in HIV prevalence
comparable, our study confirms the high incidence of have been accompanied by delays in the onset of first sex
HIV infection with data from a cohort study. and reductions in partner numbers.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-51-2048.jpg)
![Explaining continued high HIV prevalence in South Africa Hargreaves et al. S47
Our results also draw attention to the socioeconomic groups. Individual-focused interventions appear on their
patterning of HIV risk. Our finding of higher HIV own to have been insufficient to bring about population-
incidence among the least educated women was not wide change and address barriers to risk-reduction among
unexpected. Research suggests that up to the mid-1990s the most disadantaged groups. There is a strong case for
prevalent HIV infection was often more common among the wider testing and implementation of structural
individuals who were more mobile [25–27], had greater interventions that address the ‘upstream’ determinants
education [2,28–30], or were from more wealthy of HIV infection [36].
households [31,32]. More recent studies among young
people from Uganda and Zambia [2,33,34] have,
however, suggested that whereas HIV prevalence has Acknowledgements
fallen over time among the most educated, this is not so
among the least educated. More surprisingly, we found no This study was a partnership between academic
association between our measure of mobility and the risk institutions in South Africa (School of Public Health,
of new HIV infection, although power to detect any University of the Witwatersrand) and the UK (London
association was low for men (because of the relatively School of Hygiene and Tropical Medicine). The authors
small number of seroconversions) and women (because of would like to thank the Contract Laboratory Services at
low migration rates). the Johannesburg Hospital, particularly Dr Wendy
Stevens, Grant Napier, Anusha Makuraj and Dr Gwynn
With respect to sexual behaviour, migrant men reported Stevens, for assisting in the processing of laboratory
greater numbers of sexual partners but also a greater use of specimens, and the support of Jackie Hills at UCB and
condoms. Among women, lower levels of condom use Karin Botma at Omnimed for donating the collection
were found among the poorest, those with of the least device and enzyme-linked immunosorbent assays.
education and non-migrants. Migrants may be less subject
to restrictive social norms and have access to larger sexual Sponsorship: The study received financial support
networks. Male migrants may also be more likely to have from AngloAmerican Chairman’s Fund Educational
Trust, AngloPlatinum, the Department for International
greater personal income than non-migrants. Migrants Development (UK), the Ford Foundation, the Henry J.
might also come into greater contact with condoms and Kaiser Family Foundation, HIVOS, the South African
HIV-prevention materials as a result of their greater Department of Health and Welfare, and the Swedish
mobility, especially to cities where such resources are International Development Agency. J.R.H. is sup-
likely to be more commonly available. Underlying traits ported by an ESRC/MRC interdisciplinary fellowship.
such as self–confidence might also make individuals
simultaneously more likely to migrate, more attractive to Conflicts of interest: None.
sexual partners, and more likely to become ‘early
adopters’ of condoms. Whereas such issues require References
further study, our data suggest that migrants should not
be assumed to be engaged in high-risk behaviours 1. Fylkesnes K, Musonda RM, Sichone M, Ndhlovu Z, Tembo F,
Monze M. Declining HIV prevalence and risk behaviours in
although this does not preclude their being a key group in Zambia:evidence from surveillance and population based sur-
the HIV epidemic dynamics [35]. The data confirm, veys. AIDS 2001; 15:907–916.
however, that women experiencing socioeconomic 2. Kilian AH, Gregson S, Ndyanabangi B, Walusaga K, Kipp W,
Sahlmuller G, et al. Reductions in risk behaviour provide the
deprivation are among the most vulnerable to HIV in most consistent explanation for declining HIV-1 prevalence in
this rural South African setting. Uganda. AIDS 1999; 13:391–398.
3. Moulaiteye GM, Whitworth JA, Ruberantwari A, Nakiyingi JS,
Ojwiya A, Kamali A. Declining HIV-1 incidence and associated
Effective HIV prevention strategies remain an urgent prevalence over 10 years in a rural population in south-west
priority in South Africa. Strategies to date may have been Uganda: a cohort study. Lancet 2002; 360:41–46.
partly effective in reducing risk among educated and 4. Gregson S, Zhuwau T, Anderson RM, Chandiwana SK. Is there
evidence for behavior change in response to AIDS in rural
mobile members of society. It is possible that changes will Zimbabwe? Soc Sci Med 1998; 46:321–330.
emerge in all groups over time as safer-sex behaviour 5. Kamali A, Carpenter LM, Whitworth JA, Pool R, Ruberantwari A,
diffuses, perhaps leading to reductions in HIV prevalence, Ojwiya A. Seven-year trends in HIV-1 infection rates, and
changes in sexual behaviour, among adults in rural Uganda.
as witnessed in other sub-Saharan African countries. This AIDS 2000; 14:427–434.
may, however, be some way off unless there is a 6. Mekonnen Y, Sanders E, Aklilu M, Tsegaye A, Rinke De Wit TF,
et al. Evidence of changes in sexual behaviours among male
reassessment of HIV prevention strategies. Just as the factory workers in Ethiopia. AIDS 2003; 17:223–231.
HIV epidemic in the United States has increasingly 7. UNAIDS. Report on the global HIV/AIDS epidemic. Geneva:
become a disease of women and of racial and ethnic UNAIDS; 2006.
8. Department of Health. National HIV and syphillis antenatal sero-
minorities [7], so the epidemic in Africa may also prevalence survey in South Africa 2003. Pretoria: Department of
concentrate among the socially vulnerable. Our findings Health South Africa; 2003.
suggested that the least educated women in this rural 9. Shisana O, Rehle T, Simbayi L, Parker W, Zuma K, Bhana A,
et al. South African national HIV prevalence, HIV incidence,
setting reported more partners, less use of condoms, and behaviour and communication survey 2005. Cape Town: HSRC
experienced higher levels of new infection than other Press; 2005.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-52-2048.jpg)
![Household and community income, economic shocks
and risky sexual behavior of young adults: evidence
from the Cape Area Panel Study 2002 and 2005
Taryn Dinkelmana, David Lama and Murray Leibbrandtb
Objective: To describe recent trends in adolescent sexual behavior in Cape Town,
South Africa, and to determine whether household and community poverty and
negative economic shocks predict risky sexual behavior.
Data: Matched survey data on 2993 African and coloured youth from the Cape Area
Panel Study 2002 and 2005.
Main outcome measures: Sexual debut, multiple sexual partners in past year, condom
use at last sex, measured in 2002 and 2005.
Methods: We tested for changes over time in reported sexual behavior and estimate
multivariate probit models to measure the association between 2002 individual,
household and community characteristics and 2005 sexual behavior.
Results: There was a statistically significant increase in condom use and a decrease in
the incidence of multiple sexual partners between 2002 and 2005 for women aged
17–22 years. Women in households with 10% higher income were 0.53% less likely to
debut sexually by 2005; men in communities with a 10% higher poverty rate were 5%
less likely to report condom use at last sex. Negative economic shocks are associated
with a 0.04% increase in the probability of multiple partnerships for women. Education
is positively correlated with sexual debut for women and with multiple partnerships for
both sexes.
Conclusion: Trends in sexual behavior between 2002 and 2005 indicate significant
shifts towards safer practices. There is little evidence of a relationship between negative
economic shocks, household and community poverty, and risky behavior. We hypoth-
esize that the unexpected positive relationship between education and sexual debut
may be driven by peer effects in schools with substantial age mixing.
ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 7):S49–S56
Keywords: HIV, adolescence, economic resources, sexual debut, condom use,
multiple partnerships, South Africa
Introduction areas [2]. The mechanisms by which conditions of
poverty may influence sexual risk-taking behavior and
A recent United Nations publication states that ‘poverty thus the probability of contracting HIVand other sexually
increases vulnerability to HIV/AIDS’ [1], although in a transmitted diseases are, however, complex and currently
complex fashion; the HIV burden is concentrated in the not well understood.
poorest regions of the world but not always among the
poorest populations in these areas. HIV prevalence rates Researchers in public health and economics have
are indeed highest among South African youth living in hypothesized and less frequently tried to measure the
poor urban informal settlements compared with other channels from individual and community poverty to
From the aDepartment of Economics and Population Studies Center, University of Michigan, Ann Arbor, Michigan, USA, and the
b
School of Economics, University of Cape Town, Cape Town, South Africa.
Correspondence to Taryn Dinkelman, Population Studies Center, 426 Thompson Street, University of Michigan, Ann Arbor,
MI 48106, USA.
E-mail: tdinkelm@umich.edu
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S49](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-54-2048.jpg)
![S50 AIDS 2007, Vol 21 (suppl 7)
higher rates of sexual risk-taking behavior. Fenton [3], Young adults are a particularly interesting demographic
Gersovitz [4] and Sunmola [5] argued that inadequate group as they represent healthy cohorts whose future
information (often concentrated among the poor) acts as a behavior will influence the course of the HIV epidemic.
barrier to adopting safer behaviors; Cohen et al. [6] Various researchers have shown that in high seropreva-
argued that access to resources for safer sex may be costly; lence regions, a large proportion of new infections occur
MacPhail and Campbell [7] and LeClerc-Madlala [8] during adolescence [8]. Survey data that match detailed
argued that poverty directly induces sex work and individual sexual behavior measures to good measures of
informal transactional sex relationships particularly for household and community level resources are rare; panel
women, whereas Johnston and Way [9] provided data that enable us to see the evolution of sexual behaviors
descriptive evidence of these correlations and Luke for young adults are even more unusual. We used new
[10] and Luke [11] used survey data to quantify this panel data on adolescents (aged 14–22 years) in urban
relationship. Dunkel et al. [12] used data from young, rural Cape Town, South Africa, to measure the extent to which
South African men to show a strong positive correlation resources and shocks to household resources early on in
between higher socioeconomic status and the probability their lives could predict variation in sexual behaviors. We
of reporting transactional sex with casual partners. At a considered the following outcome measures that corre-
community level, Zulu et al. [13] compared non-poor spond to the A–B–C of HIV prevention campaigns:
non-slum residents with poor slum residents in Nairobi sexual debut, annual number of sexual partners and
and found significantly higher probabilities of reporting condom use at last sex.
early sexual debut, more sexual partners and a lack of
condom use among the poor slum dwellers. Economic
inequality may also operate to increase risk-taking within
communities: LeClerc-Madlala [8] posited that the Data and methods
growth of a black middle class with money has increased
the emphasis on transactional sex in some South African Data
communities. The Cape Area Panel Study (CAPS) is a representative
longitudinal study of 4752 adolescents aged 14–22 years
In many of those studies, it is difficult to isolate whether (in 2002) living in Cape Town, South Africa. The full
economic resources matter directly for behavior or sample was first interviewed in 2002 and again in 2005.
whether unobservable characteristics correlated with Most data are collected directly from the young adults.
poverty are driving factors. In addition, as the experience We use data from the household module, basic
of poverty is likely to have persistent effects on behavior demographic data and detailed information about sexual
over time, it is hard to distinguish whether current or relationships captured in both waves. We do not use data
long-term resource deprivation matters for behavior. To on what young people know about HIV and AIDS.
measure the direct effect of economic resources on Anderson and Beutel [14] reported that levels of HIV and
behavior convincingly, we would want to assign these AIDS knowledge were very high in the 2002 CAPS data.
resources randomly to households and observe the impact These panel data allow us to look at whether sexual
on behaviors. Approaching this research design with behavior is changing over time as well as how current
observational data is challenging. behaviors are related to a range of household level
variables measured earlier in the young adult’s life.
In this paper, we investigated whether household and
community incomes and negative economic shocks In order to generate an approximately equal sample of
predict risky behaviors of young adults. Focusing on African and coloured individuals, African youth were
young adults who are for the most part not yet oversampled. See Lam et al. [15] for details of sampling
working and who are just transitioning into sex methodology, initial non-response and attrition. Com-
allowed us to isolate the relationship between household pleted interviews for 2151 Africans, 1980 coloureds and
level and community resources and behavior. In the 621 whites and other races were captured in 2002. Once
absence of the random assignment of income to weights adjusting for survey design and wave 1 non-
households or communities, we used economic shocks response were applied, Africans represented 15% of this
to capture one source of unexpected variation in wave 1 sample, coloureds represented 59% and the
household resources. Although this research design does remaining races constituted 26% of the sample. In the
not identify the causal effect of economic deprivation 2005 wave 3324 of the initial 4752 sample were re-
on behavior, we advanced some way towards an interviewed, of which 2993 were African and coloured
understanding of the relationship between economic individuals (27.8% attrition rate).
resources and the risky sexual behavior of young adults.
Surprisingly, we found little evidence that differences in As initial non-response and attrition between waves were
household or community income or differences in very high for the small sample of white youth, we
economic shocks are correlated with more risk-taking excluded them from our analysis. Fifty-three per cent of
behavior. white individuals were successfully followed in 2005,](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-55-2048.jpg)
![Income and risky sexual behavior Dinkelman et al. S51
whereas attrition among coloured and African subsamples entire sample, we included a full set of age dummies in the
was substantially lower (21 and 36%, respectively). All of probit models to take out mean differences across age
our reported results are weighted with sampling weights groups in test performance. Therefore, if 22-year-old
correcting for sample design and first wave non-response. youth scored consistently above average on the test simply
Weighting for attrition between 2002 and 2005 does not because they were older and had more schooling, this
change any results substantively (results not reported). effect was absorbed in the 22-year-old dummy.
Our final matched subsample consists of 1410 African
youth and 1583 coloured youth. Methods
We compared the proportion of each sex, race and age
Variables group reporting each type of behavior in each wave for
For each individual in our analysis subsample, we used ages 17–22 years. The change in these proportions
sexual behavior information provided by the respondent between 2002 and 2005 gave us some insight into overall
in 2002 and 2005. To examine changes in average trends in behaviors. To measure the association between
behavior over time, we investigated three reported individual-level demographic data, previous income
behaviors for the group aged 17–22 years: whether the shocks, household and community resources and current
young adult has ever had sex, whether the young adult sexual behavior, we estimated probit models for each of
used a condom at last sex and whether the young adult the three binary outcome variables separately for women
had more than one sexual partner in the 12 months before and men. The results are reported as the marginal change
each survey. in the probability of a particular behavior associated with a
unit increase in each explanatory variable.
We are cautious about the reliability of reported sexual
behavior data. Misreporting is more likely when In the probit models we restricted the sample to ages
questions are more sensitive [16–19]. CAPS questions 14–18 years in 2002 for the sexual debut and condom
and survey protocols were carefully constructed to try to use outcomes, but included all ages 14–22 years in 2002
minimize the biases in these sensitive questions. In both when modeling multiple partners. There are two reasons
years, respondents were questioned without the presence for this. First, a large proportion of those aged 19 years
of any other family members as far as possible. For the and older had already sexually debuted so there was little
2005 survey, respondents could choose to fill out their variation contributed by those individuals. Second,
responses directly regarding each of their 10 most recent although multiple partnerships reflected relatively unsafe
partnerships instead of having the interviewer fill in the behavior at all ages, not using a condom at last sex was not
information. Fourteen per cent of applicants chose to unambiguously risky, especially in cases in which older
self-report. Comparing those who did with those who individuals were married or in longer-term monogamous
did not respond themselves, there was no systematic relationships. The results are presented as marginal effects
difference in the number of sex partners reported in 2005. and robust standard errors are clustered at the household
level because up to three young adults were interviewed
Restricting to the same set of ages (17–22 years) in 2002 per household.
and 2005 allowed us to compare average behavior for this
group over time. To examine how 2002 individual,
household and community-level variables were corre-
lated with behavior, we investigated these three sexual Results
behavior variables measured in 2005. The variables used
to predict individual behaviors within the probit model Summary statistics
included: age in 2002; sex; education; race; literacy and In Table 1, we present summary statistics separately by
numeracy test scores; per capita household income in race to highlight the vast differences in living environ-
2002; the presence of parents at home in 2002; and the ments of African and coloured youth. Except for age, all
proportion of households in the community below the of these differences are statistically significant across race
poverty line in the 2001 census. We also used information groups. Both groups were disadvantaged under apartheid,
on negative economic shocks experienced at the house- but coloured individuals were generally able to access
hold level between 2002 and 2005. A negative shock is better educational and work opportunities in Cape Town
defined as having occurred if the household experienced than Africans. Mean schooling was approximately ninth
a death, job loss, loss of a grant or loss of support from grade, although Africans had on average half a year less
outside the household, and if the household respondent schooling than coloureds. Africans also exhibited poorer
reported that the shock had a moderate to severe financial performance on the literacy and numeracy test. Coloured
impact on the household. youth were significantly more likely to live with their
biological mothers (82% compared with 64% for
The same literacy and numeracy test was administered to Africans) and fathers (54% compared with 35% for
each young adult in 2002 regardless of age or education Africans). Coloured households had a higher mean log
level. Although the test is not age-appropriate for the per capita income compared with African households.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-56-2048.jpg)
![Income and risky sexual behavior Dinkelman et al. S55
A third point relates to the set of economic variables. Per Taking our first two findings together, it appears that at
capita household income has a small and statistically least in Cape Town there are significant increases in
significant negative correlation with the probability of condom use and decreases in the number of sexual
female sexual debut. A 10% increase in 2002 income was partners. Changes that cannot be explained by individual
associated with a 0.6% decline in the probability of sexual behavioral change appear to be taking place.
debut between 2002 and 2005. The estimated marginal
effect of income on sexual debut was also negative for The third main finding relates to the role of education in
young men but not statistically significant at conventional predicting sexual risk behaviors. We did not see a
levels. Young men were less likely to report condom use at protective impact of grade attainment per se, although we
last sex if they lived in poorer communities: for a 10% did find that test scores were positively correlated with
increase in community poverty rate, this was a 5% safer sexual behaviors. Surprisingly, we found a significant
reduction in the probability of condom use at last sex. positive association of schooling with sexual debut for
Young women were 0.04% more likely to report multiple women and with multiple sexual partners for both men
partners if they lived in a household experiencing an and women, controlling for age, household income, and
economic shock. We tested the joint significance of all of other variables. One interpretation of our results is that
the economic variables (household per capita income, the test score variable captures some of the differences in
household shock and community poverty) and could not knowledge or ability that education usually measures,
reject the possibility that the coefficients were jointly zero therefore, youth with higher numeracy and literacy skills
in each regression. are less likely to report risky behaviors. We speculate that
the unexpected positive association between schooling
and risky behaviors may be a result of the impact of peers
within the school system. There is a great deal of grade
Discussion repetition in South Africa [20], with a wide mix of ages in
any given grade. A 17 year old in grade 11 interacts with a
Three main findings emerge from our analysis of the much more sexually active group of peers than a 17 year
panel data of 2993 Cape Town youth. First, for young old in grade 8. Further research will be required to
people aged 17–22 years, we documented large and understand why years of education may not have the
statistically significant increases in the probability of sexual protective effect that is usually hypothesized.
debut for women of both races, increased condom use at
last sex for African women, African men, and coloured
men, as well as significant reductions in the reporting of Acknowledgements
multiple sexual partnerships. Changes in household or
community-level economic resources are unlikely to This paper was written for the workshop ‘A Symposium
explain these behavioral changes, both because we for investigating the empirical evidence for under-
estimate relatively small effects of income on sexual standing vulnerability and the associations between
behavior and because there are only small improvements poverty, HIV infection and AIDS impact’. The authors
in economic conditions over this period. It is also unlikely would like to thank USAID and the Health Economics
that these differences arise from a change in social and HIV/AIDS Research Division (HEARD) for
desirability pressure towards answering sensitive questions financial support during the preparation of the paper
in particular ways; these young adults are reporting and attendance at the workshop. The data used in this
increases in risky behavior (sexual debut) at the same time paper are publicly accessible at www.caps.uct.ac.za.
as increases in protective behaviors (more condom use,
fewer multiple partnerships). Sponsorship: This work was supported by the US
National Institute of Child Health and Human Devel-
Our main interest in this paper related to whether opment (R01HD39788 and R01HD045581), the
household or community poverty variables could predict Fogarty International Center of the US National
risky behavior of young adults, and, in particular, whether Institutes of Health (D43TW000657), and the Andrew
sexual behavior is affected by unexpected income shocks. W. Mellon Foundation.
After controlling for detailed individual and family Conflicts of interest: None.
background variables, however, we found that little of
the variation in sexual behavior in 2005 was predicted by
economic variables. When household income and
References
economic shocks are significant, their marginal effects
are relatively small compared with other variables. Our 1. United Nations. Population, development and HIV/AIDS with
second finding is thus that for the sample of young adults particular emphasis on poverty: the concise report. 2005. New
in urban Cape Town, there is little evidence that York: United Nations; 2005.
2. Human Sciences Research Council. South African national HIV
community or household-level income or income shocks prevalence, HIV incidence, behavior and communication sur-
are the main factors influencing risky behavior. vey, 2005. Cape Town: HSRC Press; 2005.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-60-2048.jpg)
![HIV incidence and poverty in Manicaland, Zimbabwe:
is HIV becoming a disease of the poor?
Ben Lopmana, James Lewisb, Constance Nyamukapaa,c,
Phyllis Mushatic, Steven Chandiwanac,d,ä and Simon Gregsona,c
Background: In Zimbabwe, socioeconomic development has a complicated and
changeable relationship with HIV infection. Longitudinal data are needed to disen-
tangle the cyclical effects of poverty and HIV as well as to separate historical patterns
from contemporary trends of infection.
Methods: We analysed a large population-based cohort in eastern Zimbabwe. The
wealth index was measured at baseline on the basis of household asset ownership. The
associations of the wealth index with HIV incidence and mortality, sexual risk
behaviour, and sexual mixing patterns were analysed.
Results: The largest decreases in HIV prevalence were in the top third of the wealth index
distribution (tercile) in both men at 25% and women at 21%. In men, HIV incidence was
significantly lower in the top wealth index tercile (15.4 per 1000 person-years) compared
with the lowest tercile (27.4 per 1000 person-years), especially among young men.
Mortality rates were significantly lower in both men and women of higher wealth index.
Men of higher wealth index reported more sexual partners, but were also more likely to use
condoms. Better-off women reported fewer partners and were less likely to engage in
transactional sex. Partnership data suggests increasing like-with-like mixing in higher
wealth groups resulting in the reduced probability of serodiscordant couples.
Conclusion: HIV incidence and mortality, and perhaps sexual risk, are lower in higher
socioeconomic groups. Reduced vulnerability to infection, led by the relatively well off,
is a positive trend, but in the absence of analogous developments in vulnerable groups,
HIV threatens to become a disease of the poor.
ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 7):S57–S66
Keywords: Africa, Zimbabwe, AIDS, HIV, poverty, socio-economic
development
Introduction decline in HIV prevalence in Zimbabwe [3]. Mortality
rates are high [4], as a result of high HIV incidence in the
Similar to other countries in southern Africa, the HIV past, but this decrease cannot be explained by mortality
epidemic in Zimbabwe has a precarious relationship with alone [5]. Sexual risk behaviour is also changing; condom
socioeconomic development [1]. Zimbabwe has one of distribution has increased, young people are delaying
the more developed infrastructures in sub-Saharan Africa, their sexual debut, and there has been a reduction in the
with widespread access to education and the highest adult numbers of casual partnerships [3,6].
literacy in the region [2]. Zimbabwe is also experiencing
one of the largest national epidemics. HIV prevalence in Some investigators have suggested that as the HIV
the adult population was 20.1% in 2005, down from epidemic progresses, risk would shift from the wealthier
22.1% in 2003. There are two contributory factors to the (who, as a result of their relative wealth, are part of a larger
From the aDepartment of Infectious Disease Epidemiology, Imperial College, London, UK, the bLondon School of Hygiene and
Tropical Medicine, London, UK, the cBiomedical Research and Training Institute, Harare, Zimbabwe, and the dFaculty of Health
Sciences, University of the Witwatersrand, Johannesburg, South Africa.
Correspondence to Ben Lopman, Department of Infectious Disease Epidemiology, Imperial College London, St Mary’s Campus,
Norfolk Place, London W2 1PG, UK.
Tel: +44 020 7594 3290; fax: +44 020 7594 3282; e-mail: b.lopman@imperial.ac.uk
ä
Deceased.
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S57](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-62-2048.jpg)
![S58 AIDS 2007, Vol 21 (suppl 7)
sexual network) [7] to the poorer (who, because of their favourably with other community cohorts in rural
lower educational attainment and social position, are less African settings [6]. Enumerators were notified of deaths
empowered to change their sexual behaviour) [8,9]. It is by surviving household members or community infor-
suspected that the HIV epidemic in Zimbabwe initially mants if the household dissolved completely.
affected more mobile and more educated men as a result
of their ability to attract sexual partners, but as early as At each round, after written informed consent was given,
1998/2000 risk was similar, or perhaps slightly lower for information on demographic, socioeconomic and sexual
those with secondary education [10]. If this trend is behaviour data were collected through an interviewer-led
realized, the HIV epidemic threatens to become an questionnaire [14]. Dried blood spots were collected for
endemic disease of poverty in Zimbabwe. HIV serological testing for the purposes of research only.
Testing was performed using a highly sensitive and
The changing relationship between socioeconomics and specific antibody dipstick assay (> 99% for both) [15].
HIV must be seen in the context of sweeping
macroeconomic changes in Zimbabwe. The Zimbab- Socioeconomic status
wean economy has been in severe decline, with negative Individual wealth was measured on the basis of the asset
growth since 1997 [11]. Over the period 1997–2005, ownership of the household of residence. Data were
gross domestic product declined by more than 30%. In collected on household ownership of ‘fixed’ and ‘sellable’
2003, annual inflation was approximately 250%, and this assets. Fixed assets include water supply, toilet facilities,
has since accelerated to over 1000% per annum [11]. The electricity supply, housing structure and floor type.
economic factors that partly underlie partnership Sellable assets included ownership of radio, television,
formation [12], including behaviours ranging from sex bicycle, motorbike and automobile. Chi-squared tests
work to marriage, are likely to be highly unstable, making demonstrated significant differences of all assets (except
understanding the link between poverty and HIV automobile, which was owned by only 1.2% of house-
extremely timely yet difficult to study. holds) between towns, estates and rural areas.
A simple summed score asset ownership was created (see
Justification of ‘summed score’ as measure of wealth
Methods below). The binary and ordinal measures were each
transformed to lie between 0 and 1. For example, bike
Study population ownership conferred a score of 0 or 1, and type of floor
The Manicaland HIV/STD Prevention Project is an conferred a value of 0 for natural floor (earth/sand/dung),
ongoing population-based open cohort study. Full details 0.5 for rudimentary floor (e.g. planks/bamboo), or 1 for
of the study can be found elsewhere [6,13]. In short, the finished floor (wood/cement). The 10 variables were
study population were resident in small towns (two) added and expressed as a percentage.
forestry, tea and coffee estates (four) and rural areas (six,
including four subsistence farming and two roadside In order to augment the study power, a wealth index was
trading centres) in the province of Manicaland in eastern created by splitting the summed score into three equal
Zimbabwe. All local residents were enumerated in an groups (terciles) from the whole population. Preliminary
initial household census (conducted between July 1998 analyses demonstrated that the distribution of the wealth
and February 2000; referred to here as baseline), which index differed between towns and other areas; therefore,
was repeated 3 years later in each site (referred to here as analyses were conducted separately for towns, estate and
follow-up). Men aged 17–54 years and women aged 15– rural areas or were controlled for site type in
44 years were recruited into a cohort study of HIV multivariable regression.
transmission. A maximum of one member of each marital
group was selected for recruitment to the cohort, Analysis of HIV incidence and mortality
members of multiple married couples and all unmarried Seroconversions, defined as individuals who tested
individuals from a single household were eligible. negative at baseline and positive at follow-up, were
assumed to have been infected halfway through the
Totals of 8376 and 7102 of the households identified in period of observation. Poisson regression models were
the survey areas at baseline and at follow-up, respectively, fitted with incident infection as the outcome and wealth
were enumerated. Male and female participation rates in index tercile the explanatory variable. Models were
the individual cohort study survey were 78% (4320/5561) controlled for age and site type and are presented
and 80% (5134/6419) at baseline and 77% (3047/3958) separately for men and women.
and 80% (3972/4936) at follow-up, respectively.
Approximately 3 years after baseline 54% (2242/4142) Mortality rates were modelled using the same approach.
of the men and 66% (3265/4922) of the women who Only deaths of participants who were HIV positive at
were not known to have died were re-interviewed at baseline were included, with deaths of HIV-negative
follow-up. This loss to follow-up rate compared participants omitted. This was done for two reasons. First,](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-63-2048.jpg)
![HIV incidence and poverty in Manicaland, Zimbabwe Lopman et al. S59
deaths from non-AIDS causes are likely to differ by wealth baseline (2.7%) and 215 at follow-up (2.6%), who were
index and would therefore obscure the relationship excluded from analyses. At follow-up, the summed score
between wealth index and HIV-associated mortality. of the wealth index in men and women followed a
Second, with the aim of examining how new infections roughly normal distribution by visual assessment. The
and deaths contribute to the changing prevalence of HIV, mean summed score of the wealth index for men and
the rate of becoming infected must be compared with the women was higher in towns (0.44 and 0.43, respectively)
rate of dying from infection. than in either estates (0.34 and 0.32) or rural areas (0.34
and 0.32). Therefore, men and women in towns were
Differentials in sexual behaviour more frequently categorized in the higher wealth index
Reported sexual behaviours, collected at follow-up tercile. Rather than constructing separate wealth index
survey, were analysed for differences associated with categories for each site type, analyses are either controlled
the wealth index. The summed score of the wealth index for area of residence, or presented separately when
was modelled as a continuous variable. The influence of appropriate. There was also greater variance in wealth
the wealth index on sexual debut, the number and type of index in towns compared with estates and rural areas
partnerships, and condom usage was modelled, control- (Table 1), highlighting greater socioeconomic hetero-
ling for age and site type. geneity in urban areas. The mean summed wealth index
score did not substantially or significantly change between
Mixing patterns and wealth index baseline and follow-up in any of the site types.
Mixing patterns are not directly analysable from the
baseline and follow-up of the Manicaland cohort because Follow-up rates decreased with an increasing wealth
participants cannot be directly linked to their marital or index (66, 61, and 58%, chi-squared P < 0.0001), increas-
non-marital partners. Participants were asked whether or ing education (primary/none 70%, secondary/higher
not their last partner had secondary education. Having 55%, chi-squared P < 0.001) and being more mobile at
secondary education was significantly correlated with baseline (64 and 53%, chi-squared P < 0.0001). Follow-
the respondents’ wealth index status (R2 ¼ 0.11 and up appeared not to be directly dependent on wealth,
P < 0.0001 for men; R2 ¼ 0.14 and P < 0.0001 for however, the wealth index was not an independent
women). Therefore, we roughly approximate mixing predictor of follow-up after controlling for education and
patterns by participants’ education level with their mobility (Wald test P ¼ 0.23).
partner’s education level. We represent the degree of
assortative (like-with-like) mixing by Q. Q is 1 when Prevalence
mixing is completely assortative and 0 when completely As previously reported [6], HIV prevalence fell in the
random [16]. Results are presented separately for open cohort between baseline and follow-up. HIV
individuals aged under and over 30 years. prevalence fell in each wealth index tercile in both men
and women (Table 2). The largest decrease in prevalence
was in the highest wealth index tercile in both men at 25%
Results (compared with 11% in the poorest tercile) and women at
21% (compared with 18% in the poorest tercile).
Wealth score
In the HIV serosurvey, 9842 eligible men (aged 17–54 Incidence
years) and women (aged 15–44 years) were tested at In men, HIV incidence was lower in the top wealth index
baseline and 7728 were tested at follow-up. Complete tercile (15.4 per 1000 person-years) compared with the
wealth index data were missing from 209 of individuals at lowest tercile (27.4 per 1000 person-years; Fig. 1a–c).
Table 1. Mean and distribution of wealth index in towns, estates and rural areas in Manicaland, Zimbabwe at follow-up study (2001–2003).
Men Women
Area Wealth tercile Mean (SD) summed score N (%) Mean (SD) summed score N (%)
Town 0.44 (0.18) 0.43 (0.19)
Low 103 (19%) 117 (19%)
Med 116 (21%) 134 (22%)
High 322 (60%) 353 (58%)
Estate 0.34 (0.14) 0.32 (0.15)
Low 303 (24%) 448 (36%)
Med 436 (35%) 379 (30%)
High 513 (41%) 423 (34%)
Rural 0.34 (0.15) 0.32 (0.14)
Low 406 (29%) 877 (32%)
Med 359 (45%) 825 (30%)
High 628 (45%) 1068 (39%)](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-64-2048.jpg)
![S60 AIDS 2007, Vol 21 (suppl 7)
Table 2. HIV prevalence at baseline (1998–2001) and follow-up (2001–2003) in wealth groups and change in prevalence in the open cohort.
Sex Wealth index tercile Baseline prevalence (N) Follow-up prevalence (N) Change Pa
Men Low 0.21 (1061) 0.19 (700) À11% 0.33
Med 0.18 (1650) 0.16 (797) À13% 0.076
High 0.20 (1635) 0.16 (1292) À25% 0.004
Women Low 0.28 (1573) 0.23 (1460) À18% < 0.001
Med 0.25 (2058) 0.22 (1362) À12% 0.020
High 0.24 (1569) 0.19 (1893) À21% 0.004
a
Wald-test P value from logistic regression models controlling for age in 5-year groups and site type.
There was a significant trend of decreasing incidence by group (incidence rate ratio 0.69; 95% CI 0.50–0.95;
wealth index tercile after controlling for site type and age P ¼ 0.02, Poisson regression).
[incidence rate ratio 0.73; 95% confidence interval (CI)
0.56–0.93; P ¼ 0.03, Poisson regression]. This trend was No clear significant or monotonic trends in incidence by
even more marked in young men 17–24 years of age, in wealth index were observed in women of all ages or
whom rates in the highest wealth index group were 8.3 young women (Fig. 1d–f). Controlling for education or
per 1000 person-years and 23.3 in the lowest wealth index mobility (living outside the village in the past year) did not
(a) Men: All agesa (b) Men: 25–54 year olds (c) Men: 17–24 year oldsa
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
Low Med High Low Med High Low Med High
(d) Women: All ages (e) Women: 15–24 year olds (f) Women: 25–44 year olds
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
Low Med High Low Med High Low Med High
Fig. 1. HIV incidence by wealth tercile in Manicaland Zimbabwe, 1998–2003. Incidence is the number of new HIV infections per
1000 person-years. Person-years at risk are contributed by participants uninfected (HIV-negative) at baseline and followed up at
round 2 of the survey. Points and whiskers show the observed cumulative incidence and 95% confidence intervals. Lines and
shaded area illustrate fitted Poisson model and 95% interval controlling for age and site type.
a
Significant linear trend (likelihood ratio test P < 0.05) Poisson regression model, controlling for age and site type.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-65-2048.jpg)
![HIV incidence and poverty in Manicaland, Zimbabwe Lopman et al. S61
significantly improve the models for men or women index women between the ages of 15 and 34 years
(Wald test P > 0.35 for all tests). Mobility was not (Poisson regression Wald P ¼ 0.024). In women over 35
associated with the wealth index tercile for sex, in which years of age there was no apparent mortality trend by
education and wealth index tercile were positively wealth index. Controlling for education level or mobility
associated for men (chi-squared P < 0.0001) and women did not significantly improve the models for men or
(chi-squared P < 0.0001). women (Wald test P > 0.45 for all tests).
Mortality Sexual behaviour
Overall, 300 HIV-positive deaths were observed in the Considering the whole of the male study population,
cohort from 1998 to 2003 [4]. Mortality rates decreased men of higher wealth index were more likely to have
from 25 to 20 to 15 deaths per 1000 person-years in casual sexual partners and to have multiple partners in the
increasing wealth index terciles for men, a trend 3-year follow-up period, but were also more likely to
statistically significant after controlling for site type and report consistent condom use in their casual relationships
age (Poisson regression P ¼ 0.024; Fig. 2a–c). Although (all controlling for site type and age in logistic regression
not significant when split into young and older adulthood models; Table 3, model 1). In towns, however, men of
(35 years of age), the same trend of decreasing mortality higher wealth index did not report greater numbers of
was observed. Mortality was also lower in higher wealth partnerships but did report higher condom usage in casual
(a) Men: All agesa (b) Men: 17–34 year olds (c) Men: 35–54 year olds
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
Low Med High Low Med High Low Med High
(d) Women: All agesa (e) Women: 15–34 year oldsb (f) Women: 35–44 year olds
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
Low Med High Low Med High Low Med High
Fig. 2. Mortality rates of HIV-positive individuals by wealth tercile in Manicaland Zimbabwe, 1998–2003. Mortality rate is the
number of deaths among HIV-positive individuals per 1000 person-years in the cohort excluding HIV-negative participants who
died. Deaths among HIV-infected represents individuals leaving the population of infected and therefore is directly comparable to
the incidence of new infections. Points and whiskers show observed cumulative incidence and 95% confidence intervals. Lines
and shaded areas illustrate the fitted Poisson model and 95% interval controlling for age and site type.
a
Significant linear trend (likelihood ratio test P < 0.10) Poisson regression model, controlling for age and site type.
b
Significant linear trend (likelihood ratio test P < 0.05) Poisson regression model, controlling for age and site type.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-66-2048.jpg)
![HIV incidence and poverty in Manicaland, Zimbabwe Lopman et al. S63
likely to engage in transactional sex. In rural areas in Sexual mixing patterns
particular there were no significant (P < 0.05) associations Sexual behaviour data suggest that higher wealth index
between sexual behaviour and the wealth index. Condom men may be engaging in riskier sexual behaviours, at least
use was not associated with the wealth index in women in for certain indicators such as having casual partners.
any setting. Patterns of mixing will, however, predict the probability
of engaging with an infected partner.
Controlling for completed secondary education had no
substantive effects on the estimates of the association Both men and women in higher wealth index groups
of wealth index and the five sexual behaviours in men were more likely to have attended secondary or higher
(Table 3 and Table 4, model 2). For women, secondary education (men 61, 74, and 76%; women 44, 52, and
education was a stronger determinant of starting sex 63%; x2 < 0.001 for both sexes). The proportions who
(for under 25 years olds) than the wealth index, but had achieved secondary/higher education were much
controlling for education levels had little effect on higher in those under the age of 30 years compared with
wealth index coefficients for other indicators of sexual those aged 30 years and over (Fig. 3a,b). In higher wealth
behaviour. index groups young individuals (< 30 years) were more
(a) Education and mixing by wealth index (< 30 years) (b) Education and mixing by wealth index (< 30 + years)
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
Low Med High Low Med High Low Med High Low Med High
Men Women Men Women
(c) HIV prevalence by education (< 30 years) (d) HIV prevalence by education (30+ years)
0.3 0.5
0.4
0.2
0.3
0.2
0.1
0.1
0 0
Men Women Men Women
Fig. 3. Uneven risk resulting from sexual mixing patters: Mixing patterns and education by wealth index and HIV prevalence by
education level. (a) and (b) Education and mixing by wealth index. Secondary/higher educated; – –^– – Q [degree of
assortative (like-with-like) mixing]. (c) and (d) HIV prevalence by education. & Primary/none; & secondary/higher.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-68-2048.jpg)
![S64 AIDS 2007, Vol 21 (suppl 7)
likely to have secondary education, and mixing was poorer men. This is for two reasons. First, men of higher
increasingly assortative. This is important because in both wealth index in all sites (including towns where they do
sexes, HIV prevalence was lower among individuals with not have more partners) were more likely to use condoms
secondary education, although the difference was much in their casual partnerships. In effect, men reduce the
greater in women (secondary/higher 24.8%; none/ transmission probability if encountering an infected
primary 12.5%; chi-squared P < 0.001) than men woman. In addition, for each partnership that is formed,
(secondary/higher 11.8%; none/primary 7.6%; chi- there may be a lower probability of serodiscordance in
squared P ¼ 0.017). In other words, young men and higher wealth index groups; if partnerships are assortative
women in higher wealth index groups were more likely to (made between members of the same wealth index) and
be educated and to have an educated partner, and that HIV prevalence is lower in higher wealth index groups,
partner was less likely to be infected, with this pattern these partnerships will tend to be less risky. Given the
more pronounced in men. limitations of the present Manicaland data, we cannot
measure directly the degree to which mixing is assortative
Some 61% of women without any secondary education by wealth index. Participant reports on the level of
reported that their last partner was of the same education of their most recent partners, however, suggest
education level and 80% of women with secondary that higher wealth index men and women are markedly
education reported that their last partner had the same more likely to form partnerships with individuals with
level. Assortativeness of mixing and proportion with secondary education, and in turn, young people with
secondary education also increased with wealth index in secondary/higher education have substantially lower HIV
older participants (30þ years, Fig. 3b and d), but in this prevalence. Therefore, men and women of higher wealth
older age group men and women with secondary index are less likely to form partnerships with infected
education had a higher prevalence of HIV. Therefore, individuals. This crude measure of the sexual network
men and women in higher wealth index groups would be requires substantial refinement in two ways. First, the
more likely to contact an infected individual. In summary, level of education is only one dimension of HIV
patterns of mixing appear to confer an increased risk for prevalence. Education as a function of age, as discussed
the higher wealth index groups in the older ages but a briefly here, is another. As noted in a number of other
lower risk for the young. studies in sub-Saharan Africa, the relationship between
education and HIV vulnerability seems to be reversing,
with education becoming protective [10,18,19]. Second,
and preferably, the serostatus of each individual in a
Discussion partnership would be known to understand the degree to
which HIV has penetrated certain wealth index groups
HIV incidence was associated with poverty in men, and to what degree serosorting is occurring in new
especially young men, from 1998 to 2003 in Manicaland, partnerships. The association between HIVand education
Zimbabwe. No such trend was observed in women. reversed completely, with education being protective in
Lower HIV incidence in men of higher wealth index is young people and a risk in older groups.
partly explained and supported by other observations
from this cohort. The study was undertaken during a Analysis of mortality is one way to understand historical
period of general decline of HIV prevalence, but, overall, trends in incidence because there is approximately a 10-
the biggest decreases in prevalence occurred in higher year period between infection and death [20,21]. As with
wealth index groups. By our ‘summed score’ measure of incidence, mortality was lower in higher wealth index
the wealth index, towns were the ‘wealthiest’ of the site groups in both young men and young women, suggesting
types, but they also had the greatest variance in their that the patterns of incidence have not changed markedly
wealth index. This finding, alongside the generally higher since the estimated 10-year period when the groups
prevalence in towns, supports the suggestion that HIV currently suffering mortality became infected. Modelling
transmission may be enhanced by heterogeneity when studies of the HIV epidemic in Zimbabwe suggest that
different social or economic groups mix [17]. behaviour change began in about 1992. (Hallett et al.
unpublished information) and data from the Demo-
The relationship between reported sexual behaviour and graphic and Health Survey from as early as 1994 show
HIV incidence was not always straightforward. Men of women of higher wealth index delaying sexual debut as
higher wealth index reported having more partners and well as the more frequent use of condoms by both men
were more likely to have a casual partner. This is the same and women (Lopman et al., unpublished information).
pattern observed early in the African HIV epidemic, This suggests that behaviour change have been underway
which was used to explain the higher prevalence in the approximately 10 years before this study, with the
more mobile and relatively well off [1]. The evidence resulting impact on infection only now becoming
from the present study suggests that although higher apparent. An alternative explanation is that survival rates
wealth index men may be having more partners, they may are lower in poorer groups. If malnutrition leads to faster
be lower-risk relationships than those entered into by disease progression, as some research has suggested](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-69-2048.jpg)
![HIV incidence and poverty in Manicaland, Zimbabwe Lopman et al. S65
[22,23], and poorer groups are more malnourished, dynamic that may not be adequately represented by radio
higher mortality rates could be caused by reduced ownership, floor type, etc., or any combination of these
survival, rather than different levels of infection. variables. Finally, the summed score measure is to some
extent a marker of urban residence, as evidenced by
The present analyses have focused on incidence in order the higher mean wealth index score in towns. Levels
to understand the direction of causation between the of follow-up were comparable to other major cohort
wealth index and HIV as well as to reflect contemporary studies in Africa [6], but the wealth index was not
patterns of infection. Previous analyses have examined independently associated with a probability of follow-up,
poverty and prevalent infection, and provide an inter- so it is unlikely that these results are biased with respect
esting comparison to the incidence findings. Seropreva- to the wealth analysis. Having secondary or higher
lence was not associated with wealth index among men, education and being more mobile at baseline were,
whereas poorer women were more likely to be infected in however, associated with lower follow-up rates. If survival
the baseline survey of this cohort [24]. This is in contrast or incidence rates differed in the lost-to-follow-up groups
to lower incidence in higher wealth index men and no the analysis may be biased with respect to mobility and
association with incidence in women. This suggests a education. For example, if more educated groups left
general shift away from risk in higher wealth index the Manicaland study sites to find employment in large
groups, perhaps with the shift lagging behind in women. cities, they may have been at increased risk because of
At baseline, poor women from rural areas were more the higher prevalence in cities and the possibility of
likely to have started sex, whereas poor women from meeting new sexual partners after relocation. The group
towns were more likely to engage in transactional sex. By of migrants that were followed up, however, did not
the follow-up survey poorer women in towns were still have different levels of incidence or sexual behaviour,
more likely to engage in transactional sex, but were also but this was a small group of the total migrant population
more likely to have multiple and casual partners and to [27].
start sex younger. It may be expected that the first group
of women to be motivated and able to change behaviour Despite these limitations of the current data, we have
are relatively wealthy women in towns, and this is precisely observed a decreased risk of HIV incidence in higher
what was observed. wealth index men. Although such a trend was not
observed in women, the finding that lower wealth index
HIV risk has reduced substantially among teenagers; women engage in riskier behaviour combined with their
however, the girls who are still becoming infected have an tendency for having less-educated male partners suggests
identifiable vulnerability such as being orphaned or that future trends may follow the emerging pattern in
having experienced the death of another household men. At this advanced stage of the epidemic, a number of
member [25]. Orphaned girls or girls with an HIV- factors may contribute to infection and risk behaviour.
infected parent are more likely to drop out of school and HIV prevention activities in Zimbabwe have included
begin sex, leading to pregnancy, poor reproductive health the treatment of sexually transmitted infections, social
and HIV. So, despite not observing a general trend of marketing of condoms, voluntary counselling and testing,
wealth index and incidence in women, there is a clear education through mass media and the activities of the
causal pathway from vulnerability to leaving school, National AIDS Trust Fund (which is supported by
ultimately leading to HIV infection in young women in income tax). These initiatives, as well as fear of AIDS
this population. It has previously been observed that mortality, may have disproportionately affected those of
households experiencing a death, and particularly an higher wealth index. Risk reduction behaviour, ushered
AIDS death, disproportionably suffered the loss of the in by the relatively well off is a hopeful trend, but, in the
household head, increased healthcare expenditure, and frail Zimbabwean economy, where the poor are an
were more likely to dissolve [26]. increasing demographic, the clustering of HIV in lower
wealth strata is cause for concern.
This highlights the fact that our measure of the wealth
index may be limited in a number of ways. When
grouped into terciles, it becomes a relative measure, with
individuals categorized on the basis of the asset ownership Appendix: Justification of ‘summed score’
of their household, compared with the asset ownership of
other households. Therefore, the secular decrease in There is a high level of correlation between all binary and
wealth index likely to be occurring because of AIDS ordinal wealth variables. Therefore, exploratory analyses
mortality and the collapse of the Zimbabwean economy were undertaken to reduce the 10 assets to a simplified
[11] has not been expressed in this measure. Furthermore, measure of the wealth index [24]. A simplified measure
simplified as a relative measure, asset ownership may be a was created using multidimensional scaling analysis
crude indicator of how the wealth index is a determinant (MDS), a statistical technique for exploring similarities
of sexual behaviour. For example, falling below a certain and differences in data [28]. Starting from a correlation
poverty threshold may drive a woman to sex work; a matrix between variables, MDS is used to assign a score to](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-70-2048.jpg)
![The economic impacts of premature adult mortality:
panel data evidence from KwaZulu-Natal,
South Africa
Michael R. Cartera, Julian Mayb, Jorge Agueroc and
¨
Sonya Ravindranatha
Measuring the household level economic impacts of AIDS-related deaths is of particular
salience in South Africa, a country struggling with a legacy of poverty and economic
inequality in the midst of an HIV epidemic. Household panel data that span more than a
decade permit us to resolve many of the statistical problems that make it difficult to
determine these impacts. After allowing for the impact of demographic adjustments and
other coping strategies, we found evidence that these impacts are quite different across
different types of households, and that the largest and most persistent effects were in the
middle ranges of the South African income distribution, that is, households just above
the poverty line. Households below that level seem less severely affected, whereas
those above it seem to recover more quickly. All these results need to be treated with
caution because their statistical precision is weak.
ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 7):S67–S73
Keywords: Africa, economics
Introduction would have experienced in the absence of the death.
Second, the impacts may be heterogeneous across the
A number of recent studies have measured the economic different households that experience an AIDS-related
impacts of AIDS-related deaths on the economic death. Third, and finally, the impacts may show
wellbeing of affected households [1–3]. Measuring these differential persistence over time for households on the
impacts is of particular salience in South Africa, a country basis of their ability to cope with the death and the stress it
struggling with a legacy of poverty and economic places on household resources.
inequality in the midst of an HIV epidemic. Among
adults aged 15–49 years, the HIV prevalence rate is In an effort to deal with these difficulties, this study
estimated to be 21.5%. As the epidemic moves from employs three waves of panel data on a sample of
infection into impact, premature adult mortality rates are households from the KwaZulu-Natal province of South
increasing rapidly, with an estimated 370 000 South Africa. These data, which span the 1993–2004 period,
Africans dying of AIDS-related illness in 2003, making effectively permit this study to use each household’s
the disease the leading cause of death in almost all South trajectory in the period preceding the onset of AIDS-
African provinces [4]. It is estimated that the majority of related deaths to estimate what the household’s counter-
AIDS-related deaths have still to occur [4]. Three factual economic status would have been without such
difficulties, however, confront efforts to measure these deaths. Although the approach here is somewhat
impacts reliably. First, measuring the economic impacts of distinctive, several earlier studies have utilized panel data
an AIDS-related death requires an estimate of the to examine the dynamics of poverty status and the impact
counterfactual level of wellbeing that the household that premature deaths have on pathways into or out of
From the aProfessor, University of Wisconsin-Madison (Agricultural and Applied Economics), the bAssociate Professor, University
of KwaZulu-Natal (Development Studies), and the cAssistant Professor, University of California-Riverside (Economics), and
Graduate student, University of Wisconsin-Madison (Agricultural and Applied Economics).
Correspondence to Julian May, Associate Professor and Head of School, School of Development Studies, Memorial Tower
Building, University of KwaZulu-Natal, King George V Ave, Durban.
Tel: +27 31 2602841; fax: +27 31 2602359; e-mail: mayj@ukzn.ac.za
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S67](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-72-2048.jpg)
![S68 AIDS 2007, Vol 21 (suppl 7)
poverty [5–8]. In addition, the analysis here explores the of an AIDS-related death, or whether they become
heterogeneous impact of premature adult mortality by trapped in a permanently lower standard of living.
permitting its estimated impacts to vary according to the
households’ specific initial conditions.
Although the immediate economic effects of an AIDS- The KwaZulu-Natal Income Dynamics
related death are of course important, its long-term Study
impacts on poverty depend on whether households
recover from the death economically, or whether they fall Covering the 1993–2004 period, the KwaZulu-Natal
into a long-term poverty trap. Both the economic theory Income Dynamics Study (KIDS) allows an analysis of
of poverty traps [9] and evidence on natural disasters [10] household wellbeing both before and after the onset
suggest that severe shocks can indeed push households of AIDS-related deaths in South Africa. The first round of
below a critical level from which they cannot recover what became the KIDS data was part of the nationally
economically. Similar insights emerge from the anthro- representative survey conducted by the Project for
pological literature that shows how a poverty trap Statistics on Living Standards and Development [19].
situation can emerge for resource-poor households. The key decision makers (or ‘core’ members) of the 1354
The living standards of vulnerable households that face African and Indian households visited by the Project for
multiple shocks over time may ratchet down over time to Statistics on Living Standards and Development in the
the point at which they eventually become trapped in a KwaZulu-Natal province became the basis for the follow-
situation of structural poverty [11,12]. up KIDS surveys undertaken in 1998 and 2004 [20,21].
When subsequent survey rounds found that co-resident
In the context of AIDS-related deaths, these ideas of core members had split into separate residences, each core
repeated shocks and poverty traps have an important was interviewed along with their corresponding new
resonance. In countries in which there has been a high household. In the 2004 survey round, adult children of
prevalence of HIV, death is preceded by comparatively the core members were also interviewed. In total,
lengthy episodes of illness (and corresponding episodes of information was obtained in all three periods from at
care; Hornbrook et al. [13] have suggested framing least one eligible respondent for 74% of the original
analysis of protracted illness in terms of the costs of 1354 Project for Statistics on Living Standards and
episodes of illness and care) that present households with a Development KwaZulu-Natal households. Ethical
protracted series of shocks as the illness progresses. In the approval for study was obtained through the appropriate
case of AID-related deaths in Tanzania, the average length committees at the University of Natal and the University
of debilitating illness preceding death was 12 months of Wisconsin-Madison. Informed consent was obtained
[14]. Another Tanzanian study showed that on average, before interviews were undertaken. Although similar to
an adult experiences 17 different episodes of illness that found in other panel studies, the attrition rate in the
before dying [15]. Each of these episodes is likely to be KIDS sample needs to be kept in mind when considering
accompanied by episodes of care, which become more the results reported below. Maluccio [22,23] analysed the
costly as death approaches. Moreover, in many cultures, pattern of attrition over the 1993 to 1998 period, and
the death itself does not signal the end of the episode noted that it came from both the upper and lower tails of
because funeral celebrations are required that necessitate the livelihood distribution. Although there has not yet
further expenditure and possible indebtedness [16]. Even been a similar analysis of attrition over the 1998–2004
these expenditures may extend over several years if there period, the age-specific mortality patterns in the KIDS
are annual celebrations or customs that families observe. data are similar to those found in another study of South
Some of these changes are not unique to the HIV Africa [24]. Whereas these observations suggest that the
epidemic. Discussing the impact of malaria, Sachs and KIDS data accurately reflect the reality of AIDS-related
Malaney [17] noted the costs associated with changes in deaths, it is possible that sample attrition over the 1998–
the behavior of household members concerning decisions 2004 period disproportionately reflected households
such as schooling, child-bearing, savings and work- most severely affected by AIDS-related deaths. If that
seeking are often overlooked when measuring the indeed happened, then the results here will probably
economic impact of disease. Such changes have been understate the true impacts of those deaths.
documented in the case of AIDS-related illness or death
in Tanzania, where it has been shown that children may The KIDS data show that at ages 20–50 years, the
marry earlier, drop out of school to help support the proportion of individuals dying between the second and
family, and take on informal labor schemes [18]. These third waves was nearly three times the proportion dying
observations all suggest that the economic effects of an between the first two waves [20]. In total, 309 members of
AIDS-related death may be long lived, if not permanent. KIDS households between the ages of 20 and 50 years
Following the lead of a study of Tanzania [8], the analysis died between 1998 and 2004. Of these deaths, 74 were
here will exploit the available panel data to see whether the result of injury or accident. As we are interested in the
households recover over time from the economic impacts impact of death associated with illness, we excluded this](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-73-2048.jpg)
![Economic impacts of adult mortality Carter et al. S69
ÀyitÀ1
group and designated the remaining 235 deaths git ¼ yityitÀ1 . Under our data structure, we observe git
premature adult mortality (PAM). Premature mortality twice: once for t ¼ 1998, measuring the growth (positive
refers to death occurring before some standard age. We or negative) in wellbeing between 1993 and 1998, and
have used South African life expectancy in 2004 (51.4 once in 2004 (measuring the growth since 1998).
years) as a guideline for this standard [25]. Other studies
have used a slightly younger age group (15–50 years) on Consider the following fixed effects regression model for
the grounds that this age range is the group most at risk of this growth in economic wellbeing measure:
HIV infection [26].
git ¼ b1 hit þ d1 Sit þ d2 ½lnðyitÀ1 ÞŠ þ yi þ li þ eit (1)
where hit is a binary indicator variable that takes a value of 1
Methodology and results when family i experienced PAM between times t–1 and t,
and is 0 otherwise. We here treat the 25 households with
Evaluating the impacts of PAM on the economic status of more than one PAM as identical to those with only one
a family is difficult because we cannot observe what the PAM. Efforts to identify different effects for additional PAM
family’s status would have been counterfactually in the failed statistically, presumably because of small sample sizes.
absence of the death. The economic status of families The variable Sit signifies other unfavorable shocks that struck
unaffected by PAM may be a very bad proxy for this the household between times t–1 and t, including crop loss,
counterfactual status, especially in the case of the HIV theft, spousal abandonment, and death of an elderly house-
epidemic in which specific behaviors and situations are hold member. The terms yi, li, d and b are all parameters to
known to make infection and death more likely. be estimated, and eit is a random error term that we assume is
unrelated to the included variables.
Other studies have approached this statistical problem in
several ways. One approach is to use propensity score Consistently estimating the coefficient b1, which gives
methods to match affected with unaffected households, the impact of an adult death on the growth in wellbeing,
effectively using the latter as the counterfactual for PAM is of course our primary interest. Note that this
households [1]. Propensity score methods, however, only specification assumes that the impact of a premature
control for observable differences between affected and adult death on wellbeing is the same for all households.
unaffected households. Alternatively, with panel data it is This ‘homogenous effect’ regression model thus says that
possible to use fixed effects estimation methods that can growth in household wellbeing over time depends on a
also control for any unobserved differences between household-specific growth factor that does not change
households that do not vary over time [3,6]. over time (yi), as well as on a time-specific intercept
(li, t ¼ 98 or 04) that is assumed to be the same for
The approach here is similar to this fixed effect approach all households.
except that the three periods of KIDS data permit us to
work in rates of growth in wellbeing rather in levels of Our ability to use fixed effects panel data methods to
wellbeing. In particular, the KIDS data allow us to control for the household-specific effect is key to our
observe a family’s economic trajectory (their growth in effort to identify the impact of a prime age adult mortality
wellbeing) before the onset of the epidemic. Using this on economic wellbeing. Note that yi will capture time
information, and a few modest statistical assumptions, we invariant observable and unobservable factors that
can use fixed effects methods to predict reliably what the influence the growth in household wellbeing. It is
affected family’s economic status would have been in the precisely these unobservable differences between house-
absence of PAM. Effectively, this procedure allows each holds that make it difficult to estimate the impact of
family’s past experience to inform the counterfactual that premature adult death. Once we control for the fact that
is used to judge the impacts of PAM. households with adult deaths are likely to grow more
slowly (or perhaps more rapidly) than the typical
The KIDS data described above contain measures of household, we can be more confident in our estimate
household economic wellbeing at three points in time, of b1. More formally, failing to control for the household-
1993, 1998 and 2004. We denote the economic wellbeing specific fixed effect would tend to exaggerate the impact
of household i in time period t as, yit. Economic wellbeing of a premature death if households that suffer such deaths
is measured as total household expenditures per capita. tend on average to experience lower growth even in the
Expenditures include the imputed value of home- absence of the death. Given that the performance of the
produced food, owner-occupied housing, etc. Although South African economy improved over the 1998–2004
in principle this measure should be scaled for the period, a change that is reflected in the profile of poverty
demographic composition of the household, we have of the KIDS sample, we would expect l04 > l98. Note,
not done so here in order to maintain comparability with however, that our methodology does not account for the
the de facto per capita standard used to define poverty spillover effects of premature death. An analysis of Zambia
in South Africa. In turn, we define the growth rate of estimated that local economic growth was negatively
household wellbeing between period t–1 and t as: influenced by high concentrations of AIDS-related](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-74-2048.jpg)
![S70 AIDS 2007, Vol 21 (suppl 7)
deaths [7]. Although such macro effects may occur in different depending on the household’s initial level of
South Africa, the urbanized and well-integrated nature of wellbeing (measured as the natural logarithm of the
the South African economy makes it less likely that these household’s level of wellbeing at the beginning of the
effects can be picked up at the local community level. period, yit–1). Conventional economic theory predicts
Similar to Grimm [6], we controlled for other shocks that that d < 0, indicating that initially less well-off households
potentially affect the growth rate of household economic experience more rapid growth. Other theory suggests the
wellbeing. As measures of these shocks, we employed opposite [9]. For the purposes of this study, we are simply
binary indicator variables as to whether the household concerned to control for the impact of initial levels of
experienced the shock. The study by Grimm [6] also wellbeing on subsequent changes.
controlled for changes in household demographic
composition. We chose explicitly not to control for Table 1 displays the fixed effects estimates for the
demographic changes as we suspected that such changes homogenous effects model. As the underlying data were
are themselves coping strategies employed by families that collected through cluster design, robust standard errors
suffer PAM. Statistically, demographic changes would be were calculated that allow for intracluster correlation in
directly related to the error term eit in (1), and including it the regression errors. For the key variables of interest, the
would yield biased estimates of the effect of PAM. P value (the level of statistical significance at which it is
Although it would be possible to employ simultaneous possible to reject the hypothesis that the reported
equation methods to address the statistical endogeneity of coefficient is zero) is reported in square brackets. The
demographic changes, we prefer here to estimate reduced estimated coefficient of the PAM variable is negative, but,
form models such as (1). The parameter estimates we surprisingly, it is not statistically significant. Its value
obtained thus give us the full or bottom line effect of a (À0.21) means that PAM would be expected to lower a
PAM on household wellbeing after the household has household’s 5-year growth rate by 21%, controlling for
utilized available coping strategies (including demo- the unobserved time-invariant factors that influence each
graphic changes). household’s growth rate (yi) and other variables.
Note also that model 1 does not condition on the charac- The estimated coefficient of the initial level of wellbeing
teristics of the adult who has died (as in Yamano and Jayne is statistically significant and signals a convergent process,
[3]). Although we have no doubt that these characteristics with initially less well-off households estimated to grow
matter, we are here interested in identifying the average or faster than others. None of the shock variables are
typical effect of PAM in our South Africa data. statistically significant, although most are negative. Their
insignificance may signal that most of these shocks are of a
Finally, the basic regression model includes a term that short-term nature, and that whatever their short-term
allows the expected growth in economic wellbeing to be effects on consumption, households had largely recovered
Table 1. Fixed effects estimates of the impact of premature adult mortality on the growth rate of economic wellbeing.
Explanatory variables Homogenous effects model Heterogeneous effects model Impact persistence model
PAM impact
Common effect, b1 À0.21 [25%] 1.8 [22%] 5.4 [27%]
Differential effect, b2 – À0.370 [14%] À0.96 [25%]
Persistence of PAM effect
Common effect, b3 – – À0.10 [33%]
Differential effect, b4 – – 0.02 [35%]
Convergence, d À1.8MM À1.8MM À1.8MM
Time effects
1998 intercept, l98 10.8MM 10.4MM 10.4MM
2004 intercept, l04 11.3MM 10.9MM 10.9MM
Other shocks
Illness 0.15 0.16 0.16
Job loss À0.05 À0.05 À0.05
Lose remittances À0.12 À0.12 À0.17
Lose grant À0.02 À0.02 À0.01
Abandonment À0.34 À0.27 À0.24
Theft 0.02 0.01 0.02
Crop loss À0.27 À0.26 À0.25
Elderly death 0.33 0.32 0.32
Household fixed effects, yi Included Included Included
R2 within 0.34 0.34 0.35
PAM, Premature adult mortality. Reported significance based on robust standard errors corrected for clustering. Figures in square brackets are
P values.
M
Statistically significant at the 10% level.
MM
Statistically significant at the 5% level.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-75-2048.jpg)
![Economic impacts of adult mortality Carter et al. S71
their expected level of economics by the time of the positive effects of PAM for poorest households and
survey. The death of an elderly household member, negative effects as households become better off. The
however, is not a short-term event like an illness. The estimated coefficients are, however, not statistically
positive, but statistically insignificant, coefficient on the different from zero at conventional probability levels.
elderly death variable may seem surprising in the context We re-estimated the reported models using a lower age
of South Africa, where the death of an older person nearly cutoff to define PAM (40 and 45 years instead of 50 years).
always results in the loss of significant pension income. Its Lowering the cutoff had little effect on the magnitude of
insignificant effect may reflect the fact that the households the estimated coefficients, but did improve their statistical
were prepared (economically) for the death. One such significance. Increasing the PAM age cutoff towards 60
coping strategy may be through the shedding of the years also left the point estimates of the PAM coefficients
household members who other studies have shown tend stable, but made them even less precise. This pattern is
to immigrate into households when an elderly person consistent with the notion that the earning power of
becomes of pensionable age. adults begin to fall off as they enter their 50s. In addition,
households presumably become better prepared for an
Whereas it is common to think of a premature adult death adult death as that death becomes (statistically) more
reducing household economic wellbeing, of course it likely. This insignificance reflects the heavy demands put
need not be so, especially when large numbers of adults on the data by fixed effects procedures, as well as the
are unemployed or underemployed. Negative effects clustering of the underlying data that further reduces
could also be muted if other family members involved in the precision obtainable from a sample of the size of the
care-giving were also unemployed or underemployed at KIDS. When the impact of clustering on the standard
the time of the onset of an AIDS-related illness. In this errors is ignored, the estimated coefficients are significant
circumstance, an adult death may actually increase the at conventional levels. The estimated impacts of the other
living standards of those remaining alive in the household shock variables are qualitatively identical to those in the
as there are now fewer needs to meet from the family’s homogenous effect model.
modest resources [27]. It should be stressed that the
analysis here ignores other benefits (even those that Given that the estimated impact of PAM now depends on
are solely economic) that an individual may bring to the the household’s initial level of wellbeing, we used the
household, including support for children, their socia- estimated coefficients from Table 1 to calculate the impact
lization and education [28]. The opposite could of course of PAM on the livelihood trajectories for three typical
be the case for somewhat better-off households, in which households: one that began in the 20th percentile of the
the premature death of an adult results in a net reduction initial wellbeing distribution, another at the 50th and a
of the goods available for others. third at the 80th percentile. For each of these typical
households, we took the average of the fixed effect terms
From a statistical perspective, these observations suggest (the yi) for economically similar households. For example,
that our basic regression model (1) mixes together two for the 20th percentile household, we took the average
different regimes, one in which the immediate livelihood fixed effect estimates for all households between the 15th
effects of PAM are negative, and another in which they and the 25th percentile. A similar band was used for the
are positive. The average effect estimated in Table 1 other two household estimates. Using this estimate, plus
would, in this case, be a data-weighted average of the two the household’s initial level of wellbeing (yi93) we then
underlying regimes or regression relationships. From this calculated the predicted growth that would be expected
perspective, we see that this data-weighted average effect for such a household over the 1993–1998 period, the
is negative, but not surprisingly, it is insignificant. period before the onset of significant AIDS-related
deaths. Using this predicted growth, we then calculated
In an effort to pull these two regimes apart, and allow for the household’s predicted standard of living for 1998. To
heterogeneous PAM effects, we modified the basic fixed make this value more easily interpretable, we have divided
effect regression equation as follows: it by the poverty line such that a standard of living of 1
would imply a living standard exactly equal to the poverty
git ¼ b1 hit þ b2 ½hit lnðyitÀ1 ÞŠ þ d1 Sit þ d2 ½lnðyitÀ1 ÞŠ line, 2 a living standard double the poverty line, and so
þ yi þ lt þ eit ; (2) forth. For this analysis, we used the de facto official South
African poverty line of R322 per month per person
where the new coefficient b2 allows the PAM impact to (in year 2000 prices) [29]. Figure 1 shows these pre-PAM
change with the household’s level of initial economic well- estimates for each of the three typical households. As
being. As discussed above, we might expect b2 0 and can be seen, all but the least well-off households
b1 ! 0. experienced negative growth in wellbeing over the
1993–1998 period.
The second column in Table 1 shows the results of this
expanded, heterogeneous effects model. The impact In order to assess the predicted impact of PAM, we then
coefficients have the anticipated signs, indicating the performed the same exercise for the 1998–2004 period](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-76-2048.jpg)
![S72 AIDS 2007, Vol 21 (suppl 7)
where the new variable pit measures the passage of time (in
Without PAM
months) between the PAM death and the survey. Note that
200 this model permits the rate of recovery from PAM to vary by
Percentage of poverty line
80th Percentile household
income level. If households tend eventually to recover after
150
PAM, we would expect the effect of time-since-death to be
positive, as found in an analysis of Tanzania [8]. In contrast, a
Poverty line study of Kenya [3] did not find evidence that PAM effects
100 dissipate over time (although their period of observation was
50th Percentile household
shorter than that in Beegle et al. [8]).
50
20th Percentile household
The estimates (reported in the third column of Table 1)
are again not statistically significant at conventional levels.
0
1992 1994 1996 1998 2000 2002 2004
Although this evidence is thus a bit weak, the estimated
Year coefficients imply that a household that began in the 80th
percentile of the wellbeing distribution would have
Fig. 1. Impact of premature adult mortality on livelihood begun to recover its growth rate 5 years after the PAM.
trajectories. PAM, Premature adult mortality. Better-off households would be estimated to recover
more quickly. Less well-off households (for whom the
effects are less pronounced) are estimated to recover
for the three typical households. The starting point for
less quickly.
each household was their predicted level of wellbeing for
1998, as described above. For each household a growth
In conclusion, this paper has explored the ability of panel
rate (and resulting living standard level) was calculated
data to permit more reliable inferences on the impact of
both with and without a premature adult death. As can be
AIDS-related deaths on household economic wellbeing.
seen, the predicted impact of PAM on the household that
The results obtained, which allow us to compare the
began at the 20th percentile is slightly negative, but
economic wellbeing of AIDS-affected households with
imperceptibly so. Households at the 50th percentile in
what their wellbeing would have been in the absence of
1993 (whose wellbeing levels had collapsed over the mid-
AIDS, are less strong statistically than they might be. They
1990s) show a similar pattern. In contrast, households at
do, however, suggest a consistent story in which the
the 80th percentile show a large predicted drop in
immediate impacts of an AIDS-related death are most
wellbeing. Without PAM, the household would have
severe for somewhat better-off households. It should be
grown to a living standard in excess of 225% of the
stressed that the analysis here gives a ‘bottom line’ impact
poverty line. With PAM, the household’s wellbeing is
estimate that reflects households’ adaptation to the shock,
only 175% of the poverty line. This 50% drop is correctly
including demographic adjustments in which severely
interpreted as the impact of PAM on initially better-
affected households may send dependent members to
off households.
better-off friends and relatives. Demographic adjustments
of this sort would improve the economic wellbeing of the
Finally, our data permit us to explore whether these
remaining members, but would also hide some of the
estimated PAM impacts tend to dissipate over time. To do
effects of the AIDS shock.
this, we modified the model by including an additional
variable that indicates the number of months between the
Although somewhat at odds with the conventional
premature death and the date of the survey. This
wisdom that the HIV/AIDS crisis most severely affects
specification implies a linear relationship between time
the poorest households, our findings are consistent
since death and effects. We also estimated a more general
with the observation that premature death may actually
specification (with different dummy variables for different
reduce poverty as we measure it [27]. These findings
periods since death) and obtained qualitatively similar
also suggest that an analysis of the impacts of AIDS-
results. We also tried a specification in which we looked
related deaths needs to allow for the possibility that
for effects based on the period of time when the
these impacts could be quite variable across household
individual who eventually died first became unable to
types.
perform his ordinary economic activity. These specifica-
tions proved to be uninformative. This same variable was
Finally, the analysis here has explored the question of
also interacted with initial expenditures to give the
whether households are able to recover over time from
following ‘impact persistence model’:
the immediate impacts of an AIDS-related death. The
git ¼ b1 hit þ b2 ½hit lnðyitÀ1 ÞŠ þ b3 pit results are again statistically weak, but they suggest that
better-off households do manage to recover their rate of
þ b4 ½pit lnðyitÀ1 ÞŠ þ d1 Sit þ d2 ½lnðyitÀ1 ÞŠ
economic progress eventually. Although further investi-
þ yi þ li þ eit ð3Þ gation is needed, the results do suggest that households](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-77-2048.jpg)
![The financial impact of HIV/AIDS on poor households
in South Africa
Daryl L. Collinsa,b and Murray Leibbrandtb
Background: Rising mortality rates caused by HIV/AIDS in South Africa have sub-
stantial and lingering impacts on poor households.
Methods: This is a descriptive paper using a new dataset of daily income, expenditure
and financial transactions collected over a year from a total of 181 poor households in
South African rural and urban areas. One of the key pathways through which HIV/AIDS
impacts on household wellbeing is through the socioeconomic impacts of death, which
this dataset is especially useful in quantifying.
Results: The key impacts of death on households are funerals and the loss of income.
Funerals often cost up to 7 months of income. Nearly all households in the sample
attempt to cover such costs by holding a portfolio of funeral insurance. Despite
these efforts to insure against funeral costs, 61% of households are underinsured against
the cost of a funeral. Nearly half the sample households are dependent on a regular
wage earner, and another quarter are dependent on a grant recipient. Eighty per cent of
these households would lose over half of their monthly income should the highest
income recipient in the household die. Even by selling liquid assets, only one third of the
sample households would be able to maintain their pre-death living standards for a year
or more.
Conclusion: Death poses substantial and lingering burdens from the funerals that
surviving household members need to finance and the ongoing loss of income once
brought into the household by the deceased. These costs pose so great a threat to
households that they dominate household saving and insurance behavior.
ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 7):S75–S81
Keywords: South Africa, AIDS-related mortality, household finance, funerals,
insurance
Introduction greater destitution is revealed in household-level studies
[3–7].
The severe impact of HIV/AIDS has led to a dramatic
increase in the probability of death in South Africa’s This paper focuses on the financial impact of death at the
population. The latest forecasts from the Actuarial Society household level, using a new dataset called the Financial
of South Africa show the likelihood of death among adult Diaries, which tracks household-level cash flows over one
men jumping from 36% in 1990 to a forecasted 61% in year. The dataset used in this paper is not geared
2008, whereas the likelihood of death among adult specifically towards capturing information about HIV/
women increases from 21% in 1990 to a forecasted 53% in AIDS. As households were observed over an entire year,
2008 [1]. Given high unemployment rates in South however, the impact of illness and death revealed more
Africa, macroeconomic estimates of the impact of this details about the impacts of death on finances than other
increasing death rate are somewhat benign [2], but far survey data. These are underappreciated factors in
From the aNew York University, New York, New York, USA, and the bSouthern Africa Labour and Development Research Unit,
University of Cape Town, Cape Town, South Africa.
Correspondence and requests for reprints to Daryl Collins, Wagner School of Public Policy, New York University, 295 Lafayette
Street, New York, NY 10012, USA.
Tel: +1 914 433 9014; e-mail: dlc300@nyu.edu
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S75](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-80-2048.jpg)
![S76 AIDS 2007, Vol 21 (suppl 7)
understanding the impacts of HIV/AIDS on household 70%
wellbeing. Although the Financial Diaries did record
60%
situations revealing the burdens of caretaking and medical
expense, a key finding of our research is that these costs
50%
were dwarfed by the overwhelming financial impacts of
death [9]. The impact of death was felt the strongest with 40%
the high cost of funerals and with the loss of income from
a breadwinner. These impacts are known to households, 30%
and to a large extent they dominate the saving and
insurance decision-making of households. This paper 20%
provides a descriptive analysis that substantiates these
points and details the extent to which sample households 10%
were prepared for these financial burdens.
0%
Regular wages Grants Remittances Business Casual wages Rental income Pension UIF Agriculture
profits
Fig. 1. Source of income, by percentage of average monthly
Methods household income. Langa; Lugangeni; Diepsloot.
This paper investigates the financial impact of HIV/AIDS
on vulnerable households living in three urban and rural
areas of South Africa. The Financial Diaries dataset was period. Households earn their income from a number of
collected to examine a variety of questions about financial different sources. Figure 1 demonstrates this by reflecting
management in poor South African households. The data the average household income in each of the three areas of
can be accessed at http://www.datafirst.uct.ac.za/data_f- the study, and the percentage of average income earned
diaries.html. A sample of 181 black households was from different activities. In the urban areas, most income
selected in three of South Africa’s low income areas: comes from regular wages, from one or more people in
Langa, Cape Town (urban); Diepsloot, Johannesburg the household. Government grant income is, however,
(peri-urban); and Lugangeni, Eastern Cape (rural). A also important. There are several types of monthly
stratified selection criteria based on relative household government grants in South Africa. At the time of the
wealth was used to select households from each study, old age grants for older people above the age of 65
neighborhood in these areas. Between July 2003 and years were US$114 per month, disability grants were
December 2004, the households were interviewed on a US$114 per month, child support grants for children
fortnightly basis by a team of six field researchers. Detailed below the age of 7 years were US$26 per month and
daily income, expenditure and financial transactions, as foster care grants were US$83 per month. In the entire
well as open-ended qualitative data, were captured sample, 27% of households depend on a grant for the
during this period using a specially built and conceived majority of their income and in rural Lugangeni, grants
database. An attrition rate of 19% was experienced over account for close to 50% of average monthly household
the course of data collection, primarily in relatively income. Other sources of income were recorded but were
higher income households. The analysis presented in this less important.
paper is based on the 152 households for which there is
continuous data for the entire year. Full details regarding There is a fairly strong literature looking at the impact of
the dataset, including survey instruments, can be found the death of a working age individual on a variety of
on www.financialdiaries.com and in a preliminary household outcomes, and even some research on the
descriptive overview [8]. burden of caring for those who become ill as a result of
HIV/AIDS [5–7]. This literature has not, however,
The General Household Survey, a national South African explored the large costs of the funerals associated with
sample survey, reports national mean household expen- these deaths. Given that these funerals are frequent
diture and national mean per capita expenditure to be occurrences within poor communities, they represent
US$3287 and US$2060, respectively. The comparable daunting claims on household resources. This paper takes
figures for the black population group are US$1288 and a close descriptive look at the extent of the financial
US$814. Even allowing for comparability issues between burden that these funerals impose on households as well
these datasets, it is clear that the sample households used the behavior of households in response to these burdens.
in this paper are poor compared with average South The adequacy of funeral expense coverage is established
African households, with a mean household income of by measuring the costs and sources of funding for funerals
US$432 and mean per capita income of US$155. All and the sufficiency of funeral insurance against these costs.
dollar amounts are converted from South African Rands Although funeral insurance does dampen the financial
at a market rate of ZAR6.5 per US$, the average rate shock of death, the results show that 61% of sample
during the period of the Financial Diaries data collection households are inadequately insured for potential funeral](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-81-2048.jpg)
![Financial impact of HIV/AIDS Collins and Leibbrandt S77
costs. Within the poorest of the sample, 84% are with interest and without, and by using money remaining
inadequately insured for potential funerals in their from the two grants. She managed to pay for the funeral,
households. but was left with a significant debt that she struggled to
repay for the remainder of the year.
The financial impact after the funeral is assessed by
estimating the amount by which these households would The funerals observed in the Financial Diaries agree with
see a decline in income per capita with the loss of the main similar estimates of funeral costs from a less detailed but
income earner. The results show that 80% of the sample broader survey on funerals [10]. Most funerals appear to
would lose more than half their per capita income with cost approximately US$1500. Compared with an average
the death of the highest income earner, suggesting a household income of between US$155 and US$308 per
lingering and debilitating shock of death. This analysis is month, households can easily spend an amount com-
extended by assessing the amount of time households parable to approximately 7 months of income on one
would be able to maintain their living standards after a funeral. Roth [10] found evidence that funerals in the
death, given current liquid asset levels. With the sale of Grahamstown township cost approximately 15 times the
liquid assets, only one third of the sample households average monthly household income. It is no surprise,
would be able to maintain their pre-death living standards then, that the funeral industry in South Africa is
for a year or more. substantial. An estimated US$770 million is spent on
funerals each year, with 3000–5000 funeral parlors to
facilitate them [11].
Results Such a large one-off cost cannot be managed out of
monthly income, and some sort of financial instrument
Data from the Financial Diaries [9] recorded five funerals must be used to manage the costs. Savings instruments,
of household family members within 152 sample although helpful, are not feasible in meeting the entire
households over the study year. Although five funerals cost of the funeral, as it would take many years for most
are too few to generalize about how households manage households to save this amount. Borrowing would put
their finances to pay for funerals, they do provide a very households in severe debt, even if they were able to find
strong basis for understanding the financial impact of someone willing to lend them such a large amount
these funerals. of money.
An example of the expenses and funding sources It is therefore easy to understand why funeral insurance
demonstrates how much funerals cost and how house- dominates the financial portfolios of the poor in South
holds pay for them. Thembi (names of respondents have Africa. It is also, however, a strategy chosen to insure
been changed to protect their identity) is one of the urban against a worst case scenario rather than an optimal use of
respondents, a 50-year-old woman who lives with her 47- scarce savings to facilitate a movement out of poverty.
year-old brother. The major source of income for the There is no doubt that this behavior is driven by the fact
household was the disability grants of US$114 per month that, in the age of HIV/AIDS, working-age deaths are an
that each received, plus a part-time job that Thembi held. immediate reality in the households and communities of
Thembi belonged to a burial society but when her the poor [1]. Ten million people in South Africa have
brother died, reportedly of tuberculosis, in June 2004, she funeral insurance [12]. Of these 10 million, 8 million
was left scrambling for resources to pay for his funeral. A people belong to an informal burial society. There are an
set of consolidated accounts for the funeral is shown in estimated 80 000–100 000 burial societies in South Africa
Table 1. Of the sources of funds, only 11% came from [11].
Thembi’s burial society. The majority of the costs (54%)
were paid for through relative’s contributions. Thembi This paper distinguishes between three different forms of
was able to scrap together a bit more by borrowing, both funeral insurance: formal funeral plans with an insurance
Table 1. Sources and uses of funds for Thembi’s brother’s funeral.
Sources of funds US$ Uses of funds US$
Cash contribution from relatives 538 Undertaker 538
In-kind contribution from relatives 225 Tent 91
Burial insurance payout 154 Pots 35
Borrowed from aunt’s burial society (no interest) 154 Food 750
Borrowed from cousin’s savings club (30% per month) 92
Borrowed from cousin (no interest) 108
Leftover money from grant 92
Leftover money from brother’s grant 50
Total 1413 Total 1414](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-82-2048.jpg)
![S78 AIDS 2007, Vol 21 (suppl 7)
company; burial societies, which are informal funeral required should every person in the household need a
plans administered through a group of friends, relatives funeral.
or neighbors; and funeral parlor funeral plans. Within X
the Financial Diaries sample, funeral instruments make INSUREi ¼ FUNERAL COSTS INSUREDi =
X
up between 10 and 20% of the total number of FUNERAL COSTS REQUIREDi ð1Þ
instruments in the average household portfolio [10].
Most households have more than one burial society or
formal funeral plan, a feature that is also observed in other An example of INSURE is calculated below for a rural
South African research [11,13], as well as in Ethiopia and household. This household is headed by Mzwamadoda
Tanzania [14]. (names of respondents have been changed to protect their
identity), an older man who lives in Lugangeni with his
Despite the prioritization of funeral insurance in house- wife, Tembisa, one child and six grandchildren. Despite a
hold portfolios, however, this funeral cover rarely covers modest income (US$228 per month at the time of the
the entire cost of a funeral. The example above showed study), this household had recently joined five out of their
that the bulk of funds for the funeral come from seven plans in the past 4 years alone. All of the plans cover
contributions from extended family members. This is a Mzwamadoda, Tembisa and their two daughters. A third
consistent theme in the Financial Diaries data. Over the daughter is covered in all but three of the plans. If either
study year, 81% of sample households contributed to the Mzwamadoda or his wife were to die, US$7708 would be
funeral of a relative outside the immediate household at paid out for the funeral from all of the plans. By any
least once [9]. These contributions are substantial, often standards, this is a lot of money for a funeral. In total, the
costing up to 20% of the monthly income in some cases, sum of benefits insured for their family under all their
requiring households to borrow or dip into savings. In a plans is US$23 614, whereas the amount they would
situation in which the rate of death is increasing, as in the require to have adequate funerals for the family is only
AIDS epidemic in South Africa, relatives may become less US$9230. Therefore, the value for the INSURE variable
motivated or able to continue to contribute at the same for this household would be 2.558 (Table 2).
rate they have in the past.
INSURE is calculated household by household for the
How much insurance would it take for households to entire sample of 152 households. The results show that 61%
insure themselves against the death of a household of households are inadequately insured (with INSURE
member, without requiring help from relatives? A funeral below 1) versus 39% that are adequately insured (with
adequacy ratio is calculated from information in the INSURE above 1). The ratio of inadequately insured
Financial Diaries dataset. This calculation uses an households to adequately insured households is particularly
estimated cost of a funeral of US$1500 for an adult high in rural Lugangeni, which also has much poorer
and US$770 for a child, as well as several other estimates households than the two urban areas. Is this a choice within
based on funerals observed in the dataset. There are households, or are households too cash-constrained to
further assumptions that needed to be made for this insure themselves adequately?
estimate and we based these assumptions on what was
recorded in the Financial Diaries data and reported in This question is investigated by splitting the sample in
other papers [15]. Often benefits from burial societies or each area into three tiers based on income per capita per
funeral parlors will be in kind. As burial society costs are month, taking into account not only the number of
usually food, the value of this in-kind benefit is assumed individuals supported in the household but also income
to be US$308. Funeral parlor benefits are usually a coffin, relative to others in the area. As Figure 2 below shows,
transport and burial fees, the cost of which is estimated at only 16% of relatively low income households are
US$770. The variable INSURE is calculated as the sum of adequately insured, compared with 38% in medium
funeral costs insured over the sum of funeral costs income households and 62% in high income households.
Table 2. Mzwamadoda’s portfolio of funeral cover.
Type Member Description Monthly premium (US$) Benefits (US$)
Burial society Tembisa Pay in cash when someone dies 9.25 each time 2000
Burial society Tembisa Pay in kind when someone dies 7.70 each time 923
Burial society Mzwamadoda Pay monthly 15.40 1538
Burial society Tembisa Pay monthly 9.25 2230
Funeral plan Tembisa Pay monthly; well-known retail bank 4.60 3692
Funeral plan Tembisa Pay monthly; well-known retail bank 5.85 7385
Funeral plan Tembisa Pay monthly; unknown company 15.40 5846
Total benefits insured US$23 614
Total benefits required US$9230
INSURE 2.558](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-83-2048.jpg)
![Financial impact of HIV/AIDS Collins and Leibbrandt S79
90% 40%
80% 35%
70% 30%
60% 25%
50% 20%
40% 15%
30% 10%
20% 5%
10%
0%
100% 75%–100% 50%–75% < 50%
0%
Low income Medium income High income
Fig. 3. Percentage of householdsa that would lose income
Fig. 2. Percentage of adequately and inadequately insured from the death of the highest income earner, arranged by
households in each relative income group. Relative income is proportion of income per capita lost. aDoes not include
calculated in each area by dividing the sample into three households that are alone or are entirely dependent on
groups on the basis of per capita income. In Langa, low remittances.
income is defined as less than R577 per month, medium
income is defined as between R577 and R981, and high
income is greater than R981. In Lugangeni, low income is These observations can be generalized to quantify the loss
defined as less than R233, medium income is between R233 of income that might happen should the main income
and R639, and high income is greater than R639. In Diep- recipient die. Within the Financial Diaries sample, there
sloot, low income is defined as less than R534, medium are 128 income-generating families, 22 households that
income is between R534 and R1152, and high income is consist of one person living alone, and five households
greater than R1152. Adequately insured; inadequately that are entirely dependent on outside remittances. Only
insured. the 128 income-generating families would need to sustain
themselves should the main income earner die. Following
Bernheim et al. [16], the percentage difference between
The average level of INSURE is 2.26 in the high income per capita income with and without the main income
tier, 1.03 in the medium income tier and only 0.71 in the earner is calcuated. Figure 3 separates these households
low income tier. Therefore, households in the high or into different tiers based on the percentage of income lost
medium income levels in this sample are adequately from the death.
insured or even overinsured, whereas households in the
low income tier are underinsured. Moreover, within the The result is sobering. Over one third of these households
low income tier, only 24% have no insurance at all, so would lose 100% of their per capita household income. In
the majority (76%) of them does have insurance of some other words, the person who died was the sole income
kind, but just not enough. provider for the household and they would be left with no
income if this person died. Twenty-three per cent more
The financial impact of death does not end with the costs households would lose between 75 and 100% of their per
of the funeral. A crucial component of understanding capita household income, and another 22% would lose
the impact of death is measuring the forgone income between 50 and 75% of their per capita household
associated with the cessation of income activity as a result income. Therefore, 80% of the households in this sample
of death. As Figure 1 earlier in the paper showed, this would lose more than half of their per capita income with
income activity is not only wage earning, but also grant the death of their highest income earner.
receiving. In some situations, the household left behind
is cash-flow neutral after the death. For example, in In addition to this direct impact on household members,
Thembi’s household, discussed earlier, after her brother the Financial Diaries data show that many households
died, she was left without his income, but there was also support relatives who live outside their immediate
only one in the household now rather than two. In household. Nearly all wage earners in the sample give
another household, the main income before the death an average of 15% of their monthly income to someone
was a monthly grant of US$114 per month, split among outside the household. This amounts to contributions of
five people. After the death of this grant recipient, there between US$19 and US$35 worth of remittances every
was one less person to feed, but there was also the loss of a month. To outside households that are dependent on
main source of income. The remaining family of four had these remittances, this can be a substantial loss of income.
to get by on casual work done by the oldest daughter.
Their living standard deteriorated steadily through the How would a stock of assets or savings change this
remainder of the year, from an income per capita of picture? Only 10 households in the entire sample have life
US$70 to US$23. insurance. More common are provident or pension funds,](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-84-2048.jpg)
![Father figures: the progress at school of orphans in
South Africa
Ian M. Timaeusa and Tania Bolerb
Objective: To examine the progress in their schooling of maternal and paternal orphans
in a province of South Africa with high AIDS mortality and contrast it with that of both
children who lived in different households from their parents and children who resided
with their parents.
Methods: The KwaZulu-Natal Income Dynamics Study is a panel of households first
interviewed in 1993. The 1998 and 2004 waves of fieldwork collected 5477 reports on
children aged 8–20 years. We studied the determinants of the proportion of these
children who had completed 2þ grades fewer than expected for their year of birth using
both household fixed-effects models and difference-in-difference models fitted to
children reported on twice.
Results: Co-residence with a well-educated mother benefited children’s schooling, but
the fixed-effects models provide no evidence that maternal orphanhood or living apart
from their mother adversely affected children’s schooling. In contrast, both paternal
orphanhood and belonging to a different household from one’s father resulted in slower
progress at school. Although absence of the father was associated with household
poverty, this was not why it was associated with falling behind at school.
Discussion: Both the substantial benefits of living with their fathers for children’s
schooling and the limited importance of maternal orphanhood conflict with the results
of most studies in this issue, including those of other research in the same part of South
Africa. These findings caution against drawing general conclusions about the impact of
the AIDS epidemic from a few studies of geographically localized populations.
ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 7):S83–S93
Keywords: AIDS orphans, cohort study, educational achievement, South Africa
Introduction compare orphans with children who live in different
households from their mothers and fathers and children
One consequence of the AIDS epidemic in Africa has who reside with their parents.
been a dramatic increase in the number of children who
are orphans. The exact scale of the problem remains the In KwaZulu-Natal, HIV seroprevalence among antenatal
subject of debate, but UNAIDS estimates that were 12 clinic attendees rose from approximately 1% at the
million orphans aged less than 18 years in Africa in 2005 beginning of the 1990s to 41% in 2004 [3]. The mortality
who had lost one or both of their parents to AIDS [1]. of young adults from AIDS is already very high [4,5]. By
The prevalence of orphanhood rises rapidly with age. the time of the 2001 census, 6% of young people aged 5–
Most orphaned children are of school age and more than 19 years in the province had a dead mother and 18% a
half of all orphaned children are aged 12–17 years [2]. dead father (10% sample data).
This paper uses data from a household panel study to
examine the progress in their schooling of maternal and Some early studies of orphans in AIDS-affected African
paternal orphans in KwaZulu-Natal, South Africa, and to countries found rather equivocal evidence as to whether
From the aCentre for Population Studies, London School of Hygiene and Tropical Medicine, London, UK, and the bSection on HIV
and AIDS, UNESCO, Paris, France.
Correspondence to Ian Timaeus, Centre for Population Studies, London School of Hygiene and Tropical Medicine, Keppel Street,
London WC1E 7HT, UK.
Tel: +44 20 7299 4689; fax: +44 20 7299 4637; e-mail: ian.timaeus@lshtm.ac.uk
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S83](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-88-2048.jpg)
![S84 AIDS 2007, Vol 21 (suppl 7)
orphans were educationally disadvantaged [6,7]. Three impact of paternal orphanhood on schooling outcomes
comparative studies of Demographic and Health Survey remains statistically significant after controlling for other
(DHS) and other household survey data have, however, variables, the adverse impact of death of the mother is
found that orphans are either less likely to be enrolled in four to five times greater. The authors did not investigate
school [8,9] or are less likely to be at the correct grade for whether progress at school differs between children who
their age [10] than other children. A fourth such study live with their parents and children who have living
suggested that the effects of orphanhood are inconsistent parents but do not live in the same household as them.
and small [11]. The most sophisticated of the four studies
used household fixed-effects models to control for other
determinants of educational participation [8]. It found
that orphans are less likely to attend school than other Methods
children living in the same households. It concludes,
therefore, that their educational disadvantage cannot KwaZulu-Natal Income Dynamics Study
result solely from them being poorer than other children. The 1993 Project for Statistics on Living Standards and
Development was a nationally representative survey of
Cross-sectional surveys such as the DHS have serious households in South Africa with a design based on that of
limitations for assessment of the impact of orphanhood on the World Bank’s Living Standards Measurement Studies
children’s education. Few such surveys (and no DHS) [17,18]. The KwaZulu-Natal Income Dynamics Study
have measured money-metric poverty. In addition, if (KIDS) re-interviewed the African and Indian house-
children move between households after the death of holds from the Project for Statistics on Living Standards
their parent(s), information is seldom collected about the and Development in the province of KwaZulu-Natal in
household from which they come. It is likely that children 1998 to establish a panel study [19]. A third wave of field
are fostered into relatively advantaged households. Lower work took place in 2004 [20].
educational participation rates among orphans than other
children in the same households may thus predate the Accounts of the design and implementation of KIDS
illness and death of their parents. have been published elsewhere [19–21]. The question-
naire included a roster of household members and their
In the past few years several studies have been published characteristics and detailed modules on household
that use longitudinal designs to try and circumvent the income and expenditure. Each wave of the study
limitations of previous research on orphanhood and underwent ethical review and the fieldwork in 2004
education in Africa. The results of these investigations are was approved by the ethics committees of all three
rather mixed. Some find that the death of parents and universities involved in the project.
other adults has a large impact on primary school
enrolment or completion [12,13]; others have found In 1998, KIDS sought to interview all households
rather small effects [14]. Most of those studies, together containing ‘core’ members from 1993, that is the 1993
with some cross-sectional studies [10,15], suggest that the household head, his or her spouse and other adults who
educational disadvantages of maternal orphans are far had children in 1993. In order to refresh the panel and
greater than those of paternal orphans. improve the follow-up of children, the 2004 wave also
sought to interview ‘next-generation’ households, that is
The previous study of most relevance to the present one households established by the children of core members
examined the impact of parental death on schooling from 1993 who by 2004 were adults and had children of
outcomes using data from the Africa Centre for Health their own. It also administered a shortened questionnaire
and Population’s demographic information system to households fostering children aged less than 18 years
(ACDIS) in the Hlabisa subdistrict of northern Kwa- of core individuals.
Zulu-Natal [16]. It analysed data on nearly 20 000
children aged 6–16 years. Whereas paternal orphans in The 2004 wave of KIDS interviewed 1377 households
Hlabisa tend to live in poorer households than non- [20]. They arose from 793 of the 1354 households
orphans, no evidence exists that this results from interviewed in 1993. The fieldworkers located and
orphanhood. Moreover, paternal orphans’ low grades- interviewed 319 next-generation households (68% of
for-age are explained by their relative poverty, and they those identified) and 193 households containing foster
are no more likely than other children to drop out of children (41% of those identified). The high rate of
school. Maternal orphans, however, have significantly attrition of foster children reflects their mobility; many of
lower grades-for-age and higher dropout rates than non- them had already moved on by the time an interviewer
orphans living in the same households. The paper also visited the household identified as their home. In total,
presents cross-sectional models fitted to the 10% sample data were collected on at least one 2004 household for
data from the 2001 census for both KwaZulu-Natal and 62% of the baseline households interviewed in 1993.
South Africa as a whole [16]. The results are similar to Although the panel has suffered considerable attrition, the
those for Hlabisa. Although in the census dataset the age distribution of the resident members of the core and](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-89-2048.jpg)
![Progress at school of orphans Timaeus and Boler S85
next-generation households considered together still Africa remain at school until the age of 19 or 20 years.
approximates to that of the African and Indian population Therefore, we included these age groups in our analysis.
of KwaZulu-Natal [20].
Grade progression can be measured as either a continuous
The analysis presented in this paper was based on the 1998 variable, mean number of years behind at school, or a
outcomes of 2609 children born in 1977–1989 and the categorical one, the proportion of children who are
2004 outcomes of 2868 children born in 1984–1995, behind at school. We adopted the categorical formulation
giving a total of 5477 reports on children. for several reasons. Some children start school early and
are ahead of their birth cohort. It is unclear whether they
Definition of measures should be recoded to zero years behind at school or
KIDS distinguished between resident household mem- allowed to exert a compensating effect on the mean.
bers, who had slept there for at least 15 days in the past Second, more than half the school-age children in KIDS
month, and other members, who might have spent as had completed at least one fewer grade than they should,
little as 2 weeks in the past year residing with the often because they started school late. We are uncon-
household but nevertheless pooled resources with it when vinced that it is appropriate to consider all these children
they were present [17,19,20]. We examined the progress educationally disadvantaged. Third, the categorical
at school of both the resident and non-resident children indicator is less vulnerable to measurement errors than
who were members of panel households. the continuous indicator. Such errors would offset, at least
partly, the gains in statistical power that would be
Each wave of the study identified the parents of each obtained from working with a continuous outcome. In
household member, if they were resident or non-resident practice, modelling the two measures of educational
members of the same household as the index person, and progress led to very similar conclusions. The results
established whether they were alive or dead, if they were presented in the tables are all for the proportion of
not. Linking the responses to these questions about children who were 2þ grades behind in their schooling.
parents across the three waves reveals that sometimes, When appropriate, mention is made in the text of
when the person reported to be a parent in an earlier wave estimates for the continuous outcome.
had died, someone else was reported as the parent in a
later wave. This suggests that orphanhood is under- The economic status of households was measured using a
reported in cross-section. New orphans among those detailed series of questions on household expenditures. In
children contacted more than once were identified, particular, we used a poverty score calculated as the ratio
therefore, using the information collected in later waves of household expenditure per resident member (adjusted
on the survival of the first person reported to be the child’s for inflation) to a poverty line of R322 per month in 2000
mother or father. (US$47) [20,23].
In 2004, KIDS included detailed questions about Complete information is available on all of the
children’s schooling and administered numeracy and characteristics of children who were analysed, except
literacy tests to children aged 7–9 years. The analysis of their mothers’ education. The latter information is only
these data is reported elsewhere [22]. Less information available for waves of KIDS in which a child’s mother was
was collected on schooling in 1998, however, and only a member of the same household. Therefore, we
the question asked in successive household rosters on the modelled only whether the educational benefits of living
highest grade at school that each individual had with mothers were conditional on maternal schooling,
completed is available as an outcome if one wishes to not whether the educational outcomes of foster children
analyse the study in a way that makes full use of its and orphans were also affected by their mothers’
longitudinal nature. schooling. The educational background of the mothers
of 16 of the children who were co-resident is unknown
The South African school year coincides with the and those analyses that used this characteristic are based
calendar year. Children should start school at the on 5461 reports, not 5477.
beginning of the year in which they have their 7th
birthday and progress to the next higher grade each Statistical methods
subsequent January. Ideally, they matriculate from grade We first modelled the impact of orphanhood and
12, aged 18 years. By comparing actual progress with this separation from parents on grade progression by fitting
ideal, we established whether each child was in the correct a household random-effects model to the data [24]. Such
grade for his or her year of birth, or had fallen behind at models allow for the possibility that households may differ
school as a result of enrolling late, repeating grades or from one another in ways that affect children’s progress at
dropping out altogether. If children had completed fewer school but were not picked up by KIDS. They assume,
grades than is ideal, they are described in this paper as however, that these unmeasured characteristics of house-
‘behind in their schooling’ or ‘behind at school’. Because holds are unassociated with those that are included in the
they have fallen behind, many young people in South model explicitly.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-90-2048.jpg)
![S86 AIDS 2007, Vol 21 (suppl 7)
The second model presented is a household fixed-effects The proportion of children who were behind at school
model [24]. The model is estimated by comparing the dropped between 1998 and 2004. Although this seems
outcomes of children within the same household who encouraging, it may not accurately portray trends in
had differing characteristics or whose characteristics KwaZulu-Natal. Over time, the panel has probably
changed between waves. Such models control for all fixed become less representative of the African and Indian
characteristics of the children’s households, including population of the province. Moreover, the 2004 wave of
sources of heterogeneity in educational outcomes that KIDS followed up children in next-generation house-
KIDS did not measure explicitly. They have the holds and fostered children who would have been
disadvantage that they are less statistically efficient than excluded by design from the 1998 data.
random-effects models because no use is made of the data
on households that contained only solitary or hom- Using information on whether the first person identified
ogenous groups of children. to be the parent had died raises the estimate of the
proportion of the children who were maternal orphans in
Third, we present an individual fixed-effects model of the 2004 to 13%, compared with a cross-sectional estimate of
determinants of grade progression between 1998 and 11%, and the estimate of paternal orphanhood to 26%,
2004 among the 925 children born in 1984–1989 who compared with 22% initially. In 1998, 21% of the children
were observed in both waves. They include 286 children and in 2004 33% of them had lost at least one parent. Only
who switched between categories of the outcome just over half the children were living with their mothers
variable between the waves. on a day-to-day basis and only a minority of them were
with their fathers.
In 1998, 67% of children and in 2004 62% of them were
Results living in poor households. The rise in median household
income between the waves only benefited those children
Table 1 presents some descriptive information on the who ended up in next-generation and foster households.
school-age children and young people in KIDS house- The proportion of urban households that were poor was
holds in 1998 and 2004. Over 80% of them were enrolled half that of rural households. Only half the children who
in school and 97% of the children aged less than 17 years lived with their father belonged to poor households,
were enrolled. Many of the children were 2þ years compared with approximately 70% of other children.
behind in their schooling. Some of these children had Separation from or death of the mother was not associated
started school late or dropped out, but many of them had with child poverty.
fallen behind because they repeated grades. By the time
that they should have been in the top three grades of Table 2 presents the main results of the analysis. Looking
secondary school, the majority of children were 2þ years first at the random-effects model, Indian children had
behind in their schooling. only one quarter the odds of being behind at school in
Table 1. Characteristics of children and young people of school age who were members of KIDS households in the 1998 and 2004 wavesa.
Characteristic Coding 1998 Wave 2004 Wave
% of children 2þ years behind at school by expected grade for year of birth Expected grade 2 to 5 17 12
Expected grade 6 to 9 35 25
Expected grade 10 to 12 55 51
% Distribution according to mother’s residence and survival Resident household member 65 56
Non-resident member 10 4
Not household member 19 27
Dead 6 13
% Distribution according to father’s residence and survival Resident household member 37 30
Non-resident member 13 5
Not household member 33 40
Dead 17 26
% Distribution according to residence Rural 72 76
Urban 17 15
Metropolitan 11 9
% Indian 7 5
Household expenditure per head as % of the poverty line (R322 in 2000) 10th Percentile 28 26
Median 67 75
90th Percentile 206 292
Mean number of children per household 3.9 3.7
Number of children 2609 2868
a
Children and young people born in 1977–1989 in 1998 and born in 1984–1995 in 2004, that is those who are aged 8–20 years approximately.
See text for explanation.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-91-2048.jpg)
![Progress at school of orphans Timaeus and Boler S89
Table 3. Odds of being 2R grades behind at school for all orphans and orphans reported on the roster (household fixed-effects model, controlling
for year of birth, n U 4197 children in 580 households).
All orphans Roster orphans
Variable Coding Odds ratio 95% CI Odds ratio 95% CI
Household expenditure per head ln (poverty score) 0.84 (0.71–0.99) 0.84 (0.71–1.00)
(relative to poverty line)
Mother’s residence in household and Non-resident member 1.08 (0.74–1.57) 1.31 (0.95–1.81)
survival (reference category: resident)
Not a member 1.13 (0.89–1.42) 1.20 (0.94–1.52)
Mother dead 1.21 (0.88–1.67) 1.08 (0.78–1.50)
Father’s residence in household and Non-resident member 1.23 (0.82–1.85) 1.18 (0.81–1.72)
survival (reference category: resident)
Not a member 1.67 (1.27–2.20) 1.58 (1.21–2.06)
Father dead 1.70 (1.26–2.30) 1.64 (1.23–2.19)
Sex of child (reference category: boys) Girls 0.44 (0.37–0.53) 0.45 (0.38–0.53)
Wave of KIDS (reference category: 1998) 2004 0.59 (0.49–0.71) 0.63 (0.53–0.76)
CI, Confidence interval.
Unsurprisingly, the oldest children, particularly older manifest some years later when the affected household
boys, were least likely to remain in panel households by finally exhausts its savings and external sources of
2004. Attrition was also high among the few children assistance. Endogeneity is also a major issue. Children
who were not resident in the household that reported on may become poor because they are orphaned or be
them in 1998. No evidence exists of greater attrition of orphaned because their families are poor. In addition,
orphans, of the poor, or of urban residents. Children who adult death is a risk factor for household dissolution [25].
belonged to a different household from their mother in
1998, however, suffered higher attrition than other Such complexities suggest that only panel studies can
children. No doubt many of them left the panel hope to assess the household-level impact of the AIDS
household to move in with their mothers. epidemic, identify the determinants of vulnerability and
evaluate interventions intended to mitigate impact.
Unfortunately, panel studies have their own limitations.
They are prone to differential attrition of the poorest and
Discussion most vulnerable households. Moreover, household-based
studies alone are a poor source of data on the quality of the
One methodological issue in the study of the impact of services available to households (e.g. schools).
AIDS and other adult deaths on other household
members, including children, is that the impact of having The results presented here are based on a panel study in
an AIDS case in the household may be felt before the KwaZulu-Natal. We found that rural children were
death, when the person first becomes ill, or only become approximately twice as likely as urban children to be badly
Table 4. Risk factors for attrition between 1998 and 2004, children born in 1984–1989 (n U 1350).
Variable Coding Odds ratio 95% CI
Race (reference category: African) Indian 1.09 (0.54–2.20)
Residence in 1993 (reference category: rural) Urban 1.20 (0.76–1.90)
Metropolitan 1.09 (0.57–2.09)
Household expenditure per head in 1998 (relative to poverty line) ln (poverty score) 0.98 (0.77–1.24)
Mother’s residence and survival in 1998 (reference category: resident) Non-resident member 1.11 (0.66–1.87)
Not a household member 1.78 (1.25–2.53)
Mother dead 1.47 (0.81–2.68)
Father’s residence and survival in 1998 (reference category: resident) Non-resident member 0.75 (0.45–1.25)
Not a household member 1.00 (0.69–1.45)
Father dead 1.47 (0.92–2.33)
Year of birth (reference category: 1989) 1984 1.33 (0.90–1.95)
1985 1.32 (0.93–1.86)
1986 0.65 (0.42–1.01)
1987 0.88 (0.59–1.30)
1988 1.00 (0.67–1.50)
Sex of child (reference category: boys) Girls 0.85 (0.66–1.08)
Residence in 1998 (reference category: resident) Non-resident 2.14 (1.36–3.36)
CI, Confidence interval.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-94-2048.jpg)
![S90 AIDS 2007, Vol 21 (suppl 7)
behind at school, but that household poverty only had a way fully through the child population. Therefore, the
small impact on children’s progress at school once number of orphans in South Africa will continue to grow
residence and other confounders were allowed for. This rapidly for many years [29]. Our results suggest that this
finding suggests that the educational disadvantage of rural trend represents a serious threat to efforts to improve the
children is something that, in principle, the government levels of education of young South Africans.
could aspire to correct.
The data collected in KIDS provide only hints as to the
We also found that girls were only approximately half as mechanism by which co-resident fathers promote their
likely to be behind at school as boys. This pattern is children’s schooling. Their children were better off than
consistent with the findings of previous research in other children. The household fixed-effects model,
southern Africa [26]. It should not be taken to imply that however, suggests that changes in households’ incomes
schools in South Africa are free of attitudes and practices only exerted a small influence on whether the children in
that discriminate against girls [27]. The superior them dropped behind at school. The economic benefits
educational achievement of girls has not, however, always of co-residence with fathers thus explain little of its
received the recognition that it should in the literature on beneficial impact on children’s progress at school. In
gender and schooling in this region. addition, although co-residence with fathers was more
common in urban households, the estimated benefits of
Our findings concerning the impact of orphanhood and having a father in the household persist in the household
separation from parents on children’s schooling differ fixed-effects model. It seems unlikely, therefore, that
from those of most other studies in Africa. On the one school quality and journey times are important con-
hand, KIDS indicates that co-residence with fathers founders of the co-residence–schooling relationship.
greatly benefited children’s progress at school. On the
other, it provides no evidence that co-residence with Some of the children with fathers who were alive, but not
mothers improved children’s progress at school unless members of the same household, will have lost touch with
the mother had been to secondary school herself. The them altogether. The fact that not only orphans but
impacts of separation from and death of the father were children with absent fathers did less well at school,
similar, except that the panel analysis suggests that although many of them had never lived with their fathers,
children whose fathers died between 1998 and 2004 may suggests that, although witnessing the illness and death of
have completed fewer grades during these years than their father is obviously distressing for any child, it is the
children who became separated from their father. This lack of a father–child relationship that is more serious in
could reflect the short-term impact of the shock the longer run. As the amount of schooling completed by
of bereavement. co-resident fathers themselves did not significantly affect
whether their children were behind at school, it seems
KIDS data on children in core and next-generation unlikely that men who live with their children are
households remain broadly representative of KwaZulu- unusually strong advocates for education. It is probably
Natal [20]. They indicate that, by 2004, approximately more generic supportive and directive aspects of fathering
25% of children of school age had dead fathers. Half as that benefit children’s schooling [30–32].
many children again, 37%, were living in households to
which their father did not belong. Unfortunately, both The results presented here appear to conflict with those of
domestic situations became more common between 1998 Case and Ardington [16], who conducted a similar
and 2004. Annual deaths of adults who were aged 20–44 analysis of data from a largely rural area of KwaZulu-
years in 1998 increased fivefold during the 6 years Natal. KIDS and ACDIS are two of the most detailed and
between these two waves of KIDS [20]. Although the well-conducted longitudinal studies from anywhere in
study did not collect data on HIV status or precise causes Africa that can be used to study the impact of orphanhood
of death, this rise undoubtedly reflects the rapid growth in on children’s educational outcomes. The fact that the two
AIDS mortality in South Africa [4,28]. Only 32% of datasets yield different results is sobering.
deaths of men aged 20–44 years in 1998–2003 and 6%
of those of women resulted from injuries, and there was The apparent contradiction between the findings of the
no clear trend in the number of injury deaths across this two studies concerning mothers probably arises because
period. KIDS is too small a study to pick up the moderately sized
effects of maternal orphanhood on schooling that Case
The rate of increase in the prevalence of HIV infection in and Ardington [16] identified with much larger datasets.
South Africa has slowed since 2004 [3]. In addition, a Using a household fixed-effects model, they estimated
growing number of AIDS patients in the country now from ACDIS that maternal orphans have completed 0.12
receive antiretroviral therapy and this may be moderating years less schooling than other children, with a standard
the upward trend in mortality. Although these are error of 0.05 (see Table 5 in Case and Ardington [16]), and
welcome developments, it takes 18 years for the effect on from the 2001 census sample that maternal orphans have
orphanhood of any change in adult mortality to work its 0.22 years less schooling than other children, with a](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-95-2048.jpg)
![Progress at school of orphans Timaeus and Boler S91
standard error of 0.03 (see Table 9 in Case and Ardington should have been concentrated among those orphans
[16]). Fitting an equivalent model to KIDS suggests that who were making poor progress at school. If anything,
maternal orphans have 0.15 years less schooling than one might expect those paternal orphans who had
other children, but with a standard error of 0.09. The acquired a new ‘father’ to make better progress at school
effect is of similar magnitude but statistically insignificant. than other paternal orphans. We thus remain open-
minded as to whether underreporting of paternal
Moreover, whereas the design of the two analyses is orphanhood is the main explanation of the inconsisten-
similar, it is not identical. Case and Ardington [16] can cies between our results and those of Case and Ardington
control only for previous asset poverty, not previous [16].
income poverty. Our results did not, however, suggest
that previous poverty confounds the relationship between Another plausible explanation of why our findings about
paternal orphanhood and progress at school. Perhaps fathers differ from those obtained from the ACDIS data is
more importantly, they analysed data on children aged not bias in either set of results but that paternal
6–16 years who had been followed up in ACDIS for orphanhood has less impact on children’s schooling in
2.5 years on average. Our analysis included all children the Hlabisa subdistrict than in KwaZulu-Natal as a whole.
and young people aged 9–20 years and has a longitudinal Of course, this cannot explain why Case and Ardington
component that involves the follow-up of children aged [16] obtained much smaller estimates of the impact of
9–14 years for 6 years. Children’s schooling may suffer paternal orphanhood in the province from the 2001
most when their parent is ill and recover after census data than we do from KIDS. Some insight into the
orphanhood [14]. No evidence of this exists for maternal importance of place of residence can be obtained from
orphans in Hlabisa [16]. Nevertheless, if the impact of a KIDS, however, by examining how residence interacts
father’s death in KwaZulu-Natal emerges only in the with the father variable in the household fixed-effects
longer run, it is possible that the difference between the model in Table 2. Fitting this more complex model
follow-up periods considered by the two studies could reveals that, although having one’s father living in the
explain their contrasting findings. same household was of significant benefit to children’s
progress at school in both rural and urban areas, the
One problem that might affect the validity of the KIDS adverse impact of paternal orphanhood (but not
results is attrition of the panel. Those orphans with the separation) on children’s grade attainment was smaller
most disrupted family lives and children from the most in rural households than in urban households (odds ratio
impoverished households are probably more prone to loss 0.5, P ¼ 0.04; model not shown). Extrapolating from this
to follow-up than other children. Attrition may be less of finding, the educational benefit to a child of co-residence
a problem in ACDIS. Children who move within the area with his or her father probably varies between different
covered by ACDIS should be picked up in their new rural areas within KwaZulu-Natal. In some disadvantaged
household even though children who move outside the areas, perhaps including Hlabisa, fathers may be unable to
surveillance area will still be lost to follow-up. The do anything effective to promote their children’s
analysis of attrition in KIDS presented in Table 4 is not schooling. In such areas, the death of their fathers might
definitive. A subset of maternal orphans who make poor have little impact on children’s progress at school.
progress at school could exist that it is so difficult to track
that they are largely unrepresented in Table 4. This seems
unlikely. It also seems unlikely that KIDS has suffered Conclusion
disproportionate attrition of those paternal orphans who
are progressing well at school. Therefore, we doubt that The analysis of KIDS shows that fathers matter for
the findings from KIDS are invalidated by attrition bias. children’s schooling. Children whose fathers were
members of their household were considerably less likely
One bias that is probably more severe in both ACDIS and than other children to be behind at school. This is not
the 2001 census than in KIDS is the underreporting of because the households of orphans and other children
orphanhood. The former data sources lack the historical who were not living with their fathers were poorer than
depth that enabled us to identify unreported orphans in those of children whose fathers belonged to the same
the KIDS data. According to the 2004 wave of KIDS, household. It is the relationship that children have with
10% of children aged 6–16 years in core and next- their fathers that seems to be important. Our results are
generation households had dead mothers, compared with consistent with those of previous research that found that
9% in 2003–2004 in ACDIS, and 21% had dead fathers, maternal orphanhood has some adverse impact on
compared with only 15% in ACDIS [16]. Although it is children’s progress at school in KwaZulu-Natal, but do
impossible to determine exactly how many children not provide positive evidence in support of this finding.
should have been reported as orphans in each study, it is
likely that ACDIS has failed to detect a substantial A considerable body of research from other parts of the
minority of paternal orphans. Nevertheless, it is not world suggests that fathers (or at least father figures)
obvious why the underreporting of paternal orphanhood benefit children as well as mothers [32,33]. Although little](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-96-2048.jpg)
![Exploring the Cinderella myth: intrahousehold
differences in child wellbeing between orphans and
non-orphans in Amajuba District, South Africa
Anokhi Parikha, Mary Bachman DeSilvab, Mandisa Cakwea,
Tim Quinlana, Jonathon L. Simonb, Anne Skalickyb and
Tom Zhuwaua
Objective: To determine whether differences in wellbeing (defined by a variety of
education and health outcomes) exist between recent school-aged orphans and non-
orphans who live in the same household in a context of high HIV/AIDS mortality in
KwaZulu Natal, South Africa.
Design: The data come from the first 2 years (2004–2006) of an ongoing 3-year
longitudinal cohort study in a district in KwaZulu-Natal, the Amajuba Child Health
and Well-being Research Project. Using stratified cluster sampling based on school and
age, we constructed a cohort of 197 recent orphans and 528 non-orphans aged 9–16
years and their households and caregivers. Household heads, caregivers, and children
were interviewed regarding five domains of child wellbeing: demographic, economic,
educational, health/nutrition/lifestyle, and psychosocial status.
Methods: The analytical sample consists of 174 children (87 orphans and 87 com-
parable non-orphans who live together) at baseline and 124 children in round 2. We
estimated a linear regression model using household fixed effects for continuous
outcomes (grade adjusted for age, annual expenditure on schooling and body mass
index) and a logit model using household fixed effects for categorical variables
(malnutrition) to compare co-resident orphans and non-orphans.
Results: We found no statistically significant differences in most education, health and
labour outcomes between orphans and the non-orphans with whom they live. Paternal
orphans are more likely to be behind in school, and recent mobility has a positive effect
on schooling outcomes. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 7):S95–S103
Keywords: caregivers, children, HIV, intrahousehold, mortality, orphans,
South Africa
Introduction currently 1.2 million AIDS orphans in South Africa [2],
and this number is expected to peak at a staggering 2.3
With an antenatal seroprevalence of 40.7% in 2005, million in 2015 [3]. As such, it is no surprise that the issue
KwaZulu-Natal has the highest HIV prevalence of any of orphaning has attracted significant attention.
province in South Africa, a country with 5.1 million
individuals infected [1,2]. AIDS-related mortality is high, Much of this attention has characterized orphans as
and the consequent impact on orphaning is likely to children who are growing up without the care and
be dramatic in the years to come, irrespective of the support of their families, who have poorer learning and
expansion of the antiretroviral programme. There are knowledge levels, and who are suffering from the
From the aHealth Economics and AIDS Research Division (HEARD), University of KwaZulu Natal, Durban, South Africa, and the
b
Boston University School of Public Health, Center for International Health and Development, Boston, Massachusetts, USA.
Correspondence to Anokhi Parikh, HEARD, Private Bag X 54001, Durban, 4000, South Africa.
E-mail: anokhip@gmail.com
ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S95](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-100-2048.jpg)
![S96 AIDS 2007, Vol 21 (suppl 7)
‘absence of adults in their socialization’ [4]. The data national 2003 General Household Survey, approximately
reveal that most orphans, defined in the AIDS literature as two-thirds of all orphans in KwaZulu-Natal live in such
having one or more parents dead, in sub-Saharan Africa households. These households are of particular interest
have at least one parent living and either live with their also because the co-resident non-orphans make a natural
surviving parent or are absorbed into other families where comparison group for orphans as they both share
they have some adult supervision. household characteristics (e.g. household income, literacy
of caregiver etc.).
Nevertheless, the death of a parent (and the potentially
long illness preceding it if the parent has died of AIDS) In one of the few studies to make intrahousehold
may have various impacts on a child’s wellbeing. comparisons, Case and Ardington [15] showed that in
Numerous studies have accordingly examined the impact Hlabisa district in KwaZulu-Natal, maternal orphans
of orphaning on children, but have focused primarily on were more likely to be behind in school and had less spent
educational outcomes (commonly defined by school on their schooling, but were equally likely to be enrolled
enrolment), with few studies looking at any other aspect as the non-orphans with whom they lived. Implicit in the
of wellbeing such as health or labour outcomes. finding that orphans are worse off when compared with
other children is the notion that caregivers may prioritize
Although the empirical evidence is mixed and depends their own children over the fostered child, a Cinderella
on the location, data sources, and methods used, two approach, so to speak. For example, when facing income
broad themes emerge in the literature: that the impact of constraints, a caregiver might spend less on the fostered
orphaning depends often on which parent dies and that child than on his/her own child (although perhaps with
income is often a greater predictor of outcomes than less malice than Cinderella’s stepmother). Gender bias in
orphan status. Case et al. [5] used the Demographic and educational expenditures has been well documented in
Health Survey data from 19 countries in Africa and found Asia; a similar bias could also apply to orphans, but very
systematic differences in school attendance between little research has examined the intrahousehold aspect of
orphans and non-orphans. The finding of difference in this question to date.
attendance and enrolment between orphans and non-
orphans has been supported by other studies in rural In summary, the research to date is limited in two ways.
Kenya through longitudinal studies [6,7], and in cross- First, few studies explicitly assess intrahousehold differ-
sectional studies elsewhere in sub-Saharan Africa [8]. ences between orphans and non-orphans. Second, both
Beegle et al. [9], using a panel that follows children for intra and interhousehold comparisons have used limited
13 years, showed that maternal orphanhood was indicators such as school enrolment, grade progression,
associated with lower educational attainment and health and height, and thus do not provide a comprehensive
(as measured by height) in the long term, and paternal picture of other important aspects of child wellbeing.
orphanhood was associated with lower educational
attainment for certain groups only. Using Demographic To respond to these shortcomings, this paper compares
and Health Survey data and other nationally representa- the differences in wellbeing between orphans and non-
tive household surveys from 51 countries in Africa, orphans who live with each other using longitudinal data
Ainsworth and Filmer [10] highlighted the variation in from 2004–2006 from Amajuba District in KwaZulu-
orphan/non-orphan differentials across countries and Natal. This paper addresses the question ‘do orphans and
found that income plays a greater role in determining non-orphans living within the same household fare
school enrolment than orphaning. differently in terms of wellbeing?’ We define ‘wellbeing’
to include education, health, and labour outcomes. We
A lack of statistically significant differences in enrolment conclude with potential explanations for the observed
have been found by other researchers conducting results and trends.
longitudinal studies in East Africa [11–13]. Chatterji
et al. [12], using longitudinal data from Rwanda, showed
no differences in school enrolment and food intake
between orphans and non-orphans. Adato et al. [14] Methods
found no statistically significant differences in schooling
indicators but qualitatively did find some cases of Study area
discrimination towards orphans within the household. The Amajuba district was chosen because it included a
broad cross-section of urban, peri-urban and rural areas.
Many of these studies consider orphans and non-orphans The district has approximately 470 000 inhabitants and is
in general, but do not distinguish between orphans who poor. The economy used to be driven by the coal mining
live with other orphans and orphans who live with non- industry, but the closure of coal mines has led to high
orphans in mixed households (with some exceptions) unemployment in the region and consequently high rates
[6,7,15]. Moreover, they are an important type of of migration. Additional details are provided elsewhere
household, as according to the Statistics South Africa’s (Bachman DeSilva et al., manuscript in preparation).](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-101-2048.jpg)
![S98 AIDS 2007, Vol 21 (suppl 7)
orphan, double orphan); resident parent is whether the mately half the sample had repeated a grade once. Average
surviving parent of the orphan or the parent of the non- body mass was within normal range and was comparable
orphan is living at home (father lives at home, mother for orphans and non-orphans. Approximately 10% of the
lives at home). Hj is the child invariant household fixed children work outside the house and 91% report assisting
effect; and eijt is the error term. Several models were with chores in the household (with no differences
estimated using the different independent variables in between orphans and non-orphans). Although mobility
different combinations, but results from only one such of the sample was high, it was equally high for orphans
model are presented in this paper. and non-orphans, with approximately 30% of the sample
having moved at least once and approximately 12% in the
For categorical variables, we estimated a logit model: past 2 years. Living arrangements were different, however,
with a parent being the primary caregiver for only 39% of
PðYijt ¼ 1Þ the non-orphans and 12% of the orphans. Grandparents
¼ Fðb1 maleij þ b2 ageijt þ b3 mobilityijt were primary caregivers for 56% of the orphans and
43% of non-orphans, although this difference was not
þ b4 orphan typeijt þ b5 resident parentijt statistically significant. These findings of non-difference
remained unchanged in round 2 (results not shown).
þ Hj þ eijt Þ Table 2 shows the demographic characteristics of the
where Yij represents whether child i from household j at time sample that was lost in round 2. None of the
t is malnourished or not/works at home or not/does chores characteristics listed were significant predictors of
at home or not; and the remaining variables are as defined attrition (results not shown).
above.
Of the orphans at baseline, 13 were maternal-only
The decision to include the variables on mobility and orphans, 30 were paternal-only orphans, 26 were double
resident parent was informed by the literature. This orphans, and 19 had missing information on which parent
literature shows that fostering and mobility are high even had died. This changed to nine maternal-only orphans,
for all kinds of children in South Africa because of high 21 paternal-only orphans, 21 double orphans, and 12
levels of adult migration and children born out of with missing data in the following year. Table 3
wedlock [14,16,17]. We thus feel it is important not only summarizes the demographic and socioeconomic charac-
to introduce these variables as controls. teristics of children by orphan type.
Examining children who are co-resident using household As mentioned in the methodology section, several
fixed effects allows us to control for common household models, each controlling for a different combination of
characteristics. It is impossible to know, however, with independent variables, were estimated to examine the
these data, whether the children being compared were orphan/non-orphan differentials across different out-
indeed comparable before the death of the parent because comes. Table 4 shows results from two such models. The
we do not have information on the orphan’s character- results were consistent across specifications: paternal
istics before being orphaned (which may have themselves orphans are more likely to be behind in school than non-
been affected by orphaning). We are also unable to orphans with whom they live, they are on average a third
consider the fixed and unobserved characteristics of the of a year behind in their grade. Maternal orphans are on
child him/herself, even though this is longitudinal data, as average half a year behind in schooling, but this effect is
most of the variables of interest have remained constant statistically insignificant. Mobility within the past 2 years
over time. As a result of this limitation of the data, this is seen to have a positive effect on grade progression.
paper does not try and isolate the impact of parental death Orphanhood does not seem to have any effect on
on children. Rather, it compares orphans and non- expenditures on schooling. Recent mobility is associated
orphans on a range of indicators and tries to identify some with a substantial negative effect on schooling expendi-
causal pathways for the results. ture. The impact of parents being present at home is
insignificant (results not shown).
Bivariate analysis demonstrates no significant differences
Results in nutrition, health proxy, and labour outcome indicators
such as going to bed hungry the previous night, being sick
At baseline, bivariate analysis of sociodemographic in the past 6 months, and working both within and
characteristics shows few differences between the 87 outside the house (see Table 1). Analysis of BMI,
orphans and 87 non-orphans in the analytical sample presented in Table 4, shows that BMI is lower for orphans.
(Table 1). There are no statistically significant differences This is also robust when controlling for mobility. As
between the demographic, educational, health, or labour malnutrition also manifests itself as obesity, however,
outcomes between orphans and co-resident non- lower BMI is not necessarily informative, especially in
orphans. Whereas attendance was near 100%, approxi- adolescents. Logistic analysis (Table 5) shows that](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-103-2048.jpg)
![Wellbeing of co-resident orphans and non-orphans Parikh et al. S101
Table 4. Educational and health outcomes for orphans and co-resident non-orphans (household fixed effects).
Grade normalized Grade normalized Annual expenditure Annual expenditure Body mass
for age for age on school on school index
Age (in years) À0.06b À0.07b À2.18 À1.59 0.77b
(0.01) (0.01) (38.25) (37.40) (0.10)
Sex (male) À0.02b À0.02b 23.19b 25.75b À3.11b
(0.00) (0.00) (8.74) (8.46) (0.45)
Changed residence in past 2 years 0.05b À170.11b
(0.02) (55.93)
Changed residence more than 2 years ago À0.03 106.43a
(0.02) (48.28)
Maternal-only orphan À0.04 À0.03 À11.14 À54.29 1.5a
(0.02) (0.02) (60.97) (60.74) (0.73)
Paternal-only orphan À0.04a À0.04b À5.15 6.53 À1.12a
(0.01) (0.01) (40.88) (40.25) (0.48)
Double orphan 0 À0.01 À61.42 À27.76 À1.71b
(0.01) (0.01) (42.74) (42.27) (0.50)
Constant 1.24b 1.25b 120.89 84.97 11.29b
(0.04) (0.04) (113.95) (110.35) (1.32)
Observations 296 296 298 298 298
Number of field code 87 87 87 87 87
R-squared 0.28 0.32 0.04 0.12 0.42
a
Significant at 5%.
b
Significant at 1%.
Standard errors in brackets.
Variable with missing orphan type included but not shown.
maternal and double orphans are at greater odds of being coefficient on the father being a resident within the
malnourished than non-orphans but this is not statistically household is insignificant (results not shown), suggesting
significant. Maternal and paternal orphans are at greater that the impact of paternal orphanhood may be caused by
odds of doing chores within the house and at lower the fact that the death of a father could be an economic
odds of working outside the house compared with co- shock that may have, at some point, resulted in children
resident non-orphans, but again these differences are not dropping out of school. We, unfortunately, have no way
statistically significant. to test this hypothesis and can only offer it as a
potential explanation.
What could explain the lack of overall differences
Discussion between orphans and non-orphans? First, temporal issues
may be driving the result. It is important to remember
The results show some statistical differences in edu- that we are merely looking at incident (recent) orphans,
cational outcomes and no differences for health and i.e. children that have lost at least one of their parents in
labour outcomes between orphans and non-orphans who the 6 months before the survey and in the same year the
live in the same households. survey was conducted. This may not be sufficient time to
see a large effect on children, or the households may have
Case and Ardington [15], in their intrahousehold analysis effective short-term coping mechanisms to mitigate the
also set in KwaZulu Natal (albeit in a poorer district), effect on children [9]. On the other hand, one can argue
found that maternal orphans are ‘on average, 0.12 of a that the critical period for an orphan child is the terminal
year behind in their schooling and have 7% less spent on illness period as a result of the trauma of seeing a parent
their education’ compared with the non-orphans with wasting away and sometimes having to miss school in
whom they live. Although the differences in expenditure order to attend to sick parent(s), and that the impacts may
are moderate, the magnitude of the difference between diminish over time.
orphans and non-orphans in terms of schooling is small:
0.12 of a year behind equates to orphans being behind by Second, Case and Ardington [15] have argued that
a little over a month. They found no difference for differences in outcomes between orphans and non-
paternal orphans. Our results, on the other hand, show orphans are driven by the tendency to live with distant or
that paternal orphans are behind in school. unrelated caregivers. All the orphans in the sample are
either living with close relatives (aunt or grandmother) or
Given the fact that a large proportion of South African their surviving parent, as in a study from the same
fathers are absent or not linked to the household [16], the province by Adato et al. [14]. Therefore, they are likely to
significant effect of paternal orphanhood is curious. The receive the same support as the non-orphans with whom](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-106-2048.jpg)
![S102 AIDS 2007, Vol 21 (suppl 7)
Table 5. Health and labour outcomes for orphans and co-resident non-orphans (logistic model with household fixed effects).
Child is malnourished Child works outside the house Child does household chores
Odds ratio 95% CI Odds ratio 95% CI Odds ratio 95% CI
Age (in years) À0.29 (À0.81–0.23) 0.01 (À0.31–0.32) À0.14 (À0.46–0.18)
Sex (male) 0.01 (À1.50–1.52) 0.11 (À1.37–1.59) À1.54 (À3.33–0.24)
Maternal-only orphan 0.48 (À1.88–2.83) À0.24 (À2.28–1.81) 0.69 (À1.77–3.15)
Paternal-only orphan À0.18 (À2.29–1.92) À0.03 (À1.41–1.35) 0.25 (À1.22–1.72)
Double orphan 0.56 (À1.13–2.24) À0.72 (À2.41–0.96) À0.95 (À2.61–0.72)
Observationsa 54 86 96
Number of groups 16 24 26
CI, Confidence interval. aMultiple positive/negative outcomes were encountered within groups and thus the groups were dropped from
the regression resulting in fewer numbers of observations. Standard errors in brackets. Variable with missing orphan type included but not
shown.
they live. With time, destination households may become moved houses in the preceding year, 13 orphans and 10
oversaturated and could struggle to absorb more children non-orphans. Moreover, in round 2, sample attrition was
and this may change. South Africa’s extensive social grants equally high for orphans and non-orphans. Therefore,
system potentially mitigates against this phenomenon and although there may be endogeneity in placement
assist families in coping. decisions, we do not believe it is disproportionately so
for orphans when compared with non-orphans. Child
Third, in Amajuba District’s context of high adult migration is a historical/cultural phenomenon, and
migration, having a parent alive does not equate to the fostering literature shows how children have lived away
presence of a parent at home, thus orphanhood itself may from their ‘nuclear’ families (although this may be
not be associated with lower educational or health exacerbated by AIDS mortality) [17].
outcomes. The majority of both orphans and non-
orphans live without parents present at home. Table 1 Fifth, it is possible that the indicators used may not be
indicates that only 38.64% of non-orphans have parents as sensitive to differences, particularly because the orphans
primary caregivers, and a large proportion of both were so recently orphaned. There may be some
orphans and non-orphans live with their grandparents. limitations of BMI, but in general it is difficult to
Even single parent orphans tend not to live with their identify good measures of health of older/school-aged
surviving parent. Migration for employment was the most children because this age group is generally very healthy
frequently cited reason for parents’ not living at home. (self-reported or otherwise). In terms of schooling, there
Furthermore, approximately a third of the fathers were maybe differences in performance within a grade that are
not living at home because they were not married to the not captured by these instruments. Whether an orphan
mother. This figure also calls into question the role of child is truly learning, as opposed to progressing, like
biological parents (especially fathers) in caregiving, and non-orphaned children, may not be fully captured by
supports other literature that shows that the absence of the data.
fathers is high in South Africa, with 55% of fathers being
absent in rural South Africa in 2002 [16]. Current The study has some limitations worth mentioning that
definitions of orphan inaccurately privilege the biological may bias the results towards the null. First, the tests have
parent in a context in which even non-orphans do not low power as a result of the relatively small sample size,
live with their parents. This calls into question our and this may contribute to not finding statistically
thinking on the category of orphan in South Africa. significant effects. The attrition in round 2 only further
reduced the sample. Comparisons with much larger
Fourth, when orphans have moved from their original studies should be made with this in mind.
households (i.e. they were fostered into the survey
household, which has non-orphans), there may be Second, the study sample was drawn from a random
endogeneity in placement decisions, in that orphans sample of schools in the district. Using schools as our sole
are strategically moved to better-off households and this recruitment source for study participants was both
may bias the results towards the null. The positive and methodological and practical. Drawing a sample of
significant coefficient on recent mobility (within the past school-going ‘recent’ orphans and non-orphans intro-
2 years) supports the idea that children are often moved duces a sampling bias that potentially biases the
for schooling. It is important to note, however, that intrahousehold results towards the null as worse-off
mobility is equally high for both orphans and non- orphans may have been excluded from the sample. What
orphans. Table 1 shows how there is no statistically is important to note is that school enrolment rates in
significant difference between orphans’ and non-orphans’ KwaZulu-Natal are extremely high. The national 2003
mobility. Of the 174 children at baseline, 23 children General Household Survey conducted by Statistics South](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-107-2048.jpg)
![Wellbeing of co-resident orphans and non-orphans Parikh et al. S103
Africa showed school enrolment rates to be 97.5% for United States National Institutes of Health under its
non-orphans and 95% for orphans. Upon further analysis African Partnerships programme (grant R29
of this survey, we found orphans who are not enrolled HD43629).
neither worse off than the non-orphans with whom they
Conflicts of interest: None.
live nor are they are worse off compared with enrolled
orphans in terms of health and labour outcomes. Their
schooling outcomes do differ, however, and this may bias References
the results towards the null.
1. Department of Health. National HIV and Syphilis Antenatal
Sero-Prevalence Survey in South Africa 2005. Pretoria: National
Finally, Evans and Miguel [6] argued that studies that do Department of Health, South Africa; 2006.
not take into consideration child fixed effects are likely to 2. UNAIDS. 2006 Report on the global AIDS pandemic. Geneva:
UNAIDS; 2006.
be biased to the null because they are unable to account 3. Dorrington R, Bradshaw D, Johnson L, Budlender D. The
for omitted variable bias as well as endogeneity because demographic impact of HIV/AIDS in South Africa: national
we do not have information on the child’s characteristics indicators for 2004. Cape Town: Centre for Actuarial Studies,
South Africa Medical Research Council, Actuarial Society of
before orphaning. The results therefore serve as a lower South Africa; 2004.
bound on the differences between orphans and non- 4. Meintjes H, Giese S. Spinning the epidemic: the making of
orphans. mythologies of orphanhood in the context of AIDS. Childhood
2006; 13:407–430.
5. Case A, Paxson C, Ableidinger J. Orphans in Africa: parental
death, poverty and school enrolment. Demography 2004;
41:483–508.
6. Evans D, Miguel E. Orphans and schooling in Africa: a long-
Conclusion itudinal analysis. Demography 2007; 44:35–57.
7. Yamano T, Jayne TS. Working-age adult mortality and primary
We have shown a lack of systematic difference in school attendance in rural Kenya. Econ Dev Cultural Change
2005; 53:619–653.
education and health outcomes between orphans and co- 8. Bicego G, Rutstein S, Johnson K. Dimensions of the emerging
resident non-orphans, with a few exceptions. Policy orphan crisis in sub-Saharan Africa. Soc Sci Med 2003;
responses (and literature) do not always distinguish 56:1235–1247.
9. Beegle K, Weerdt JD, Dercon S. Orphanhood and the long-run
between orphans living with other orphans or orphans impact on children. Oxford: Centre for Studies of African
living in mixed households, even though they may live in Economies; 2006.
completely different circumstances. Comparing the 10. Ainsworth M, Filmer D. Inequalities in children’s schooling:
AIDS, orphanhood, poverty and gender. World Dev 2006;
results presented in this paper with the results in the 34:1099–1128.
literature reveals the heterogeneity of the category of 11. Lloyd CB, Blanc AK. Children’s schooling in sub-Saharan
orphans. The contradiction of the results of Case and Africa: the role of fathers, mothers and others. Popul Dev
Rev 1996; 22:265–298.
Ardington [15] highlights the need to refrain from 12. Chatterji M, Dougherty L, Ventimiglia T, Mulenga Y, Jones A,
generalizing to the country based on district or even Mukaneza A, et al. The well-being of children affected by HIV/
province level results. The paper thus illustrates the need AIDS in Lusaka, Zambia and Gitarama Province, Rwanda:
findings from a study. Working paper no. 2. Washington, DC:
for context-specific approaches that pay attention to Community REACH; 2005.
definitions, as opposed to sweeping global responses to 13. Crampin A, Floyd S, Clynn J, Madise N, Nyondo A, Khondwe M,
the crisis. The remarkable similarity of orphans’ and et al. The long-term impact of HIV and orphanhood on the
mortality and physical well-being of children in rural Malawi.
non-orphans’ living arrangement calls into question the AIDS 2003; 17:389–397.
category of orphan in South Africa. Evidence from 14. Adato M, Kadiyala S, Roopnaraine T, Biermayr-Jenzano P,
Norman A. Children in the shadow of AIDS: studies of vulner-
the study thus far challenges the Cinderella assertion in able children and orphans in three provinces in South Africa.
the case of orphans in South Africa. Unstable households Washington, DC: IFPRI; 2005.
with absent fathers are common and mobility is high, as 15. Case A, Ardington C. The impact of parental death on school
outcomes: longitudinal evidence from South Africa. Demogra-
also suggested by other literature. Reframing the phy 2006; 43:401–420.
discussion around orphanhood to be appropriate to 16. Posel D, Devey R. The demographics of fathers in South Africa:
South Africa’s social context is warranted. an analysis of survey data, 1993–2002. In: Richter L, Morrell R,
editors. Baba: men and fatherhood in South Africa. Cape Town:
HSRC Press; 2005. pp. 38–52.
Sponsorship: This project was funded by the National 17. Madhavan S. Fosterage patterns in the age of AIDS: continuity
Institute of Child Health and Development of the and change. Soc Sci Med 2004; 58:1443–1454.](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-108-2048.jpg)
![Investigating the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact[1]](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-110-2048.jpg)
![Investigating the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact[1]](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-111-2048.jpg)
![Investigating the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact[1]](https://image.slidesharecdn.com/investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact1-100929062506-phpapp01/75/Investigating-the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact-1-112-2048.jpg)
Investigating the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact[1]
This document introduces a supplement issue on poverty, HIV, and AIDS in Southern Africa. It notes that over 25 years since AIDS was first reported, it has become one of the most studied diseases in history. However, the social, economic, behavioral, and psychological drivers of HIV spread remain less understood, as do the long-term consequences of increased illness and death in poor communities. While early fears predicted rapid HIV spread outside of Africa, the virus has spread diversely. Some populations have not seen generalized heterosexual epidemics. This supplement aims to provide empirical evidence on vulnerability, associations between poverty, HIV, and AIDS impact through multiple studies and perspectives.
More Related Content
![Civil society-brief-on-international-legal-obligations[1]](https://cdn.slidesharecdn.com/ss_thumbnails/civil-society-brief-on-international-legal-obligations1-100824085138-phpapp02-thumbnail.jpg?width=640&height=640&fit=bounds)
Similar to Investigating the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact[1]
Investigating the-empirical-evidence-for-understanding-vulnerability-and-the-associations-between-poverty-hiv-infection-and-aids-impact[1]
- 1. Volume 21 Supplement7 November 2007 Poverty, HIV and AIDS: Vulnerability and Impact in Southern Africa Editors: Stuart Gillespie Robert Greener Jimmy Whitworth Alan Whiteside Sponsored by UNAIDS, RENEWAL and HEARD This publication was made possible through support provided by the Joint United Nations Programme on HIV/AIDS (UNAIDS), and through additional grants to the Regional Network on AIDS, Livelihoods and Food Security (RENEWAL), facilitated by the Interna- tional Food Policy Research Institute (IFPRI), from Irish Aid, SIDA and USAID. Support to HEARD (the Health Economics and HIV/ AIDS Research Division of the University of KwaZulu-Natal, South Africa) was provided by a DFID Research Partner’s Consortium and a Joint Financing Agreement involving SIDA, Royal Netherlands Embassy, Irish Aid, UNAIDS and DFID.
- 2. www.aidsonline.com EDITORS Jay A Levy (Editor-in-Chief, San Francisco) Brigitte Autran (Paris) Roel A Coutinho (Amsterdam) John P Phair (Chicago) EDITORIAL BOARD P Aggleton, London (2008) J Goedert, Rockville (2007) M-L Newell, London (2009) AA Ansari, Atlanta (2009) F Gotch, London (2009) G Pantaleo, Lausanne (2008) T Boerma, Geneva (2009) M-L Gougeon, Paris (2007) M Peeters, Montpellier (2009) M Bulterys, Atlanta (2008) R Gray, Baltimore (2009) D Pieniazek, Atlanta (2009) S Butera, Atlanta (2009) A Greenberg, Washington (2007) G Poli, Milan (2008) A Buvé, Antwerp (2008) S Gregson, London (2008) B Polsky, New York (2009) A Carr, Sydney (2007) S Grinspoon, Boston (2009) M Prins, Amsterdam (2008) M Carrington, Bethesda (2008) A Grulich, Sydney (2009) B Richardson, Seattle (2009) B Clotet, Badalona (2007) D Havlir, San Francisco (2008) CA Rietmeijer, Denver (2007) B Conway, Vancouver (2007) NA Hessol, San Francisco (2009) Y Rivière, Paris (2009) H Coovadia, Natal (2008) A Hill, London (2007) S Rowland-Jones, Oxford (2008) A Cossarizza, Modena (2007) JP Ioannidis, Ioannina (2007) C Sabin, London (2007) D Costagliola, Paris (2008) C Katlama, Paris (2009) H Schuitemaker, Amsterdam (2008) B Cullen, Durham (2007) D Katz, London (2008) Y Shao, Beijing (2008) E Daar, Los Angeles (2008) D Katzenstein, Stanford (2009) V Soriano, Madrid (2009) F Dabis, Bordeaux (2009) HA Kessler, Chicago (2007) S Spector, La Jolla (2008) J del Amo, Alicante (2007) S Kippax, Sydney (2008) S Strathdee, La Jolla (2008) E Delwart, San Francisco (2009) D Kuritzkes, Boston (2007) M Tardieu, Paris (2008) T Folks, Atlanta (2009) J Lundgren, Hvidovre (2009) P van de Perre, Montpellier (2009) A Fontanet, Paris (2008) D Margolis, Chapel Hill (2009) C van der Horst, Chapel Hill (2009) M French, Perth (2007) J-P Moatti, Marseille (2008) C Wanke, Boston (2007) A Ghani, London (2009) R Montelaro, Pittsburgh (2007) D Wolday, Addis Ababa (2008) J Glynn, London (2007) RL Murphy, Chicago (2007) Statistical advisers: VT Farewell (University College London, London), F Lampe, A Cozzi Lepri, A Mocroft, AN Phillips C Sabin, C Smith, Z Fox, W Bannister (Royal Free and University College Medical School, London). AIMS AND SCOPE AIDS publishes papers reporting original scientific, clinical, epidemiological, and social research which are of a high standard and contribute to the overall knowledge of the field of the acquired immune deficiency syndrome. The Journal publishes Original Papers, Concise Communications, Research Letters and Correspondence, as well as invited Editorial Reviews and Editorial Comments.
- 3. Contents Introduction Investigating the empiricalevidence for understanding vulnerability and the associations between poverty, HIV S1 infection and AIDS impact Stuart Gillespie, Robert Greener, Alan Whiteside and James Whitworth Is poverty or wealth driving HIV transmission? S5 Stuart Gillespie, Suneetha Kadiyala and Robert Greener HIV infection does not disproportionately affect the poorer in sub-Saharan Africa S17 Vinod Mishra, Simona Bignami-Van Assche, Robert Greener, Martin Vaessen, Rathavuth Hong, Peter D. Ghys, J. Ties Boerma, Ari Van Assche, Shane Khan and Shea Rutstein The socioeconomic determinants of HIV incidence: evidence from a longitudinal, population-based study in rural S29 South Africa Till Bärnighausen, Victoria Hosegood, Ian M. Timaeus and Marie-Louise Newell Explaining continued high HIV prevalence in South Africa: socioeconomic factors, HIV incidence and sexual S39 behaviour change among a rural cohort, 2001–2004 James R. Hargreaves, Christopher P. Bonell, Linda A. Morison, Julia C. Kim, Godfrey Phetla, John D.H. Porter, Charlotte Watts and Paul M. Pronyk Household and community income, economic shocks and risky sexual behavior of young adults: evidence from the S49 Cape Area Panel Study 2002 and 2005 Taryn Dinkelman, David Lam and Murray Leibbrandt HIV incidence and poverty in Manicaland, Zimbabwe: is HIV becoming a disease of the poor? S57 Ben Lopman, James Lewis, Constance Nyamukapa, Phyllis Mushati, Steven Chandiwana and Simon Gregson The economic impacts of premature adult mortality: panel data evidence from KwaZulu-Natal, South Africa S67 Michael R. Carter, Julian May, Jorge Agüero and Sonya Ravindranath The financial impact of HIV/AIDS on poor households in South Africa S75 Daryl L. Collins and Murray Leibbrandt Father figures: the progress at school of orphans in South Africa S83 Ian M. Timaeus and Tania Boler Exploring the Cinderella myth: intrahousehold differences in child wellbeing between orphans and non-orphans in S95 Amajuba District, South Africa Anokhi Parikh, Mary Bachman DeSilva, Mandisa Cakwe, Tim Quinlan, Jonathon L. Simon, Anne Skalicky and Tom Zhuwau S104 List of contributors © Wolters Kluwer Health | Lippincott Williams & Wilkins
- 4. AIDS AIDS (ISSN 0269-9370)is published at 16522 Hunters Green Parkway, Current AIDS Literature, Current Awareness in Biological Sciences, Hagerstown, MD 21740. Business offices are located at 530 Walnut Current Contents, Excerpta Medica, Index Medicus/MEDLINE, Street, Philadelphia, PA 19106-3621. Correspondence should be Laboratory Performance Information Exchange System, Research addressed to the production office: AIDS, 250 Waterloo Road, London Alert, Science Citation Index, Scisearch, Telegen Abstracts, Biosis, SE1 8RD, UK. Embase and PsycInfo. Publishing Editor: Phil Daly (Phil.Daly@wolterskluwer.com) © 2007 Lippincott Williams & Wilkins: All rights reserved; no part Production Editor: Ranadi Johnston of this publication may be reproduced, stored in a retrieval system or (Ranadi.Johnston@wolterskluwer.com) transmitted in any form or by any means, electronic, mechanical, Supplements and Special Projects Manager: Bridie Selley photocopying, recording or otherwise without either the prior written (bridie.selley@wolterskluwer.com) permission of the publisher or a licence permitting restricted photo- copying issued in the UK by the Copyright Licensing Authority and in Editorial Project Coordinator: Anna Rioland the USA by the Copyright Clearance Center. (anna.rioland@wolterskluwer.com) Applications for permission should be addressed to the Permissions Advertising: For further information contact Christopher Ploppert, Department, Lippincott Williams & Wilkins, 351 West Camden Street, Tel: +1 215 521 8570; Fax: +1 215 521 8411; Baltimore, MD 21201, USA. Fax: +1 410 528 8550; e-mail: Christopher.Ploppert@wolterskluwer.com e-mail: journalpermissions@lww.com (USA and Canada) or Dick Bower, The Point of Difference Ltd. Tel: +44 (0)20 8542 3200; Disclaimer: Although every effort is made by the publisher and Fax: +44 (0)20 8543 3810; editorial board to see that no inaccurate or misleading data, opinion or e-mail: pointofdiff@btinternet.com (rest of world). statement appear in this journal, they wish to make it clear that the data This journal accepts advertising and publishes supplements on behalf of and opinions appearing in the articles and advertisements herein are the academic and corporate sponsors. Supplements are normally supplied to responsibility of the contributor or advertiser concerned. Accordingly, subscribers at no extra charge (enquiries should be directed to e-mail: the publisher, the editorial board and their respective employees accept supplements@pharma.wkhealth.com). Article offprints can be ordered no liability for the consequences of any such inaccurate or misleading by authors (prices are available from the Offprints Administrator, data, opinion or statement. e-mail: offprint@lww.co.uk) or larger quantities of reprints can be Drugs and drug dosages: Readers are advised that new methods ordered to meet corporate requirements (enquiries should be directed and techniques described involving drug usage should be followed to Christopher Bassett, e-mail: Christopher.Bassett@wolterskluwer.com). only in conjunction with drug manufacturers’ own published literature. Author guidelines: The guidelines are available in the January issue of Paper type ECF – Typeset by Thomson Digital, Noida Special the journal and the journal’s web site at www.aidsonline.com Economic Zone, Noida, India; and printed by Page Bros, Norwich, Norfolk, UK on recyclable elemental chlorine free paper from Publishing information: AIDS is indexed and abstracted by sustainable forests meeting the requirements of ISO 9706 and EN Cambridge Scientific Abstracts, Chemical Abstracts Service, ISO 14001. Address for subscription information, orders, or change of address: be paid by personal cheque or charge. Charges are accepted on VISA, (except Japan, India, Bangladesh, Sri Lanka, Nepal and Pakistan) Lippincott, MasterCard, American Express, Diners Card, or Discover cards. Please Williams & Wilkins, 16522 Hunters Green Parkway, Hagerstown, MD have the card number and expiration date available when calling. Prices 21740; call +1 301 223 2300 outside North America, or +1 800 638 subject to change without notice. Copies will be replaced without 3030 in USA, Canada and Mexico; fax +1 301 223 2320 e-mail: charge if the publisher receives a request within 90 days of the mailing customerservice@lww.com. In Japan, contact Wolters Kluwer Health date, both in the US and worldwide. Individual and institutional Japan Co. Ltd., 3-23-14 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan. subscribers receive supplements to AIDS at no extra cost. Members of Tel: +81 3 5689 5400; fax: +81 3 5689 5402. In India, Bangladesh, Sri the International AIDS Society are entitled to reduced subscription Lanka, Nepal and Pakistan: contact Globe Publication Pvt. Ltd. B-13, rates. See application form in journal. 3rd FL, A Block, Shopping Complex, Naraina Vihar, Ring Road, New Delhi 110028, India. Tel: +91 11 5793211; fax: +91 11 579 8876. Site licences are available to all institutions.Visit www.lww.com/ librarians.htm to download the site licence most appropriate to your Annual subscription rates worldwide (volume 21, 18 issues): organization. For additional information, contact Marylou O’Connor $491.00 Individual, $1,742.00 Institution (This includes the handling at +1 215 521 8452. In the U.S. and Canada, contact Heidi Alexander at fee; The Canadian GST tax of 7% will be added to the subscription +1 800 326 1685. price of all orders shipped to Canada. Lippincott Williams & Wilkins’ GST Identification Number is 130876246. Other sales taxes are added where applicable.) Please add $36.00 for Airfreight for AIDS (USPS No.: 000-933) is published every three weeks by shipping outside Europe (Airfreight delivery usually occurs within Lippincott Williams & Wilkins and distributed in the US by DSW, 7 to 21 days.) 75 Aberdeen Road, Emigsville, PA 17318-0437. Periodicals postage paid at Emigsville, PA. Subscriptions outside United States must be prepaid. Payment must be POSTMASTER: send address changes to AIDS, PO Box 1550, in US funds drawn on a US bank. Subscriptions at the personal rate must Hagerstown, MD 21741.
- 6. Investigating the empiricalevidence for understanding vulnerability and the associations between poverty, HIV infection and AIDS impact Stuart Gillespiea, Robert Greenerb, Alan Whitesidec and James Whitworthd AIDS 2007, 21 (suppl 7):S1–S4 It is just over 25 years since the first cases of AIDS were were dead, killed in the First World War. It is only in the reported. Over this quarter-century, AIDS has become past decade that the last of these spinsters has died. The one of most highly studied diseases in history. There impacts of AIDS will take even longer to work through have been significant medical advances in understanding the population. the consequences of HIV infection and treating AIDS, as is well documented in many journals, including AIDS. Second, HIV is diverse in its spread. Early fears that the The complex and place-specific social, economic, virus would spread rapidly outside Africa have not behavioural and psychological drivers of the spread of materialized. For example, the UNAIDS 2006 ‘Report HIV remain less well delineated. The consequences of on the global AIDS epidemic’ estimated that there were increased illness and death in poor countries and commu- 5.7 million people living with HIV in India. In July 2007, nities are still unfolding. this was revised downward to 2.5 million, reflecting much less spread of the infection than had been feared [2]. In 2000, HIV was placed firmly on the global development Similar downward revisions of estimates have been made agenda by UN Security Council Resolution 1308, which in China. In a recent book, James Chin [3] argued that stated: ‘the spread of HIV can have a uniquely devastating there are many populations in which heterosexual impact on all sectors and levels of society’. A year later, in epidemics will not occur in the general population and July 2001, there was a UN General Assembly Special the epidemic will remain confined to specific risk groups. Session on HIV/AIDS. Since then our understanding of Chin’s examples of where the potential for HIVepidemics the epidemic and its potential impacts has deepened. This has been overstated are primarily from Asia, and in supplement, written by social scientists, looks at how particular China and the Philippines. This is not to socioeconomic determinants drive HIV spread and how understate the individual tragedy of each infection, but AIDS illness and mortality is impacting on communities. rather to recognize that there are countries where AIDS will have a considerable impact and others where its It is helpful to locate the contents of this supplement in importance can be downgraded. the context of the history of the epidemic. There are three overarching points to be made in introduction. First, the It is not just globally that there is wide variation. In epidemic is complex both in terms of what is driving it mainland sub-Saharan Africa HIV prevalence in adults and the effects it has. It has been described as a ‘long wave ranges from 0.7% in Mauritania to 33.4 % in Swaziland. event’. It takes years for the epidemic to spread through The hardest-hit countries are all in southern Africa; these society and generations for the full impact to be felt. A are shown in Fig. 1, the so-called ‘red’ countries. Adult recent book highlights the nature of such long wave HIV prevalence exceeds 20% in four of these countries: events [1]. ‘Singled out: how two million women Swaziland, Lesotho, Botswana and Zimbabwe. South survived without men after the First World War’ describes Africa, Namibia, Zambia, Mozambique, and Malawi all how in the United Kingdom a generation of women were have adult prevalence rates in the range of 10–20% [2]. unable to marry, as the men they would have partnered These countries are the focus of this supplement. From the aInternational Food Policy Research Institute, Geneva, Switzerland, the bJoint United Nations Programme on HIV/AIDS, Geneva, Switzerland, the cHealth Economics and HIV/AIDS Research Division, University of KwaZulu-Natal, South Africa, and the dWellcome Trust, London, United Kingdom Correspondence to Alan Whiteside, Health Economics and HIV/AIDS Research Division, University of KwaZulu-Natal, Block J418 Westville, University Road Westville, Private Bag XS4001, Durban, 4000, South Africa. Fax: +27 (31) 260 25 87; e-mail: whitesid@ukzn.ac.za ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S1
- 7. S2 AIDS 2007, Vol 21 (suppl 7) deficiency virus (HIV) was identified as the cause. The number of cases rose rapidly across the United States and was quickly identified in Europe, Australia, New Zealand and Latin America. In central Africa, health workers were observing new illnesses such as Kaposi’s sarcoma (a cancer) in Zambia, cryptococcosis (an unusual fungal infection) in Kinshasa, and there were reports of ‘slim disease’ and unexpectedly high rates of death in Lake Victoria fishing villages in Uganda [6–8]. These illnesses were occurring in heterosexual adults, not just gay men, individuals with haemophilia, blood transfusion recipients, and intravenous drug users, who formed the main groups at risk in developed countries. By 1982, cases were being seen in the partners and infants of those infected [8,9]. The initial response of public health specialists, epide- miologists and scientists was to try to identify what was causing the disease and to understand how it was spreading. This would inform prevention strategies and Fig. 1. Map of adult HIV prevalence in Africa. 20–34%; medical interventions. Early responses were therefore 10–< 20%; 5–< 10%; 1–< 5%; < 1%. predominantly scientific and technical in nature. Third, social science faces problems in addressing the It soon became apparent, however, that this was not phenomenon of HIVand its consequences. The epidemic enough, and attention shifted to understanding why is only 25 years old, which means that it, and its effects, are people were being exposed. This led to early knowledge still unfolding. Social science relies on assessing what has attitude and practice surveys, which sought to understand happened. This is done through surveys and panel data, high-risk behaviours [3] p.73. This emphasis on and sometimes the picture is at odds with what we expect. prevention gained momentum because medical scientists For example in the 1980s it was suggested, on the basis of had not yet discovered drugs that could cure, or even slow, models, that AIDS would cause economies to grow more the progress of the disease. Initial optimism for developing slowly than otherwise would be the case. In 2007, at the an effective vaccine soon faded and is now seen to be individual country level, this does not seen to have many years, if not decades, away. occurred. Uganda had the worst epidemic in the world during the early 1990s yet managed consistent economic Internationally, the World Health Organization (WHO) growth estimated at 6.5% per annum from 1991 to 2002. took the lead in response to HIV in 1986; teams visited Botswana’s growth rate over the same period was 5.6%. most developing countries to establish short and South Africa has seen steady growth since 1999. Yet it is medium-term AIDS programmes, which then evolved only through longitudinal and cross-sectional studies that into national AIDS programmes [10]. International we can hope to understand the impact of the disease. responses to HIV were, however, limited and character- Longitudinal panel data give a picture of what has ized by denial, underestimation, and oversimplification. happened in a population over the period for which the HIV was not placed high on the agenda of any other data are collected. An alternative is to gather cross- United Nations agency. Although life expectancy was sectional data: if we can understand what has happened in plummeting in certain African countries, for example, Uganda will it help predict what might happen in the United Nations Development Programme waited Lesotho? The one thing we have not been good at is until 1997 to take this into account in calculating its predicting the future, although UNAIDS made a brave human development index [11]. attempt at this through its ‘AIDS in Africa: three scenarios to 2025’ report launched in March 2005 [4]. By the 1990s there was a new perspective developing, as interest in the individual, social, and economic milieux that lead to vulnerability to HIV infection began to grow. Academics and programme officers increasingly recog- A brief history of 25 years of response nized that social justice, poverty and equity issues were driving the uneven spread of the virus within and 1981–1996 between communities and societies [12–15]. The AIDS epidemic was recognized in 1981, initally among gay men in New York and San Francisco [5]. It was 1996–2007 officially named ‘acquired immune deficiency syndrome’ In 1996, there were major changes in response to HIV, (AIDS) in July 1982, and in 1983 the human immuno- reflecting and reflected in the scholarship of the time. In
- 8. Introduction Whiteside etal. S3 the 1994 book ‘AIDS in Africa’ of 33 chapters only three inequity, long-term concurrent partnerships, the lack of were on preventive strategies and four on socioeconomic male circumcision, and the prevalence of co-infections impact, the rest were scientific or epidemiological [16]. are factors that have been identified and need further By 1996, when the second edition of ‘AIDS in the world’ examination. There are no easy solutions to curbing the was published, of 41 chapters only approximately 18 were spread of the epidemic. There are countries, outside pure science [17]. southern Africa, where the epidemic appears to be under control: Uganda brought early hope to Africa by showing In 1996, the new UN agency charged with coordinating how high levels of political commitment and com- the response to the epidemic, UNAIDS, began operations munity-led responses can work to stabilize HIV in Geneva. This was significant as it acknowledged that prevalence. In other locations, such as Tanzania, infection the international health body the WHO was not able to rates peaked at a lower level than those currently seen in respond to the epidemic in all its facets, and there needed most of southern Africa. to be international coordination for an exceptional disease. At the XIth International AIDS Conference in The focus of this supplement is on bringing together and Vancouver, the arrival of new drugs in developed understanding the data on the socioeconomic dimensions countries to treat AIDS was announced, and mortality of the epidemic. It came out of a meeting sponsored by among those being treated plummeted. UNAIDS and hosted by the Health Economics and HIV/AIDS Research Division of the University of At the XIIIth International AIDS Conference in KwaZulu-Natal held in Durban from 16 to 18 October Durban, South Africa, in July 2000, Nelson Mandela, 2006. The aim of the symposium was to bring together closed the conference with a call for drugs to be made people, especially those involved in field research, to share accessible to all. Since then, the response to AIDS has knowledge and experience and to address gaps in our been dominated by new initiatives for making treatment understanding of the spread of HIV and impact of AIDS. accessible, especially in developing countries. The price In particular, we were looking for community- of drugs has fallen dramatically with the manufacture of based longitudinal studies currently being carried out generic drugs.1 In 2001, United Nation’s Secretary in Africa. General, Kofi Annan, called for spending on AIDS to be increased 10-fold in developing countries, and the The outputs of this meeting were to be a review of the Global Fund for AIDS, TB and Malaria was established. main longitudinal socioeconomic data collections in The same year, President George W. Bush announced Africa with a bearing on HIV, the publication of the the Presidential Emergency Plan for AIDS Relief participants’ best papers, and an opportunity to network (PEPFAR) targeting 15 developing countries. In 2003, and share ideas. the WHO and UNAIDS proclaimed the ‘3 by 5’ plan, to treat 3 million people in poor countries by the end The meeting was a qualified success in that papers were of 2005. presented and we have this interesting and thought- provoking supplement. There are, however, a number of Over the decade from 1996 to 2006, more financial caveats, and these cut to the heart of the issues we are resources than ever before were made available for the dealing with. South African research and papers response to AIDS, with emphasis increasingly on making dominate. Of the 11 papers we publish, eight are from treatment available in developing countries. In 1996, South Africa, two compare data from across the continent there was approximately US$300 million for HIV/AIDS and one is from Zimbabwe. This is also true of the in low and middle-income countries; by 2006, this authors, the vast majority are either South African or increased to US$8.3 billion. It is noteworthy that this based in the developed world. Clearly, there are real issues response, largely a result of treatment becoming with developing capacity in African countries. The global available and affordable, led to a ‘remedicalization’ of emphasis is on delivery not research, but, as this HIV/AIDS. supplement shows, quality data and good science are essential. It is not clear why southern Africa has been so hard hit by HIV. Socioeconomic variables, cultural factors and sexual Of the ten papers we publish, seven are from South Africa behaviour all play a role. Poverty, income inequality, sex two compare data from across the continent and one is from Zimbabwe. This is a good spread. What do the papers tell us? Put simply, the causes and consequences of 1 Presentation by Peter Graaf of the HIV/AIDS Department of the the epidemic are complex and policy needs to take this WHO to an ‘Informal technical consultation on the relevance and into account. modalities of implementation of an observatory for HIV commodities in Africa’ organized by Health Economics and HIV/AIDS Research Division (HEARD), University of KwaZulu Natal, the World Health Although poor individuals and households are likely to be Organization, and Swedish/Norwegian HIV/AIDS Team on 25 June hit harder by the downstream impacts of AIDS than their 2007. less poor counterparts, their chances of being exposed to
- 9. S4 AIDS 2007, Vol 21 (suppl 7) HIV in the first place are not necessarily greater than References wealthier individuals or households. It is too simplistic to refer to AIDS as a ‘disease of poverty’. As an infectious 1. Nicholson V. Singled out: how two million women survived disease, it is appropriate that the primary core response to without men after the First World War. London: Viking; 2007. HIV focuses on public health prevention strategies and on 2. UNAIDS. 2006 Report on the Global AIDS epidemic. 2006. Available at: http://www.unaids.org/en/HIV_data/2006Global- medical treatment and care. But if we are to make further Report/default.asp. Accessed: September 2007. strides in combating the epidemic we need broad-based 3. Chin J. The AIDS pandemic: the collision of epidemiology with prevention, that is, prevention that deals with the political correctness. Oxford: Radcliffe Publishing; 2006. 4. UNAIDS. AIDS in Africa: three scenarios to 2025. Geneva: contextual environment and the underlying socio- UNAIDS; 2005. economic, behavioural and psychological drivers of the 5. Centers for Disease Control and Prevention. MMWR Morb epidemic. Like the virus, these strategies need to cut Mortal Wkly Rep. across all socioeconomic strata of society. 6. Bayley A. Aggressive Kaposi’s sarcoma in Zambia. Lancet 1984; ii:1318–1320. 7. Hooper E. The river: a journey back to the source of HIV and On the downstream side, although AIDS impoverishes AIDS. London: Allen Lane/The Penguin Press; 1999. Copyright households, its effects are not uniform. Again, appropriate Edward Hooper 2000. 8. Iliffe J. The African AIDS epidemic: a history. Oxford: James responses need to take account of the context-specificity Currey; 2006. and dynamic nature of the stresses, shocks and local 9. Shilts R. And the band played on: people politics and the AIDS responses brought by AIDS, so that mitigation measures epidemic. London: Viking; 1988. are appropriately designed. 10. Mann J, Tarantola D, editors. Government national AIDS pro- grams, Chap. 30. In: AIDS in the world II. Oxford: Oxford University Press; 1996. Finally, as is always the case with a publication, there are 11. Whiteside A, Barnett T, George G, Van Niekerk A. Through a people who need to be thanked. In Durban, Marisa glass, darkly: data and uncertainty in the AIDS debate. In: Developing world bioethics, issue 3. Oxford: Blackwell Publish- Casale took charge of organizing the meeting. UNAIDS ers Ltd.; 2003. sponsored both the meeting and publication. Alan 12. Whiteside A. AIDS – socio-economic causes and conse- Whiteside’s time was largely supported through a DFID quences. Occasional paper no 28. Economic Research Unit, University of Natal, Durban; 1993. Research Partners Consortium grant. Stuart Gillespie’s 13. Gruskin S, Hendriks A, Tomasevski K. Human rights and the time was supported by the RENEWAL programme response to HIV/AIDS. In: AIDS in the world II. Edited by Mann through support from Irish Aid and the Swedish J, Tarantola D. Oxford: Oxford University Press; 1996. International Development Cooperation Agency, and 14. Loewenson R, Whiteside A. Social and economic issues of HIV/ AIDS in southern Africa: a review of current research. SAfAIDS by UNAIDS. We also acknowledge the extensive inputs 1997;. of Suneetha Kadiyala of the International Food Policy 15. Barnett T, Whiteside A. HIV/AIDS and development: case studies Research Unit throughout the preparation of this and a conceptual framework. Eur J Dev Res 1999; 11:200–234. 16. Essex M, Mboup S, Kanki PJ, Kalengayi MR. AIDS in Africa. New supplement. York: Raven Press; 1994. 17. Mann J, Tarantola D, editors. AIDS in the world II. Oxford: Conflicts of interest: None. Oxford University; 1996.
- 10. Is poverty orwealth driving HIV transmission? Stuart Gillespiea, Suneetha Kadiyalab and Robert Greenerc Evidence of associations between socioeconomic status and the spread of HIV in different settings and at various stages of the epidemic is still rudimentary. Few existing studies are able to track incidence and to control effectively for potentially confounding factors. This paper reviews the findings of recent studies, including several included in this volume, in an attempt to uncover the degree to which, and the pathways through which, wealth or poverty is driving transmission in sub-Saharan Africa. We investigate the question of whether the epidemic is transitioning from an early phase in which wealth was a primary driver, to one in which poverty is increasingly implicated. The paper concludes by demonstrating the complexity and context-specificity of associ- ations and the critical influence of certain contextual factors such as location, sex and age asymmetries, the mobility of individuals, and the social ecology of HIV trans- mission. Whereas it is true that poor individuals and households are likely to be hit harder by the downstream impacts of AIDS, their chances of being exposed to HIV in the first place are not necessarily greater than wealthier individuals or households. What is clear is that approaches to HIV prevention need to cut across all socioeconomic strata of society and they need to be tailored to the specific drivers of transmission within different groups, with particular attention to the vulnerabilities faced by youth and women, and to the dynamic and contextual nature of the relationship between socio- economic status and HIV. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins AIDS 2007, 21 (suppl 7):S5–S16 Keywords: socioeconomic status, poverty, inequality, HIV, gender, prevention Introduction to have better access to reproductive healthcare, condom use is generally low in Africa and other parts of the Evidence of the association between HIV transmission developing world. Pre-existing sexual behaviour patterns and socioeconomic status is mixed [1–3]. Although early (from ‘pre-HIV’ times) therefore make the richer and the studies tended to find positive correlations between better educated more vulnerable to HIV infection, economic resources, education and HIV infection [4,5], especially in the early stages of the epidemic, when as the epidemic has progressed, it has increasingly been information about the virus and how to protect oneself is assumed that this relationship is changing. Evidence of the usually low [6,8]. At a later stage, however, it has been degree, type and dynamics of the influence of socio- argued that individuals with higher socioeconomic status economic factors on rates of HIV transmission in different tend to adopt safer sexual practices, once the effects of settings and at various stages of the AIDS epidemic is, AIDS-related morbidity and mortality become more however, still rudimentary. This paper seeks to bring apparent, adding greater credibility to HIV prevention together what is known on this, drawing especially on the messages [9,10]. findings of some recent studies, including several in this supplement. Another currently postulated dynamic is that poverty (possibly itself fuelled by AIDS) is increasingly placing In most countries, relatively rich and better educated men individuals from poor households at greater risk of and women have higher rates of partner change because exposure to HIV via the economically driven adoption of they have greater personal autonomy and spatial mobility risky behaviours. Poverty and food insecurity are thought [4,6,7]. Although the richer and better educated are likely to increase sexual risk taking, particularly among women From the aInternational Food Policy Research Institute, Geneva, Switzerland, the bInternational Food Policy Research Institute, Washington, DC, USA, and the cJoint United Nations Programme on HIV/AIDS, Geneva, Switzerland. Correspondence and requests for reprints to Stuart Gillespie, International Food Policy Research Institute, c/o UNAIDS, 20 Avenue Appia, CH-1211 Geneva 27, Switzerland. E-mail: s.gillespie@cgiar.org ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S5
- 11. S6 AIDS 2007, Vol 21 (suppl 7) who may engage in transactional sex to procure food Does poverty increase exposure to HIV? for themselves and their children. Women’s economic dependence on their partners may also make it difficult for At the country level there is a weak positive relationship them to insist on safer sex (e.g. condom use). In addition, between national wealth and HIV prevalence across poor people are more likely to be food insecure and countries in sub-Saharan Africa, where higher prevalence malnourished. Malnutrition is known to weaken the is seen in the wealthier countries of southern Africa immune system, which in turn may lead to a greater risk of (Fig. 1). Strong urban–rural economic linkages, good HIV transmission in any unprotected sexual encounter transport links and high professional mobility may (although this remains under-researched). This strand of translate into both higher incomes and higher HIV literature on HIV transmission in Africa stresses the reversal incidence. National poverty rates, on the other hand, do in the distribution of the epidemic across population not show a strong association with HIV prevalence subgroups as the epidemic advances within countries, with (Fig. 2). There is, however, a clear and significant pattern those of lower socioeconomic status experiencing a higher of association between income inequality and HIV subsequent rate of HIV transmission. prevalence across countries; countries with greater inequality have higher HIV prevalence, especially in We aim to present an overview of the findings of key sub-Saharan Africa but also to a lesser extent in Asia and recent African studies (primarily 2004–2007) examining Latin America (Fig. 3). the relationship between economic resources/status and the risk of HIV infection (see Table 1). The starting point Household level evidence that poverty is a major driver of was the evidence presented in this supplement on this the epidemic is rather mixed. It is important, however, to relationship, but our search then expanded to draw upon note that most studies focus on relative poverty in the other recent literature from sub-Saharan Africa where the context of generalized chronic poverty. In most cases, it is epidemic is most severe. only the highest one or two quintiles (or possibly three in middle-income southern African countries) that can be First, PUBMED and ECONLIT searches (2004–2007) thought of as representing the non-poor, using the were used to identify all studies addressing the link standard poverty line definitions, or the US$1 or US$2 between socioeconomic status (poverty and education in per day measures adopted for the purpose of global particular) and the risk of HIV. Searches were limited to comparison. Comparisons are thus between ‘wealthier’ English language and Africa. Keywords pertaining to the and ‘poorer’ groups. explanatory variables were ‘poverty’, ‘wealth’, ‘socio- economic status’, ‘socioeconomic’, ‘education’ and Studies adopting ethnographic methodologies suggest ‘education level’. Keywords pertaining to the outcome that material poverty increases the risks of contracting variable of interest were ‘HIV risk’, ‘HIV transmission’, HIV mainly through the channel of high-risk behaviour ‘sexual behaviour’ and ‘HIV prevalence’. Studies on adoption. The respondents of an ethnographic study in special groups of populations such as truck drivers and the southern province of Zambia [26] identified frequent uniformed services have been excluded. Conceptual/ droughts and limited wage labour opportunities, after the theoretical papers have not been included in the review of post-economic liberalization closure of companies, as the the association between socioeconomic status, poverty, ‘push’ factors behind the increasing resort of women to education and the risk of heterosexual HIV transmission, transactional sex. In a qualitative study in Malawi [27] although such studies have been used from a reference certain social groups were found to continue to engage in perspective. Quantitative studies with only descriptive high-risk behaviours despite knowing the risks. They did statistics have been excluded. Sixteen of the 49 retrieved so, the authors contend, to affirm their social identity and articles were thus excluded. In addition, a Dissertation to deny that ‘anything they do makes a difference to what Abstracts Online search and a Google Scholar search were they perceive as a life of powerlessness and despair’ (p. 17). also conducted to identify pertinent recent grey literature. The ‘culture of poverty’, as documented by Lewis [28] in Whenever possible, the authors of such papers that met Latin America, may thus be as significant as material the above criteria were contacted for the latest drafts and poverty in motivating risky behaviours. updates on the status of their articles. The findings from several recent quantitative surveys that As such, this overview is intended to complement earlier investigated the relationship between economic depri- reviews examining this relationship [23,24]. It then seeks vation and the adoption of high-risk behaviours are to delve deeper into the pathways and interactions that generally consistent with much of the qualitative research contextualize the link between wealth/poverty and [29–31], although there are important differences heterosexual HIV transmission risk. We stress at the between behaviours and regarding the influence of outset that we are not reviewing evidence of the gender in different contexts [12,14,32]. downstream impacts of AIDS on poverty, a subject that has been comprehensively covered recently elsewhere Employing the Cape Area Panel Study, which surveys [23–25]. individual youths aged 14–22 years in Cape Town, South
- 12. Table 1. Recentquantitative studies examining the relationship between HIV and socioeconomic status. Study Objective Study design and statistical analyses Key findings Dinkelman et al. [11] Estimate if sexual debut between 2002 Cape Area Panel Study data that surveyed Household income negatively associated with sexual and 2005, number of recent partners 4752 boys and girls, 14–22 years of age debut, and economic shocks positively associated and lack of condom use at last sex in Cape Town, South Africa (2002–2005). with multiple partnerships among girls. Community in 2005 is affected by household Multivariate probit models poverty rates predict earlier sexual debut and income constraints and income shocks. higher rates of unprotected recent sex for boys. Schooling positively associated with a significant condom use, but negatively associated with multiple partners for both boys and girls. Weiser et al. [12] Studies the association between food Cross-sectional population-based survey of Food insufficiency associated with inconsistent insufficiency (not having enough food 1255 adults in Botswana and 796 adults condom use with a non-primary partner, sex to eat over the previous 12 months) in Swaziland. exchange, intergenerational sexual relationships, and and inconsistent condom use, sex Multivariable logistic regression analyses, lack of control in sexual relationships. For men, exchange, and other measures of risky sex. clustered by country, and stratified by sex. food insufficiency was associated with increase in the odds of unprotected sex only. Higher educated women, but not men, were less likely to report high-risk behaviours. Johnson and Way [13] Investigates the association between Cross-sectional, 2003 Kenya Demographic Wealth was positively related to HIV-positive demographic, social, behavioural, and Health Survey. serostatus for both men and women. Women and biological variables and HIV Multivariate logistic regression model with primary education were nearly twice as likely serostatus in Kenya. stratified by sex. to be HIV positive as those with no education. Sexual behaviour factors were not significantly associated with HIV serostatus. Nii-Amoo Dodoo et al. [14] Examines the relationship between Quantitative data are drawn from the Although poverty was significantly associated with HIV-related sexual activity outcomes, Demographic & Health Surveys (DHS) the examined sexual outcomes in all settings, the specifically age at first sex and multiple and qualitative data from the Sexual urban poor are significantly more likely than their sexual partnerships, and socioeconomic Networking and Associated Reproductive rural counterparts to have an early sexual debut deprivation amenities index, (based on and Social Health Concerns study. and a greater incidence of multiple sexual partnerships. asset index and amenities index) in rural Multivariate Cox regressions. The disadvantage of the urban poor is accentuated and urban Kenya. for married women; those in Nairobi’s slums are at least three times as likely to have multiple sexual Poverty, wealth, HIV transmission Gillespie et al. partners as their rural counterparts. Lopman et al. [15] Studies the association between wealth Manicaland, Zimbabwe HIV/STD Prevention The greatest decrease in HIV prevalence occurred in index (based on household asset ownership) Project’s population-based open cohort the highest wealth index tercile in both men and and HIV incidence, HIV mortality, sexual (baseline between 1998 and 2001 and women. In men (but not women), HIV incidence risk behaviour, and sexual mixing patterns. follow-up between 2001 and 2003). was lowest in the top wealth index tercile. Mortality Multivariate logistics and Poisson regression rates were significantly lower in both men and women models. of higher wealth index. Men of higher wealth index reported more sexual partners, but were also more likely to use condoms, controlling for age and site type. Better-off women reported fewer partners and were less likely to engage in transactional sex. Hargreaves et al. [16] To assess the evidence that HIV incidence Prospective cohort of 1967 individuals Among men, there was little evidence that HIV rates and sexual behaviour patterns differed (14–35 years of age) in Limpopo province, seroconversion was associated with any by wealth, education and migration. South Africa (2001 and 2004). socioeconomic factor. Among women, HIV Multivariate logistic regression models, seroconversion was negatively associated with stratified by sex. education, but not wealth or migration. Migrant men more often reported multiple partners. Migrant and more educated individuals of both sexes, and women from wealthier households, reported higher levels of condom use. Mishra et al. [17] Examines the association between wealth Cross-sectional nationally representative In all eight countries, adults in the wealthiest quintiles (index based on household ownership surveys from eight sub-Saharan African have higher prevalence of HIV than those in the of consumer durables) and HIV serostatus countries conducted during 2003–2005. poorer quintiles, but the positive association of 15–49-year-old individuals. Multivariate logistic regression models, between wealth and HIV status was statistically stratified by sex. insignificant in multivariate models. (continued overleaf ) S7
- 13. S8 AIDS Table 1. (continued ) Study Objective Study design and statistical analyses Key findings 2007, Vol 21 (suppl 7) Barnighausen et al. [18] ¨ Investigates the effect of educational Longitudinal data (2003–2005) on 3325 adults Belonging to a household in the middle attainment, household wealth categories from Africa Centre Demographic Information wealth category increased the risk of (based on a ranking of households on an System in KwaZulu-Natal, South Africa. HIV seroconversion. One additional grade assets index scale) and total household Semiparametric and parametric survival models. of educational attainment reduced the expenditure, on HIV incidence. hazard of HIV seroconversion by approximately 7%. Urban residence was associated with a 65% increase in the hazard of HIV seroconversion. Chapoto and Jayne [19] To determine the ex-ante socioeconomic Nationally representative panel data set of Relatively non-poor men (ranked by characteristics of individuals who died 18 821 individuals from 5420 households assets levels) were 43% more likely in their prime age (15–59 years) surveyed between 2001 and 2004. to die than poor men. Poor and non-poor in Zambia. Multivariate probit models, stratified by sex women were equally likely to die. No clear and assets. relationship observed between education attainment and probability of prime-age mortality. Poor women with business income were 15% less likely, and non-poor women with business income 7% more likely, to die than those without business income. Kirimi and Jayne [20] Estimates the potentially changing Nationwide data set of 5755 individuals Over time, the probability of disease-related relationship over time between from 1500 Kenyan rural households death declined for both men and women. household and individual-level collected in 1997, 2000, 2002 and 2004. A reversal in the effect of education on death indicators of poverty and subsequent Multivariate probit models, stratified by sex. was observed, with more educated women death of prime-age adults in Kenya. and men, and particularly younger ones, being at greater risk of death. Although weak, there is also a delayed but significant negative effect of landholding size and asset value on male mortality. Glynn et al. [9] Investigates the associations between Cross-sectional population-based survey No association between schooling and HIV schooling and both HIV and herpes conducted in 1997–1998 in four African infection and a significant negative association simplex 2 infection and risky behaviours cities including approximately with herpes simplex 2 in women observed in in Cotonou (Benin), Yaounde (Cameroon), 2000 adults in each city. Kisumu or Ndola,. In Yaounde, women with Kisumu (Kenya) and Ndola (Zambia). Multivariate models, stratified by sex. more schooling were less likely to be HIV positive. Similar association observed among men in Cotonou for herpes simplex 2. In all cities, those with more education tended to report less risky sexual behaviours. De Walque et al. [10] Investigates the association between Population-based cohort followed between In 1989/90, there was no significant relationship changing HIV prevalence, condom 1989/1990 and 1999/2000. between education and HIV prevalence. use and education in rural south-west Multivariate and bivariate (condom versus In 1999–2000 women aged 18–29 years Uganda. education) analyses. with post-primary education were at significantly lower risk of HIV-1 infection than women with no education. Condom use increased during the study period and this increase has been concentrated among more educated individuals. Luke [21] To study the trade-off between transfers Cross-sectional survey of Luo men aged Men’s income was not significantly associated and condom use at last sexual intercourse 21–45 years in Kisumu, Kenya. with condom use. Having an adolescent in non-commercial, non-marital sexual Multivariate models including male fixed female partner does not have a significant relationships in Kenya. effects models. effect on condom use. For every Ksh500, approximately the mean amount given in transfers per partnership, the probability of condom use decreased by approximately 8%. Trade-off between transfers and condom use does not vary between adolescents and adult women.
- 14. Poverty, wealth, HIVtransmission Gillespie et al. S9 level of gender inequality, age is protective. Similarly, the not always significant. Conditional on gender inequality, Africa (2002–2005), Dinkelman et al. [11] show that for effect of gender inequality for women decreased with the share of young women who live in poverty in the was associated with a 1% increase in the probability girls, sexual debut appears to be earlier in poor increasing household assets, although this effect was A one standard deviation increase in gender inequality of being HIV positive for young women. For a given households, especially those who have experienced an in inherited land, the total amount of transfers community did not increase the probability of Economic status was positively and significantly economic shock (a death, illness or job loss). A recent increases by Ksh10 on average. Wealth was associated with both the giving of transfers and the amount. For every additional acre additional year of education increased the cross-sectional study in Kenya found asset poverty to be not correlated with condom use. Each significantly related to risky sexual outcomes, such as early sexual debut, multiple sexual partnerships, in all three residential settings studied [14]. In a study in probability of condom use by Botswana and Swaziland [12], although protective in individual HIV infection. unadjusted analyses, controlling for other variables, approximately 3.4%. income was not associated with intergenerational sex and a lack of control in sexual relationships among women. Wealthier men reported having more sex exchange [adjusted odds ratios (aOR) 1.94, 95% confidence interval (CI) 1.59–2.37] but were also more likely to report condom use (aOR 0.78, 95% CI 0.72– 0.84). Another recent cross-sectional study of Luo men aged 21–45 years of age in urban Kisumu, Kenya, found male economic status, controlling for age and education, to Cross-sectional survey of Luo men aged and Housing Census, Kenya Poverty be positively associated with transactional sex and the Three sources of cross-sectional data: value of transfers [22]. For every Ksh1000 in male Health Survey, 1999 Population 21–45 years in Kisumu, Kenya. 2003 Kenya Demographic and income, the probability of giving a transfer in the past month increases approximately 1%, and the total amount Multivariate probit models of transfers increases Ksh29 (US$0.40). Wealth (income Multivariate models. and inherited land) was not, however, correlated with condom use, suggesting that larger transfers are not being Map (2003). given by wealthier men as an incentive for condom-free (riskier) sex. Two prospective cohort studies examining the relation- ship between economic resources and high-risk sexual behaviours are presented in this volume. In a 3-year women and adult men within an individual’s follow-up study (baseline between 1998 and 2001 and Examines the relationship between HIV status women’s poverty status on individual HIV inherited land), transfers, and non-marital follow-up between 2001 and 2003) in Manicaland, Empirical investigation of the connection between economic status (income and and gender inequality between young non-commercial, sexual relationships Zimbabwe, Lopman et al. [15], found wealthier men community and to examine young reporting more sexual partners, but also more frequent use of condoms, controlling for age and site type. This relationship became insignificant, however, after con- trolling for education level, in addition to age and site type, suggesting that the effect of wealth is at least partly status in Kenya. the result of differences in education across wealth levels. Better-off women reported fewer partners and were less in Kenya. likely to engage in transactional sex, adjusting for age, education level and site type. Hargreaves et al. [16] in Limpopo, South Africa (2001–2004) found women, but not men, from wealthier households reporting higher levels of condom use (aOR comparing household ‘doing OK’ with ‘very poor’ 2.03, 95% CI 1.29–3.20). Beegle and Ozler (unpublished) Using Demographic and Health Survey (DHS) data from eight countries, Mishra et al. [17] found a positive association between an asset-based wealth index and HIV status. This relationship was stronger for women, and it was clear that HIV prevalence was generally lower among Luke [22] the poorest individuals in these countries. This is partly accounted for by an association of wealth with other
- 15. S10 AIDS 2007, Vol 21 (suppl 7) 35% Swaziland 30% Botswana 25% Lesotho HIV prevalence 20% Zimbabwe Namibia Southern Africa Zambia R squared = 0.2952 Mozambique South Africa 15% not significant Malawi Central African Republic 10% Gabon Côte d'Ivoire Tanzania Kenya E&W Africa Uganda 5% R squared = 0.0000 Angola not significant Sierra leone Ethiopia 0% US$100 US$1 000 US$10000 GDP per capita (PPP, logarithmic scale) Fig. 1. HIV and per-capita gross domestic product in Africa. Sources: Economic data from UNDP Human Development Report 2006; HIV prevalence data from UNAIDS Epi Update, May 2006. underlying factors. Wealthier individuals tend to live in likely than the poorest women to be HIV positive [13]. urban areas where HIV is more prevalent, they tend to be Similar findings were reported in Tanzania [33] and in more mobile, more likely to have multiple partners, more Burkina Faso [34]. likely to engage in sex with non-regular partners, and they live longer; all factors that may present greater Studies of cross-sectional associations between HIV lifetime HIV risks. On the other hand, however, they serostatus and socioeconomic status (such as those above tend to be better educated, with better knowledge of HIV and the cross-sectional studies featured in another prevention methods, and are more likely to use condoms; comprehensive review [1]) suffer from important factors that reduce their risk compared with poorer limitations: They are unable to distinguish between the individuals. Controlling for these associations, however, effect of economic status on HIV infection and the effect does not reverse the conclusion: there is no apparent of HIV infection on economic status, and they are unable association between low wealth status and HIV. to control for the fact that individuals from richer households may survive longer with HIV, and are thus Using data from the cross-sectional, population-based more likely to be present in the population to be tested, 2003 Kenya Demographic and Health Survey, a recent thereby increasing HIV prevalence rates. study found increased wealth to be positively related to HIV infection, with the effect being stronger for women In a cross-sectional study, it is thus conceivable to find a than men; the wealthiest women being 2.6 times more positive association between economic status and HIV 25% Botswana Lesotho Zimbabwe 20% Namibia South Africa Southern Africa R squared = 0.0996 Zambia Mozambique not significant HIV prevalence 15% Malawi Central African Republic 10% E&W Africa Côte d'Ivoire Uganda Tanzania R squared = 0.0307 Kenya not significant 5% Cameroon Nigeria Rwanda Burundi Ghana Ethiopia Gambia Mali Burkina Faso Niger Senegal Mauritania Sierra Leone Madagascar 0% 0 10 20 30 40 50 60 70 80 Percentage below US$1 per day Fig. 2. HIV and poverty in Africa. Sources: Economic data from UNDP Human Development Report 2006; HIV prevalence data from UNAIDS Epi Update, May 2006.
- 16. Poverty, wealth, HIVtransmission Gillespie et al. S11 35% Swaziland 30% 2 R = 0.4881 P = 0.005% 25% Botswana Lesotho HIV prevalence Zimbabwe Namibia 20% South Africa Zambia Mozambique 15% Malawi Central African Republic 10% Tanzania Uganda Côte d'Ivoire Kenya Cameroon 5% Rwanda Nigeria Burundi Ghana Mali Ethiopia Senegal Niger 0% 0.25 0.35 0.45 0.55 0.65 0.75 GINI coefficient Fig. 3. HIV and income inequality in Africa. Sources: Economic data from UNDP Human Development Report 2006; HIV prevalence data from UNAIDS Epi Update, May 2006. infection, even if higher economic status protects transmission. Those studies are nationally representative individuals from acquiring HIV. Both the above-stated rural household panel surveys, unlike the studies reviewed limitations can be overcome by using prospective cohorts above (in which the national level surveys are cross- to track HIV incidence. This volume presents three such sectional and longitudinal cohorts are limited to studies with differing results: (i) Lopman et al. [15] in provinces). Although they do not directly measure Manicaland, Zimbabwe, reported a significantly lower HIV prevalence or incidence, they do employ innovative male HIV incidence (between baseline in 1998/2000 and methodologies to infer the extent of HIV-related prime follow-up in 2001/2003) in the wealthiest asset tercile age adult mortality. (15.4/1000 person-years) compared with the lowest tercile (27.4/1000 person-years), controlling for age and A nationally representative rural panel data survey (2001– site of residence. This trend was even more marked in 2004) in Zambia [19], sought to determine the ex ante young men under 17–24 years of age. No such socioeconomic characteristics of individuals who died in association between wealth and HIV seroconvesion was their prime age (15–59 years). When ranked by asset observed among women. Mortality rates were signifi- levels, relatively wealthier men were 43% more likely to cantly lower in both men and women of higher wealth die of disease-related causes than men in poor households, groups. They also found a decrease in HIV prevalence with no clear association among women. across all asset wealth groups during the study period, with the largest decrease in the wealthiest tercile for both In contrast, a nationwide rural panel survey (1997–2004) men at 25% and women at 21%. (ii) Controlling for place in Kenya [20], performing similar analyses to the above, of residence, migration status, partnership status, sex and reported men and women from relatively asset-poor age, a study in rural KwaZulu Natal by Barnighausen et al. ¨ households to be more likely to die than those from [18] found that individuals from households in the middle wealthier households. The authors also found a shift in asset wealth tercile had a significantly higher hazard of the relationship between landholding size and prime-age HIV seroconversion (1.7 times that of the poorest tercile), mortality in which no significant association was observed whereas there was no significant difference between the between 1997 and 2000, but in both the 2000–2002 and wealthiest and poorest terciles. Per capita household 2000–2004 periods, access to more land was associated expenditures on the other hand did not significantly with reduced male mortality. influence the hazard of HIV seroconversion. (iii) In a study of HIV incidence in Limpopo Province of South Africa between 2001 and 2004, Hargreaves et al. [16] did not find a statistically significant association between HIV Does education reduce exposure to HIV? seroconversion and economic status (assessed through participatory wealth ranking methods) in either men Education is one of the most studied socioeconomic or women. factors in the context of AIDS epidemics. Although education and economic resources are often jointly A few other longitudinal studies have added to our determined, empirical evidence has shown that education understanding of socioeconomic differentials in HIV predicts health independently of income [35].
- 17. S12 AIDS 2007, Vol 21 (suppl 7) A systematic review in 2002 of 27 studies [7], mostly especially among women [16,18]. Hargreaves et al. [16] cross-sectional, with data predominantly collected before found that among women (but not men) HIV 1996, found that increased schooling was either not seroconversion was negatively associated with education associated with HIV infection or was associated with an (aOR comparing attended secondary school versus increased risk of HIV infection among men and women none/primary 0.49, 95% CI 0.28–0.85; comparing from both rural and urban communities in Africa. As the those completing secondary school versus none/primary epidemic within countries has advanced, the evidence 0.25, 95% CI 0.12–0.53). Barnighausen et al. [18] ¨ suggests a shift towards a reduced relative risk of HIV reported that one additional grade of educational infection among adults, especially younger women, who attainment reduced the hazard of HIV seroconversion have a secondary education [9,10,36]. by approximately 7%. The hypothesis that the ability to process and access In sum, a relatively clear picture emerges for education, the information is one channel through which education majority of studies suggest that education is increasingly affects health outcomes has been examined in a study in associated with less risky behaviours. Sustained efforts to Uganda [10], in which changes in association between improve education levels as well as targeted and tailored schooling levels, HIV prevalence, and condom use were messages on HIV prevention efforts can yield positive estimated among a population-based rural cohort in results. Masaka District between 1989/1990 and 1999/2000. During the early years of the epidemic in 1990, there was no robust relationship between HIV and years of education for either sex for all individuals older than Poverty and HIV: pathways and 17 years of age. By 2000, however, each additional year of interactions education was found to lower the risk of being HIV positive significantly among 18–29-year-old women Links between socioeconomic conditions, such as wealth (aOR 0.863, 95% CI 0.77–0.96). Condom use was found and education, and HIV risk and vulnerability are clearly to be positively associated (using bivariate analysis only) complex, perhaps too complex for a single explanation. A with schooling levels between 1995 and 2000, with the major analytical challenge is to define the causal pathways gradient between higher educational achievement and operating from distal socioeconomic factors to proximal greater condom use being steeper for women than men individual behaviours and ultimately physiological (chi-square for trend of odds in 1996/1997, 69.10 for factors. Different socioeconomic factors may affect health men and 82.13 for women, and in 1999/2000, 103.01 for at different times in the life course [40,41], operating at men and 164.18 for women). different levels (e.g. individual, household and neigh- bourhoods) [42,43] and through different causal pathways One study in Cote d’Ivoire [37] found more highly [44,45]. educated people to be more likely to engage in multiple sexual partnerships, although they were also more likely The sections below highlight some of the more important to use condoms, thus offsetting some of the risk of factors and processes that condition the relationship exposure to HIV. Similar observations of a higher between poverty, wealth and HIV. Here we focus on the probability of condom use among the more educated key issues of gender inequality, mobility and social have been reported elsewhere [38,39]. ecology. Malnutrition is another potentially important conditioning factor affecting the risk of HIV infection. A cross-sectional study in Botswana and Swaziland found Given space limitations, the reader is referred to other that higher educated women were less likely to report a reviews and ongoing work in this area [24]. lack of control in sexual relationships (aOR 0.36, 95% CI 0.36–0.37), were less likely to report inconsistent condom use (aOR 0.72, 95% CI 0.57–0.91) and Gender and economic asymmetries intergenerational sex (aOR 0.68, 95% CI 0.53–0.86). The issue of gender is front and central to any discussion No association between risk behaviours and education of HIV and poverty. Women’s dependence on men’s among men was observed [12]. Studies employing economic support throughout much of the developing longitudinal rural panel datasets from Zambia, Kenya world means that women’s personal resources, including and Ethiopia have shown a pattern of negative association their sexuality, has economic potential. Economic between educational attainment and disease-related asymmetries within a couple are reinforced by various mortality [19] (A. Chapoto et al., unpublished). contextual factors, such as family and peer pressures, social and economic institutions and pervasive and deeply As with economic status, few studies have prospectively entrenched sex-based inequalities. Social norms in many investigated the relationship between education and HIV sub-Saharan African contexts, for example, permit (and incidence. Two such studies presented in this volume even encourage) men to engage in sex with multiple found a significantly protective effect of education, partners, with much younger partners, and to dominate
- 18. Poverty, wealth, HIVtransmission Gillespie et al. S13 sexual decision-making. In a study of four communities in HIV prevalence becomes even stronger. Finally, they a southern province of Zambia [26] respondents blamed show how the relationship between inequality and HIV is women. Women were perceived to move around and stronger when inequality is generated more by higher ‘give love for money’; women who some believe could proportions of richer men than poorer women. otherwise work hard and do not need to have sex for money. The fact that men, often much older than girls/ Using a combination of data sources on HIV status at the women, pay for sex was rarely mentioned as a cause of individual level and poverty and inequality measures at the problem. the community level, a study in Kenya (K. Beegle, B. Ozler, unpublished data) found, conditional on a set of Pre and extramarital sex may involve the male to female individual and community characteristics, gender transfer of material resources, such as money and gifts. inequality between young women and adult men to be Such exchanges may take the form of commercial sex or significantly correlated with the individual’s HIV-positive more informal transactional sex, which is common in status. This effect is stronger for young women, especially high HIV contexts [26,46,47]. In a study of young Luo in western Kenya where HIV prevalence is highest, and is men in Kenya [21], male to female transfers were given in robust to various definitions of economic inequality three-quarters of recent non-marital partnerships, and between young women and older men. transfers were substantial. Men on average provided approximately US$8.50 (Ksh600) to each non-marital In Botswana and Swaziland, food insufficiency among partner in the past month, equivalent to 9% of a male women was found to be significantly associated with mean monthly income. The author also reported a inconsistent condom use with a non-primary partner negative and significant relationship between the value of (aOR 1.73, 95% CI 1.27–2.36), sex exchange (aOR transfer and reported consistent condom use [21]. For 1.84, 95% CI 1.74–1.93), intergenerational sexual every transfer of (monetary or non-monetary) Ksh500, relationships (aOR 1.46, 95% CI 1.03–2.08), and lack approximately the mean amount given in transfers per of control in sexual relationships (aOR 1.68, 95% CI non-marital partnership, the probability of condom use 1.24–2.28). For men, food insufficiency was associated decreased approximately 8%. with only a 14% increase in the odds of reporting unprotected sex, and was not associated with other risky Evidence points to significant positive associations between sexual behaviours [12]. Although food insufficiency is larger age differences between partners, the value of certainly influenced by income, it is a distinct entity with economic transactions and unsafe sexual behaviours [46– different causes and consequences; there are many steps 48]. In South Africa, low socioeconomic status has been between an aggregated household income variable and found not only to increase female odds of exchanging sex the ability of an individual woman to access, control and for money or goods, but also to raise female chances of use income to buy food. A specific focus on protecting experiencing coerced sex, and male and female odds of and promoting access to food may thus decrease exposure having multiple sexual partners. It also lowers female to HIV, especially among women. chances of abstinence, female and male age at sexual debut, condom use at last sex, and communication with most Mobility recent sexual partner about sensitive topics. Low socio- The link between mobility and the spread of HIV is economic status has more consistent negative effects on determined by the structure of the migration process, the female than on male sexual behaviours; it also raises the conditions under which it occurs, including poverty, female risk of early pregnancy [48]. exploitation, separation from families and partners, and separation from the sociocultural norms that guide A few interesting recent studies have suggested that behaviours within communities [49]. Mobility can increased economic inequality between men and women increase vulnerability to high-risk sexual behaviour as leads to partnerships that are riskier in terms of HIV migrants’ multilocal social networks create opportunities exposure. In one (B. Penman, B. Ozler, K. Beegle, S. Baird, for sexual networking. Mobility also makes individuals unpublished data), a basic model of HIV epidemiology more difficult to reach for preventive, care or treatment was combined with population demographic processes, services. factoring in the marital and economic status of sexually active heterosexual individuals. Using a few simple There is convincing empirical evidence of a link between assumptions regarding partnership patterns, the data human mobility and the risk of HIV transmission. In sub- generate a clear correlation between gender inequality Saharan Africa, the risk of HIV infection has been found (defined by economic inequality between young women to be higher near roads, and among individuals who and older men) and HIV prevalence in a completely either have personal migration experience or have sexual susceptible population after 20–25 years. As expected, if partners who are migrants [18,49–53]. rich men or poor women contribute a higher share of their respective populations to the high sexual activity In eastern and southern Africa, mining, plantations and group, then the relationship between sex inequality and related agricultural industries (typically producing tea,
- 19. S14 AIDS 2007, Vol 21 (suppl 7) coffee, tobacco, sugar cane, and rice) are often associated rates in the Cape Area Panel Study by Dinkelman et al. [11] with situations of significant risk. Risks may be enhanced significantly predicted earlier sexual debut for girls and by regularized single-sex migration as in the case of boys, and higher rates of unprotected recent sex for boys. southern African mines [53]; high and seasonal demands for agriculture labour on estates; workers moving on their The structural context of labour arrangements also own, sometimes from considerable distances and lodged contributes to the demand for transactional sex. Although in single-sex dormitories; long and often irregular pay considered very arduous and physically demanding, cane intervals; and a dependent population of occasional or cutting jobs, for example, command higher monthly commercial sex workers from nearby villages or further wages than most permanent positions. In a Zambian afield [26,54,55]. Ownership structures, the national study [26], in the two worker compounds that make up policy environment, and the economics of the industries the study area in Mazabuka, there was widespread are all important drivers of HIV transmission risk. awareness that married women sleep with cane cutters to access resources they either need or want. The social ecology of HIV The socio-ecological systems perspective of disease Social cohesion and social capital are other important transmission fosters a deliberate analysis of the dynamics conditioning factors. A higher HIV risk has been found to of population patterns of health and wellbeing at each be significantly associated with structural factors related to level of biological, ecological and social organization [45]. the community in a study in Limpopo, South Africa [57]. Although most attention has been paid to the socio- Such factors included easier access to a trading centre, economic conditions of individuals and their households, higher proportions of short-term residents, and lower relatively little attention has been paid to the socio- levels of social capital (particularly significant among ecological conditions that shape norms, behaviours and men); the latter being an index based on social network access to various resources. membership and responses to questions on levels of trust, reciprocity, solidarity in a time of crisis, collective action For example, in a Tanzanian study [50], community (positive) and local serious and violent crime rate characteristics, such as the type of economic activity, ratio (negative). In other words, HIV prevalence was higher of bar girls to men, share of migrants, and distance to big in settings in which the social order had broken down, or cities have all been found to correlate positively with HIV had never been established in the first place. Among men, seroprevalence; traits that are usually associated with higher HIV prevalence was also seen among communities higher income. with easier access to a local mine, a higher density and activity of local bars, a higher numbers of sex workers per Slum populations may be particularly vulnerable. In village, and lower proportions of outmigrants. More Kenya, for example, slum residence has been found to be research is needed on social cohesion and HIV risk. unique in its adverse impact on sexual outcomes, presumably because monetary currency is central to existence in cities where difficult economic circumstances coerce women to use sex as a means of survival [56]. Conclusions Using two separate indicators of deprivation, a Kenyan study has shown that although poverty is significantly In conclusion, this paper has drawn up recent studies to associated with the examined sexual outcomes in all examine what is known about the degree to which, and settings, the urban poor were significantly more likely the ways in which, socioeconomic status is associated than their rural counterparts to have an early sexual with HIV transmission. The notion that poverty is the debut and multiple sexual partnerships, even among main driver of HIV transmission is too simplistic. Relative married women [14]. Complementing their quantitative wealth appears to have a mixed influence on HIV risk evidence with qualitative research, the authors posited depending on an array of contextual factors. Gender that, beyond purely economic factors, other social inequality appears to be particularly important. In the conditions contributed to higher levels of sexual activity most comprehensive multicountry, cross-sectional study in the very poor slum communities. Young children were to date [17], the residual effect of wealth was found to be socialized into sexual behaviour because of: a lack of statistically insignificant after controlling for variables alternate recreational opportunities; residential con- such as urban residence, age, education and differences in ditions that precluded privacy for adult sexual activity; sexual behaviour. There are very few cohort studies that and role modeling by adults who either transacted sex for are able to relate wealth or poverty to the incidence of money or were more generally involved in casual sexual HIV, and they tend to be rurally based. One such study, activity. reported here, shows the highest risk of infection in the middle wealth group. Consistent with these findings, Barnighausen et al. [18] ¨ showed that living in urban and peri-urban areas increases Education in general appears to be protective with regard the hazard of HIV seroconversion. Community poverty to HIV risk, and the interaction effects between
- 20. Poverty, wealth, HIVtransmission Gillespie et al. S15 education and wealth could be very positive; when and women, and to the dynamic and contextual nature individuals have resources and the ability to use those of the relationship between socioeconomic status and resources, they can act on safeguarding their sexual health. HIV. In investigating the relationship between poverty, wealth and HIV it is important to state the following Sponsorship: Financial support was provided by the qualifications: (i) Many of the studies presented in this Joint United Nations Programme on HIV/AIDS volume and elsewhere suffer from important limitations (UNAIDS) and the RENEWAL programme of the such as: low statistical power (especially when measuring International Food Policy Research Institute. HIV incidence); high attrition rates (especially of Conflicts of interest: None. educated, mobile men); difficulty in tracking certain individuals (e.g. mobile commercial sex workers, truck drivers); a paucity of longitudinal nationally representa- References tive panels tracking HIV incidence; and a lack of a 1. Wojciki JM. Socioeconomic status as a risk factor for HIV comprehensive measure of economic status. (ii) On the infection in women in East, Central and Southern Africa: a latter, different aspects of economic status are likely to systematic review. J Biosoc Sci 2005; 37:1–36. behave differently; e.g. assets, income, expenditures, cash 2. Nyindo M. Complementary factors contributing to the rapid spread of HIV-1 in sub-Saharan Africa: a review. East Afr Med J flows, sources and control of income, and their 2005; 82:40–46. intrahousehold and sex differentials. Most studies here 3. Buve A, Lagarde E, Carael M, Rutenberg N, Ferry B, Glynn JR, use an asset-based index of household wealth. (iii) Most et al. Study Group on Heterogeneity of HIV Epidemics in African Cities. Interpreting sexual behavior data: validity issues studies also examine relative not absolute wealth. Many of in the multicentre study on factors determining the differential the wealthiest groups in affected communities may spread of HIV in four African cities. AIDS 2001; 15 (Suppl. 4): actually fall below an absolute poverty line. (iv) Patterns S117–S126. 4. Ainsworth M, Semali I. Who is most likely to die of AIDS? are not uniform across, or even within, countries, and a Socioeconomic correlates of adult deaths in Kagera Region, variety of socioeconomic factors and processes are likely Tanzania. In: Confronting AIDS: evidence for developing world. Edited by Ainsworth M, Fransen L, Over M. Brussels: European to be at play in complex and interlinked ways. (v) The Commission; 1998. relationship between wealth and HIV is dynamic and may 5. Gregson S, Waddell H, Chandiwana S. School education and change over time. Most studies are cross-sectional in HIV control in sub-Saharan Africa: from discord to harmony? J Int Dev 2001; 13:467–485. nature, not longitudinal, and thus ‘snapshots’ at a single 6. Gregson S, Garnett GP, Nyamukapa CA, Hallett TB, Lewis JJ, point in time. It is important therefore to track the rates of Mason PR, et al. HIV decline associated with behaviour change new infections. If incidence is higher in groups moving in eastern Zimbabwe. Science 2006; 311:664–666. 7. Hargreaves JR, Glynn JR. Educational attainment and HIV-1 out of poverty, this implies a greater need to ‘HIV-proof ’ infection in developing countries: a systematic review. Trop poverty reduction so as to make the options for increasing Med Int Health 2002; 7:489–498. wealth less risky. (vi) The role and influence of social 8. Anderson RM, May RM, Boily MC, Garnett GP, Rowley JT. The spread of HIV-1 in Africa – sexual contact patterns and the capital, social cohesion and community-level structural predicted demographic-impact of AIDS. Nature 1991; 352:581– factors is under-researched, little understood but poten- 589. tially very important; especially given the known ´ 9. Glynn JR, Carael M, Buve A, Anagonou S, Zekeng L, Kahindo M, et al. Does increased general schooling protect against HIV association between different forms of inequality and infection? A study in four African cities. Trop Med Int Health risk and vulnerability. (vii) Likewise, the literature is 2004; 9:4–14. somewhat biased towards explaining the relationship 10. De Walque D, Nakiyingi-Miiro JS, Busingye J, Whitworth JA. 2005. Changing association between schooling levels and HIV- between socioeconomic conditions and HIV through 1 infection over 11 years in a rural population cohort in south- high-risk behaviour adoption pathways, with less west Uganda. Trop Med Int Health 2005; 10:993–1001. attention being paid to the ways in which pre-existing 11. Dinkelman T, Lam D, Leibbrandt M. Household and commu- nity income, economic shocks and risky sexual behavior of health and nutritional status may have compromised the young adults: evidence from the Cape Area Panel Study 2002 immune status of individuals [58]. and 2005. AIDS 2007; 21 (Suppl. 7):S49–S56. 12. Weiser SD, Leiter K, Bangsberg DR, Butler ML, Percy-de Korte F, Hlanze Z, et al. Food insufficiency is associated with high risk In summary, when examining the interplay between sexual behavior among women in Botswana and Swaziland. wealth or poverty and HIV transmission, there is no PLoS Medicine 2007; 4:e260. simple explanation, no magic bullet. AIDS cannot 13. Johnson K, Way A. Risk factors for HIV infection in a national adult population: evidence from 2003 Kenya Demographic and accurately be termed a ‘disease of poverty’. Although Health Survey. J Acquir Immune Defic Syndr 2006; 42:627–636. it is true that poor individuals and households are likely 14. Nii-Amoo Dodoo F, Zulu EM, Ezeh AC. Urban–rural differ- ences in the socioeconomic deprivation – sexual behavior link to be hit harder by the downstream impacts of AIDS, in Kenya. Soc Sci Med 2007; 64:1019–1031. their chances of being exposed to HIV in the first place 15. Lopman B, Lewis J, Nyamukapa C, Mushati P, Chandiwana S, are not necessarily greater than wealthier individuals Gregson S. HIV incidence and poverty in Manicaland, Zimbabwe: is HIV becoming a disease of the poor? AIDS or households. What is clear is that approaches to HIV 2007; 21 (Suppl. 7):S57–S66. prevention need to cut across all socioeconomic strata 16. Hargreaves JR, Bonell CP, Morison LA, Kim JC, Phetla G, Porter of society, and they need to be tailored to the specific JDH, et al. Explaining continued high HIV prevalence in South Africa: socioeconomic factors, HIV incidence and sexual drivers of transmission within different groups, with behaviour change among a rural cohort, 2001–2004. AIDS particular attention to the vulnerabilities faced by youth 2007; 21 (Suppl. 7):S39–S48.
- 21. S16 AIDS 2007, Vol 21 (suppl 7) 17. Mishra et al. AIDS 2007; 21 (Suppl. 7):S17–S28. 40. Smith GD, Hart C, Blane D, Hole D. Adverse socioeconomic 18. Barnighausen T, Hosegood V, Timaeus IM, Newell ML. The ¨ conditions in childhood and cause specific adult mortality: socioeconomic determinants of HIV incidence: evidence from prospective observational study. BMJ 1998; 316:1631– a longitudinal, population-based study in rural South Africa. 1635. AIDS 2007; 21 (Suppl. 7):S29–S38. 41. Galobardes B, Lynch JW, Davey SG. Childhood socioeconomic 19. Chapoto A, Jayne T. Socio-economic characteristics of indivi- circumstances and cause-specific mortality in adulthood: duals afflicted by AIDS-related prime-age mortality in Zambia. Systematic review and interpretation. Epidemiol Rev 2004; In: AIDS, poverty and hunger: challenges and responses. Edited 26:7–21. by Gillespie SR. Washington DC: International Food Policy 42. Yen IH, Syme SL. The social environment and health: a dis- Research Institute; 2006. cussion of the epidemiologic literature. Annu Rev Publ Health 20. Kirimi L, Jayne T. Poverty, gender and working-age adult 1999; 20:287–308. mortality: evidence from rural Kenya. Working paper. Nairobi: 43. Pickett KE, Pearl M. Multilevel analyses of neighbourhood Egerton University/Tegemeo Institute; 2007. socioeconomic context and health outcomes: a critical review. 21. Luke N. Exchange and condom use in informal sexual relation- J Epidemiol Commun H 2001; 55:111–122. ships in urban Kenya. Econ Dev Cult Change 2006; 54:319–348. 44. Link BG, Phelan J. Social conditions as fundamental causes of 22. Luke N. Economic status, informal exchange, and sexual risk in disease. J Health Soc Behav 1995; 35:80–94. Kisumu, Kenya. Econ Dev Cult Change 2007; In press. 45. Krieger N. Theories of social epidemiology in the 21st cen- 23. Greener R. The impact of HIV/AIDS on poverty and inequality. tury: an ecosocial perspective. Int J Epidemiol 2001; 30:668– Chapter 5. In: Haacker M, editor. The macroeconomics of HIV/ 677. AIDS. Washington DC: International Monetary Fund; 2004. 46. Luke N. Confronting the ‘sugar daddy’ stereotype: age and 24. Gillespie SR, Kadiyala S. HIV/AIDS and food and nutrition economic asymmetries and risky sexual behavior in urban security: from evidence to action. Food policy review no. 7. Kenya. Int Fam Plan Perspect 2005; 31:6–14. Washington DC: International Food Policy Research Institute; 47. Luke N. Age and economic asymmetries in the sexual relation- 2005. ships of adolescent girls in sub-Saharan Africa. Stud Family 25. Gillespie SR, editor. AIDS, poverty and hunger: challenges and Plann 2003; 34:67–86. responses. Washington DC: International Food Policy Research 48. Hallman K. Socioeconomic disadvantage and unsafe sexual Institute; 2006. behaviors among young women and men in South Africa. Policy 26. Byron E, Gillespie SR, Hamazakaza P. Local perceptions of risk Research Division working paper no. 190. New York: Popula- and HIV prevention in southern Zambia. RENEWAL working tion Council; 2004. paper. 2006. Available at: www.ifpri.org/renewal. Accessed: 49. Crush J, Frayne B, Grant M. Linking migration, HIV/AIDS and October 2007. urban food security in southern and eastern Africa. RENEWAL 27. Bryceson D, Fonseca J. An enduring or dying peasantry: inter- working paper. Washington DC: International Food Policy active impact of famine and HIV/AIDS in rural Malawi. In: Research Institute; 2006. Available at: http://www.ifpri.org/ AIDS, poverty and hunger: challenges and responses. Edited by renewal/renewalpub.asp. Accessed: October 2007. Gillespie SR. Washington DC: IFPRI; 2006. 28. Lewis O. Five families: Mexican case studies in the culture of 50. Bloom SS, Urassa M, Ng’weshemi J, Boerma JT. Community poverty. New York: Basic Books; 1976. effects on the risk of HIV infection in rural Tanzania. Sex 29. Tladi LS. Poverty and HIV/AIDS in South Africa: an empirical Transm Infect 2002; 78:261–266. contribution. SAHARA J 2006; 3:369–381. 51. Zuma K, Gouws E, Williams B, Lurie M. Risk factors for HIV 30. Brook DW, Morojele NK, Zhang C, Brook JS. South African infection among women in Carletonville, South Africa: migra- adolescents: pathways to risky sexual behavior. AIDS Educ Prev tion, demography and sexually transmitted diseases. Int J STD 2006; 18:259–272. AIDS 2003; 14:814–817. 31. Kaufman C, Clark S, Manzini N, May J. Communities, oppor- 52. Boerma JT, Gregson S, Nyamukapa C, Urassa M. Understand- tunities, and adolescents’ sexual behavior in KwaZulu-Natal, ing the uneven spread of HIV within Africa: comparative study South Africa. Stud Fam Plann 2004; 35:261–274. of biologic, behavioral, and contextual factors in rural popula- 32. Kimuna S, Djamba Y. Wealth and extramarital sex among men tions in Tanzania and Zimbabwe. Sex Transm Dis 2003; in Zambia. Int Fam Plan Perspect 2005; 31:83–89. 30:779–787. 33. Tanzania Commission for AIDS (TACAIDS), National Bureau of 53. Lagarde E, Schim M, Enel C, Holmgren B, Dray-Spira R, Pison G, Statistics (NBS), and ORC Macro. Tanzania HIV/AIDS Indicator et al. Mobility and the spread of human immunodeficiency Survey 2003–04. Calverton, Maryland USA: TACAIDS, NBS, virus into rural areas of West Africa. Int J Epidemiol 2003; and ORC Macro; 2005. 32:744–752. 34. Lachaud JP. HIV prevalence and poverty in Africa: micro- and 54. Ngwira N, Bota S, Loevinsohn M. HIV/AIDS, Agriculture and macro-econometric evidences applied to Burkina Faso. J Health food security in Malawi: background to action. RENEWAL Econ 2007; 26:483–504. Epub ahead of print 17 November working paper no. 1. The Hague: International Service for 2006. National Agricultural Research; and Lilongwe: Ministry of 35. Deaton A. Health, inequality, and economic development. Agriculture and Irrigation; 2001. J Econ Lit 2003; 41:113–158. 55. Kisamba Mugerwa W, Nduhura D. Background paper on HIV/ 36. Michelo C, Sandoy IF, Fylkesnes K. Marked HIV prevalence AIDS and agriculture in Uganda. Washington DC: RENEWAL; declines in higher educated young people: evidence from 2002. Available at: www.ifpri.org/renewal. Accessed: October population-based surveys (1995–2003) in Zambia. AIDS 2007. 2006; 20:1031–1038. 56. Zulu EM, Dodoo FN, Ezeh AC. Sexual risk taking in the slums of 37. Cogneau D, Grimm M. Socioeconomic status, sexual behavior, Nairobi, Kenya, 1993–98. Popul Stud – J Demogr 2002; and differential AIDS mortality: evidence from Cote D’Ivoire. 56:311–323. Popul Res Pol Rev 2006; 25:393–407. 57. Pronyk PM, Morison LA, Euripodou R, Phetla G, Hargreaves JR, 38. Font A, Puigpinos R, Chichango IE, Cabrero N, Borrell C. AIDS- Kim JC, et al. The role of structural factors in explaining related knowledge and behaviors in Mozambique. Rev Epide- variations in community HIV prevalence: a study in rural South miol Sante Pub 2006; 54:341–354. Africa. Sex Transm Infect 2007; In press. 39. Sandoy IF, Michelo C, Siziya S, Fylkesnes K. Associations 58. Stillwaggon E. The ecology of poverty, nutrition, parasites and between sexual behaviour change in young people and decline vulnerability to HIV/AIDS. In: AIDS, poverty and hunger: in HIV prevalence in Zambia. BMC Pub Health 2007; 7:60. challenges and responses. Edited by Gillespie SR. Washington Published online 2007 April 23 DC: International Food Policy Research Institute; 2006.
- 22. HIV infection doesnot disproportionately affect the poorer in sub-Saharan Africa Vinod Mishraa, Simona Bignami-Van Asscheb, Robert Greenerc, Martin Vaessena, Rathavuth Honga, Peter D. Ghysc, J. Ties Boermad, Ari Van Asschee, Shane Khana and Shea Rutsteina Background: Wealthier populations do better than poorer ones on most measures of health status, including nutrition, morbidity and mortality, and healthcare utilization. Objectives: This study examines the association between household wealth status and HIV serostatus to identify what characteristics and behaviours are associated with HIV infection, and the role of confounding factors such as place of residence and other risk factors. Methods: Data are from eight national surveys in sub-Saharan Africa (Kenya, Ghana, Burkina Faso, Cameroon, Tanzania, Lesotho, Malawi, and Uganda) conducted during 2003–2005. Dried blood spot samples were collected and tested for HIV, following internationally accepted ethical standards and laboratory procedures. The association between household wealth (measured by an index based on household ownership of durable assets and other amenities) and HIV serostatus is examined using both descriptive and multivariate statistical methods. Results: In all eight countries, adults in the wealthiest quintiles have a higher prevalence of HIV than those in the poorer quintiles. Prevalence increases mono- tonically with wealth in most cases. Similarly for cohabiting couples, the likelihood that one or both partners is HIV infected increases with wealth. The positive association between wealth and HIV prevalence is only partly explained by an association of wealth with other underlying factors, such as place of residence and education, and by differences in sexual behaviour, such as multiple sex partners, condom use, and male circumcision. Conclusion: In sub-Saharan Africa, HIV prevalence does not exhibit the same pattern of association with poverty as most other diseases. HIV programmes should also focus on the wealthier segments of the population. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins AIDS 2007, 21 (suppl 7):S17–S28 Keywords: Africa, AIDS, HIV, poverty, sexual behaviour, wealth Introduction health status, including nutrition, morbidity and mortality, and of healthcare utilization [1–3]. Consistent The relationship between socioeconomic status and with these findings, there is evidence of an inverse health is well documented. There is ample evidence that relationship between socioeconomic status and the risk wealthier populations do better on most measures of of sexually transmitted infections (STI), such as herpes, From the aMacro International Inc., Calverton, Maryland, USA, the bUniversity of Montreal, Montreal, Canada, the cJoint United Nations Programme on HIV/AIDS, Geneva, Switzerland, the dWorld Health Organization, Geneva, Switzerland, and the eHEC Montreal, Montreal, Canada. Correspondence to Vinod Mishra, DHR Division, Macro International Inc., 11785 Beltsville Drive, Calverton, MD 20705, USA. Tel: +1 301 572 0220; fax: +1 301 572 0999; e-mail: vinod.mishra@macrointernational.com ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S17
- 23. S18 AIDS 2007, Vol 21 (suppl 7) chlamydia, gonorrhoea, syphilis, and bacterial vaginosis SOCIAL AND EPIDEMIOLOGICAL CONTEXT [4–13]. Although much of this evidence is from Social attitudes and practices, level of economic development, stage of HIV/AIDS epidemic, HIV prevalence, availability and access to prevention and treatment methods developed countries, it is reasonable to expect that poverty increases vulnerability to HIV in the same UNDERLYING PROXIMATE manner in low and middle-income countries. It is indeed FACTORS FACTORS OUTCOME often argued that poverty is the root cause of the spread Sexual behaviour of HIV [14]. A recent article in the Lancet argued Abstinence that ‘[s]ince poverty plays a role in creating an Sexual debut Multiple sexual partners environment in which individuals are particularly Concurrent partners susceptible and vulnerable to HIV/AIDS, poverty Wealth Non-regular partners Commercial sex reduction will undoubtedly be at the core of a sustainable Partner faithfulness Type of sexual activity Age solution to HIV/AIDS’ [15]. Analogous views have been Sex/gender (MSM, anal, oral, etc.) Ethnicity expressed in numerous public statements and publi- Religion Transmission cofactors Urban/rural residence cations, and guide HIV prevention efforts in several Geographical region Other STI countries. Education Male circumcision HIV status Nutritional status Occupation Exposure to media At the global level, there is evidence of a positive Marital status Treatment & care Duration in union correlation between countries’ HIV prevalence and Mobility Treatment & care Alcohol use ART poverty, as measured by per capita income, income HIV knowledge & attitudes inequality, or absolute poverty [16]. The HIVepidemic in Other risk factors sub-Saharan Africa represents a notable exception to this Condom use general pattern. On the one hand, at the macro level Injecting drug use Medical injections African nations with high HIV prevalence, such as South Blood transfusion Skin cutting/tattooing Africa and Botswana, tend to be the wealthier countries in the region [17,18]. On the other hand, at the individual level, wealth has been found to be positively associated with HIV serostatus [19–21]. Reviews of the existing literature about the association between socioeconomic Fig. 1. Association between wealth status and HIV preva- status and HIV infection indicate that only a few studies lence: a conceptual framework. have found a negative association, whereas most have found a positive or no association [22,23]. To account for and epidemiological context. For wealth to have an effect this finding, it has been argued that a greater prevalence of on HIV incidence and prevalence, the underlying factors risky sexual behaviours among the wealthier may increase must affect one or more of the proximate factors, which their vulnerability to HIV infection, whereas better in turn affect either the rate of infection or the duration of nutritional status, greater access to healthcare, and greater infectivity with HIV [24] (a detailed discussion of this use of antiretroviral drugs may improve their survival if conceptual framework and possible pathways between infected [21]. wealth and HIV status is provided elsewhere [25]). In this study, using data from eight recent population- We analysed data from six Demographic and Health based, nationally representative surveys with HIV Surveys (DHS; Kenya, Malawi, Lesotho, Cameroon, testing in sub-Saharan Africa, we conducted an in- Ghana, and Burkina Faso) and two AIDS Indicator depth analysis of the association between household Surveys (AIS; Tanzania and Uganda) with linked HIV test wealth status and HIV prevalence in sub-Saharan Africa. results. These surveys, conducted during the period Our aim is to identify what specific characteristics and 2003–2005, collected sociodemographic and behavioural behaviours of the wealthier are associated with HIV data as well as blood samples for HIV testing from infection, and to what extent confounding factors such as nationally representative samples of adult women and place of residence and other risk factors mediate this men. Table 1 gives basic information about the eight association. surveys. The sampling design and survey implementation procedures for each country are described in detail in the individual country survey reports [26–33]. Although the age ranges of adults included in the surveys varied across Methods countries (see Table 1), for consistency the present analysis is limited to men and women aged 15–49 years in We conceptualized the association between wealth and each country. HIV as being influenced by a host of underlying background factors and mediated by several proximate In all surveys, HIV testing was carried out using dried factors (Fig. 1). We viewed this relationship as transitional blood spot samples collected on a special filter paper using in nature, operating within and depending on the social capillary blood from a finger prick, except in Uganda
- 24. Wealth and HIVin sub-Saharan Africa Mishra et al. S19 Table 1. Number of men and women interviewed and tested for HIV, Demographic and Health Surveys/AIDS Indicator Surveys with linked HIV testing. Country (year) No. eligible for No. Interview No. eligible for No. tested HIV response No. ever Sex (age group, years) interview interviewed response rate (%) HIV testing for HIV rate (%) had sex Kenya 2003 Male (15–54) 4183 3578 86 4183 2941 70 3038 Female (15–49) 8717 8195 94 4303 3285 76 6784 Tanzania 2003/04 Male (15–49) 6194 5659 91 6194 4774 77 4690 Female (15–49) 7154 6863 96 7154 5973 83 5963 Uganda 2004/05 Male (15–59) 9905 8830 89 9905 8298 84 7390 Female (15–59) 11 454 10 826 95 11 454 10 227 89 9483 Malawi 2004 Male (15–54) 3797 3261 86 3797 2404 63 2863 Female (15–49) 12 229 11 698 96 4071 2864 70 10 397 Lesotho 2004/05 Male (15–59) 3305 2797 85 3305 2246 68 2291 Female (15–49) 7522 7095 94 3758 3032 81 5917 Cameroon 2004 Male (15–59) 5676 5280 93 5676 5098 90 4424 Female (15–49) 11 304 10 656 94 5738 5287 92 9280 Ghana 2003 Male (15–59) 5345 5015 94 5345 4274 80 3861 Female (15–49) 5949 5691 96 5949 5311 89 4807 Burkina Faso 2003 Male (15–59) 3984 3605 90 3984 3418 86 2769 Female (15–49) 12 952 12 477 96 4575 4223 92 10 911 where venous blood was collected. Participation in HIV by means of an index based on household ownership of testing was voluntary and, before collecting blood samples consumer durables (such as a television and a bicycle; for HIV testing, each selected participant was asked to materials used for housing construction; and the avail- provide informed consent to the testing [34]. Informed ability of amenities such as electricity, source of drinking consent was obtained separately for the questionnaire water, and type of toilet facility) that tend to be correlated interview. In each country, HIV testing was conducted in with household economic status. The index, constructed a central laboratory by following a standard testing using principal components analysis, is a composite algorithm designed to maximize the sensitivity and measure of the cumulative living standard of a household, specificity of HIV test results, and an approved quality which places individual households on a continuous scale assurance and quality control plan [35]. The testing of relative wealth [36,37]. The wealth index is divided algorithm used two HIV enzyme immunosorbent assays, into population quintiles, with the lowest quintile based on different antigens. All discordant samples that representing the poorest 20% and the highest quintile were positive on the first test and negative on the second representing the wealthiest 20% of households within test were retested with the same enzyme immunosorbent each country. The wealth index defined in this manner assays, and if still discordant, were resolved by Western captures well the relative economic status within each blot testing. These steps were also repeated for 5–10% of country, and it correlates strongly with the health and randomly selected samples that tested negative on the first wellbeing of people [37]. test. For external quality assessment, a subset of dried blood spot samples (usually approximately 5%) was Using the conceptual framework illustrated in Fig. 1, we retested at an outside reference laboratory using the systematically examined the association between wealth same algorithm. and HIV infection. For each country, we first descrip- tively evaluated whether household wealth was associated In order to ensure confidentiality, the HIV test results with key risk behaviours and protective factors, including were anonymously linked to individual and household those that may increase the risk of HIV exposure [age at questionnaire information through bar codes, after first sexual intercourse, age at first cohabitation, number scrambling the household and cluster identifiers [35]. of times married, duration in current union, polygamy, All HIV testing procedures were reviewed by the ethical partner living elsewhere (for women only), number of review boards of Macro International Inc. (a US-based lifetime and recent sexual partners, sexual intercourse company that provides technical assistance to DHS/AIS with a non-regular (non-marital, non-cohabiting) part- surveys around the world), and the host country. ner], and those that may be associated with an increased risk of transmission per exposure [condom use with the As DHS/AIS surveys do not include direct questions on last non-regular partner and consistent condom use, income or expenditure, we measured household wealth alcohol use at last sexual intercourse, reported STI or STI
- 25. S20 AIDS 2007, Vol 21 (suppl 7) symptoms, circumcision (for men only)]. We also the exception of men in Ghana and Lesotho where HIV examined the association with knowledge of the prevalence in the highest wealth quintile is slightly lower respondent’s HIV status, and knowledge of HIV than in the lowest quintile. In most cases, HIV prevalence prevention methods. increases monotonically with household wealth status, with the notable exception of Ghana where there is an We used multivariate logistic regression to measure the inverted U-shaped relationship between the two. independent relationship between household wealth and Similarly, for cohabiting couples the likelihood that HIV status after controlling for underlying and mediating one or both partners is HIV infected increases proximate factors. In particular, for women and men aged with household wealth, with the wealthiest 20% of 15–49 years who reported ever having sex, we estimated couples being two to seven times more likely to have five alternative logistic regression models. The first model HIV than the poorest 20%. The only exception is estimated unadjusted effects of household wealth on HIV Lesotho, where this ratio is smaller (1.2) but in the same prevalence. The second model added controls for several direction. underlying background factors, including age, ethnicity, religion, urban/rural residence, and geographical region Wealthier men and women tend to be more educated, of residence. The third model additionally controlled for more mobile, and more likely to live in urban areas, education, occupation, media exposure, marital status, where HIV is more prevalent. Wealthier men and duration in union, number of years at current place of women are also more likely to be older, regularly residence, alcohol use at last sex in the previous 12 exposed to mass media, to be working in professional or months, knowledge of HIV prevention methods, and service jobs than poorer individuals (not shown). In knowledge of own HIV status. The fourth model added a addition, consistently across countries, wealthier indi- set of proximate factors that were likely to mediate the viduals tend to start cohabiting at an older age than relationship between the underlying background factors poorer individuals, with an average age difference and HIV prevalence (as indicated in Fig. 1). These between the highest and lowest wealth quintile of 2– included age at first sexual intercourse, the number of 4 years in most cases (Tables 3 and 4). Wealthier men lifetime sexual partners (replaced with whether the and women are less likely to be in a polygamous union respondent had two or more partners in the previous than poorer men and women. Knowledge of HIV 12 months in Kenya, Ghana, Burkina Faso, and prevention (being faithful to one’s regular partner and Malawi, where information on lifetime partners was using condoms) increases with household wealth for not available), reported STI or STI symptoms in the both men and women in all countries, except for men in previous 12 months, circumcision (for men only), and Tanzania, Malawi, and Cameroon and for women consistent condom use in the previous 12 months. Finally, in Lesotho, where there is little difference in such the fifth model added a control for community-level knowledge by wealth status. Wealthier men are more wealth, computed by averaging the household wealth likely to report having had two or more sexual partners scores in each cluster. We also fitted a similar set of models in the past 12 months than poorer men, with the notable to the data for cohabiting couples to examine the exception of Tanzania where the pattern is reversed. association between household wealth and the likelihood Wealthier men also tend to have more lifetime sexual that one or both partners was HIV positive. Results from partners and are more likely to have sex with non-regular only the first model (unadjusted) and the fifth model partners than poorer men. In all countries, wealthier (adjusted for all potential confounders and mediating men and women are more likely to use condoms factors) are presented. The analysis accounts for the than poorer individuals. Also, wealthier men are more complex survey design of the DHS/AIS to estimate likely than poorer men to be circumcised, except in efficient regression coefficients and robust standard errors Lesotho. adjusting for intracluster correlation and by using sampling weights. In Table 5 we present unadjusted and adjusted odds of HIV infection by wealth quintile. Unadjusted odds indicate that, in all countries except Lesotho and Ghana, men belonging to the highest wealth quintile are more Results likely to be HIV infected than those belonging to the lowest wealth quintile. In Lesotho and Ghana, there is The overall HIV prevalence in the eight countries an inverted U-shaped relationship between household considered ranges from 1.8% in Burkina Faso to 23.5% in wealth and HIV prevalence among men; in other Lesotho, with women having a higher HIV prevalence terms, the odds of HIV infection peak in the middle than men in all countries except Burkina Faso (Table 2). wealth quintile. Higher HIV prevalence with increasing In all countries, HIV prevalence tends to be much higher household wealth is also observed for women; the odds of among adults belonging to the wealthiest 20% of HIV infection are two to five times greater in the highest households than among those from the poorest 20%. wealth quintile than in the lowest wealth quintile This pattern holds for men and women separately, with (statistically significant in all countries). This suggests a
- 26. Wealth and HIVin sub-Saharan Africa Mishra et al. S21 Table 2. HIV prevalence among men and women aged 15–49 years, and among cohabiting couples, by household wealth status, Demographic and Health Surveys/AIDS Indicator Surveys with linked HIV testing. HIV prevalence Country/wealth index Men Women Total (men and women) Cohabiting couples (either/both HIVþ) Kenya 4.6 8.7 6.7 11.1 Lowest 3.4 3.9 3.6 8.0 Second 4.2 8.5 6.5 11.0 Middle 2.2 7.1 4.8 9.7 Fourth 4.3 9.7 7.1 9.9 Highest 7.3 12.2 9.8 16.5 Number 2851 3151 6001 1116 Tanzania 6.3 7.7 7.0 10.7 Lowest 4.1 2.8 3.4 5.2 Second 4.3 4.7 4.5 7.7 Middle 4.3 6.8 5.6 9.9 Fourth 7.7 10.9 9.4 13.8 Highest 9.5 11.4 10.5 17.6 Number 4994 5753 10 747 2219 Uganda 5.0 7.5 6.4 8.1 Lowest 4.0 4.8 4.5 4.9 Second 4.2 6.6 5.5 6.6 Middle 5.1 6.7 6.0 8.4 Fourth 5.9 7.0 6.5 9.4 Highest 5.5 11.0 8.6 11.0 Number 7515 9391 16 906 3882 Malawi 10.2 13.3 11.8 16.7 Lowest 4.4 10.9 8.3 7.2 Second 4.6 10.3 7.6 10.2 Middle 12.1 12.7 12.4 19.7 Fourth 11.7 14.6 13.2 19.5 Highest 14.9 18.0 16.4 26.7 Number 2465 2686 5150 1324 Lesotho 19.3 26.4 23.5 32.5 Lowest 18.3 19.6 19.1 27.3 Second 16.8 27.9 23.3 30.5 Middle 23.7 25.5 24.6 37.0 Fourth 21.6 27.3 25.0 34.8 Highest 14.8 28.9 24.3 33.6 Number 2012 3031 5043 593 Cameroon 4.1 6.8 5.5 7.4 Lowest 1.4 3.1 2.4 2.7 Second 2.2 4.1 3.2 4.6 Middle 4.7 8.1 6.5 9.5 Fourth 5.3 9.4 7.4 11.4 Highest 5.3 8.0 6.6 12.3 Number 4672 5227 9900 2027 Ghana 1.5 2.7 2.2 4.2 Lowest 1.4 1.4 1.4 2.8 Second 1.5 2.7 2.2 3.4 Middle 2.0 4.0 3.1 5.4 Fourth 1.4 3.0 2.2 4.7 Highest 1.1 2.4 1.9 5.1 Number 4045 5097 9142 1790 Burkina Faso 1.9 1.8 1.8 3.1 Lowest 1.4 0.9 1.1 0.8 Second 2.9 1.1 1.9 4.7 Middle 1.3 1.5 1.4 2.6 Fourth 0.4 1.7 1.1 1.6 Highest 2.7 3.4 3.1 5.8 Number 3065 4086 7151 2230 stronger positive effect of wealth on HIV infection among selected proximate factors, and community-level wealth women than among men. are controlled for in alternative models (not shown). Even with all underlying and proximate factors The strong, positive association between wealth and controlled, however, the odds of HIV infection HIV infection in the unadjusted models is diminished remain greater than one in the highest wealth quintile considerably when a number of underlying factors, in four of the eight countries considered for men,
- 27. S22 AIDS Table 3. Selected behaviours of men aged 15–49 years, by household wealth status, Demographic and Health Surveys/AIDS Indicator Surveys with linked HIV testing. Country/ Median Median % Married % With % In % With 3þ % With 2þ % Had sex with % Used condom % Used % Used % With % % Know % Know wealth age at age at more than 10þ years polygamous lifetime sex sex partners in a non-regular with non-regular condom alcohol at STI/STI Circumcised own HIV about HIV 2007, Vol 21 (suppl 7) status 1st sex cohabitation once in union union partners past 12 months partner partner consistently last sex symptoms status prevention Kenya 17.0 25.1 14 51 4 n/a 17 11 46 16 n/a 3 84 14 75 Lowest 17.0 23.5 14 56 10 n/a 15 9 25 9 n/a 5 76 10 71 Second 16.4 24.7 18 55 4 n/a 17 11 37 15 n/a 3 83 11 74 Middle 16.7 24.5 13 54 3 n/a 14 8 25 11 n/a 2 89 11 72 Fourth 17.3 25.8 11 53 4 n/a 15 11 46 18 n/a 2 87 12 74 Highest 17.4 25.9 15 45 3 n/a 19 13 67 21 n/a 3 84 23 82 Number 2507 1825 1752 1615 3363 n/a 2380 1613 171 2379 n/a 2822 3355 3343 3113 Tanzania 18.7 24.1 30 51 5 66 27 23 53 19 12 6 70 13 69 Lowest 18.3 22.9 32 49 5 68 30 26 33 11 13 7 59 6 68 Second 18.6 23.3 33 51 7 66 28 24 46 12 10 6 55 9 69 Middle 18.9 23.1 38 57 8 63 27 24 56 15 14 6 58 11 68 Fourth 18.6 24.8 28 54 4 66 27 25 66 21 14 7 75 15 70 Highest 18.8 26.1 23 44 3 68 24 19 65 30 11 5 92 22 70 Number 4304 3291 3313 3000 5656 4556 4181 2999 694 4171 4181 4682 5649 5656 5646 Uganda 18.4 21.9 27 62 11 67 29 18 54 14 28 21 25 11 73 Lowest 18.5 21.3 27 66 14 61 23 11 44 7 41 16 19 5 65 Second 18.4 21.3 28 64 10 65 25 15 38 10 30 18 19 7 70 Middle 18.3 21.7 30 66 13 68 28 16 45 9 31 20 25 6 72 Fourth 18.3 21.7 28 60 12 70 31 20 46 13 25 25 25 10 73 Highest 18.3 23.7 22 54 9 71 36 26 80 28 16 23 33 21 80 Number 5940 4678 4870 4223 8010 6393 5642 4228 760 5635 5634 6567 8003 8010 7939 Malawi 18.5 22.9 23 49 6 n/a 12 7 46 15 n/a 6 21 15 65 Lowest 18.6 22.5 20 45 5 n/a 10 5 31 14 n/a 3 18 10 64 Second 18.4 22.1 25 48 7 n/a 11 6 45 12 n/a 7 23 8 67 Middle 18.5 22.6 22 43 7 n/a 13 9 39 10 n/a 7 21 14 64 Fourth 18.3 22.2 26 60 7 n/a 13 8 45 12 n/a 6 23 14 64 Highest 18.6 25.3 21 45 4 n/a 11 6 67 24 n/a 5 20 25 64 Number 2464 1877 2030 1936 3114 n/a 2402 1894 133 2402 n/a 2713 3032 3098 3057 Lesotho 19.0 25.5 6 49 n/a 66 30 31 39 27 13 12 47 10 65 Lowest 20.0 24.3 6 49 n/a 63 31 32 18 10 18 13 70 5 53 Second 19.2 24.5 4 46 n/a 61 27 30 19 18 9 16 55 7 57 Middle 18.7 25.5 4 47 n/a 65 30 36 30 25 10 14 49 9 62 Fourth 18.9 26.7 6 47 n/a 66 30 31 56 31 12 9 40 13 68 Highest 18.8 25.9 7 55 n/a 71 33 29 65 44 14 10 28 12 76 Number 1753 1246 1083 952 n/a 1916 1742 903 282 1742 1742 1987 2489 2325 2325 Cameroon 18.3 24.9 39 49 5 80 40 39 55 28 25 9 93 14 75 Lowest 19.3 22.2 45 56 14 70 35 19 40 8 34 6 75 5 73 Second 18.9 23.9 40 48 7 76 38 31 37 11 27 6 87 6 75 Middle 18.3 24.5 40 51 6 79 41 44 46 23 23 8 97 10 77 Fourth 18.3 25.7 37 49 3 81 38 45 62 35 24 11 98 15 74 Highest 17.8 27.1 37 44 2 85 46 51 69 45 22 12 98 24 76 Number 3591 2638 2703 2271 4815 3949 3660 2214 857 3638 3658 3957 4811 4777 4776 Ghana 20.0 24.6 28 58 6 n/a 15 14 44 19 16 4 95 8 78 Lowest 20.1 24.3 20 57 12 n/a 14 10 27 8 14 5 83 3 69 Second 20.0 23.8 30 62 6 n/a 11 11 38 12 16 5 96 3 77 Middle 19.6 23.7 29 62 6 n/a 16 15 42 17 16 4 98 6 79 Fourth 20.1 24.4 31 55 3 n/a 16 18 37 26 15 5 98 9 80 Highest 20.1 27.2 29 55 3 n/a 19 17 66 28 16 3 99 14 83 Number 3422 2738 2489 2228 4529 n/a 2905 2227 318 2906 2904 3373 4529 4497 4497 Burkina Faso 20.5 25.5 32 54 12 n/a 24 12 72 29 n/a 4 90 n/a 71 Lowest 21.0 24.8 27 49 11 n/a 13 8 48 10 n/a 3 88 n/a 63 Second 20.8 25.2 34 56 15 n/a 23 10 33 16 n/a 2 83 n/a 74 Middle 20.6 24.2 36 55 16 n/a 22 9 77 23 n/a 3 90 n/a 72 Fourth 20.5 24.8 40 61 18 n/a 22 8 72 23 n/a 6 91 n/a 66 Highest 19.6 27.8 27 48 5 n/a 31 22 93 52 n/a 5 97 n/a 76 Number 2332 1769 1689 1636 3209 n/a 2014 1631 194 2014 n/a 2373 3209 n/a 2696 STI, Sexually transmitted infection.
- 28. Table 4. Selectedbehaviours of women aged 15–49 years, by household wealth status, Demographic and Health Surveys/AIDS Indicator Surveys with linked HIV testing. Country/ Median Median % Married % With % In % Partner % With 3þ % With 2þ sex % Had sex with % Used condom % Used % Used % With % Know % Know wealth age at age at more than 10þ years polygamous living lifetime sex partners in a non-regular with non-regular condom alcohol at STI/STI own HIV about HIV status 1st sex cohabitation once in union union elsewhere partners past 12 months partner partner consistently last sex symptoms status prevention Kenya 17.8 19.7 7 54 10 22 n/a 3 2 20 5 n/a 4 13 67 Lowest 16.5 17.8 11 60 18 20 n/a 3 3 6 2 n/a 4 5 59 Second 17.0 19.0 8 55 12 25 n/a 2 3 18 4 n/a 5 9 63 Middle 17.3 19.3 6 57 9 27 n/a 2 2 14 3 n/a 4 12 66 Fourth 18.2 20.2 6 57 8 24 n/a 2 1 38 5 n/a 5 13 68 Highest 18.8 22.0 7 43 6 15 n/a 3 2 34 11 n/a 4 23 74 Number 6339 4648 5752 4919 8195 4914 n/a 5709 4906 110 5705 n/a 6778 8070 7056 Tanzania 17.6 18.7 19 54 6 n/a 32 6 5 38 11 16 5 13 63 Lowest 16.7 17.9 24 55 8 n/a 35 7 7 27 5 19 7 5 56 Second 16.9 18.2 22 55 6 n/a 35 7 6 35 7 15 5 6 61 Middle 17.4 18.2 21 56 9 n/a 34 7 5 35 8 18 6 7 64 Fourth 17.7 18.8 18 58 6 n/a 28 5 4 58 13 16 4 14 66 Highest 18.6 20.4 12 48 3 n/a 31 5 3 49 22 15 5 27 67 Number 5373 3993 5176 4354 6863 n/a 5949 5289 4362 204 5284 5286 6853 6863 6801 Uganda 16.8 17.7 23 61 21 n/a 31 4 3 49 9 32 33 13 64 Lowest 16.9 17.5 23 65 22 n/a 23 4 2 36 5 47 23 6 49 Second 16.6 17.4 24 65 20 n/a 27 3 3 28 6 36 30 9 57 Middle 16.7 17.4 25 63 22 n/a 28 3 3 40 5 36 34 8 62 Fourth 16.7 17.8 23 60 22 n/a 33 4 3 55 8 28 36 11 67 Highest 16.8 18.6 18 52 18 n/a 40 5 4 70 19 20 38 25 77 Number 7755 5822 7720 6290 9941 n/a 8549 7387 6349 183 7376 7368 8596 9941 9801 Malawi 17.2 17.9 23 50 11 45 n/a 1 1 26 5 n/a 8 13 76 Lowest 16.7 17.7 31 50 13 31 n/a 2 1 33 3 n/a 10 8 66 Second 16.9 17.8 25 49 13 38 n/a 1 1 14 3 n/a 9 11 71 Middle 17.0 17.6 25 49 11 46 n/a 1 1 30 4 n/a 8 12 76 Fourth 17.4 17.8 21 52 12 54 n/a 1 1 12 5 n/a 8 13 78 Highest 18.1 18.8 12 48 6 52 n/a 1 1 54 10 n/a 8 16 80 Number 9306 6436 9728 8312 11 698 8305 n/a 9087 8004 67 9081 n/a 10 354 11 513 10 960 Lesotho 18.6 19.1 3 54 n/a 45 n/a 11 12 33 19 12 15 13 50 Wealth and HIV in sub-Saharan Africa Mishra et al. Lowest 17.9 18.1 4 45 n/a 31 n/a 14 15 18 6 14 16 9 47 Second 18.3 18.4 3 51 n/a 38 n/a 13 14 28 9 15 18 10 50 Middle 18.5 18.8 3 52 n/a 46 n/a 10 10 30 15 12 16 10 50 Fourth 18.7 19.2 3 56 n/a 54 n/a 10 10 39 20 11 16 12 52 Highest 19.2 20.3 4 64 n/a 52 n/a 10 10 50 34 10 12 22 49 Number 5385 3922 4722 3709 n/a 3694 n/a 4981 3704 437 4980 4981 1990 6638 6640 Cameroon 16.4 17.6 23 54 20 22 43 8 14 39 15 19 12 12 63 Lowest 15.7 15.8 24 64 35 11 21 2 4 15 3 24 5 2 45 Second 15.6 16.5 26 56 24 18 35 6 9 20 6 19 9 4 56 Middle 16.2 17.6 26 54 21 25 46 7 15 29 11 19 13 8 62 Fourth 16.8 18.1 22 50 15 28 52 10 21 43 22 18 14 15 70 Highest 17.7 20.7 16 45 9 28 59 12 23 50 28 17 17 24 77 Number 7972 5720 8096 7166 10 656 7139 9252 8060 6570 929 7923 8047 9278 10 352 10 422 Ghana 18.1 19.4 27 63 14 30 n/a 2 3 17 8 10 8 7 70 Lowest 17.5 18.7 22 63 29 19 n/a 0 2 18 3 9 7 3 64 Second 17.6 18.7 29 65 18 28 n/a 2 3 0 5 11 8 5 67 Middle 17.8 18.9 31 66 16 33 n/a 1 3 7 5 9 7 6 72 Fourth 18.1 19.3 28 59 9 39 n/a 1 4 12 12 8 8 9 71 Highest 19.1 21.7 23 62 5 31 n/a 3 5 34 15 11 11 11 74 Number 4543 3531 4075 3549 5691 3531 n/a 3863 3545 115 3863 3862 4805 5590 5597 Burkina Faso 17.4 17.7 12 61 37 8 n/a 1 1 38 9 n/a 5 n/a 64 Lowest 17.2 17.4 14 63 33 8 n/a 1 0 26 3 n/a 1 n/a 52 Second 17.4 17.5 14 61 43 5 n/a 1 0 40 3 n/a 2 n/a 60 Middle 17.4 17.7 11 62 46 7 n/a 1 1 40 4 n/a 2 n/a 61 Fourth 17.4 17.7 12 61 49 10 n/a 2 1 43 6 n/a 3 n/a 68 Highest 17.9 18.7 9 57 19 11 n/a 3 2 37 23 n/a 14 n/a 74 Number 9701 7427 10 140 9655 12 477 9626 n/a 8168 9635 87 8167 n/a 10 910 n/a 8747 STI, Sexually transmitted infection. S23
- 29. S24 AIDS Table 5. Odds ratio estimates of effects of wealth status on the likelihood of being HIV infected among men and women aged 15–49 years who ever had sex, and on the likelihood of one or both partners being HIV infected among cohabiting couples, Demographic and Health Surveys/AIDS Indicator Surveys with linked HIV testing. Country/ Men Women Cohabiting couples 2007, Vol 21 (suppl 7) wealth status Unadjusted Adjusted Unadjusted Adjusted Unadjusted Adjusted OR (95% CI; P value) OR (95% CI; P value) OR (95% CI; P value) OR (95% CI; P value) OR (95% CI; P value) OR (95% CI; P value) Kenya Lowesta 1.00 1.00 1.00 1.00 1.00 1.00 Second 1.28 (0.60, 2.73; 0.516) 1.73 (0.81, 3.70; 0.158) 2.26 (1.21, 4.23; 0.011) 1.82 (0.88, 3.73; 0.105) 1.42 (0.66, 3.02; 0.367) 1.57 (0.65, 3.80; 0.318) Middle 0.70 (0.24, 2.03; 0.505) 1.02 (0.34, 3.09; 0.967) 1.98 (1.11, 3.55; 0.022) 2.13 (1.05, 4.33; 0.036) 1.23 (0.60, 2.54; 0.575) 1.86 (0.77, 4.45; 0.166) Fourth 1.20 (0.59, 2.46; 0.611) 1.64 (0.65, 4.13; 0.295) 2.76 (1.58, 4.81; 0.000) 1.86 (0.86, 4.03; 0.114) 1.26 (0.57, 2.76; 0.564) 2.20 (0.88, 5.52; 0.093) Highest 2.07 (1.07, 4.04; 0.032) 2.35 (0.81, 6.87; 0.117) 3.36 (1.90, 5.95; 0.000) 1.76 (0.63, 4.89; 0.280) 2.28 (1.14, 4.54; 0.019) 1.65 (0.40, 6.86; 0.491) Number 2266 2048 2734 2217 1083 1015 Tanzania Lowesta 1.00 1.00 1.00 1.00 1.00 1.00 Second 1.01 (0.59, 1.75; 0.963) 0.85 (0.48, 1.50; 0.576) 1.65 (1.02, 2.69; 0.042) 1.55 (0.94, 2.56; 0.089) 1.52 (0.74, 3.15; 0.255) 1.42 (0.67, 3.02; 0.361) Middle 1.09 (0.66, 1.78; 0.738) 0.85 (0.50, 1.43; 0.533) 2.61 (1.54, 4.41; 0.000) 2.29 (1.31, 4.00; 0.004) 2.00 (0.99, 4.04; 0.053) 1.91 (0.90, 4.04; 0.090) Fourth 2.00 (1.21, 3.30; 0.007) 1.41 (0.81, 2.48; 0.227) 4.40 (2.87, 6.75; 0.000) 3.51 (2.12, 5.80; 0.000) 2.93 (1.44, 5.95; 0.003) 2.83 (1.27, 6.30; 0.011) Highest 2.35 (1.48, 3.74; 0.000) 1.56 (0.74, 3.29; 0.247) 5.07 (3.32, 7.75; 0.000) 3.39 (1.73, 6.64; 0.000) 3.90 (1.99, 7.62; 0.000) 3.96 (1.59, 9.89; 0.003) Number 3948 3847 5214 5149 2220 2174 Uganda Lowesta 1.00 1.00 1.00 1.00 1.00 1.00 Second 1.08 (0.70, 1.68; 0.717) 0.99 (0.61, 1.60; 0.954) 1.36 (0.98, 1.89; 0.063) 0.98 (0.68, 1.41; 0.928) 1.37 (0.86, 2.18; 0.186) 1.14 (0.69, 1.87; 0.617) Middle 1.24 (0.82, 1.90; 0.309) 0.97 (0.61, 1.54; 0.893) 1.35 (0.97, 1.88; 0.078) 0.96 (0.65, 1.41; 0.835) 1.76 (1.12, 2.75; 0.014) 1.14 (0.70, 1.87; 0.596) Fourth 1.50 (1.02, 2.19; 0.037) 1.06 (0.69, 1.62; 0.805) 1.46 (1.02, 2.07; 0.037) 0.98 (0.66, 1.44; 0.902) 2.01 (1.34, 3.01; 0.001) 1.42 (0.91, 2.22; 0.118) Highest 1.43 (0.93, 2.20; 0.104) 0.89 (0.50, 1.60; 0.706) 2.44 (1.80, 3.31; 0.000) 1.41 (0.90, 2.20; 0.133) 2.37 (1.51, 3.72; 0.000) 1.30 (0.70, 2.40; 0.412) Number 6141 5755 8094 7538 3949 3672 Malawi Lowesta 1.00 1.00 1.00 1.00 1.00 1.00 Second 1.08 (0.47, 2.48; 0.864) 0.85 (0.33, 2.19; 0.742) 0.89 (0.56, 1.42; 0.626) 1.00 (0.59, 1.69; 0.998) 1.46 (0.65, 3.28; 0.363) 1.29 (0.52,3.17; 0.582) Middle 2.93 (1.37, 6.25; 0.006) 2.44 (0.99, 6.02; 0.052) 1.15 (0.77, 1.73; 0.498) 1.14 (0.69, 1.91; 0.604) 3.17 (1.44, 6.96; 0.004) 3.06 (1.29,7.25; 0.011) Fourth 2.98 (1.39, 6.37; 0.005) 2.49 (0.98, 6.34; 0.056) 1.41 (0.95, 2.08; 0.090) 1.51 (0.92, 2.49; 0.104) 3.11 (1.40, 6.92; 0.005) 2.84 (1.17,6.92; 0.022) Highest 4.12 (1.89, 8.94; 0.000) 2.82 (1.02, 7.76; 0.045) 1.96 (1.28, 3.00; 0.002) 1.19 (0.63, 2.25; 0.586) 4.68 (2.07, 10.61; 0.000) 2.43 (0.94,6.26; 0.066) Number 2031 1948 2609 2458 1297 1268 Lesotho Lowesta 1.00 1.00 1.00 1.00 1.00 1.00 Second 0.91 (0.59, 1.41; 0.678) 1.07 (0.64, 1.78; 0.790) 1.49 (1.07, 2.07; 0.017) 1.42 (0.97, 2.07; 0.069) 1.17 (0.69, 1.97; 0.564) 1.42 (0.74, 2.74; 0.294) Middle 1.47 (0.96, 2.24; 0.077) 1.65 (0.90, 3.02; 0.103) 1.42 (1.01, 1.98; 0.041) 1.38 (0.92, 2.07; 0.122) 1.56 (0.89, 2.75; 0.123) 2.07 (0.91, 4.75; 0.084) Fourth 1.43 (0.93, 2.18; 0.100) 1.84 (0.93, 3.65; 0.081) 1.56 (1.12, 2.19; 0.009) 1.59 (1.00, 2.54; 0.050) 1.42 (0.76, 2.64; 0.270) 1.42 (0.53, 3.78; 0.482) Highest 0.80 (0.49, 1.30; 0.363) 0.82 (0.36, 1.86; 0.628) 1.82 (1.31, 2.52; 0.000) 1.52 (0.89, 2.62; 0.126) 1.35 (0.68, 2.65; 0.390) 1.46 (0.43, 5.00; 0.543) Number 1593 1420 2541 2308 586 554 Cameroon Lowesta 1.00 1.00 1.00 1.00 1.00 1.00 Second 1.42 (0.59, 3.41; 0.432) 1.24 (0.49, 3.11; 0.654) 1.32 (0.72, 2.42; 0.376) 0.79 (0.41, 1.55; 0.496) 1.74 (0.66, 4.57; 0.260) 1.11 (0.37, 3.28; 0.853) Middle 2.86 (1.34, 6.13; 0.007) 2.48 (1.08, 5.67; 0.031) 2.82 (1.71, 4.66; 0.000) 1.46 (0.81, 2.63; 0.208) 3.80 (1.65, 8.75; 0.002) 2.16 (0.85, 5.49; 0.105) Fourth 3.65 (1.84, 7.27; 0.000) 3.29 (1.33, 8.09; 0.010) 3.34 (2.01, 5.55; 0.000) 1.23 (0.59, 2.56; 0.574) 4.65 (2.00, 10.82; 0.000) 2.40 (0.84, 6.82; 0.100) Highest 3.61 (1.76, 7.40; 0.000) 3.22 (1.12, 9.31; 0.030) 2.97 (1.81, 4.87; 0.000) 0.94 (0.42, 2.08; 0.871) 5.10 (2.21, 11.76; 0.000) 3.04 (0.88, 10.48; 0.079) Number 3802 3743 4556 4320 2014 1959 Ghana Lowesta 1.00 1.00 1.00 1.00 1.00 1.00 Second 1.10 (0.41, 2.95; 0.845) 1.11 (0.40, 3.07; 0.845) 1.98 (1.02, 3.82; 0.042) 1.43 (0.73, 2.79; 0.302) 1.20 (0.51, 2.84; 0.673) 0.86 (0.35, 2.14; 0.750) Middle 1.36 (0.58, 3.18; 0.483) 1.07 (0.36, 3.21; 0.904) 3.03 (1.67, 5.47; 0.000) 1.96 (1.04, 3.69; 0.037) 1.98 (0.90, 4.38; 0.092) 1.31 (0.53, 3.24; 0.555) Fourth 1.00 (0.41, 2.44; 0.992) 0.59 (0.14, 2.44; 0.462) 2.37 (1.23, 4.56; 0.010) 1.10 (0.47, 2.57; 0.817) 1.69 (0.68, 4.17; 0.257) 0.71 (0.21, 2.44; 0.590) Highest 0.79 (0.28, 2.27; 0.664) 0.42 (0.07, 2.66; 0.356) 2.12 (1.12, 4.03; 0.021) 0.95 (0.32, 2.83; 0.929) 1.87 (0.76, 4.61; 0.175) 0.59 (0.11, 3.24; 0.543) Number 2825 2739 4505 4301 1814 1811 (continued overleaf )
- 30. Wealth and HIVin sub-Saharan Africa Mishra et al. S25 months, knowledge of prevention methods, knowledge of own HIV status, age at first sexual intercourse, number of lifetime sexual partners (replaced with whether the respondent had two or more (1.89, 18.16; 0.002) Adjusted models for individual men and women estimate effects of household wealth status on the likelihood that the individual is HIV positive, controlling for age, ethnicity, religion, urban/rural residence, geographical region of residence, education, occupation, media exposure, marital status, duration in union, number of years in current place of residence, alcohol use at last sex in past 12 the respondent had two or more partners in the previous 12 months in Kenya, Ghana, Burkina Faso, and Malawi for both spouses; and in Lesotho for the female partner), circumcision status of the male infection symptoms in past 12 months, circumcision (men only), consistent condom use in past 12 months, and community-level wealth score (computed by averaging the individual household wealth Adjusted models for cohabiting couples estimate effects of household wealth status on the likelihood that one or both partners is HIV positive, controlling for wife’s age, age gap between spouses, urban/rural residence, geographical region, wife’s education, education gap between spouses, union type, duration in union, number of lifetime sexual partners for each spouse (replaced with whether partners in the previous 12 months in Kenya, Ghana, Burkina Faso, and Malawi for both men and women; and in Lesotho for women), reported sexually transmitted infection or sexually transmitted (0.91, 8.53; 0.072) (0.21, 6.74; 0.851) (0.08, 8.93; 0.898) and in seven out of the eight countries considered for women, but lose statistical significance in most cases. Results are similar for cohabiting couples. In all but one country the unadjusted odds of one or both partners being HIV infected are two to seven times greater among 2145 5.86 2.79 1.18 0.86 1.00 couples in the highest wealth quintile than among those in the lowest wealth quintile. Adding controls for selected (1.96, 17.97; 0.002) (0.93, 10.60; 0.065) (1.94, 28.07; 0.003) underlying factors, proximate factors, and community- (0.40, 8.73; 0.420) level wealth progressively diminishes the strength of this association. With all factors controlled for, the odds of one or both partners being HIV infected remain greater than one in six out of the eight countries considered, but statistically significant at the 5% level in only one country (Tanzania). 2157 5.94 3.14 1.88 7.37 1.00 0.959) 0.595) 0.851) 0.685) Discussion 3.17; 4.02; 3.43; 3.57; This study found that, contrary to evidence for other (0.30, (0.45, (0.36, (0.14, infectious diseases and theoretical expectations, HIV prevalence is not disproportionately higher among adults living in poorer households in sub-Saharan 3565 0.97 1.35 1.11 0.72 1.00 Africa. In all eight countries included in the present analysis, wealthier men and women tend to have a higher prevalence of HIV than poorer individuals. In (1.66, 11.24; 0.003) (0.38, 3.09; 0.878) (0.59, 3.89; 0.387) (0.68, 5.54; 0.219) most cases, the positive association between wealth status and HIV is considerably diminished when a number of underlying factors (such as education, urban/ rural residence, and community wealth) and some of the partner, consistent condom use in past 12 months, and community-level wealth score. behavioural and biological pathways (proximate factors, such as sexual risk taking, condom use, and male 3624 1.09 1.52 1.93 4.32 1.00 circumcision) are taken into account. The results indicate that much of the positive association between wealth and HIV is caused by these underlying or (0.79, 10.23; 0.111) (0.25, 6.00; 0.799) (0.01, 3.10; 0.259) (0.06, 2.82; 0.360) mediating factors. Even after accounting for these various factors, however, in most countries wealthier adults remain at least as likely as poorer individuals to be infected with HIV, if not more. The results are similar for cohabiting couples. 2129 1.00 2.84 1.23 0.21 0.40 Our analysis indicates that several factors may be res- ponsible for the observed higher HIV prevalence among wealthier individuals in these countries. First, the 0.042) 0.720) 0.225) 0.167) CI, Confidence interval; OR, odds ratio. wealthier are more likely to live in urban areas and to live in wealthier communities, where HIV is more 6.82; 4.01; 2.28; 6.29; prevalent. Wealthier adults, especially men, tend to be (1.03, (0.38, (0.03, (0.73, more mobile, more likely to have multiple partners, and more likely to engage in sex with non-regular partners, scores in each cluster). behaviours that tend to be associated with a higher HIV Reference category. 2157 1.00 2.66 1.24 0.26 2.14 prevalence. On the other hand, wealthier men and women tend to be more educated and have a greater knowledge of HIV prevention methods. As such, they may be more likely Burkina Faso to receive healthcare, to use condoms (both with non- Number Lowesta Highest Second Middle Fourth regular partners and consistently with all partners), and less likely to use alcohol when having sex. Wealthier men are a .
- 31. S26 AIDS 2007, Vol 21 (suppl 7) more likely to be circumcised, which may reduce their risk to underreport and men tend to exaggerate their of HIV infection. Also, wealthier adults may live longer premarital and extramarital sexual activity [39]. Epide- with HIV than poorer individuals as a result of their better miological studies in Africa have also observed weak health and nutritional status. associations between self-reported risky sexual behaviour and HIV status [40]. The findings of our study may be Women are less likely than men to report having multiple biased to the extent that men and women misreport their partners and non-regular partners. We found that the number of sexual partners, sex with non-regular partners, positive association between wealth status and HIV condom use, and other related behaviours, or to the prevalence tends to be stronger for women than for men extent that the degree of misreporting is different across in most countries, suggesting disproportionately greater the wealth quintiles. vulnerability of women in the wealthier groups. A fourth limitation is that the surveys included in the There are several limitations of this study that should be analysis did not collect data on concurrent partnerships kept in mind when interpreting our findings. One and sexual networks. We were thus unable to examine the important limitation is that DHS/AIS surveys do not extent to which wealthier individuals are more likely to collect data on household income or expenditure, which engage in such complex patterns of sexual relations, would traditionally be included in an assessment of wealth which may increase the risk of HIV infection in Africa by status. The assets-based wealth index used here is only a allowing the virus to spread rapidly to others [41–45]. proxy indicator for household economic status [36]. In addition, wealth index scores cannot be compared across Moreover, because of the cross-sectional nature of the countries both because the level and distribution of data used in this study, endogeneity might bias our results wealth differs from one country to another and because at several levels. First, when considering the effect of the choice of assets included in the construction of the wealth status on HIV prevalence we do not allow for the index varies somewhat from country to country. In spite opposite, detrimental effect of HIV infection on wealth of these issues, in developing-country settings, the wealth status, which is well established [46,47]. Excluding HIV- index has been shown to produce superior results and positive individuals who reported being seriously ill for equal or greater distinctions in health outcomes than three or more months in the previous 12 months (in household expenditure-based measures [37]. Moreover, Tanzania, Uganda, Cameroon, and Malawi, where such as income and expenditure measures can be volatile and information was collected) had virtually no effect on the temporary, wealth status (which results from the observed associations between wealth and HIV status accumulation of income) is a preferred measure to relate (data not shown). Second, if HIV-positive adults were to HIV prevalence (which results from an accumulation aware of their serostatus, they might have adjusted their of incidence). sexual and reproductive behaviour. We tested this by excluding HIV-positive individuals who were previously A second limitation is differential non-response in the tested and received the result, which made little difference surveys considered. Non-response rates for HIV testing to the observed associations between wealth and HIV tend to be higher among the wealthier, urban, and more status (data not shown). Third, when infected with HIV, educated adults, who also tend to have higher HIV wealthier individuals are likely to survive longer than prevalence. Previous research has, however, indicated that poorer individuals because of better nutrition and access in these surveys differential non-response has small and to healthcare. Cross-sectional data used in this study did insignificant effects on the observed HIV prevalence, so not allow taking into account such selective survival of any bias caused by differential non-response by wealth wealthier respondents. A lack of information on the status should be small [38]. In any case, if there were no availability and access to treatment and care (antiretroviral differential non-response by wealth status, the positive drugs in particular) further limited the possibility of association between wealth status and HIV prevalence disentangling this effect. Antiretroviral therapy coverage would be even stronger. In addition, the surveys was, however, still very low at the time of the survey data considered exclude population groups that are difficult collection in most countries. to locate or interview, most notably the homeless. The observed positive association between wealth status and Finally, for many HIV-positive adults, the infection may HIV prevalence may be overestimated to the extent that have preceded their sexual and other behaviours recorded the homeless are poorer and have higher HIV prevalence in the survey, which may have biased some of the than those included in the survey. Given that the associations. Moreover, the strength and direction of proportion of the homeless in the total population tends the relationship between wealth status and HIV to be small, any effect of excluding this group on the prevalence and the roles of risk behaviours and protective observed associations is likely to be small. factors are likely to change over time, depending on the stage and spread of the epidemic [48]. Cross-sectional data Another limitation is that our analysis is based on self- used in our study do not allow the examination of causal reported behaviours. There is evidence that women tend effects and these transitional phenomena.
- 32. Wealth and HIVin sub-Saharan Africa Mishra et al. S27 In conclusion, this study found a positive association 2. Adler NE, Newman K. Socioeconomic disparities in health: pathways and policies. Health Aff (Millwood) 2002; 21:60– between household economic status and HIV prevalence 76. among adult men and women in sub-Saharan Africa. 3. Fotso JC, Kuate-Defo B. Socioeconomic inequalities in early Accounting for various underlying factors and proxi- childhood malnutrition and morbidity: modification of the household-level effects by the community SES. Health Place mate determinants explains much of this positive 2005; 11:205–225. association, but in most cases wealthier adults remain 4. Ellen JM, Kohn RP, Bolan GA, Shiboski S, Krieger N. Socio- at least as likely as poorer individuals to be HIV infected. economic differences in sexually transmitted disease rates among black and white adolescents, San Francisco, 1990 to We found that HIV prevalence does not follow the same 1992. Am J Public Health 1995; 85:1546–1548. pattern of association with poverty within countries in 5. Fleming DT, McQuillan GM, Johnson RE, Nahmias AJ, Aral SO, sub-Saharan Africa as most other diseases. Although Lee FK, et al. Herpes simplex virus type 2 in the United States, 1976 to 1994. N Engl J Med 1997; 337:1105–1111. poverty reduction is an essential strategy to improve 6. Lacey CJN, Merrick DW, Bensley DC, Fairley I. Analysis of the health and combat the HIV epidemic, our analysis sociodemography of gonorrhoea in Leeds, 1989-93. BMJ 1997; 314:1715–1718. suggests that HIV prevention, care, and treatment 7. Holtgrave DR, Crosby RA. Social capital, poverty, and income programmes should also be focused on the better-off inequality as predictors of gonorrhoea, syphilis, chlamydia and segments of the population. Focusing on the most AIDS cases rates in the United States. Sex Transm Infect 2003; 79:62–64. important modes of exposure will probably be more 8. Kyriakis KP, Hadjivassiliou M, Paparizos VA, Flemetakis A, effective than focusing broadly on poverty reduction Stavrianeas N, Katsambas A. Incidence determinants of gonor- [49,50]. It will also be important to extend programmes rhea, chlamydial genital infection, syphilis and chancroid in attendees at a sexually transmitted disease clinic in Athens, to the rural areas where a majority of the population in Greece. Int J Dermatol 2003; 42:876–881. sub-Saharan Africa resides. 9. Miller GC, McDermott R, McCulloch B, Fairley CK, Muller R. Predictors of the prevalence of bacterial STI among young disadvantaged Indigenous people in north Queensland, It is important to stress that our findings do not imply that Australia. Sex Transm Infect 2003; 79:332–335. the poor are not disproportionately affected by HIV when 10. Wald A. Synergistic interactions between herpes simplex virus they do get infected. Poverty reduction is an extremely type-2 and human immunodeficiency virus epidemics. Herpes 2004; 11:70–76. important goal in itself for many reasons, and it will 11. Chawla R, Bhalla P, Garg S, Meghachandra Singh M, Bhalla K, certainly help combat the HIV epidemic in the long run Sodhani P, et al. Community based study on sero-prevalence of and deal with its many adverse consequences. syphilis in New Delhi (India). J Commun Dis 2004; 36:205– 211. 12. Uuskula A, Nygard JF, Kibur-Nygard M. Syphilis as a social disease: experience from the post-communist transition period in Estonia. Int J STD AIDS 2004; 15:662–668. 13. Bukusi EA, Cohen CR, Meier AS, Waiyaki PG, Nguti R, Njeri JN, Acknowledgements Holmes KK. Bacterial vaginosis: risk factors among Kenyan women and their male partners. Sex Transm Dis 2006; 33:361– 367. The authors wish to thank the numerous people who 14. Fitzgerald DW, Behets F, Caliendo A, Roberfroid D, Lucet C, provided comments on the two presentations and on Fitzgerald JW, Kuykens L. Economic hardship and sexually earlier drafts of this paper. transmitted diseases in Haiti’s rural Artibonite Valley. Am J Trop Med Hyg 2000; 62:496–501. 15. Fenton L. Preventing HIV/AIDS through poverty reduction: Sponsorship: Funding for this research was provided the only sustainable solution? Lancet 2004; 364:1186–1187. by the Joint United Nations Programme on HIV/AIDS 16. Bloom DE, Sevilla J. Health, wealth, AIDS, and poverty. In: (UNAIDS) (no. HQ/05/423963). Additional support Report of the Asia-Pacific Ministerial Meeting. 9–10 October 2001. Melbourne, Australia: The Australian Government Over- was provided by the United States Agency for seas Aid Programme. International Development (USAID) through the MEA- 17. Whiteside A. Poverty and HIV/AIDS in Africa. Third World Q SURE DHS project (no. GPO-C-00-03-00002-00). 2002; 23:313–332. 18. UNAIDS. Report on the global AIDS epidemic. Geneva: The preliminary findings of this research were pre- UNAIDS; 2006. 19. Menon R, Wawer MJ, Konde-Lule JK, Sewankambo NK, Li C. sented at the US President’s Emergency Plan for AIDS The economic impact of adult mortality on households in Rakai Relief (PEPFAR) Annual Meeting in Durban, South district, Uganda. In: Ainsworth M, Fransen L, Over M, editors. Africa, on 12–15 June 2006, and at the International Confronting AIDS: evidence from the developing world. Brus- AIDS Economic Network Meeting in Toronto, Canada, sels: European Commission; and Washington, DC: The World Bank; 1998. pp. 321–335. on 11–12 August 2006. 20. Kirunga CT, Ntozi JP. Socio-economic determinants of HIV serostatus: a study of Rakai District, Uganda. Health Transit Rev The views expressed are those of the authors and do 1997; 7 (Suppl.):175–188. not necessarily reflect the views of UNAIDS, USAID, 21. Shelton JD, Cassell MM, Adetunji J. Is poverty or wealth at the root of HIV? Lancet 2005; 366:1057–1058. the United States Government, or the organizations 22. Ainsworth M, Semali I. Who is most likely to die of AIDS? with which the authors are affiliated. Socioeconomic correlates of Adult Deaths in Kagera Region, Tanzania. In: Ainsworth M, Fransen L, Over M, editors. Con- fronting AIDS: evidence from the developing world. Brussels: References European Commission; and Washington, DC: The World Bank; 1998. pp. 95–109. 1. Kuate-Defo B. Effects of socioeconomic disadvantage and wo- 23. Wojciki JM. Socioeconomic status as a risk factor for HIV men’s status on women’s health in Cameroon. Soc Sci Med infection in women in East, Central and Southern Africa: a 1997; 44:1023–1042. systematic review. J Biosoc Sci 2005; 37:1–36.
- 33. S28 AIDS 2007, Vol 21 (suppl 7) 24. Boerma JT, Weir SS. Integrating demographic and epidemio- 37. Rutstein SO, Johnson K. The DHS wealth asset index. DHS logical approaches to research on HIV/AIDS: the proximate- comparative report no. 6. Calverton, Maryland, USA: ORC determinants framework. J Infect Dis 2005; 191 (Suppl. 1):S61– Macro; 2004. S67. 38. Mishra V, Vaessen M, Boerma JT, Arnold F, Way A, Barrere B, 25. Mishra V, Bignami S, Greener R, Vaessen M, Hong R, Ghys P, et al. HIV testing in national population-based surveys: experi- et al. A study of the association of HIV infection with wealth in ence from the Demographic and Health Surveys. Bull WHO sub-Saharan Africa. DHS working papers no. 31. Calverton, 2006; 84:537–545. Maryland, USA: Macro International Inc.; 2007. 39. Zaba B, Pisani E, Slaymaker E, Boerma JT. Age at first sex: 26. Central Bureau of Statistics (CBS), Ministry of Health (MOH), understanding recent trends in African demographic surveys. and ORC Macro. Kenya demographic and health survey 2003. Sex Transm Infect 2004; 80 (Suppl. 2):ii28–ii35. Calverton, Maryland, USA: CBS, MOH, and ORC Macro; 2004. ´ 40. Buve A, Lagarde E, Carael M, Rutenberg N, Ferry B, Glynn JR, ¨ 27. Ghana Statistical Service (GSS), Noguchi Memorial Institute for et al. Study Group on Heterogeneity of HIV Epidemics in Medical Research (NMIMR), and ORC Macro. Ghana demo- African Cities. Interpreting sexual behaviour data: validity graphic and health survey 2003. Calverton, Maryland, USA: issues in the multicentre study on factors determining the GSS, NMIMR, and ORC Macro; 2004. differential spread of HIV in four African cities. AIDS 2001; ´ 28. Institut National de la Statistique et de la Demographie (INSD), 15 (Suppl. 4):S117–S126. and ORC Macro. Burkina Faso demographic and health survey 41. Hudson CP. AIDS in rural Africa: a paradigm for HIV-1 pre- 2003 [in French]. Calverton, Maryland, USA: INSD and ORC vention. Int J STD AIDS 1996; 7:236–243. Macro; 2004. 42. Morris M, Kretzschmar M. Concurrent partnerships and the 29. Tanzania Commission for AIDS (TACAIDS), National Bureau of spread of HIV. AIDS 1997; 11:641–648. Statistics (NBS), and ORC Macro. Tanzania HIV/AIDS indicator 43. Lagarde E, Auvert B, Chege J, Sukwa T, Glynn JR, Weiss HA, survey 2003–04. Calverton, Maryland, USA: TACAIDS, NBS, et al. Condom use and its association with HIV/sexually trans- and ORC Macro; 2005. mitted diseases in four urban communities of sub-Saharan 30. Institut National de la Statistique (INS) and ORC Macro. Africa. AIDS 2001; 15 (Suppl. 4):S71–S78. Cameroun demographic and health survey 2004 [in French]. 44. Halperin DT, Epstein H. Concurrent sexual partnerships help to Calverton, Maryland, USA: INS and ORC Macro; 2005. explain Africa’s high HIV prevalence: implications for preven- 31. National Statistical Office (NSO), and ORC Macro. Malawi tion. Lancet 2004; 364:4–6. demographic and health survey 2004. Calverton, Maryland, 45. Helleringer S, Kohler HP. Social networks, perceptions of risk, USA: NSO and ORC Macro; 2005. and changing attitudes towards HIV/AIDS: new evidence from 32. Ministry of Health and Social Welfare (MOHSW), Bureau of a longitudinal study using fixed-effects analysis. Popul Stud Statistics (BOS), and ORC Macro. Lesotho demographic and 2005; 59:265–282. health survey 2004. Calverton, Maryland, USA: MOHSW, BOS, 46. Piot P, Bartos M, Ghys PD, Walker N, Schwartlander B. The and ORC Macro; 2005. global impact of HIV/AIDS. Nature 2001; 410:968–973. 33. Ministry of Health (MOH) and ORC Macro. Uganda HIV/AIDS 47. UNAIDS. Report on the global HIV/AIDS epidemic. Geneva: sero-behavioural survey 2004–2005. Calverton, Maryland, UNAIDS; 2000. USA: MOH and ORC Macro; 2006. 48. Hargreaves J, Boler T. Girl power: the impact of girls’ education 34. ORC Macro. Anemia and HIV testing field manual: demo- on HIV and sexual behaviour. Johannesburg, South Africa: graphic and health surveys. Calverton, Maryland, USA: ORC ActionAid International; 2006. Macro; 2005. 49. Pisani E, Garnett GP, Grassly NC, Brown T, Stover J, Hankins C, 35. ORC Macro. HIV testing laboratory manual: demographic and et al. Back to basics in HIV prevention: focus on exposure. BMJ health surveys. Calverton, Maryland, USA: ORC Macro; 2005. 2003; 326:1384–1387. 36. Filmer D, Pritchett L. Estimating wealth effects without 50. Gouws E, White PJ, Stover J, Brown T. Short term estimates of expenditure data – or tears: an application to educational adult HIV incidence by mode of transmission: Kenya and enrollments in states of India. Demography 2001; 38:115– Thailand as examples. Sex Transm Infect 2006; 82 (Suppl. 3): 132. iii51–iii55.
- 34. The socioeconomic determinantsof HIV incidence: evidence from a longitudinal, population-based study in rural South Africa Till Barnighausena,b, Victoria Hosegooda,c, Ian M. Timaeusa,c and ¨ Marie-Louise Newella,d Background: Knowledge of the effect of socioeconomic status on HIV infection in Africa stems largely from cross-sectional studies. Cross-sectional studies suffer from two important limitations: two-way causality between socioeconomic status and HIV serostatus and simultaneous effects of socioeconomic status on HIV incidence and HIV-positive survival time. Both problems are avoided in longitudinal cohort studies. Methods: We used data from a longitudinal HIV surveillance and a linked demo- graphic surveillance in a poor rural community in KwaZulu-Natal, South Africa, to investigate the effect of three measures of socioeconomic status on HIV incidence: educational attainment, household wealth categories (based on a ranking of households on an assets index scale) and per capita household expenditure. Our sample comprised of 3325 individuals who tested HIV-negative at baseline and either HIV-negative or -positive on a second test (on average 1.3 years later). Results: In multivariable survival analysis, one additional year of education reduced the hazard of acquiring HIV by 7% (P ¼ 0.017) net of sex, age, wealth, household expenditure, rural vs. urban/periurban residence, migration status and partnership status. Holding other factors equal, members of households that fell into the middle 40% of relative wealth had a 72% higher hazard of HIV acquisition than members of the 40% poorest households (P ¼ 0.012). Per capita household expenditure did not sig- nificantly affect HIV incidence (P ¼ 0.669). Conclusion: Although poverty reduction is important for obvious reasons, it may not be as effective as anticipated in reducing the spread of HIV in rural South Africa. In contrast, our results suggest that increasing educational attainment in the general population may lower HIV incidence. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins AIDS 2007, 21 (suppl 7):S29–S38 Keywords: Africa, economics, epidemiology, HIV incidence, longitudinal study, socioeconomic factors, surveillance Introduction 5.5 million adults and children living with HIV and an estimated 320 000 deaths as a result of HIV/AIDS in 2005 The HIV epidemic remains one of the greatest health and [1]. From the beginning of the HIVepidemic, researchers development challenges facing sub-Saharan Africa. South have tried to understand its relationship with socio- Africa bears a substantial brunt of the HIVepidemic, with economic status. Reviews of studies in Africa show a wide From the aAfrica Centre for Health and Population Studies, University of KwaZulu-Natal, Mtubatuba, South Africa, the b Department of Population and International Health, Harvard School of Public Health, Boston, Massachusetts, USA, the cCentre for Population Studies, London School of Hygiene and Tropical Medicine, London, UK, and the dCentre for Paediatric Epidemiology and Biostatistics, Institute of Child Health, University College London, London, UK. Correspondence to Till Barnighausen, Africa Centre for Health and Population Studies, University of KwaZulu-Natal, PO Box 198, ¨ Mtubatuba 3935, South Africa. E-mail: tbarnighausen@africacentre.ac.za ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S29
- 35. S30 AIDS 2007, Vol 21 (suppl 7) variety of relationships between HIV and socioeconomic has been published, and there have only been a few status [2–5]. Most of the studies, however, examine longitudinal studies reporting on the effect of socio- cross-sectional associations between HIV serostatus and economic status on the hazard of HIV seroconversion in socioeconomic status and thus suffer from two other African countries [17–21]. With one exception, important limitations. these longitudinal studies were conducted in cohorts of sexually active young urban women attending antenatal First, cross-sectional studies are usually unable to care or family planning services [17,18,20], or in cohorts distinguish between the effect of socioeconomic status of urban factory workers [19,21]. Relationships between on HIV infection and the effect of HIV infection on socioeconomic status and HIV incidence in these socioeconomic status [6–8]. Several pathways have been subpopulations may differ from those in the general suggested through which a decrease in socioeconomic population, or may not be detectable because of limited status may increase the risk of HIV infection. Malnu- variation in socioeconomic status [22]. trition and micronutrient deficiencies, which are more common among the poor than in the rest of society can No clear pattern of the relationships between socio- disrupt the integrity of the vaginal epithelium, increasing economic status and HIV seroconversion risk emerges its permeability to HIV [9]. Ulcerative genital diseases, from these studies. In urban women attending antenatal which are associated with low socioeconomic status, care or family planning services, the risk of HIV increase the transmission probability of HIV [7]. Women seroconversion has been found not to be associated with of low socioeconomic status may be economically education [18], to decrease with the educational dependent on male partners, limiting their ability to attainment of women’s partners [20], and to decrease negotiate condom use in relationships, or forcing them to with women’s incomes [17]. In studies in urban factory sell sex for money [2,10]. On the other hand, an increase workers, the risk of HIV seroconversion was positively in socioeconomic status may increase the risk of HIV associated with occupational status [19] and educational infection because wealthy or educated people have more attainment [21]. The only published study using resources with which to attract and maintain multiple longitudinal, population-based data to examine the partners [11]. association between socioeconomic status and HIV incidence in Africa found that in rural Uganda HIV status may also be an important determinant of educational attainment is not a significant predictor of socioeconomic status. HIV-related diseases can limit HIV incidence [23]. people’s ability to work, which may decrease socio- economic status, especially among individuals who work Wojcicki [4] concluded from a review of studies on in economic sectors in which income is closely linked to socioeconomic status and HIV infection in sub-Saharan productivity associated such as agriculture or the informal Africa that ‘‘further studies should use multiple measures of sector [12]. In addition, HIV and its associated diseases socioeconomic status’’ because different dimensions of may lead to substantial expenditure for health care, socioeconomic status influence HIV infection risk in decreasing household wealth [12–14]. different ways. We report findings about the effect of three different measures of socioeconomic status (educational A second problem of cross-sectional studies is that they attainment, household wealth categories and household usually cannot distinguish between the effect of socio- expenditure) on HIV incidence from a longitudinal HIV economic status on HIV incidence and the effect of surveillance of the general population in a rural community socioeconomic status on survival with HIV. Previous in KwaZulu-Natal, South Africa. studies have shown that HIV survival time increases with socioeconomic status [15], possibly because diet or access to antiretroviral therapy improve with socioeconomic status [4,16]. In a cross-sectional study, it is thus possible to Methods find a positive association between socioeconomic status and HIV infection, even if higher socioeconomic status Study area protects individuals from acquiring HIV. We used data from the longitudinal population-based HIV surveillance conducted by the Africa Centre for Both problems – two-way causality between socio- Health and Population Studies, University of KwaZulu- economic status and HIV serostatus and the simultaneous Natal, and from the Africa Centre Demographic effects of socioeconomic status on HIV incidence and on Information System (ACDIS) to investigate the relation- HIV-positive survival time – are avoided when a cohort ship between socioeconomic status and the hazard of HIV of HIV-negative individuals is followed over time and the seroconversion. ACDIS has been collecting demographic hazards of HIV seroconversion are compared across data since January 2000 and socioeconomic data since individuals of different socioeconomic status in survival February 2001 [24]. The ACDIS demographic surveil- analyses. To date, no such longitudinal study on lance area (DSA) is located in the rural district of socioeconomic status and HIV incidence in South Africa Umkhanyakude in northern KwaZulu-Natal, South
- 36. The socioeconomic determinantsof HIV incidence Barnighausen et al. ¨ S31 Africa. It covers 435 square kilometres and a total resident were contacted and refused to consent to an HIV test population of approximately 86 000 Zulu-speaking (n ¼ 2551)]. In comparison with the individuals included people (June 2003). Although the study took place in in the sample, individuals for whom information on the an overall rural community, the area includes an urban independent variables used in this study was available, but township and periurban areas (informal settlements with a a second HIV test was not, were not significantly different population density of more than 400 people per square at the 5% confidence level with regard to per capita kilometer). In 2001, 61% of households in the ACDIS household expenditure, rural vs. urban/periurban place area had a toilet and only 38% had access to piped water. of residence or the probability of having a partner at The unemployment rate in the same year was 30% [25]. baseline. They were, however, slightly younger (25.4 vs. 26.4 average years of age, P < 0.001), slightly more Data collection educated (8.2 vs. 7.7 average educational grade attained, Teams of two trained fieldworkers visited each eligible P < 0.001), wealthier (a value of À0.073 vs. À0.352 on individual in his or her household. Fieldworkers revisited the household assets index scale, P < 0.001), more likely households up to four times to contact absentees. If a to be male (49% vs. 40%, P < 0.001), more likely to subject no longer lived at the household, the field worker have migrated out of the DSA between the two handed the case to a specially trained tracking team that surveillance rounds (14% vs. 7%, P < 0.001) and less made up to 10 attempts to find the individual in his or her likely to be married (9% vs. 11%, P ¼ 0.001) than the new residence. After written informed consent, the field individuals in the sample. Table 1 shows the character- workers collected blood by finger stick and prepared istics of the 3325 individuals included in the sample. dried blood spots for HIV testing according to the Joint Even though the study took place in an overall rural United Nations Programme on HIV/AIDS (UNAIDS) community, a substantial proportion of participants (28%) and World Health Organization (WHO) Guidelines for lived in either an urban or a periurban area (Table 1). Using HIV Testing Technologies in Surveillance [26]. HIV status was determined by antibody testing with a Although in comparison with South Africa as whole, the broad-based HIV-1/HIV-2 enzyme-linked immunosor- average indicators of socioeconomic status in this bent assay (ELISA; Vironostika, Organon Teknika, community are low [25], their dispersion within the Boxtel, the Netherlands) followed by a confirmatory community is quite wide (see Table 1). For example, in ELISA (GAC-ELISA; Abbott, Abbott Park, Illinois, our sample the 10th percentile of educational attainment USA). All covariates used in this study were collected by was 2nd grade, whereas the 90th percentile was 12th the ACDIS demographic surveillance system conducted grade, and the 10th percentile of daily total per capita between January 2003 and September 2004 [25]. household expenditure was 1.2 South African Rand (ZAR), whereas the 90th percentile was 7.2 ZAR, Sample suggesting that there is sufficient variation in these two Our sample includes all individuals who met the measures of socioeconomic status to warrant investigating following criteria: they were age-eligible for inclusion their effects on HIV incidence. in the HIV surveillance (women between 15 and 49 years of age and men between 15 and 54 years of age) both Table 1. Sample characteristics (N U 3325). during the first surveillance round (from June 2003 to N (%)a December 2004) and during the second round (from January to December 2005); they were residents in the Sex ACDIS DSA during the first round of the HIV Male 1317 (40%) Female 2008 (60%) surveillance and either residents in the DSA or non- Mean (SD) age (years) 26.4 (11.0) resident household members during the second round; Mean (SD) grade of educational attainment 7.7 (3.4) they tested HIV-negative during the first round and tested Wealth category either HIV-negative or HIV-positive during the second Poorest 40% 1330 (40%) Middle 40% 1330 (40%) round; and data on all independent variables used in this Wealthiest 20% 665 (20%) analysis were available at the time of the HIV test in the Mean (SD) daily total per capita 5.37 (30.50) first round. On average the time between the first and the household expenditures (ZAR) Place of residence second HIV test was 1.3 years. Rural 2396 (72%) Urban/periurban 929 (28%) Of the 8952 participants in the population-based HIV Migration status surveillance who were HIV-negative during the first Migrant 231 (7%) Non-migrant 3094 (93%) round of testing and still eligible to be tested during the Partnership status second round, information on at least one of the Not married, without partner 1687 (51%) independent variables used in this study was missing for Married 374 (11%) 1597 individuals, and information on HIV status during Not married, with partner 1264 (38%) the second round of testing was missing for 4030 [either a Or mean (SD) where indicated. SD, standard deviation; ZAR, South because individuals could not be contacted (n ¼ 1479) or African rand.
- 37. S32 AIDS 2007, Vol 21 (suppl 7) Independent variables spending for household members (on shopping, rent, The focus of our analyses is the socioeconomic clothes, water, fuel, electricity, health, transport, religious determinants of HIV incidence. We investigated the activities, telephones, cell phones, payments for goods relationship of HIV infection with three measures of bought by hire-purchase or lay-bye, funerals, life socioeconomic status: educational attainment, household insurance, and school), as well as spending for individuals wealth and expenditure. Increased educational attainment outside the household (money, goods and food). We used (the highest education grade that an individual has daily total household expenditure and divided it by the completed within a country’s educational system) has number of members in each household to adjust for been hypothesized to lead to a lower risk of HIV infection, household size [34]. The resulting variable, i.e. daily total because it improves the ability to understand and act on per capita household expenditure, is henceforth referred to health promotion messages and because it is associated with as household expenditure. We logarithmically transformed increased exposure to school-based HIV prevention the household expenditure variable to reduce skewness. programmes or increased access to health services [27]. As expected, educational attainment, the household assets Wealth and expenditure are likely to capture different index, and household expenditure were positively financial aspects of social status. Households generate correlated. However, the three measures of socioeconomic wealth through saving of income after spending money status were not very highly correlated (with Pearson’s on consumption. There is commonly greater variation in correlation coefficient at 0.256 between educational wealth than in expenditure because wealth is accumulated attainment and the household assets index; at 0.174 and because some expenditure on basic items such as food between educational attainment and the log transform of and clothing are indispensable for human survival. Wealth household expenditure; and at 0.380 between the house- may thus be a more sensitive measure of the long-term hold assets index and the log transform of household social position of a household and may capture influences expenditure), reinforcing the theoretical conrideration of social status on the risk of HIV infection better than that each captures a different aspect of socioeconomic status expenditure. and cannot be used in place of one of the other two in multiple regression analysis. We used a household assets index as a measure of wealth. As shown by Morris and colleagues [28], household assets In addition to the three measures of socioeconomic status, indices are valid proxies for wealth in health surveys in we controlled for a number of variables that have been rural Africa. Following Filmer and Pritchett [29], we used found to be associated with HIV infection in the first principal component obtained in a principal cross-sectional surveys in South Africa (sex [5,35–37], component analysis of information on house ownership, age [38–41], rural vs. urban/periurban residence [36], water source, energy, toilet type, electricity and 27 migration status [5,39,41] and partnership status [5,37,39, household assets as an assets index. The assets included 41–43]). All independent variables were measured at items that can be used for consumption, production or baseline and assumed to be time-invariant. both, such as beds, bicycles, tables, telephones, television sets, sewing machines, block makers, wheelbarrows, Statistical analysis tractors, cattle, and other livestock. We categorized In order to investigate the effects of explanatory variables households as either belonging to the poorest 40%, the on the time to HIV seroconversion, we used semipara- middle 40% or the wealthiest 20% on the assets index metric and parametric survival models in the following scale. We chose these three categories of relative wealth, proportional hazards specification [44]. because they have been found to capture wealth effects hðt; Xi Þ ¼ h0 ðtÞ Á expðXi bÞ (1) well in a number of studies in poor provinces of South Africa [30,31], including studies investigating the effect of wealth on health [32,33]. where h0(t) is the baseline hazard function that is obtained when all explanatory variables are equal to 0. A unit Household expenditure captures the short-term financial change of an independent variable in this model leads to a liquidity of the members of a household and should thus constant parallel shift of the baseline hazard function. If be a better measure of the influences of current the baseline hazard function is left unspecified, the model consumption of costly services on HIV infection than is the semiparametric Cox proportional hazards model wealth. For example, in a cash-strapped period, an (CPHM). individual may not have sufficient funds to seek treatment for sexually transmitted diseases (STD) whose presence Although parametric models that specify the functional increases the risk of HIV transmission, or may not form of the baseline hazard function have the disadvan- be able to travel to a place where condoms are available. tage that they may lead to inconsistent estimates if the We measured total household expenditure by summing baseline function is misspecified, they will be more spending across all categories on which expenditure efficient than the semiparametric model if the base- information is available in the ACDIS, including line hazard is correctly specified. We thus estimated
- 38. The socioeconomic determinantsof HIV incidence Barnighausen et al. ¨ S33 parametric models in addition to the CPHM. First, we Results used the exponential model During 4352 person-years at risk, 131 of the 3325 hðt; Xi Þ ¼ expðaÞ Á expðXi bÞ (2) individuals in our sample became seropositive. The overall incidence of HIV infection was 3.0 per 100 which assumes that the hazard function is constant over person-years (95% confidence interval 2.5–3.6). Table 2 time. Next, we estimated the Weibull model shows the unadjusted hazard ratios when we examined hðt; Xi Þ ¼ p Á tpÀ1 Á expðaÞ Á expðXi bÞ (3) separately the effects on time to seroconversion of sex; age, age2 and age3; educational attainment; wealth categories; household expenditure; place of residence; which allows the hazard function to increase (p > 1) or migration status; and partnership status in CPHM. In decrease (p < 1) monotonically over time. The Weibull these separate regressions, female sex, age, belonging to model includes the exponential model as a special case the middle wealth category, urban/periurban place of (p ¼ 1). residence, having migrated out of the DSA between the first and the second round of the HIV surveillance, and Finally, we estimated random effects generalizations to the not being married but having a partner were positively proportional hazards models, frailty models, which allow associated with the hazard of HIV seroconversion (all for variability in unobserved individual-level factors that P < 0.010). Educational attainment and household is unaccounted for by the independent variables included expenditure were not significantly associated with the in the models [45]. The unobservable individual effect hazard of HIV seroconversion (both P ! 0.556). (frailty), v, is considered a random variable over the population that multiplicatively enters the hazard func- By experimentation we found that a third-order poly- tion in the above proportional hazards specification, i.e. nomial age specification provided a good fit for the hðt; Xi jvi Þ ¼ vi Á hðt; Xi Þ ¼ vi Á ho ðtÞ Á expðXi bÞ (4) relationship between age and time to HIV seroconversion. In order to reduce multicollinearity we expressed age as its The random variable v is assumed to be positive, to be deviation from its mean [46]. When we adjusted for sex and distributed independently of t and X, and to follow a age (in the third-order polynomial specification) two of the gamma distribution with unit mean (required for relationships from the individual regressions that do not identification) and finite variance (u). If u is not significantly adjust for any other factor changed significantly (Table 2). different from zero, individual heterogeneity is not First, the hazard ratio of educational attainment changed important and it is appropriate to estimate the non-frailty from 0.99 (P ¼ 0.785) to 0.93 (P ¼ 0.022). Holding age models. and sex constant, each additional year of educational Table 2. Unadjusted hazard ratios of HIV seroconversion and hazard ratios adjusted for sex and age. HR (s.e.) P value aHR (s.e.) P value Sex Male 1 – – Female 1.8314 (0.3616) 0.002 – – Age (years) 0.9954 (0.0162) 0.778 – – Age2 0.9915 (0.0020) < 0.001 – – Age3 1.0003 (0.0001) 0.002 – – Educational attainment (years) 0.9931 (0.0251) 0.785 0.9338 (0.0279) 0.022 Wealth category Poorest 40% 1 1 Middle 40% 1.8525 (0.3716) 0.002 1.8688 (0.3750) 0.002 Wealthiest 20% 1.0247 (0.2789) 0.928 1.0427 (0.2844) 0.878 Daily total per capita household expenditures (ZAR, ln) 0.9156 (0.1370) 0.556 0.9591 (0.1415) 0.777 Place of residence Rural 1 1 Urban/periurban 1.5852 (0.2836) 0.010 1.6272 (0.2923) 0.007 Migration status Migrant 1 1 Non-migrant 0.4387 (0.1067) 0.001 0.4764 (0.1172) 0.003 Partnership status Not married, without partner 1 1 Married 0.8437 (0.3065) 0.640 0.8224 (0.3505) 0.646 Not married, with partner 2.1285 (0.3946) < 0.001 1.5992 (0.3680) 0.041 HR, hazard ratio; aHR, hazard ratio adjusted for sex and age; s.e., standard error; ZAR, South African rand; Ln, natural Logarithm.
- 39. S34 AIDS 2007, Vol 21 (suppl 7) attainment reduced the hazard of HIV seroconversion by significant determinant of HIV seroconversion in any 7%. Second, the hazard ratio for the group of unmarried of the models. Urban residence was associated with a 65% individuals with a partner changed from 2.13 (P < 0.001) increase in the hazard of HIV seroconversion (P ¼ 0.012; to 1.60 (P ¼ 0.041). Table 3, IIB). Individuals who remained residents in the ACDIS DSA between the two rounds of the HIV Table 3 shows estimation results of the semiparametric surveillance faced approximately half the hazard of HIV CPHM and the parametric exponential and Weibull seroconversion of those who migrated out of the DSA regression models in their proportional hazard specifica- between the two rounds (P ¼ 0.006). Once all other tion (expressions 1, 2 and 3, respectively). We tested the variables were controlled for, the hazard ratio of proportional hazards assumption for all variables jointly unmarried individuals with a partner at baseline remained and for each variable individually using the tests proposed borderline significant (P ¼ 0.074) and the partnership by Grambsch and Therneau [47]. The null hypothesis variables jointly increased the model fit with borderline that the hazard rates are proportional could not be significance (P ¼ 0.099). rejected at the 10% confidence level in any of the tests. A unit change in one of the independent variables leads to a In order to test whether the effects of education and wealth proportional shift of the hazard rate. on HIV incidence are modified by sex, we added in turn education-sex and wealth category-sex interaction terms The sizes and significance levels of the adjusted hazard to all the models reported in Table 3. None of the ratios were very similar in all three estimations. Whereas interaction terms was significant at the 5% confidence the CPHM leads to consistent estimates and is more level. Furthermore, we added the square of the educational flexible than the parametric models, it is less efficient than attainment variable to the regression models IB, IIB and the appropriate parametric model. In the Weibull IIIB in order to investigate whether the effect of education regression, the null hypothesis that P ¼ 1 could not be on HIV incidence is non-linear. In none of the models was rejected (Table 3), i.e. the hazard function is neither the added term significant; we thus did not include it in the increasing nor decreasing over time. The exponential final regression equations. Similarly, we replaced the log estimation is thus preferred over the Weibull estimation. transform of the household expenditure variable in models In order to check for unobserved individual heterogen- IB, IIB and IIIB with alternative functional forms (linear, eity, we estimated the exponential model (Table 3, IIB) linear and square). In none of the alternative regression and the Weibull model (Table 3, IIIB) including an equations was any of the household expenditure terms individual random effect, or frailty, y (see equation 4). significant at the 5% confidence level. The null hypothesis that the individual random effect is equal to zero was not rejected at the 10% significance level. The exponential estimation without frailty is thus the preferred parametric model and was used for the Discussion description of results below. We show that educational attainment significantly In multivariable survival analysis, belonging to the middle reduces the hazard of becoming infected with HIV in 40% of households as ranked by the assets index increased a poor rural community in South Africa when controlling the hazard of HIV seroconversion by a factor of for sex, age, wealth, household expenditure, place of approximately 2 (P ¼ 0.001; Table 3, IIA). Controlling residence, migration status and partnership status. The for place of residence, migration status and partnership protective effect of education shown in this study differs status in addition to sex and age reduced the size of the from the findings of previous studies that suggest that hazard ratio (to 1.72) but the effect remained significant educational attainment is not significantly associated, or (P ¼ 0.012; Table 3, IIB). To test whether our finding that positively associated, with the risk of HIV infection [27]. belonging to a household in the middle wealth category increases the risk of HIV incidence is robust to a change in The differences between our results and the findings of the choice of household wealth categories, we repeated previous studies may be caused by methodological issues the regressions in Table 3 with households categorized such as how they control for confounding. We find that into wealth tertiles on the assets index scale. The educational attainment is not correlated with the risk of alternative categorization changed the sizes and signifi- HIV seroconversion in univariable analysis but that its cant levels of all coefficients only slightly. In particular, protective effect against HIV seroconversion becomes when we replaced the wealth variables in model IIB with apparent once sex and age are controlled for. In the case of variables capturing wealth tertiles, the coefficient of the South Africa, the relationship between educational middle wealth category became 1.62 (P ¼ 0.037). attainment and time to HIV seroconversion is likely to be confounded by sex and age. For example, HIV One additional grade of educational attainment reduced seroconversion risk decreases with age above certain peak the hazard of HIV seroconversion by approximately 7% ages in women and men [48], whereas average (Table 3, IIB). Household expenditure was not a educational attainment decreases with age in older age
- 40. Table 3. Multipleregression models of the hazard of HIV seroconversion. Cox estimation Exponential estimation Weibull estimation Model IA Model IB Model IIA Model IIB Model IIIA Model IIIB Independent variables aHR (s.e.) P value aHR (s.e.) P value aHR (s.e.) P value aHR (s.e.) P value aHR (s.e.) P value aHR (s.e.) P value Sex Male 1 1 1 1 1 1 Female 2.0024 (0.4101) 0.001 1.9356 (0.4049) 0.002 2.0004 (0.4098) 0.001 1.9294 (0.4037) 0.002 1.9993 (0.4095) 0.001 1.9284 (0.4035) 0.002 Age (years) 0.9663 (0.0178) 0.063 0.9664 (0.0192) 0.085 0.9660 (0.0178) 0.061 0.9662 (0.0192) 0.083 0.9661 (0.0178) 0.061 0.9963 (0.0192) 0.084 Age2 0.9894 (0.0021) < 0.001 0.9917 (0.0023) < 0.001 0.9894 (0.0021) < 0.001 0.9917 (0.0023) < 0.001 0.9894 (0.0021) < 0.001 0.9917 (0.0023) < 0.001 Age3 1.0004 (0.0001) < 0.001 1.0003 (0.0001) 0.002 1.0004 (0.0001) < 0.001 1.0003 (0.0001) 0.002 1.0004 (0.0001) < 0.001 1.0003 (0.0001) 0.002 Education (years) 0.9252 (0.0285) 0.011 0.9287 (0.0289) 0.017 0.9255 (0.0284) 0.012 0.9290 (0.0288) 0.017 0.9256 (0.0284) 0.012 0.9291 (0.0288) 0.018 Wealth category Poorest 40% 1 1 1 1 1 1 Middle 40% 2.0334 (0.4203) 0.001 1.7200 (0.3707) 0.012 2.0184 (0.4170) 0.001 1.7182 (0.3701) 0.012 2.0200 (0.4173) 0.001 1.7183 (0.3701) 0.012 Wealthiest 20% 1.2652 (0.3761) 0.429 0.9335 (0.2990) 0.830 1.2403 (0.3690) 0.469 0.9249 (0.2965) 0.808 1.2436 (0.3700) 0.464 0.9262 (0.2969) 0.811 Daily total per capita 0.9514 (0.1545) 0.759 0.9319 (0.1498) 0.661 0.9522 (0.1548) 0.763 0.9335 (0.1502) 0.669 0.9522 (0.1548) 0.763 0.9335 (0.1502) 0.669 The socioeconomic determinants of HIV incidence Barnighausen et al. household expenditures (ZAR, ln) Place of residence Rural – 1 – 1 – 1 Urban/periurban – – 1.6837 (0.3367) 0.009 – – 1.6526 (0.3304) 0.012 – – 1.6574 (0.3315) 0.012 Migration status Migrant – 1 – 1 – 1 Non-migrant – – 0.4988 (0.1230) 0.005 – – 0.5066 (0.1248) 0.006 – – 0.5052 (0.1245) 0.006 Partnership status Not married, without partner – 1 – 1 – 1 Married – – 0.9113 (0.3912) 0.829 – – 0.9093 (0.3906) 0.825 – – 0.9095 (0.3906) 0.825 Not married, with partner – – 1.4885 (0.3386) 0.080 – – 1.5011 (0.3413) 0.074 – – 1.4992 (0.3409) 0.075 ln (p) À0.0340 À0.0422 p 0.9665 0.9587 N 3325 3325 3325 3325 3325 3325 Time at risk (person-years) 4352 4352 4352 4352 4352 4352 Seroconversions (number) 131 131 131 131 131 131 Log pseudolikelihood À1008 À999 À660 À650 À660 À650 AIC 2033 2021 1338 1327 1339 1329 BIC 2082 2095 1393 1406 1400 1420 aHR, adjusted hazard ratio; s.e., standard error; ZAR, South African rand; Ln, natural logarithm; AIC, Akaike information criterion; BIC, Bayesian information criterion. S35 ¨
- 41. S36 AIDS 2007, Vol 21 (suppl 7) groups. The age-specific pattern of education reflects Finally, we find that other covariates (sex, age, place of secular changes in South African education policy, such as residence, migration status and partnership status) the South African Schools Act of 1996 that abolished influence the hazard of HIV seroconversion as expected racial segregation in schools [49]. For example, the matric based on previous studies [35–38,40,41]. Studies of risky pass rate (i.e. attainment of grade 12) increased from 40% sexual behaviour in Africa have shown striking differ- in the late 1990s to 68% in 2005 [50]. ences between women and men [51–54]. In as far as education and wealth effects on HIV incidence are Alternatively, education effects may differ by the stage of conveyed by sexual behaviour we expected to find, but the HIVepidemic. Most of the published studies that have did not, that the effects of education and wealth on HIV examined the relationship between education and HIV incidence are modified by sex. It is possible that pathways infection were conducted in early stages of the epidemic from education and wealth to HIVacquisition that are not [27], when educational attainment may have been sex-specific (e.g. malnutrition) are relatively more positively associated with HIV infection, for example important in explaining our findings than sex-specific because the more educated had more partners in any pathways, or that after controlling for sex and other given period of time than the less educated. In contrast, as factors different pathways in women and men have similar the epidemic matured, the more educated may have effects on HIV incidence. Given that our sample includes adopted HIV risk-reducing behaviours more quickly fewer men than women, however, it is also possible that than the less educated because they were more exposed to our study lacks the power to detect a sex differential in the health promotion messages or more empowered to effects of education and wealth on HIV incidence. negotiate protective behaviours with sexual partners [27]. Another possible limitation of our study are uncontrolled We also show that in this overall poor community it is not selection effects because of selection into the baseline the members of the asset-poorest households who are at sample, because of missing information on independent highest risk of HIV acquisition but people who live in variables, or because of attrition between the first and the households belonging to the middle category of relative second round of the HIV surveillance. Whereas selection wealth. Recent analyses of cross-sectional surveys of HIV on observed factors that are associated with HIV serostatus in Africa have shown that the poor do not have seroconversion will bias estimates of HIV incidence (unless the highest HIV prevalence [11,22]. Our longitudinal different selection biases balance each other out), study provides evidence about the causal effect of relative coefficient estimates in multiple regression will be wealth on the risk of HIV acquisition. First, our results consistent if the observed factors that determine selection obtain when important other determinants of acquisition are included as independent variables in the regression of HIV that may be correlated with wealth are controlled equation [55,56]. Our regression equations thus control for for, particularly urban residence and migration status. selection on sex, age, education, wealth, household Second, unlike analyses of cross-sectional surveys, we can expenditure, rural versus urban/periurban place of rule out the possibility that the positive HIV status of residence, migration status and partnership status. As these study participants caused his or her household to fall into characteristics are among the most commonly observed poverty because of the loss of employment or increased correlates of HIV infection in South Africa [57], it is expenses related to disease. Third, unlike results from possible that our model adequately controls for selection cross-sectional surveys, our findings cannot have been effects. We cannot completely rule out that selection on caused by a wealth gradient in the survival with HIV. unobserved characteristics that are associated with the risk of contracting HIV affect our findings. One possibility to Our third main finding is that household expenditure adjust for selection on unobservable factors are Heckman- does not seem to influence the hazard of HIV type selection models, which are not as well developed for seroconversion in this population. In the ACDIS DSA, survival analysis as they are, for example, for ordinary least government clinics distribute condoms for free and squares regression or probit regression, and whose provide basic health services, including treatment for performance commonly depends on the existence of a sexually transmitted diseases, for free. One plausible valid and relevant exclusion restriction, i.e. a variable that is explanation of our result is thus that access to services that a significant predictor of selection into the sample, but not can help prevent HIV transmission does not depend on independently associated with the time to HIV serocon- households’ short-term ability to pay. Alternatively, it is version [58–60]. Future studies are needed to investigate possible that financial liquidity does improve access (for the effect of selection on unobservable factors in analysis example, because transport to government clinics is costly of the socioeconomic determinants of HIV incidence in or because private health care providers offer some this community. services that are effective in reducing HIV transmission that are not available at government clinics), but that In sum, our results provide little support for the assertion access does not translate into actual utilization of such that ‘‘reducing poverty will be at the core of a long-term, services (for example, because individuals who could sustainable solution to reducing HIV/AIDS’’ [61]. access them do not believe that they are effective). Although poverty reduction is important for obvious
- 42. The socioeconomic determinantsof HIV incidence Barnighausen et al. ¨ S37 reasons, it may not be as effective as anticipated in reducing 14. Booysen FL, Mafereka R. The impact of HIV/AIDS on house- the spread of HIV in rural South Africa. In contrast, hold savings in two Free State communities. In: Conference on Reducing Poverty and Inequality: how can Africa be included? increasing educational attainment in the general popu- Oxford: Oxford University, Centre for Study of African Econo- lation, whatever the precise pathways of the effect, may mies; 2006. lower HIV incidence. 15. Braitstein P, Brinkhof MW, Dabis F, Schechter M, Boulle A, Miotti P, et al. Mortality of HIV-1-infected patients in the first year of antiretroviral therapy: comparison between low- income and high-income countries. Lancet 2006; 367:817– 824. 16. Wood E, Montaner JS, Chan K, Tyndall MW, Schechter MT, Acknowledgement Bangsberg D, et al. Socioeconomic status, access to triple therapy, and survival from HIV-disease since 1996. AIDS The authors wish to thank the fieldworkers and 2002; 16:2065–2072. 17. Bulterys M, Chao A, Habimana P, Dushimimana A, Nawrocki P, supervisors at the Africa Centre for Health and Saah A. Incident HIV-1 infection in a cohort of young women in Population Studies for their excellent work in the HIV Butare, Rwanda. AIDS 1994; 8:1585–1591. surveillance and the Demographic Information System. 18. Kapiga SH, Lyamuya EF, Lwihula GK, Hunter DJ. The incidence of HIV infection among women using family planning methods They would also like to thank the Africa Centre in Dar es Salaam, Tanzania. AIDS 1998; 12:75–84. community for their participation in the surveys. 19. Mayala M, Minlangu M, Nzila N, Mama A, Jingu M, Mundele L, et al. Prevalence and incidence of HIV-1 infection among employees of a large textile business and their wives in Sponsorship: The research reported in this paper was Kinshasa, 1991–1996 [in French]. Rev Epidemiol Sante Pub- supported by the Wellcome Trust through grant lique 2001; 49:117–124. GR065377/Z/01/B for the HIV surveillance and grant 20. Mbizvo MT, Kasule J, Mahomed K, Nathoo K. HIV-1 serocon- GR065377/Z/01/H for the Africa Centre Demographic version incidence following pregnancy and delivery among women seronegative at recruitment in Harare, Zimbabwe. Information System. Some of the methods used in this Cent Afr J Med 2001; 47:115–118. analysis were devised and tested at workshops run by 21. Senkoro KP, Boerma JT, Klokke AH, Ng’weshemi JZ, Muro AS, the Wellcome Trust-funded ALPHA network (grant Gabone R, Borgdorff MW. HIV incidence and HIV-associated GR075886/Z/04/Z). mortality in a cohort of factory workers and their spouses in Tanzania, 1991 through 1996. J Acquir Immune Defic Syndr 2000; 23:194–202. Conflicts of interest: None. 22. de Walque D. Who gets AIDS and how? The determinants of HIV infection and sexual behaviors in Burkina Faso, Cameroon, Ghana, Kenya, and Tanzania. Policy research working paper References series 3844. The World Bank: Washington, DC; 2006. 23. Wawer MJ, Serwadda D, Gray RH, Sewankambo NK, Li C, 1. UNAIDS. 2006 Report on the global AIDS epidemic, annex 2: Nalugoda F, et al. Trends in HIV-1 prevalence may not reflect HIV and AIDS estimates and data. Geneva: Joint United Nations trends in incidence in mature epidemics: data from the Rakai Programme on HIV/AIDS; 2006. population-based cohort, Uganda. AIDS 1997; 11:1023–1030. 2. Buve A, Bishikwabo-Nsarhaza K, Mutangadura G. The spread 24. Hosegood V, Timaeus IM. Household composition and dy- and effect of HIV-1 infection in sub-Saharan Africa. Lancet namics in KwaZulu-Natal, South Africa: mirroring social reality 2002; 359:2011–2017. in longitudinal data collection. In: van der Walle, E., Sharpe, 3. Nyindo M. Complementary factors contributing to the rapid M.E., editors. African households: censuses and surveys. spread of HIV-I in sub-Saharan Africa: a review. East Afr Med J Champaign, IL: University of Illinois Press; 2006. pp. 58–77. 2005; 82:40–46. 25. Case A, Ardington C. Socioeconomic factors. Africa Centre 4. Wojcicki JM. Socioeconomic status as a risk factor for HIV Demographic Information System: sociodemographic and infection in women in East, Central and Southern Africa: a health events. Africa Centre Population Studies Group. Mtuba- systematic review. J Biosoc Sci 2005; 37:1–36. tuba: Africa Centre for Health and Population Studies; 2004. 5. Johnson L, Budlender D. HIV risk factors: a review of demo- pp. 68–78. graphic, socioeconomic, biomedical and behavioural determi- 26. UNAIDS/WHO. Guidelines for using HIV testing technologies in nants of HIV prevalence in South Africa. CARE monograph no. 8. surveillance: selection, evaluation, and implementation. Cape Town: Centre for Actuarial Research, University of Cape Geneva: UNAIDS/WHO; 2001. Town; 2002. 27. Hargreaves JR, Glynn JR. Educational attainment and HIV-1 6. Braveman PA, Cubbin C, Egerter S, Chideya S, Marchi KS, infection in developing countries: a systematic review. Trop Metzler M, Posner S. Socioeconomic status in health research: Med Int Health 2002; 7:489–498. one size does not fit all. JAMA 2005; 294:2879–2888. 28. Morris SS, Carletto C, Hoddinott J, Christiaensen LJ. Validity of 7. Mosley A. Does HIV or poverty cause AIDS? Biomedical and rapid estimates of household wealth and income for health epidemiological perspectives. Theor Med Bioeth 2004; 25:399– surveys in rural Africa. J Epidemiol Commun Health 2000; 421. 54:381–387. 8. Myer L, Ehrlich RI, Susser ES. Social epidemiology in South 29. Filmer D, Pritchett LH. Estimating wealth effects without Africa. Epidemiol Rev 2004; 26:112–123. expenditure data – or tears: an application to educational 9. Stillwaggon E. HIV/AIDS in Africa: fertile terrain. J Dev Studies enrollments in States of India. Demography 2001; 38:115–132. 2002; 38:1–22. 30. van der Berg S, Louw M. Changing patterns of South African 10. Booysen FL, Summerton J. Poverty, risky sexual behaviour, and income distribution: towards time series estimates of distribu- vulnerability to HIV infection: evidence from South Africa. tion and poverty. S Afr J Econ 2004; 72:546–572. J Health Popul Nutr 2002; 20:285–288. 31. Gyekye A, Akinboade O. A profile of poverty in the Limpopo 11. Shelton JD, Cassell MM, Adetunji J. Is poverty or wealth at the province of South Africa. East Afr Soc Sci Res Rev 2003; 19:89– root of HIV? Lancet 2005; 366:1057–1058. 109. 12. Greener R. The impact of HIV/AIDS on poverty and inequality. 32. Blaauw D, Penn-Kekana L. Socio-economic inequalities and In: Haacker, M., editor. The macroeconomics of HIV/AIDS. maternal health in South Africa. Johannesburg: Health Systems Washington, DC: International Monetary Fund; 2004. pp. 167– Development Programme. Centre for Health Policy, University 181. of Witswatersrand; 2002. 13. Bachmann MO, Booysen FL. Health and economic impact of 33. Booysen FL. Using demographic and health surveys to measure HIV/AIDS on South African households: a cohort study. BMC poverty – an application to South Africa. J Stud Econ Econo- Public Health 2003; 3:14. metr 2002; 26:53–70.
- 43. S38 AIDS 2007, Vol 21 (suppl 7) 34. World Bank. Poverty in Africa – standardized welfare indicators. 47. Grambsch PM, Therneau TM. Proportional hazards tests and 2007. Available at: http://www4.worldbank.org/afr/poverty/ diagnostics based on weighted residuals. Biometrika 1994; measuring/Indicators/definitions_en.htm. Accessed: 12 June 81:515–526. 2006. 48. Barnighausen T, Tanser F, Mbizana C, Gqwede Z, Herbst K, ¨ 35. Auvert B, Ballard R, Campbell C, Carael M, Carton M, Fehler G, Newell ML. HIV incidence in a rural community in KwaZulu- et al. HIV infection among youth in a South African mining Natal, South Africa. In: 3rd South African AIDS Conference. town is associated with herpes simplex virus-2 seropositivity Durban, South Africa, 5–8 June, 2007 [Abstract 737]. and sexual behaviour. AIDS 2001; 15:885–898. 49. Maithufi IP. The South African Schools Act of 1996: a break 36. Connolly C, Shisana O, Colvin M, Stoker D. Epidemiology of with the past. In: Robinson, J.A., editor. The law of children and HIV in South Africa–results of a national, community-based minors in South Africa. South Africa: Butterworth; 1997. survey. S Afr Med J 2004; 94:776–781. 50. SouthAfrica.info. Education in South Africa. 2006. Available at: 37. Pettifor AE, Rees HV, Kleinschmidt I, Steffenson AE, MacPhail C, http://www.southafrica.info/ess_info/sa_glance/education/edu- Hlongwa-Madikizela L, et al. Young people’s sexual health in cation.htm. Accessed: 15 January 2007. South Africa: HIV prevalence and sexual behaviors from a 51. Babalola S. Gender differences in the factors influencing con- nationally representative household survey. AIDS 2005; sistent condom use among young people in Tanzania. Int J 19:1525–1534. Adolesc Med Health 2006; 18:287–298. 38. Dunkle KL, Beksinska ME, Rees VH, Ballard RC, Htun Y, Wilson 52. Boer H, Mashamba MT. Gender power imbalance and differ- ML. Risk factors for HIV infection among sex workers in ential psychosocial correlates of intended condom use among Johannesburg, South Africa. Int J STD AIDS 2005; 16:256– male and female adolescents from Venda, South Africa. Br J 261. Health Psychol 2007; 12:51–63. 39. Lurie MN, Williams BG, Zuma K, Mkaya-Mwamburi D, Garnett 53. Dunkle KL, Jewkes RK, Nduna M, Levin J, Jama N, Khuzwayo N, G, Sturm AW, et al. The impact of migration on HIV-1 et al. Perpetration of partner violence and HIV risk behaviour transmission in South Africa: a study of migrant and nonmi- among young men in the rural Eastern Cape, South Africa. AIDS grant men and their partners. Sex Transm Dis 2003; 30:149– 2006; 20:2107–2114. 156. 54. Jewkes R, Dunkle K, Nduna M, Levin J, Jama N, Khuzwayo N, 40. Shisana O, Hall EJ, Maluleke R, Chauveau J, Schwabe C. HIV/ et al. Factors associated with HIV sero-status in young rural AIDS prevalence among South African health workers. S Afr South African women: connections between intimate partner Med J 2004; 94:846–850. violence and HIV. Int J Epidemiol 2006; 35:1461–1468. 41. Zuma K, Gouws E, Williams B, Lurie M. Risk factors for HIV 55. Wooldridge JM. Econometric analysis of cross section and panel infection among women in Carletonville, South Africa: migra- data. Cambridge, MA: MIT Press; 2001. tion, demography and sexually transmitted diseases. Int J STD 56. Vella F. Estimating models with sample selection bias: a survey. AIDS 2003; 14:814–817. J Hum Resources 1998; 33:127–169. 42. Ahmed S, Lutalo T, Wawer M, Serwadda D, Sewankambo NK, 57. Shisana O, Rehle T, Simbayi LC, Parker W, Zuma K, Bhana A, Nalugoda F, et al. HIV incidence and sexually transmitted et al. South African National HIV prevalence, HIV incidence, disease prevalence associated with condom use: a population behaviour and communication survey. Cape Town: HSRC Press; study in Rakai, Uganda. AIDS 2001; 15:2171–2179. 2005. 43. Shisana O, Zungu-Dirwayi N, Toefy Y, Simbayi LC, Malik S, 58. Greene WH. Econometric analysis.. Upper Saddle River, NJ: Zuma K. Marital status and risk of HIV infection in South Prentice Hall; 2002. Africa. S Afr Med J 2004; 94:537–543. 59. Heckman JJ. Sample selection bias as a specification error. 44. Cox DR. Regression models and life tables. J Roy Stat Soc 1972; Econometrica 1979; 47:153–161. 34:187–220. 60. Ahn HT, Powell JL. Semiparametric estimation of censored 45. Gutierrez RG. Parametric frailty and shared frailty survival selection models with nonparametric selection mechanism. models. Stata J 2002; 2:22–44. J Econometr 1993; 58:3–29. 46. Kutner MH, Nachtsheim CJ, Neter J, Li W. Applied linear 61. Fenton L. Preventing HIV/AIDS through poverty reduction: the statistical models. 5th ed. Boston: McGraw-Hill; 2004. only sustainable solution? Lancet 2004; 364:1186–1187.
- 44. Explaining continued highHIV prevalence in South Africa: socioeconomic factors, HIV incidence and sexual behaviour change among a rural cohort, 2001–2004 James R. Hargreavesa,b, Christopher P. Bonella, Linda A. Morisona, Julia C. Kima,b, Godfrey Phetlab, John D.H. Portera, Charlotte Wattsa and Paul M. Pronyka,b Objectives: To estimate HIV incidence and explore evidence for changing sexual behaviour over time among men and women belonging to different socioeconomic groups in rural South Africa. Design and methods: A cohort study conducted between 2001 and 2004; 3881 individuals aged 14–35 years enumerated in eight villages were eligible. At least three household visits were made to contact each eligible respondent at both timepoints. Sexual behaviour data were collected in structured, respondent-focused interviews. HIV serostatus was assessed using an oral fluid enzyme-linked immunosorbent assay at each timepoint. Results: Data on sexual behaviour were available from 1967 individuals at both timepoints. A total of 1286 HIV-negative individuals at baseline contributed to the analysis of incidence. HIV incidence was 2.2/100 person-years among men and 4.9/ 100 person-years in women, among whom it was highest in the least educated group. Median age at first sex was lower among later birth cohorts. A higher number of previously sexually active individuals reported having multiple partners in the past year in 2004 than 2001. Condom use with non-spousal partners increased from 2001 to 2004. Migrant men more often reported multiple partners. Migrant and more educated individuals of both sexes and women from wealthier households reported higher levels of condom use. Discussion: HIV incidence is high in rural South Africa, particularly among women of low education. Some risky sexual behaviours (early sexual debut, having multiple sexual partners) are becoming more common over time. Condom use is increasing. Existing HIV prevention strategies have only been partly effective in generating population-level behavioural change. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins AIDS 2007, 21 (suppl 7):S39–S48 Keywords: Education, HIV infection, migration, poverty, sexual behaviour, South Africa Introduction behaviour [4–6] from a number of sub-Saharan African countries. In contrast, antenatal surveillance data and In recent years, there have been reports of decreases in repeated national HIV prevalence surveys from South HIV prevalence [1,2], HIV incidence [3] and sexual risk Africa suggest a continued rise in HIV prevalence despite From the aLondon School of Hygiene and Tropical Medicine, London, UK, and the bRural AIDS and Development Action Research Programme, Acornhoek, South Africa. Correspondence to James R. Hargreaves, Infectious Disease Epidemiology Unit, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK. Tel: +44 020 7927 2955; fax: +44 020 7637 4314; e-mail: james.hargreaves@lshtm.ac.uk ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S39
- 45. S40 AIDS 2007, Vol 21 (suppl 7) extensive efforts to reduce sexual risk behaviour [7–10]. reporting. Household wealth was assessed through a HIV prevalence data are, however, an uncertain guide to participatory wealth-ranking technique based on com- incidence because prevalence is affected both by HIV munity informant rankings of each household’s wealth incidence and changing mortality patterns. Despite this, repeated three times [13]. few studies have reported direct measures of HIV incidence in any South African population. There is an Oral fluid samples were collected using the OraSure urgent need to understand better the pattern of new HIV collection device (UCB Group, Belgium) and analysed infections in South Africa and whether this is associated with the Vironostika HIV Uni-Form II assay (bioMerieux, with changes in sexual behaviour. France). HIV data from one interviewer at baseline raised quality concerns and were excluded from the analysis We report data from a cohort study conducted in rural (n ¼ 168). Samples testing negative at baseline were South Africa between 2001 and 2004, as part of a cluster- included in the analysis of HIV incidence. Of these, randomized trial of a microfinance and training inter- 34% were stored for slightly longer than recommended by vention reported elsewhere [11]. The current paper has the manufacturer before analysis, but were included in the three objectives: to estimate HIV incidence among a rural analysis after checking that their inclusion did not bias the South African cohort; to explore evidence for changing measure of incidence. sexual behaviour between 2001 and 2004 in this study population; and to assess the evidence that HIV incidence Statistical analysis rates and sexual behaviour patterns differed across Data were entered into an Access database (Microsoft, socioeconomic groups identified on the basis of wealth, California, USA) with statistical analysis conducted using education and temporary migrancy. Stata version 9 (Stata Corp., College Station, Texas, USA). The key exposure variable was date of interview (baseline predominantly in 2001 and follow-up pre- Methods dominantly in 2004). The exposure period was recorded as the duration between the first and last interview, or half Setting of this for those who HIV seroconverted. Household The study was conducted in Limpopo province in South wealth was assessed at baseline and identified households Africa’s north east. Poverty remains widespread in the study as ‘very poor’, ‘poor, but a bit better off ’ or ‘doing OK’. area [12,13] and unemployment rates exceed 40% [14]. Temporary migrancy status was assessed on the basis of There are high levels of labour migration, with 60% of whether an individual was sleeping in the home at the adult men and 25% of women residing away from home for time of interview at both timepoints and was coded as a more than 6 months per year [15]. Few households have binary measure. Educational attainment was coded into land or livestock sufficient to support livelihoods. three categories (no or primary education only; attended but did not complete secondary education; completed Data collection secondary or postsecondary study). This measure used Ethical approval for the study was granted by committees data collected at follow-up rather than baseline because at the London School of Hygiene and Tropical Medicine education had not changed for most except the very and the University of Witswatersrand. young in whom the later data were considered more relevant to our outcomes. Two hundred dwellings were randomly sampled in each of eight study villages in 2001. A household roster was Data from baseline and follow-up surveys were compiled assembled including all individuals identified as house- to analyse age of sexual debut by a survival approach with hold members by the household head, regardless of censoring at the current age for those not yet sexually whether they were currently sleeping at the dwelling, in active because there were many individuals in this order to account for high levels of temporary labour category. The number of sexual partnerships during the migration in South Africa. Individuals of both sexes aged past 12 months (including spousal and non-spousal 14–35 years were eligible for inclusion in the cohort. partnerships) was explored via a binary variable (> 1 sexual partner; 1 or no partner). Condom use at last sex Data were collected by trained female fieldworkers with a non-spousal partner was analysed as a binary through face-to-face structured interviews conducted in characteristic of sexual partnerships, as opposed to the local language (Sepedi). Witnessed verbal consent was individuals, and was recorded for the three most recent obtained from all subjects. Attempts were made to partnerships from the past 12 months for each respondent. maximize response and follow-up rates by instituting a full-time field office and making repeated efforts to trace HIV incidence rates by age at baseline, sex and migrants. Effort was also made to ensure accurate socioeconomic status were calculated among those reporting through the use of limited recall times, a who were HIV negative at baseline. Logistic regression respondent-oriented interview and stressing confidenti- was used to assess whether socioeconomic status variables ality, anonymity and the importance of honesty in were associated with HIV seroconversion. In order to
- 46. Explaining continued highHIV prevalence in South Africa Hargreaves et al. S41 assess whether there was evidence for any change in age of (73.5%) were succesfully interviewed at baseline and 1967 sexual debut over time, we used Cox regression to assess (68.8%) of those interviewed at baseline had data available the evidence that the rate of sexual debut differed at follow-up. Inability to contact the interviewee was between those aged 14 and 19 years at baseline (many of the primary reason for lack of interview (16% at whose sexual debut occurred after 2001) and those aged baseline, 19% at follow-up), refusal being rare (3%, 3%) 20–35 years (whose sexual debut mostly occurred before and missing data accounting for most other exclusions. In 2001). After this, the influence of household wealth on addition, 44 individuals died during the follow-up age of first sex was investigated. We did not investigate the period, and 371 migrated and could not be traced. association between current educational or mobility Men were less likely to have complete follow-up, as were status and age at first sex because these socioeconomic men and women who were older, married, sleeping away factors were likely to have changed since the time at from the home or had more education (Table 1). The which first sex occurred. average time between baseline and follow-up was 3.1 years. Among those succesfully followed up, valid data on The proportion reporting more than one sexual partner HIV serostatus was collected at both timepoints on 1396 during the past 12 months was calculated at both 2001 individuals, of whom 1286 were HIV negative at baseline, and 2004. To limit selection biases and residual these contributing to the analysis of HIV incidence. confounding as a result of cohort ageing, we restricted the analysis to individuals who had data available at both There were 34 seroconversions among men and 108 timepoints and had been previously sexually active, among women. HIV incidence was 2.2/100 person-years because we expected age to be a strong determinant of [95% confidence interval (CI) 1.5–3.0] for men and 4.9/ first sex, but less strongly associated with the number of 100 person-years (95% CI 4.0–5.9) for women. HIV partners in the past year among those who had already incidence in the age group 15–24 years at baseline was 2.0 started sexual activity. To explore the influence of time (1.3–3.0) for men and 4.7 (3.7–6.0) for women. and socioeconomic factors, a dataset was constructed Incidence was lowest in the youngest age group among containing a record for each individual at each timepoint both sexes and higher among women than men at all ages with the temporary migration variable being allowed to (Table 2). Among men, there was little evidence that HIV vary with time. A logistic regression model, specifying seroconversion was associated with any socioeconomic individual-level clustering via population-averaged gen- factor. Among women, HIV seroconversion was signifi- eral estimating equations was constructed for each sex cantly less common among those with higher levels of separately, with multiple partners during the previous 12 education [adjusted odds ratio (aOR) comparing months as the outcome variable. attended secondary school versus none/primary 0.49, 95% CI 0.28–0.85; comparing those completing Analysis of condom use at last sex included data on all non- secondary school versus none/primary 0.25, 95% CI spousal partnerships reported at both baseline and follow- 0.12–0.53]. There was less evidence for differing HIV up and was thus limited to individuals reporting a non- incidence by marital status, trial arm, household wealth or spousal partner at each time point. Logistic regression, temporary migrancy. employing general estimating equations and specifying individual clustering, was used to estimate the effects of Among men, the median age at first sex was 16 years for date of interview and socioeconomic status variables. those aged 14–19 years at baseline compared with 17 years for those aged over 20 years at baseline (Fig. 1a; All analyses were stratified by sex. Variables considered as hazard ratio 0.60, 95% CI 0.50–0.72). Among women, potential confounders of the effect of time or socio- earlier first sex was also signficantly more often reported economic status variables on outcome characteristics by those aged 14–19 years at baseline (median 16 years) were: age; marital status; village-pair; trial arm; and than the older group (17 years, hazard ratio 0.77, 95% CI (among women for sexual behaviour outcomes) ever 0.67–0.89). Household wealth was not significantly having had a child. For the partnership-level analysis, associated with age at first sex among either sex. models were also adjusted for a measure of frequency of sex during the previous 12 months (more than five times, Men were more likely to report multiple partners in the five or fewer times). When confounders varied over time past year than women at both timepoints (aOR 5.14 95%, this was accounted for in the model. For each analysis CI 4.06–6.53; Fig. 1b and Table 3). Among men, having interaction terms were fitted between the time period had multiple partners during the previous year tended to variable and each of the socioeconomic status variables. be most common among those aged 20–25 years. The number of previously sexually active men reporting multiple partners in the past year increased between the Results baseline and follow-up interviews (Fig. 1b; aOR 1.34, 95% CI 1.02–1.77). Ever having been married, trial arm, Some 1482 households were succesfully enumerated, household wealth and educational attainment were not identifying 3881 eligible 14–35 year olds. Of these, 2858 associated with having had multiple partners in the past
- 47. S42 AIDS 2007, Vol 21 (suppl 7) Table 1. Sociodemographic differences between individuals interviewed at both timepoints and those not included in the final analysis in a rural South African cohort study, 2001–2004. Men Women Eligible individ- Eligible individ- uals not included uals not included Interviewed at both in the final Interviewed at both in the final timepoints analysis timepoints analysis P value P value N % N % (chi square) N % N % (chi square) 767 (41.2%) 1094 1200 (59.4%) 819 Age group (years) 14–19 426 55.5 292 26.7 523 47.5 259 31.6 20–25 187 24.4 384 35.1 332 27.7 273 33.3 26–35 154 20.1 418 38.2 < 0.01 345 28.8 287 35.0 < 0.01 Marital status Never married 670 87.4 792 85.3 851 70.9 447 74.1 Married during follow-up 46 6.0 0 0.0 84 7.0 1 0.2 Ever married at baseline 51 6.7 137 14.8 < 0.01 265 22.1 155 25.7 < 0.01 Household wealth Very poor 226 29.6 326 33.0 343 28.9 253 31.0 Poor, but a bit better off 423 55.4 632 58.2 657 55.4 449 55.0 Doing OK 115 15.1 128 11.8 0.116 186 15.7 114 14.0 0.435 Migrancy status Non-migrant 570 79.4 324 36.6 922 85.9 237 44.4 Becomes migrant 72 10.0 137 15.5 73 6.8 79 14.8 Returns home 21 2.9 207 23.4 19 1.8 120 22.5 Migrant both timepoints 55 7.7 218 24.6 < 0.01 59 5.5 98 18.4 < 0.01 Educational attainment None/primary only 112 14.6 197 18.0 162 13.5 152 18.6 Attended secondary 484 63.1 584 53.4 799 66.6 457 55.8 Completed secondary 171 22.3 313 28.6 < 0.01 239 19.9 210 25.6 < 0.01 Among those included in the final analysis there were missing data on household wealth (17 individuals) and migrancy (176). Among those not included there were missing data on marital status (381 individuals), household wealth (11) and migrancy (493).
- 48. Explaining continued highHIV prevalence in South Africa Hargreaves et al. S43 Table 2. HIV incidence rates among men and women in a rural South African cohort study 2001–2004, by socioeconomic factors. Men Women HIVþ/pyar Rate/100 pyar aOR HIVþ/pyar Rate/100 pyar aOR All 34/1578 2.2 (1.5–3.0) – 108/2196 4.9 (4.0–5.9) – Age at baseline (years) 14–19 13/959 1.4 1 41/1139 3.6 1 20–25 13/340 3.8 2.86 (1.24–6.58) 34/504 6.7 2.32 (1.39–3.87) 26–35 8/279 2.9 1.70 (0.58–4.97) 133/552 6.0 2.55 (1.40–4.66) Marital status Never married 27/1418 1.9 1 79/1575 5.0 1 Married during follow-up 5/81 6.2 2.28 (0.72–7.21) 6/140 4.3 0.57 (0.23–1.43) Ever married at baseline 2/79 2.5 0.82 (0.16–4.26) 23/481 4.8 0.55 (0.29–1.02) Trial arm Control 21/785 2.7 1 49/1125 4.4 1 Intervention 13/793 1.6 0.70 (0.33–1.48) 59/1070 5.5 1.32 (0.87–2.01) Household wealth at baseline Very poor 11/446 2.5 1 35/574 6.1 1 Poor, but a bit better off 13/854 1.5 0.56 (0.24–1.33) 61/1223 5.0 0.84 (0.53–1.33) Doing OK 10/276 3.6 1.42 (0.56–3.64) 12/366 3.3 0.54 (0.27–1.11) Migrancy status Non-migrant 20/1199 1.7 1 77/1692 4.6 1 Becomes migrant 5/157 3.2 1.75 (0.60–5.13) 7/144 4.8 1.08 (0.46–2.53) Returns home 2/31 6.4 3.43 (0.55 –21.41) 3/27 11.3 2.87 (0.70–11.75) Migrant at both timepoints 4/100 4.0 1.49 (0.43–5.12) 7/88 8.2 1.47 (0.60–3.61) Educational attainment at follow-up None/attended primary only 4/233 1.7 1 24/265 9.1 1 Attended secondary 22/999 2.2 1.57 (0.51–4.85) 71/1495 4.7 0.49 (0.28–0.85) Completed secondary 8/346 2.3 1.23 (0.35–4.36) 13/436 3.0 0.25 (0.12–0.53) aOR, Adjusted odds ratio for seroconversion comparing socioeconomic categories adjusted for age, village pair, trial arm and marital status; pyar, person-years at risk. year. There was some evidence that migrant men were sex was more common among partnerships reported at more likely to report multiple partners (aOR versus non- follow-up than at baseline (Fig. 1c; aOR 1.43, 95% CI migrants 1.51, 95% CI 1.03–2.20). 1.07–1.92). Men who had ever been married more often reported condom use than those who had not, although Among previously sexually active women, having had this was not statistically significant (aOR 1.50, 95% CI multiple partners in the past year was most common 0.73–3.09). Condom use was also more frequent in among the youngest age group and was least common sexual relationships in which sex occurred fewer than five among women who had ever been married. As was the times during the previous year (aOR 2.20, 95% CI 1.63– case for men, there was some evidence for an increase 2.97). Condom use tended to be reported more often by over time, adjusting for age and other potential migrants than non-migrants (aOR 1.46, 95% CI 0.98– confounders, in the number of women reporting multiple 2.19) and by those of increasing educational status (aOR partnerships (Fig. 1b; aOR 2.09, 95% CI 1.39–3.17). comparing completed secondary with none/primary Having had multiple partners was not associated with education 2.91, 95% CI 1.73–4.90) but was not household wealth (aOR for household ‘doing OK’ versus associated with mens’ household wealth (aOR compar- ‘very poor’ aOR 1.13, 95% CI 0.62–2.05), migrancy ing household ‘doing OK’ with ‘very poor’ 1.20, 95% CI (aOR 1.05, 95% CI 0.53–2.07) or education (aOR 0.69, 0.75–1.92). 95% CI 0.36–1.30). There was some evidence that living in a village receiving the intervention was associated with Among the 2547 non-spousal partnerships reported by a lower chance of having had multiple partners in the past women, condom use was most often reported by the year (aOR 0.66, 95% CI 0.46–0.93). There was little youngest women. There was strong evidence that evidence of interaction between interview date and condom use was reported by women more commonly socioeconomic status variables for either sex. at follow-up than at baseline (Fig. 1; aOR 1.46, 95% CI 1.14–1.87). Condom use at last sex was more commonly Condom use at last sex within a partnership was more reported by women who had ever been married (aOR often reported when the reporting partner was male than 1.81, 95% CI 1.04–3.14) and in non-spousal relationships female at both timepoints (aOR 1.24, 95% CI 1.01–1.52; in which sex occurred less frequently (aOR 1.45, 95% CI Fig. 1c and Table 4). Among the 1686 non-spousal 1.11–1.89). Condom use was less commonly reported by partnerships reported by men, condom use was most women who reported previously ever having had a child commonly reported when the man was aged 20–25 (aOR 0.72, 95% CI 0.53–0.99). Condom use at last years. There was strong evidence that condom use at last sex was more commonly reported by women from
- 49. S44 AIDS 2007, Vol 21 (suppl 7) (a) Male Female 1.00 0.75 14--19 years in 2001 20--35 years in 2001 0.50 0.25 0.00 10 15 20 25 30 35 10 15 20 25 30 35 Age (years) (b) Male Female % 40 30 Baseline (mainly 2001) Follow-up (mainly 2004) 20 10 0 15 20 25 30 35 15 20 25 30 35 Age (years) (c) % 100 Male Female 80 60 Baseline (mainly 2001) Follow-up (mainly 2004) 40 20 0 15 20 25 30 35 15 20 25 30 35 Age (years) Fig. 1. Age patterns of sexual behaviour by timeperiod among males and females in a rural South African cohort 2001–4. (a) Survival analysis of age at first sex, by sex and birth cohort; (b) Percentage of previously sexually active individuals reporting more than one sexual partner during the previous 12 months (3-year average), by sex and time-period of survey; (c) Percentage of non-spousal sexual partnerships reporting condom use at last sex (3-year average), by sex and time-period of survey. households of greater wealth (aOR comparing household Discussion ‘doing OK’ with ‘very poor’ 2.03, 95% CI 1.29–3.20), those who had completed secondary education (aOR We report data from a cohort study conducted in rural compared to none/primary only 2.25, 95% CI 1.34– South Africa between 2001 and 2004. HIV incidence was 3.78) and migrants (aOR 1.48, 95% CI 0.98–2.23). high among both men and women. Among both sexes There was little evidence of interaction between inter- there was evidence that age of first sexual intercourse view date and socioeconomic status variables among declined over time, whereas, if anything, having had either sex. multiple sexual partnerships during the previous year was
- 50. Explaining continued highHIV prevalence in South Africa Hargreaves et al. S45 Table 3. Multiple partnerships among those previously sexually active reported by men and women in 2001 and 2004 in a rural South African cohort, by socioeconomic factors. Men reporting multiple partners in previous year Women reporting multiple partners in previous year 2001 2004 2001 2004 n/N (%) n/N (%) aOR n/N (%) n/N (%) aOR Interview date Baseline (2001) 129/495 (26.1) – 1 54/943 (5.7) – 1 Follow-up (2004) – 158/493 (32.1) 1.34 (1.02–1.77) – 82/943 (8.7) 2.09 (1.39–3.17) Age at baseline (years) 14–19 44/180 (24.4) 20/62 (32.3) 1 23/278 (8.3) 19/73 (26.0) 1 20–25 52/167 (311) 84/214 (39.3) 1.40 (0.96–2.06) 17/323 (5.3) 31/390 (8.0) 0.44 (0.27–0.73) 26þ 33/148 (22.3) 54/217 (24.9) 0.78 (0.51–1.18) 14/342 (4.1) 32/480 (6.7) 0.41 (0.23–0.71) Marital status Never married 114/444 (25.7) 136/401 (33.9) 1 48/680 (7.1) 59/622 (9.0) 1 Ever married 15/51 (29.4) 22/92 (23.9) 1.01 (0.62–1.64) 6/263 (2.3) 23/321 (7.2) 0.71 (0.45–1.10) Trial arm Control 66/247 (26.7) 85/245 (34.7) 1 23/478 (4.8) 58/478 (12.1) 1 Intervention 63/248 (25.4) 73/248 (29.4) 0.80 (0.60–1.08) 31/467 (6.7) 24/465 (5.2) 0.66 (0.46–0.93) Ever had a child No – – – 22/282 (7.8) 20/159 (12.6) 1 Yes – – – 32/652 (4.9) 62/780 (8.0) 0.87 (0.57–1.34) Household wealth Very poor 38/141 (27.0) 47/142 (33.1) 1 12/276 (4.4) 22/276 (8.0) 1 Poor, but a bit better off 73/278 (26.3) 84/277 (30.3) 0.91 (0.65–1.27) 35/513 (6.8) 43/513 (8.4) 1.23 (0.81–1.87) Doing OK 17/75 (22.7) 27/73 (37.0) 0.95 (0.58–1.54) 7/143 (4.9) 13/143 (9.1) 1.13 (0.62–2.05) Migrancy status Non-migrant 108/424 (25.5) 118/383 (30.8) 1 50/865 (5.8) 61/761 (8.0) 1 Migrant 21/71 (29.6) 31/71 (43.7) 1.51 (1.03–2.20) 4/78 (5.1) 6/73 (8.2) 1.05 (0.53–2.07) Educational attainment at follow-up None/primary only 22/73 (30.1) 18/73 (24.7) 1 11/138 (8.0) 9/137 (6.6) 1 Attended secondary 72/286 (25.2) 93/285 (32.6) 1.02 (0.65–1.60) 35/602 (5.8) 61/602 (10.1) 0.97 (0.58–1.64) Completed secondary 35/136 (25.7) 47/135 (34.8) 1.12 (0.68–1.84) 8/203 (3.9) 12/204 (5.9) 0.69 (0.36–1.30) aOR, Adjusted odds ratio comparing socioeconomic groupings across both timepoints, adjusted for interview date, age, marital status, village pair and trial arm, and, for women only, ever had a child. more commonly reported in 2004 than in 2001. Condom represented in the final sample, and these individuals were use at last sex with non-spousal partners was, however, more likely to be migrants and well educated. It is possible more commonly reported in 2004. that their sexual behaviour or risk of HIV infection differed from those included in the study; if so, our Regarding socioeconomic patterns among these out- estimates may have been biased. As the cohort was ageing, comes, HIV incidence among men was not associated with there may have been residual confounding by age for socioeconomic factors, but among women infections outcomes in which age was an important determinant. occurred fastest among the least educated. Sexually active Many authors have also pointed to the difficulties migrant men more often reported multiple sexual partners, inherent in capturing accurate sexual behaviour infor- but migrant and more educated men also reported more mation in one-off interviews, and it is likely that some common condom use with non-spousal partners. Among misreporting occurred [16,17]. If such misreporting sexually active women, having had multiple sexual partners varied between individuals from different socioeconomic in the past year was not associated with socioeconomic groups or at different timepoints, this may also have factors, but women who were migrants, from wealthier produced some bias, although this is difficult to assess. households and with higher levels of education were more Misreporting of age of sexual debut might have differed likely to report condom use at last sex with a non-spousal with respect to age at baseline because of the likely greater partner. There was little evidence that the strength of time intervals involved in recall for older participants. association between socioeconomic variables and sexual Therefore, the finding of lower age at first sex among later behaviours had changed over time. age cohorts should be treated with some caution. The strengths of the study included explicit attempts to Another limitation of the study was relatively low maximize follow-up rates and ensure accurate reporting. statistical power, particularly with respect to HIV Furthermore, important potential confounders such as incidence analyses and interaction tests. Finally, although age, childbirth and partnership characterstics were our educational exposure may have been relatively simple adjusted for in the analysis. Nevertheless, the study had to record, our assessment of migrancy did not identify limitations. A proportion of eligible individuals were not migrations at times other than when surveys were
- 51. S46 AIDS 2007, Vol 21 (suppl 7) Table 4. Condom use at last sex within a non-spousal sexual partner reported by men and women in 2001 and 2004 in a rural South African cohort, by socioeconomic factors. Men reporting condom use at last sex with non-spousal Women reporting condom use at last sex with partner non-spousal partner 2001 2004 2001 2004 n/N (%) n/N (%) aOR n/N (%) n/N (%) aOR Interview date Baseline (2001) 327/1002 (32.6) – 1 253/915 (27.7) – 1 Follow-up (2004) – 272/684 (39.8%) 1.43 (1.07–1.92) – 237/723 (32.8) 1.46 (1.14–1.87) Age (years) 14–19 101/321 (31.5) 26/66 (39.4) 1 117/340 (34.4) 36/82 (43.9) 1 20–25 139/411 (33.8) 114/261 (43.7) 1.26 (0.86–1.85) 87/361 (24.1) 87/298 (29.2) 0.60 (0.43–0.84) 26þ 87/270 (32.2) 132/357 (37.0) 0.91(0.59–1.40) 40/214 (22.9) 114/343 (33.2) 0.56 (0.37–0.86) Marital status Never married 312/964 (32.4) 242/629 (38.5) 1 240/880 (27.3) 214/658 (32.5) 1 Ever married 15/38 (39.5) 30/55 (54.6) 1.50 (0.73–3.09) 13/35 (37.1) 23/65 (35.4) 1.81 (1.04–3.14) Trial arm Control 173/473 (36/6) 142/347 (40.9) 1 110/422 (26.1) 107/373 (28.7) 1 Intervention 154/529 (29.1) 130/337 (38.6) 0.64 (0.47–0.88) 143/443 (29.0) 130/371 (37.1) 1.25 (0.94–1.66) Ever had a child No – – – 122/365 (33.4) 79/18 (42.0) 1 Yes – – – 127/543 (23.4) 158/532 (29.7) 0.72 (0.53–0.99) Household wealth Very poor 100/291 (34.4) 79/203 (38.9) 1 67/274 (24.5) 65/223 (29.2) 1 Poor, but a bit better off 177/588 (30.1) 143/380 (37.6) 0.87 (0.61–1.24) 137/508 (27.4) 131/395 (33.2) 1.26 (0.91–1.75) Doing OK 49/119 (41.2) 50/100 (50.0) 1.20 (0.75–1.92) 45/125 (36.0) 37/94(39.4) 2.03 (1.29–3.20) Migrancy status Non-migrant 251/807 (31.1) 202/531 (38.0) 1 215/794 (27.1) 178/583 (30.5) 1 Migrant 76/195 (39.0) 58/119 (48.7) 1.46 (0.98–2.19) 38/120 (31.7) 32/84 (38.1) 1.48 (0.98–2.23) Educational attainment at follow-up None/primary only 39/143 (23.9) 24/76 (27.9) 1 31/134 (23.1) 26/90 (28.9) 1 Attended secondary 178/563 (31.6) 160/418 (38.3) 2.06 (1.73–4.90) 150/583 (25.7) 152/477 (31.9) 1.31 (0.83–2.05) Completed secondary 110/276 (39.9) 88/180 (48.9) 2.91 (1.73–4.90) 72/197 (36.6) 57/156 (37.8) 2.25 (1.34–3.78) Frequency of sexual intercourse in past year More than 5 time 194/639 (30.4) 179/522 (34.3) 1 149/632 (23.6) 172/554 (31.1) 1 Five or fewer times 133/363 (36.6) 93/162 (57.4) 2.20 (1.63–2.97) 104/283 (36.8) 65/169 (38.5) 1.45 (1.11–1.89) aOR, Adjusted odds ratio comparing socioeconomic groupings across both timepoints, adjusted for interview date, age, marital status, village pair, trial arm, frequency of sexual intercourse and, for women only, ever had a child. conducted. This is likely to have resulted in an These figures shed light on why HIV prevalence is not, underestimation of associations involving migrancy. uniquely among sub-Saharan African countries, decreasing Furthermore, the assessment of household wealth in in South Africa. Our estimates of incidence do not suggest developing countries is complex [18]. Our participatory a declining epidemic, being higher, for example, than approach had high internal consistency [13], but a low annual HIV incidence measures among adult men and level of correlation with an indicator based on multiple women in Uganda in the mid-1990s (1.72% per annum, assets (J. Hargreaves, L. Morison, J. Gear, J.D.H. Porter, 95% CI 1.38–2.16 and 1.69% per annum, 95% CI 1.38– M.B. Makhubele, J.C. Kim, et al., in preparation). 2.08, respectively) [19,20]. This is particularly worrying given that previous studies have suggested wide inter- This study provides direct measures of annual HIV provincial variation in adult HIV incidence within South incidence from a South African population, among men Africa (0.5–4.2%), with Limpopo, the province under (2.2%, 95% CI 1.5–3.0) and women (4.9%, 95% CI 4.0– study here, lying only at the midpoint of this range [9]. 5.9) aged 14–35 years at baseline. National estimates from cross-sectional research employing a detuned enzyme- Furthermore, our research suggests that although linked immunosorbent assay that detects infections in the condom use has increased over time, young people, if past 180 days have previously estimated HIV incidence anything, may be initiating sex earlier and the proportion among 15–24 year olds at 0.8% per annum for men reporting multiple partners has, if anything, increased. (compared with 2.0% for this age group in this study) and These data confirm findings from recent cross-sectional 6.5% for women (compared with 6.0%), with the overall studies in South Africa [9,10,21], and stand in contrast estimate for Limpopo province among 15–49 year olds at to the experience of Uganda [20], Kenya [22] and 2.4% per annum [9]. Although not always directly Zimbabwe [23,24], where reductions in HIV prevalence comparable, our study confirms the high incidence of have been accompanied by delays in the onset of first sex HIV infection with data from a cohort study. and reductions in partner numbers.
- 52. Explaining continued highHIV prevalence in South Africa Hargreaves et al. S47 Our results also draw attention to the socioeconomic groups. Individual-focused interventions appear on their patterning of HIV risk. Our finding of higher HIV own to have been insufficient to bring about population- incidence among the least educated women was not wide change and address barriers to risk-reduction among unexpected. Research suggests that up to the mid-1990s the most disadantaged groups. There is a strong case for prevalent HIV infection was often more common among the wider testing and implementation of structural individuals who were more mobile [25–27], had greater interventions that address the ‘upstream’ determinants education [2,28–30], or were from more wealthy of HIV infection [36]. households [31,32]. More recent studies among young people from Uganda and Zambia [2,33,34] have, however, suggested that whereas HIV prevalence has Acknowledgements fallen over time among the most educated, this is not so among the least educated. More surprisingly, we found no This study was a partnership between academic association between our measure of mobility and the risk institutions in South Africa (School of Public Health, of new HIV infection, although power to detect any University of the Witwatersrand) and the UK (London association was low for men (because of the relatively School of Hygiene and Tropical Medicine). The authors small number of seroconversions) and women (because of would like to thank the Contract Laboratory Services at low migration rates). the Johannesburg Hospital, particularly Dr Wendy Stevens, Grant Napier, Anusha Makuraj and Dr Gwynn With respect to sexual behaviour, migrant men reported Stevens, for assisting in the processing of laboratory greater numbers of sexual partners but also a greater use of specimens, and the support of Jackie Hills at UCB and condoms. Among women, lower levels of condom use Karin Botma at Omnimed for donating the collection were found among the poorest, those with of the least device and enzyme-linked immunosorbent assays. education and non-migrants. Migrants may be less subject to restrictive social norms and have access to larger sexual Sponsorship: The study received financial support networks. Male migrants may also be more likely to have from AngloAmerican Chairman’s Fund Educational Trust, AngloPlatinum, the Department for International greater personal income than non-migrants. Migrants Development (UK), the Ford Foundation, the Henry J. might also come into greater contact with condoms and Kaiser Family Foundation, HIVOS, the South African HIV-prevention materials as a result of their greater Department of Health and Welfare, and the Swedish mobility, especially to cities where such resources are International Development Agency. J.R.H. is sup- likely to be more commonly available. Underlying traits ported by an ESRC/MRC interdisciplinary fellowship. such as self–confidence might also make individuals simultaneously more likely to migrate, more attractive to Conflicts of interest: None. sexual partners, and more likely to become ‘early adopters’ of condoms. Whereas such issues require References further study, our data suggest that migrants should not be assumed to be engaged in high-risk behaviours 1. Fylkesnes K, Musonda RM, Sichone M, Ndhlovu Z, Tembo F, Monze M. Declining HIV prevalence and risk behaviours in although this does not preclude their being a key group in Zambia:evidence from surveillance and population based sur- the HIV epidemic dynamics [35]. The data confirm, veys. AIDS 2001; 15:907–916. however, that women experiencing socioeconomic 2. Kilian AH, Gregson S, Ndyanabangi B, Walusaga K, Kipp W, Sahlmuller G, et al. Reductions in risk behaviour provide the deprivation are among the most vulnerable to HIV in most consistent explanation for declining HIV-1 prevalence in this rural South African setting. Uganda. AIDS 1999; 13:391–398. 3. Moulaiteye GM, Whitworth JA, Ruberantwari A, Nakiyingi JS, Ojwiya A, Kamali A. Declining HIV-1 incidence and associated Effective HIV prevention strategies remain an urgent prevalence over 10 years in a rural population in south-west priority in South Africa. Strategies to date may have been Uganda: a cohort study. Lancet 2002; 360:41–46. partly effective in reducing risk among educated and 4. Gregson S, Zhuwau T, Anderson RM, Chandiwana SK. Is there evidence for behavior change in response to AIDS in rural mobile members of society. It is possible that changes will Zimbabwe? Soc Sci Med 1998; 46:321–330. emerge in all groups over time as safer-sex behaviour 5. Kamali A, Carpenter LM, Whitworth JA, Pool R, Ruberantwari A, diffuses, perhaps leading to reductions in HIV prevalence, Ojwiya A. Seven-year trends in HIV-1 infection rates, and changes in sexual behaviour, among adults in rural Uganda. as witnessed in other sub-Saharan African countries. This AIDS 2000; 14:427–434. may, however, be some way off unless there is a 6. Mekonnen Y, Sanders E, Aklilu M, Tsegaye A, Rinke De Wit TF, et al. Evidence of changes in sexual behaviours among male reassessment of HIV prevention strategies. Just as the factory workers in Ethiopia. AIDS 2003; 17:223–231. HIV epidemic in the United States has increasingly 7. UNAIDS. Report on the global HIV/AIDS epidemic. Geneva: become a disease of women and of racial and ethnic UNAIDS; 2006. 8. Department of Health. National HIV and syphillis antenatal sero- minorities [7], so the epidemic in Africa may also prevalence survey in South Africa 2003. Pretoria: Department of concentrate among the socially vulnerable. Our findings Health South Africa; 2003. suggested that the least educated women in this rural 9. Shisana O, Rehle T, Simbayi L, Parker W, Zuma K, Bhana A, et al. South African national HIV prevalence, HIV incidence, setting reported more partners, less use of condoms, and behaviour and communication survey 2005. Cape Town: HSRC experienced higher levels of new infection than other Press; 2005.
- 53. S48 AIDS 2007, Vol 21 (suppl 7) 10. Shisana O, Simbayi L. South African national HIV prevalence, 23. Gregson S, Garnett GP, Nyamukapa CA, Hallett TB, Lewis JJ, behavioral risks and mass media. Household survey 2002. Cape Mason PR, et al. HIV decline associated with behavior change Town: Human Sciences Research Council; 2002. in eastern Zimbabwe. Science 2006; 311:664–666. 11. Pronyk PM, Hargreaves JR, Kim JC, Morison LA, Phetla G, Watts 24. Mahomva A, Greby S, Dube S, Mugurungi O, Hargrove J, Rosen C, et al. Effect of a structural intervention for the prevention of D, et al. HIV prevalence and trends from data in Zimbabwe, intimate-partner violence and HIV in rural South Africa: a 1997-2004. Sex Transm Infect 2006; 82 (Suppl. 1):i42–i47. cluster randomised trial. Lancet 2006; 368:1973–1983. 25. Lurie MN, Williams BG, Zuma K, Mkaya-Mwamburi D, Garnett 12. Rose D, Charlton KE. Prevalence of household food poverty in G, Sturm AW, et al. The impact of migration on HIV-1 trans- south Africa: results from a large, nationally representative mission in South Africa: a study of migrant and nonmigrant survey. Pub Health Nutr 2003; 5:383–389. men and their partners. Sex Transm Dis 2003; 30:149–156. 13. Hargreaves JR, Morison LA, Gear J, Porter JDH, Makhubele MB, 26. Mnyika KS, Klepp KI, Kvale G, Nilssen S, Kissila PE, Ole-King’ori Kim JC, et al. Hearing the voices of the poor’’: assigning poverty N. Prevalence of HIV-1 infection in urban, semi-urban and lines on the basis of local perceptions of poverty; a quantitative rural areas in Arusha region, Tanzania. AIDS 1994; 8:1477– analysis of qualitative data from participatory wealth ranking 1481. in rural South Africa. World Dev 2007; 35:212–229. 27. Glynn JR, Ponnighaus J, Crampin AC, Sibande F, Sichali L, 14. Lestrade-Jefferis J. Housing, and household access to services Nkhosa P, et al. The development of the HIV epidemic in and facilities. In: Udjo E, editor. The people of South Africa Karonga District, Malawi. AIDS 2001; 15:2025–2029. population census 1996. Pretoria: Statistics South Africa; 2000. 28. Fylkesnes K, Masonda RM, Kasumba K, Ndhlovu Z, Mluanda F, 15. Collinson MA, Tollman SM, Kahn K, Clark SJ, Garenne M. Kaetano L, Chipaila CC. The HIV epidemic in Zambia: socio- Highly prevalent circular migration: households, mobility demographic prevalence patterns and indications of trends and economic status in rural South Africa. In: Tienda M, Findley among childbearing women. AIDS 1997; 11:339–345. SE, Tollman SM, Preston-Whyte E, editors. African migration in 29. Grosskurth H, Mosha F, Todd J, Senkoro K, Newell J, Klokke A, comparative perspective. Johannesburg: Wits University Press; et al. A community trial of the impact of improved sexually 2006. transmitted disease treatment on the HIV epidemic in rural 16. Catania JA, Gibson DR, Chitwood DD, Coates TJ. Methodolo- Tanzania: 2. Baseline survey results. AIDS 1995; 9:927– gical problems in AIDS behavioural research:influences on 934. measurement error and participation bias in studies of sexual behavior. Psychol Bull 1990; 108:339–362. 30. Smith J, Nalagoda F, Wawer MJ, Serwadda D, Sewankambo N, 17. Gregson S, Mushati P, White PJ, Mlilo M, Mundandi C, Konde-Lule J, et al. Education attainment as a predictor of HIV Nyamukapa C. Informal confidential voting interview methods risk in rural Uganda: Results from a population-based study. Int and temporal changes in reported sexual risk behaviour for J STD AIDS 1999; 10:452–459. HIV transmission in sub-Saharan Africa. Sex Transm Infect 31. Bloom SS, Urassa M, Isingo R, Ng’weshemi J, Boerma JT. 2004; 80 (Suppl. 2):ii36–ii42. Community effects on the risk of HIV infection in rural Tan- 18. Falkingham J, Namazie C. Measuring health and poverty: a zania. Sex Transm Infect 2002; 78:261–266. review of approaches to identifying the poor. London: DFID; 32. Chao A, Bulterys M, Musanganire F, Habimana P, Nawrocki P, 2002. Taylor E, et al. Risk factors associated with prevalent HIV-1 19. Hargreaves JR, Morison LA, Gear J, Porter JDH, Makhubele MB, infection among pregnant women in Rwanda. National Uni- Kim JC, et al. The assessment of household wealth in health versity of Rwanda–Johns Hopkins University AIDS Research studies in developing countries; a comparison of participatory Team. Int J Epidemiol 1994; 23:371–380. wealth ranking and survey techniques from rural South Africa. 33. de Walque D, Nakiyingi-Miiro JS, Busingye J, Whitworth JA. Emerging Themes in Epidemiology 2007, 4. Changing association between schooling levels and HIV-1 20. Stoneburner RL, Low-Beer D. Population-level HIV declines infection over 11 years in a rural population cohort in and behavioral risk avoidance in Uganda. Science 2004; south-west Uganda. Trop Med Int Health 2005; 10:993–1001. 304:714–718. 34. Michelo C, Sandoy IF, Fylkesnes K. Marked HIV prevalence 21. Pettifor AE, Rees HV, Kleinschmidt I, Steffenson AE, MacPhail C, declines in higher educated young people: evidence from Hlongwa-Madikizela L, et al. Young people’s sexual health in population-based surveys (1995–2003) in Zambia. AIDS South Africa: HIV prevalence and sexual behaviors from a 2006; 20:1031–1038. nationally representative household survey. AIDS 2005; 19: 35. Quinn TC. Population migration and the spread of types 1 and 2 1525–1534. human immunodeficiency viruses. Proc Natl Acad Sci 1994; 22. Cheluget B, Baltazar G, Orege P, Ibrahim M, Marum LH, Stover 91:2407–2414. J. Evidence for population level declines in adult HIV preva- 36. Blakenship KM, Bray SJ, Merson MH. Structural interventions lence in Kenya. Sex Transm Infect 2006; 82 (Suppl. 1):i21–i26. in public health. AIDS 2000; 14 (Suppl. 1):S11–S21.
- 54. Household and communityincome, economic shocks and risky sexual behavior of young adults: evidence from the Cape Area Panel Study 2002 and 2005 Taryn Dinkelmana, David Lama and Murray Leibbrandtb Objective: To describe recent trends in adolescent sexual behavior in Cape Town, South Africa, and to determine whether household and community poverty and negative economic shocks predict risky sexual behavior. Data: Matched survey data on 2993 African and coloured youth from the Cape Area Panel Study 2002 and 2005. Main outcome measures: Sexual debut, multiple sexual partners in past year, condom use at last sex, measured in 2002 and 2005. Methods: We tested for changes over time in reported sexual behavior and estimate multivariate probit models to measure the association between 2002 individual, household and community characteristics and 2005 sexual behavior. Results: There was a statistically significant increase in condom use and a decrease in the incidence of multiple sexual partners between 2002 and 2005 for women aged 17–22 years. Women in households with 10% higher income were 0.53% less likely to debut sexually by 2005; men in communities with a 10% higher poverty rate were 5% less likely to report condom use at last sex. Negative economic shocks are associated with a 0.04% increase in the probability of multiple partnerships for women. Education is positively correlated with sexual debut for women and with multiple partnerships for both sexes. Conclusion: Trends in sexual behavior between 2002 and 2005 indicate significant shifts towards safer practices. There is little evidence of a relationship between negative economic shocks, household and community poverty, and risky behavior. We hypoth- esize that the unexpected positive relationship between education and sexual debut may be driven by peer effects in schools with substantial age mixing. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins AIDS 2007, 21 (suppl 7):S49–S56 Keywords: HIV, adolescence, economic resources, sexual debut, condom use, multiple partnerships, South Africa Introduction areas [2]. The mechanisms by which conditions of poverty may influence sexual risk-taking behavior and A recent United Nations publication states that ‘poverty thus the probability of contracting HIVand other sexually increases vulnerability to HIV/AIDS’ [1], although in a transmitted diseases are, however, complex and currently complex fashion; the HIV burden is concentrated in the not well understood. poorest regions of the world but not always among the poorest populations in these areas. HIV prevalence rates Researchers in public health and economics have are indeed highest among South African youth living in hypothesized and less frequently tried to measure the poor urban informal settlements compared with other channels from individual and community poverty to From the aDepartment of Economics and Population Studies Center, University of Michigan, Ann Arbor, Michigan, USA, and the b School of Economics, University of Cape Town, Cape Town, South Africa. Correspondence to Taryn Dinkelman, Population Studies Center, 426 Thompson Street, University of Michigan, Ann Arbor, MI 48106, USA. E-mail: tdinkelm@umich.edu ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S49
- 55. S50 AIDS 2007, Vol 21 (suppl 7) higher rates of sexual risk-taking behavior. Fenton [3], Young adults are a particularly interesting demographic Gersovitz [4] and Sunmola [5] argued that inadequate group as they represent healthy cohorts whose future information (often concentrated among the poor) acts as a behavior will influence the course of the HIV epidemic. barrier to adopting safer behaviors; Cohen et al. [6] Various researchers have shown that in high seropreva- argued that access to resources for safer sex may be costly; lence regions, a large proportion of new infections occur MacPhail and Campbell [7] and LeClerc-Madlala [8] during adolescence [8]. Survey data that match detailed argued that poverty directly induces sex work and individual sexual behavior measures to good measures of informal transactional sex relationships particularly for household and community level resources are rare; panel women, whereas Johnston and Way [9] provided data that enable us to see the evolution of sexual behaviors descriptive evidence of these correlations and Luke for young adults are even more unusual. We used new [10] and Luke [11] used survey data to quantify this panel data on adolescents (aged 14–22 years) in urban relationship. Dunkel et al. [12] used data from young, rural Cape Town, South Africa, to measure the extent to which South African men to show a strong positive correlation resources and shocks to household resources early on in between higher socioeconomic status and the probability their lives could predict variation in sexual behaviors. We of reporting transactional sex with casual partners. At a considered the following outcome measures that corre- community level, Zulu et al. [13] compared non-poor spond to the A–B–C of HIV prevention campaigns: non-slum residents with poor slum residents in Nairobi sexual debut, annual number of sexual partners and and found significantly higher probabilities of reporting condom use at last sex. early sexual debut, more sexual partners and a lack of condom use among the poor slum dwellers. Economic inequality may also operate to increase risk-taking within communities: LeClerc-Madlala [8] posited that the Data and methods growth of a black middle class with money has increased the emphasis on transactional sex in some South African Data communities. The Cape Area Panel Study (CAPS) is a representative longitudinal study of 4752 adolescents aged 14–22 years In many of those studies, it is difficult to isolate whether (in 2002) living in Cape Town, South Africa. The full economic resources matter directly for behavior or sample was first interviewed in 2002 and again in 2005. whether unobservable characteristics correlated with Most data are collected directly from the young adults. poverty are driving factors. In addition, as the experience We use data from the household module, basic of poverty is likely to have persistent effects on behavior demographic data and detailed information about sexual over time, it is hard to distinguish whether current or relationships captured in both waves. We do not use data long-term resource deprivation matters for behavior. To on what young people know about HIV and AIDS. measure the direct effect of economic resources on Anderson and Beutel [14] reported that levels of HIV and behavior convincingly, we would want to assign these AIDS knowledge were very high in the 2002 CAPS data. resources randomly to households and observe the impact These panel data allow us to look at whether sexual on behaviors. Approaching this research design with behavior is changing over time as well as how current observational data is challenging. behaviors are related to a range of household level variables measured earlier in the young adult’s life. In this paper, we investigated whether household and community incomes and negative economic shocks In order to generate an approximately equal sample of predict risky behaviors of young adults. Focusing on African and coloured individuals, African youth were young adults who are for the most part not yet oversampled. See Lam et al. [15] for details of sampling working and who are just transitioning into sex methodology, initial non-response and attrition. Com- allowed us to isolate the relationship between household pleted interviews for 2151 Africans, 1980 coloureds and level and community resources and behavior. In the 621 whites and other races were captured in 2002. Once absence of the random assignment of income to weights adjusting for survey design and wave 1 non- households or communities, we used economic shocks response were applied, Africans represented 15% of this to capture one source of unexpected variation in wave 1 sample, coloureds represented 59% and the household resources. Although this research design does remaining races constituted 26% of the sample. In the not identify the causal effect of economic deprivation 2005 wave 3324 of the initial 4752 sample were re- on behavior, we advanced some way towards an interviewed, of which 2993 were African and coloured understanding of the relationship between economic individuals (27.8% attrition rate). resources and the risky sexual behavior of young adults. Surprisingly, we found little evidence that differences in As initial non-response and attrition between waves were household or community income or differences in very high for the small sample of white youth, we economic shocks are correlated with more risk-taking excluded them from our analysis. Fifty-three per cent of behavior. white individuals were successfully followed in 2005,
- 56. Income and riskysexual behavior Dinkelman et al. S51 whereas attrition among coloured and African subsamples entire sample, we included a full set of age dummies in the was substantially lower (21 and 36%, respectively). All of probit models to take out mean differences across age our reported results are weighted with sampling weights groups in test performance. Therefore, if 22-year-old correcting for sample design and first wave non-response. youth scored consistently above average on the test simply Weighting for attrition between 2002 and 2005 does not because they were older and had more schooling, this change any results substantively (results not reported). effect was absorbed in the 22-year-old dummy. Our final matched subsample consists of 1410 African youth and 1583 coloured youth. Methods We compared the proportion of each sex, race and age Variables group reporting each type of behavior in each wave for For each individual in our analysis subsample, we used ages 17–22 years. The change in these proportions sexual behavior information provided by the respondent between 2002 and 2005 gave us some insight into overall in 2002 and 2005. To examine changes in average trends in behaviors. To measure the association between behavior over time, we investigated three reported individual-level demographic data, previous income behaviors for the group aged 17–22 years: whether the shocks, household and community resources and current young adult has ever had sex, whether the young adult sexual behavior, we estimated probit models for each of used a condom at last sex and whether the young adult the three binary outcome variables separately for women had more than one sexual partner in the 12 months before and men. The results are reported as the marginal change each survey. in the probability of a particular behavior associated with a unit increase in each explanatory variable. We are cautious about the reliability of reported sexual behavior data. Misreporting is more likely when In the probit models we restricted the sample to ages questions are more sensitive [16–19]. CAPS questions 14–18 years in 2002 for the sexual debut and condom and survey protocols were carefully constructed to try to use outcomes, but included all ages 14–22 years in 2002 minimize the biases in these sensitive questions. In both when modeling multiple partners. There are two reasons years, respondents were questioned without the presence for this. First, a large proportion of those aged 19 years of any other family members as far as possible. For the and older had already sexually debuted so there was little 2005 survey, respondents could choose to fill out their variation contributed by those individuals. Second, responses directly regarding each of their 10 most recent although multiple partnerships reflected relatively unsafe partnerships instead of having the interviewer fill in the behavior at all ages, not using a condom at last sex was not information. Fourteen per cent of applicants chose to unambiguously risky, especially in cases in which older self-report. Comparing those who did with those who individuals were married or in longer-term monogamous did not respond themselves, there was no systematic relationships. The results are presented as marginal effects difference in the number of sex partners reported in 2005. and robust standard errors are clustered at the household level because up to three young adults were interviewed Restricting to the same set of ages (17–22 years) in 2002 per household. and 2005 allowed us to compare average behavior for this group over time. To examine how 2002 individual, household and community-level variables were corre- lated with behavior, we investigated these three sexual Results behavior variables measured in 2005. The variables used to predict individual behaviors within the probit model Summary statistics included: age in 2002; sex; education; race; literacy and In Table 1, we present summary statistics separately by numeracy test scores; per capita household income in race to highlight the vast differences in living environ- 2002; the presence of parents at home in 2002; and the ments of African and coloured youth. Except for age, all proportion of households in the community below the of these differences are statistically significant across race poverty line in the 2001 census. We also used information groups. Both groups were disadvantaged under apartheid, on negative economic shocks experienced at the house- but coloured individuals were generally able to access hold level between 2002 and 2005. A negative shock is better educational and work opportunities in Cape Town defined as having occurred if the household experienced than Africans. Mean schooling was approximately ninth a death, job loss, loss of a grant or loss of support from grade, although Africans had on average half a year less outside the household, and if the household respondent schooling than coloureds. Africans also exhibited poorer reported that the shock had a moderate to severe financial performance on the literacy and numeracy test. Coloured impact on the household. youth were significantly more likely to live with their biological mothers (82% compared with 64% for The same literacy and numeracy test was administered to Africans) and fathers (54% compared with 35% for each young adult in 2002 regardless of age or education Africans). Coloured households had a higher mean log level. Although the test is not age-appropriate for the per capita income compared with African households.
- 57. S52 AIDS 2007, Vol 21 (suppl 7) Table 1. Descriptive statistics of matched Africans and coloured young adults. African Coloured Full sample Variables measured in 2002 Proportion female 0.54 0.51 0.52 Age in 2002 17.93 (2.51) 17.72 (2.45) 17.79 (2.47) Years of schooling 8.89 (2.20) 9.44 (2.06) 9.27 (2.12) Resides with biological mother 0.64 0.82 0.76 Resides with biological father 0.35 0.54 0.48 Literacy and numeracy test z-score À0.46 (0.87) 0.18 (0.83) À0.02 (0.89) Log per capita household income 5.60 (0.95) 6.57 (0.87) 6.28 (1.00) Household income imputed 0.04 0.05 0.05 Community poverty rate (2001 census) 0.45 0.17 0.25 Variables measured between 2002 and 2005 Negative economic shock 0.24 0.16 0.18 No. young adults 1410 1583 2993 No. households 999 1184 2183 The sample consists of all African and coloured young adults interviewed in 2002 and again in 2005. All statistics (means and standard deviation in parentheses) are weighted by the individual youth weight that corrects for sample design and non-response in the first wave. Household shock variable is ¼ 1 if any adult in the household died between 2002 and 2005, or if the houshold experienced a moderate or large financial shock as a result of any of the following between 2002 and 2005: job loss; loss of a grant; loss of financial support from outside of the house or other reason. All variables are statistically significantly different (P ¼ 0.05) across race groups, except for age, which is not different across the two groups. On average, youth lived in communities in which 25% of Table 2 shows the percentage of each race, sex and 2-year households were below the 2001 poverty line, but this age group reporting each of three sexual behaviors in percentage was substantially higher for Africans (45%). 2002 and 2005. Note that for Table 2 we did not follow Eighteen per cent of these young adults lived in the same individuals over time, but simply looked at households experiencing a serious economic shock the cross-section of respondents in a given age group in between 2002 and 2005. Whereas shocks were observed each wave. As the original sample of 14–22 year olds was in households in all income quintiles, they were 17–25 years of age in 2005, we looked at the ages from somewhat more prevalent in the poorest quintiles (results 17 to 22 years, the ages that overlap in the two waves. The not shown). Almost one in five African youth lived in a first panel shows the percentage reporting having ever had household that experienced an economic shock between sex at the time of the 2002 or 2005 interview. The overall 2002 and 2005. Across all variables, African youth lived in pattern is an increase in sexual activity between 2002 and significantly poorer households and communities. 2005. Across all groups, young adults aged 17 to 22 are Table 2. Percentage of Cape Area Panel Study respondents in three categories of sexual behavior, 2002 and 2005. African woman African man Coloured woman Coloured man Age (years) 2002 2005 2002 2005 2002 2005 2002 2005 A: Ever had sex 17–18 60.4 71.6MM 59.2 64.6 22.2 30.7MM 35.9 37.4 19–20 84.4 87.8 79.5 87.8M 52.1 56.3 61.0 68.8 21–22 88.0 96.4MMM 86.7 88.1 70.7 63.4 67.5 77.2M All 76.5 86.8MMM 74.9 80.1M 44.7 51.1MM 51.7 61.8MMM N 511 529 418 411 524 594 451 534 B: Condom use at last sex 17–18 58.9 77.3MMM 75.3 79.9 30.5 33.7 74.3 82.8 19–20 51.5 70.5MMM 76.5 85.0 28.1 40.7M 74.9 61.9MM 21–22 41.4 67.3MMM 68.8 87.5MMM 17.9 30.5MM 55.9 63.3 All 50.4 71.7MMM 72.0 84.7MMM 24.9 33.5MM 68.2 65.3 N 371 420 296 293 221 272 214 294 C: Multiple sex partners in past year 17–18 23.1 12.7M 56.8 39.6MM 16.3 4.7M 48 25.1MM 19–20 20.5 7.2MMM 52.5 41.9 4.7 2.4 48 26.9MMM 21–22 22.3 12.2MM 63.3 31.9MMM 8.9 5.6 43.2 19.2MMM All 21.8 10.8MMM 56.7 36.5MMM 7.7 3.7M 47.8 24.5MMM N 339 406 270 277 194 246 186 253 Asterisks indicate significance level for test of equality between 2002 and 2005 percentages: M0.1; MM0.05; MMM0.01. Sample includes all African and coloured respondents in 2002 who were followed in 2005. Ever had sex ¼ 1 if young person reported ever having had sex in that year’s interviews. Condom use at last sex -1 if young person reported using a condom at last sex. Multiple partners in past year ¼ 1 if young person reported more than one sex partner in the 12 months before the survey. Only respondents who have ever had sex are included in the definitions of condom use and multiple sex partners.
- 58. Income and riskysexual behavior Dinkelman et al. S53 more likely to report sexual debut in the later period. reported lower rates of condom use than African women African girls aged 17 to 18 report the largest increases in at every age, and there is less evidence of an increase in sexual debut: 60% of this group reported ever having sex condom use over time for young coloured women. in 2002, compared with 72% in 2005. At the same time, Higher rates of marriage at young ages cannot explain this parts B and C of Table 2 indicate significant increases in significantly lower rate of condom use among coloured safer sex practices. Condom use at last sex for male and women, because only 4–5% of African and coloured female Africans is significantly higher in 2005 than in women were married by ages 17–22 years in 2002. In 2002 across all age groups, except young men aged 17 and 2005, only 3.4% of coloured women and 1.6% of African 18 years. For African young women, these increases are women aged 17–22 years were married. Part C of Table 2 very large, approximately 20% or higher for each age shows the changing prevalence of multiple sexual partners group. There is also some evidence of increased condom by age, race and sex. use among coloured young women, although both the initial level and the increase between waves is There is a fairly consistent pattern of decreasing unsafe smaller for coloured young women than for African sexual behavior for all groups. Among African young young women. women, 22% of 17–22 year olds reported having multiple sexual partners (not necessarily concurrently) in the past The changes in condom use between 2002 and 2005 are 12 months in 2002, compared with 11% in 2005. For shown graphically in Figure 1, using single years of age African young men, this decrease was even larger: 55% of from 17 to 22. Reported condom use by African women African young men reported multiple partners in 2002, increased at every age between 2002 and 2005. The falling to 37% in 2005. Coloured young men also showed proportion of 17-year-old African women who reported a decline in the incidence of multiple partnerships across using a condom at last sex rose from 50% in 2002 to 82% all age groups. in 2005. In contrast, coloured women consistently Probit regressions Africans Table 3 presents probit results analysing the determinants 100% of sexual debut between 2002 and 2005, condom use at 90% 82% most recent sex in 2005, and multiple partners in the past 80% 75% 74% year in 2005. For condom use and multiple partner 67% 70% 70% 65% 65% outcomes, we included a dummy variable for whether 60% 57% sexual debut had occurred by 2002, to capture differences 50% 50% 45% 43% in behavior between those who made an early versus a late 41% 40% sexual debut. 30% 20% Three main points emerge from these results. First, 10% African and coloured behavior is statistically significantly 0% different on all outcomes except for male sexual debut in 17 18 19 20 21 22 2005. These differences are large but do not consistently Coloureds reflect more risky behavior on the part of African youth. 100% Compared with coloured women, African women had a 90% 33.6% higher probability of sexual debut and an 8.4% 80% higher prevalence of multiple partners, controlling for the 70% other variables included in the probits. At the same time, 60% African women had a 52.6% higher probability of using a 50% 46% condom at last sex. As the sample changed across outcome 40% 40% 37% 35% 32% 36% variables, the group of girls in column (1) is a subset of the 31% 30% 24% 23% girls in column (3). Second, higher levels of education are 20% 20% 15% 13% associated with more unsafe behavior for women and 10% men. Controlling for age and the other variables in 0% Table 3, those with more schooling were more likely to 17 18 19 20 21 22 have had sex and more likely to report multiple partners. Age Higher scores on the 2002 literacy and numeracy test Fig. 1. Percentage of women reporting condom use at last were associated with a statistically significant lower sex, Cape Area Panel Study 2002 and 2005. Only girls who probability of sexual debut and a lower likelihood of reported having had sex before are included in this sample. multiple partnerships for both sexes. A young adult Percentages are weighted to correct for sampling design and with one standard deviation higher score on the test had a wave 1 non-response. See Table 2 for overall sample sizes. 5% lower probability of sexual debut between 2002 2002; 2005. and 2005.
- 59. S54 AIDS 2007, Vol 21 (suppl 7) Table 3. Probit regressions for reports of three sexual behavior variables in 2005 – marginal effects. Women Men (1) (2) (3) (4) (5) (6) Variables 2002 Ever had sex Condom at last sex Multiple partners Ever had sex Condom at last sex Multiple partners African 0.336MMM (0.061) 0.529MMM (0.074) 0.0849MMM (0.024) 0.097 (0.080) 0.272MMM (0.061) 0.139MM (0.059) Age 15 years 0.235MMM (0.063) À0.124 (0.11) 0.041 (0.057) 0.152MM (0.065) 0.040 (0.083) À0.156MMM (0.057) Age 16 years 0.233MMM (0.070) À0.123 (0.11) 0.000 (0.036) 0.349MMM (0.064) À0.018 (0.088) À0.072 (0.072) Age 17 years 0.349MMM (0.072) À0.070 (0.12) 0.012 (0.045) 0.379MMM (0.066) À0.163M (0.097) À0.075 (0.073) Age 18 years 0.167M (0.096) À0.159 (0.13) À0.004 (0.036) 0.373MMM (0.073) À0.078 (0.10) À0.197MMM (0.053) Years of education 0.0520MM (0.023) 0.016 (0.024) 0.00838MM (0.0041) 0.011 (0.022) 0.024 (0.018) 0.0311MMM (0.011) Mother at home À0.043 (0.059) À0.014 (0.062) 0.004 (0.014) À0.036 (0.069) 0.047 (0.061) 0.012 (0.040) Father at home À0.062 (0.049) 0.041 (0.056) 0.003 (0.014) À0.019 (0.052) À0.0868M (0.045) À0.006 (0.037) Test score À0.0650M (0.034) 0.0877MM (0.035) À0.011 (0.0080) À0.0590M (0.033) 0.029 (0.029) À0.019 (0.022) Log per capita household income À0.0535M (0.029) 0.049 (0.031) À0.004 (0.0066) À0.031 (0.031) À0.008 (0.027) À0.009 (0.022) Household shock (2002–2005) 0.076 (0.057) 0.004 (0.062) 0.0377M (0.019) 0.048 (0.062) 0.011 (0.055) 0.065 (0.048) Community poverty rate À0.104 (0.22) À0.375 (0.29) À0.088 (0.056) 0.317 (0.26) À0.459M (0.24) 0.002 (0.17) Ever had sex by 2002 À0.022 (0.058) 0.000 (0.014) À0.016 (0.048) 0.036 (0.042) Age 19 years 0.037 (0.055) À0.172MMM (0.062) Age 20 years À0.026 (0.025) À0.170MMM (0.064) Age 21 years À0.007 (0.035) À0.230MMM (0.048) Age 22 years À0.0358M (0.019) À0.204MMM (0.055) Observations 686 532 952 545 450 760 Mean of outcome variable 0.47 0.51 0.06 0.49 0.74 0.28 Pseudo R-squared 0.14 0.13 0.10 0.10 0.07 0.05 Robust standard errors in brackets, clustered at the household level for multiple observations in the same household: P < 0.01MMM; P < 0.05MM; P < 0.1M. Results are marginal effects from probit models, evaluated at sample means. Outcome variables are binary dependent variables measured in 2005: Ever had sex ¼ 1 if respondent had made sexual debut between 2002 and 2005 (conditional on not having had sex by 2002); Condom use at last sex ¼ 1 if young person reported using a condom at last sex; Multiple partners ¼ 1 if young adult reported more than one sex partner in the 12 months before the 2005 survey. All independent variables measured in 2002 except for household shock, which is measured between 2002 and 2005. Sample for column 1 and 4 is respondents aged 14–18 years in 2002 who reported not having sex in 2002 interview. Sample for columns 3 and 6 is the full sample of 14–22-year-old respondents in wave 1 who had ever had sex by 2005. Coefficients that are statistically significantly different across male and female regressions at 5% level: African (for outcome Ever had sex), Father at home (for outcome Condom use at last sex), and age dummies 15, 19 and 21 (for outcome Multiple partners in past year).
- 60. Income and riskysexual behavior Dinkelman et al. S55 A third point relates to the set of economic variables. Per Taking our first two findings together, it appears that at capita household income has a small and statistically least in Cape Town there are significant increases in significant negative correlation with the probability of condom use and decreases in the number of sexual female sexual debut. A 10% increase in 2002 income was partners. Changes that cannot be explained by individual associated with a 0.6% decline in the probability of sexual behavioral change appear to be taking place. debut between 2002 and 2005. The estimated marginal effect of income on sexual debut was also negative for The third main finding relates to the role of education in young men but not statistically significant at conventional predicting sexual risk behaviors. We did not see a levels. Young men were less likely to report condom use at protective impact of grade attainment per se, although we last sex if they lived in poorer communities: for a 10% did find that test scores were positively correlated with increase in community poverty rate, this was a 5% safer sexual behaviors. Surprisingly, we found a significant reduction in the probability of condom use at last sex. positive association of schooling with sexual debut for Young women were 0.04% more likely to report multiple women and with multiple sexual partners for both men partners if they lived in a household experiencing an and women, controlling for age, household income, and economic shock. We tested the joint significance of all of other variables. One interpretation of our results is that the economic variables (household per capita income, the test score variable captures some of the differences in household shock and community poverty) and could not knowledge or ability that education usually measures, reject the possibility that the coefficients were jointly zero therefore, youth with higher numeracy and literacy skills in each regression. are less likely to report risky behaviors. We speculate that the unexpected positive association between schooling and risky behaviors may be a result of the impact of peers within the school system. There is a great deal of grade Discussion repetition in South Africa [20], with a wide mix of ages in any given grade. A 17 year old in grade 11 interacts with a Three main findings emerge from our analysis of the much more sexually active group of peers than a 17 year panel data of 2993 Cape Town youth. First, for young old in grade 8. Further research will be required to people aged 17–22 years, we documented large and understand why years of education may not have the statistically significant increases in the probability of sexual protective effect that is usually hypothesized. debut for women of both races, increased condom use at last sex for African women, African men, and coloured men, as well as significant reductions in the reporting of Acknowledgements multiple sexual partnerships. Changes in household or community-level economic resources are unlikely to This paper was written for the workshop ‘A Symposium explain these behavioral changes, both because we for investigating the empirical evidence for under- estimate relatively small effects of income on sexual standing vulnerability and the associations between behavior and because there are only small improvements poverty, HIV infection and AIDS impact’. The authors in economic conditions over this period. It is also unlikely would like to thank USAID and the Health Economics that these differences arise from a change in social and HIV/AIDS Research Division (HEARD) for desirability pressure towards answering sensitive questions financial support during the preparation of the paper in particular ways; these young adults are reporting and attendance at the workshop. The data used in this increases in risky behavior (sexual debut) at the same time paper are publicly accessible at www.caps.uct.ac.za. as increases in protective behaviors (more condom use, fewer multiple partnerships). Sponsorship: This work was supported by the US National Institute of Child Health and Human Devel- Our main interest in this paper related to whether opment (R01HD39788 and R01HD045581), the household or community poverty variables could predict Fogarty International Center of the US National risky behavior of young adults, and, in particular, whether Institutes of Health (D43TW000657), and the Andrew sexual behavior is affected by unexpected income shocks. W. Mellon Foundation. After controlling for detailed individual and family Conflicts of interest: None. background variables, however, we found that little of the variation in sexual behavior in 2005 was predicted by economic variables. When household income and References economic shocks are significant, their marginal effects are relatively small compared with other variables. Our 1. United Nations. Population, development and HIV/AIDS with second finding is thus that for the sample of young adults particular emphasis on poverty: the concise report. 2005. New in urban Cape Town, there is little evidence that York: United Nations; 2005. 2. Human Sciences Research Council. South African national HIV community or household-level income or income shocks prevalence, HIV incidence, behavior and communication sur- are the main factors influencing risky behavior. vey, 2005. Cape Town: HSRC Press; 2005.
- 61. S56 AIDS 2007, Vol 21 (suppl 7) 3. Fenton L. Preventing HIV/AIDS through poverty reduction: 12. Dunkle K, Jewkes R, Mzikazi N, Jama N, Levin J, Sikweyiya Y, the only sustainable solution. Lancet 2004; 364:1186– Koss M. Transactional sex with casual and main partners 1187. among young South African men in the rural Eastern Cape: 4. Gersovitz M. The HIV epidemic in four African countries seen prevalence, predictors and associations with gender-based through the Demographic and Health Surveys. J Afr Economies violence. Soc Sci Med 2007; 65:1235–1248. 2005; 4:191–246. 13. Zulu EM, Nii-Amoo Dodoo F, Chika-Ezeh A. Sexual risk-taking in the slums of Nairobi, Kenya 1993–98. Popul Stud 2002; 5. Sunmola AM. Sexual practices, barriers to condom use and its 56:311–323. consistent use among long distance truck drivers in Nigeria. 14. Anderson KG, Beutel AM. HIV/AIDS prevention knowledge AIDS Care 2005; 17:208–221. among youth in Cape Town, South Africa. J Soc Sci 2007; 6. Cohen D, Scribner R, Bedimo R, Farley TA. Cost as a barrier to 3:143–150. condom use: the evidence for condom subsidies in the United 15. Lam D, Seekings J, Sparks M. The Cape Area Panel Study: States. Am J Pub Health 1999; 89:567–568. overview and technical documentation for waves 1-2-3. Uni- 7. MacPhail C, Campbell C. ‘I think condoms are good but, Aai, versity of Cape Town, Cape Town; December 2006. I hate those things’: condom use among adolescents and young 16. Smith T. Discrepancies between men and women in reporting people in a South African township. Soc Sci Med 2001; number of sexual partners: a summary from four countries. Soc 52:1613–1627. Biol 1992; 39:203–211. 8. LeClerc-Madala S. Youth, HIV/AIDS and the importance of 17. Nnko S, Boerma JT, Urassa M, Mwaluko G, Zaba B. Secretive sexual culture and context. Soc Dynam 2002; 28:20–41. females or swaggering males? An assessment of the quality of sexual partnership reporting in rural Tanzania. Soc Sci Med 9. Johnston K, Way A. Risk factors for HIV infection in a national 2004; 59:299–310. adult population: evidence from the 2003 Kenya Demographic 18. Becker S, Hossain M, Thompson E. Disagreement in spousal Health Survey. J Acquir Immune Defic Syndr 2006; 42:627– reports of current contraceptive use in sub-Saharan Africa. J 637. Biosoc Sci 2006; 38:779–796. 10. Luke N. Age and economic asymmetries in the sexual relation- 19. Tourangeau R, Smith T. Asking sensitive questions: the impact ships of adolescent girls in sub-Saharan Africa. Stud Family of data collection mode, question format and question context. Plann 2003; 34:67–86. Pub Opin Quart 1996; 60:75–304. 11. Luke N. Exchange and condom use in informal sexual relation- 20. Anderson KG, Case A, Lam D. Causes and consequences of ships in urban Kenya. Econ Dev Cult Change 2006; 54:319– schooling outcomes in South Africa: evidence from survey 344. data. Soc Dynam 2001; 27:37–59.
- 62. HIV incidence andpoverty in Manicaland, Zimbabwe: is HIV becoming a disease of the poor? Ben Lopmana, James Lewisb, Constance Nyamukapaa,c, Phyllis Mushatic, Steven Chandiwanac,d,ä and Simon Gregsona,c Background: In Zimbabwe, socioeconomic development has a complicated and changeable relationship with HIV infection. Longitudinal data are needed to disen- tangle the cyclical effects of poverty and HIV as well as to separate historical patterns from contemporary trends of infection. Methods: We analysed a large population-based cohort in eastern Zimbabwe. The wealth index was measured at baseline on the basis of household asset ownership. The associations of the wealth index with HIV incidence and mortality, sexual risk behaviour, and sexual mixing patterns were analysed. Results: The largest decreases in HIV prevalence were in the top third of the wealth index distribution (tercile) in both men at 25% and women at 21%. In men, HIV incidence was significantly lower in the top wealth index tercile (15.4 per 1000 person-years) compared with the lowest tercile (27.4 per 1000 person-years), especially among young men. Mortality rates were significantly lower in both men and women of higher wealth index. Men of higher wealth index reported more sexual partners, but were also more likely to use condoms. Better-off women reported fewer partners and were less likely to engage in transactional sex. Partnership data suggests increasing like-with-like mixing in higher wealth groups resulting in the reduced probability of serodiscordant couples. Conclusion: HIV incidence and mortality, and perhaps sexual risk, are lower in higher socioeconomic groups. Reduced vulnerability to infection, led by the relatively well off, is a positive trend, but in the absence of analogous developments in vulnerable groups, HIV threatens to become a disease of the poor. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins AIDS 2007, 21 (suppl 7):S57–S66 Keywords: Africa, Zimbabwe, AIDS, HIV, poverty, socio-economic development Introduction decline in HIV prevalence in Zimbabwe [3]. Mortality rates are high [4], as a result of high HIV incidence in the Similar to other countries in southern Africa, the HIV past, but this decrease cannot be explained by mortality epidemic in Zimbabwe has a precarious relationship with alone [5]. Sexual risk behaviour is also changing; condom socioeconomic development [1]. Zimbabwe has one of distribution has increased, young people are delaying the more developed infrastructures in sub-Saharan Africa, their sexual debut, and there has been a reduction in the with widespread access to education and the highest adult numbers of casual partnerships [3,6]. literacy in the region [2]. Zimbabwe is also experiencing one of the largest national epidemics. HIV prevalence in Some investigators have suggested that as the HIV the adult population was 20.1% in 2005, down from epidemic progresses, risk would shift from the wealthier 22.1% in 2003. There are two contributory factors to the (who, as a result of their relative wealth, are part of a larger From the aDepartment of Infectious Disease Epidemiology, Imperial College, London, UK, the bLondon School of Hygiene and Tropical Medicine, London, UK, the cBiomedical Research and Training Institute, Harare, Zimbabwe, and the dFaculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa. Correspondence to Ben Lopman, Department of Infectious Disease Epidemiology, Imperial College London, St Mary’s Campus, Norfolk Place, London W2 1PG, UK. Tel: +44 020 7594 3290; fax: +44 020 7594 3282; e-mail: b.lopman@imperial.ac.uk ä Deceased. ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S57
- 63. S58 AIDS 2007, Vol 21 (suppl 7) sexual network) [7] to the poorer (who, because of their favourably with other community cohorts in rural lower educational attainment and social position, are less African settings [6]. Enumerators were notified of deaths empowered to change their sexual behaviour) [8,9]. It is by surviving household members or community infor- suspected that the HIV epidemic in Zimbabwe initially mants if the household dissolved completely. affected more mobile and more educated men as a result of their ability to attract sexual partners, but as early as At each round, after written informed consent was given, 1998/2000 risk was similar, or perhaps slightly lower for information on demographic, socioeconomic and sexual those with secondary education [10]. If this trend is behaviour data were collected through an interviewer-led realized, the HIV epidemic threatens to become an questionnaire [14]. Dried blood spots were collected for endemic disease of poverty in Zimbabwe. HIV serological testing for the purposes of research only. Testing was performed using a highly sensitive and The changing relationship between socioeconomics and specific antibody dipstick assay (> 99% for both) [15]. HIV must be seen in the context of sweeping macroeconomic changes in Zimbabwe. The Zimbab- Socioeconomic status wean economy has been in severe decline, with negative Individual wealth was measured on the basis of the asset growth since 1997 [11]. Over the period 1997–2005, ownership of the household of residence. Data were gross domestic product declined by more than 30%. In collected on household ownership of ‘fixed’ and ‘sellable’ 2003, annual inflation was approximately 250%, and this assets. Fixed assets include water supply, toilet facilities, has since accelerated to over 1000% per annum [11]. The electricity supply, housing structure and floor type. economic factors that partly underlie partnership Sellable assets included ownership of radio, television, formation [12], including behaviours ranging from sex bicycle, motorbike and automobile. Chi-squared tests work to marriage, are likely to be highly unstable, making demonstrated significant differences of all assets (except understanding the link between poverty and HIV automobile, which was owned by only 1.2% of house- extremely timely yet difficult to study. holds) between towns, estates and rural areas. A simple summed score asset ownership was created (see Justification of ‘summed score’ as measure of wealth Methods below). The binary and ordinal measures were each transformed to lie between 0 and 1. For example, bike Study population ownership conferred a score of 0 or 1, and type of floor The Manicaland HIV/STD Prevention Project is an conferred a value of 0 for natural floor (earth/sand/dung), ongoing population-based open cohort study. Full details 0.5 for rudimentary floor (e.g. planks/bamboo), or 1 for of the study can be found elsewhere [6,13]. In short, the finished floor (wood/cement). The 10 variables were study population were resident in small towns (two) added and expressed as a percentage. forestry, tea and coffee estates (four) and rural areas (six, including four subsistence farming and two roadside In order to augment the study power, a wealth index was trading centres) in the province of Manicaland in eastern created by splitting the summed score into three equal Zimbabwe. All local residents were enumerated in an groups (terciles) from the whole population. Preliminary initial household census (conducted between July 1998 analyses demonstrated that the distribution of the wealth and February 2000; referred to here as baseline), which index differed between towns and other areas; therefore, was repeated 3 years later in each site (referred to here as analyses were conducted separately for towns, estate and follow-up). Men aged 17–54 years and women aged 15– rural areas or were controlled for site type in 44 years were recruited into a cohort study of HIV multivariable regression. transmission. A maximum of one member of each marital group was selected for recruitment to the cohort, Analysis of HIV incidence and mortality members of multiple married couples and all unmarried Seroconversions, defined as individuals who tested individuals from a single household were eligible. negative at baseline and positive at follow-up, were assumed to have been infected halfway through the Totals of 8376 and 7102 of the households identified in period of observation. Poisson regression models were the survey areas at baseline and at follow-up, respectively, fitted with incident infection as the outcome and wealth were enumerated. Male and female participation rates in index tercile the explanatory variable. Models were the individual cohort study survey were 78% (4320/5561) controlled for age and site type and are presented and 80% (5134/6419) at baseline and 77% (3047/3958) separately for men and women. and 80% (3972/4936) at follow-up, respectively. Approximately 3 years after baseline 54% (2242/4142) Mortality rates were modelled using the same approach. of the men and 66% (3265/4922) of the women who Only deaths of participants who were HIV positive at were not known to have died were re-interviewed at baseline were included, with deaths of HIV-negative follow-up. This loss to follow-up rate compared participants omitted. This was done for two reasons. First,
- 64. HIV incidence andpoverty in Manicaland, Zimbabwe Lopman et al. S59 deaths from non-AIDS causes are likely to differ by wealth baseline (2.7%) and 215 at follow-up (2.6%), who were index and would therefore obscure the relationship excluded from analyses. At follow-up, the summed score between wealth index and HIV-associated mortality. of the wealth index in men and women followed a Second, with the aim of examining how new infections roughly normal distribution by visual assessment. The and deaths contribute to the changing prevalence of HIV, mean summed score of the wealth index for men and the rate of becoming infected must be compared with the women was higher in towns (0.44 and 0.43, respectively) rate of dying from infection. than in either estates (0.34 and 0.32) or rural areas (0.34 and 0.32). Therefore, men and women in towns were Differentials in sexual behaviour more frequently categorized in the higher wealth index Reported sexual behaviours, collected at follow-up tercile. Rather than constructing separate wealth index survey, were analysed for differences associated with categories for each site type, analyses are either controlled the wealth index. The summed score of the wealth index for area of residence, or presented separately when was modelled as a continuous variable. The influence of appropriate. There was also greater variance in wealth the wealth index on sexual debut, the number and type of index in towns compared with estates and rural areas partnerships, and condom usage was modelled, control- (Table 1), highlighting greater socioeconomic hetero- ling for age and site type. geneity in urban areas. The mean summed wealth index score did not substantially or significantly change between Mixing patterns and wealth index baseline and follow-up in any of the site types. Mixing patterns are not directly analysable from the baseline and follow-up of the Manicaland cohort because Follow-up rates decreased with an increasing wealth participants cannot be directly linked to their marital or index (66, 61, and 58%, chi-squared P < 0.0001), increas- non-marital partners. Participants were asked whether or ing education (primary/none 70%, secondary/higher not their last partner had secondary education. Having 55%, chi-squared P < 0.001) and being more mobile at secondary education was significantly correlated with baseline (64 and 53%, chi-squared P < 0.0001). Follow- the respondents’ wealth index status (R2 ¼ 0.11 and up appeared not to be directly dependent on wealth, P < 0.0001 for men; R2 ¼ 0.14 and P < 0.0001 for however, the wealth index was not an independent women). Therefore, we roughly approximate mixing predictor of follow-up after controlling for education and patterns by participants’ education level with their mobility (Wald test P ¼ 0.23). partner’s education level. We represent the degree of assortative (like-with-like) mixing by Q. Q is 1 when Prevalence mixing is completely assortative and 0 when completely As previously reported [6], HIV prevalence fell in the random [16]. Results are presented separately for open cohort between baseline and follow-up. HIV individuals aged under and over 30 years. prevalence fell in each wealth index tercile in both men and women (Table 2). The largest decrease in prevalence was in the highest wealth index tercile in both men at 25% Results (compared with 11% in the poorest tercile) and women at 21% (compared with 18% in the poorest tercile). Wealth score In the HIV serosurvey, 9842 eligible men (aged 17–54 Incidence years) and women (aged 15–44 years) were tested at In men, HIV incidence was lower in the top wealth index baseline and 7728 were tested at follow-up. Complete tercile (15.4 per 1000 person-years) compared with the wealth index data were missing from 209 of individuals at lowest tercile (27.4 per 1000 person-years; Fig. 1a–c). Table 1. Mean and distribution of wealth index in towns, estates and rural areas in Manicaland, Zimbabwe at follow-up study (2001–2003). Men Women Area Wealth tercile Mean (SD) summed score N (%) Mean (SD) summed score N (%) Town 0.44 (0.18) 0.43 (0.19) Low 103 (19%) 117 (19%) Med 116 (21%) 134 (22%) High 322 (60%) 353 (58%) Estate 0.34 (0.14) 0.32 (0.15) Low 303 (24%) 448 (36%) Med 436 (35%) 379 (30%) High 513 (41%) 423 (34%) Rural 0.34 (0.15) 0.32 (0.14) Low 406 (29%) 877 (32%) Med 359 (45%) 825 (30%) High 628 (45%) 1068 (39%)
- 65. S60 AIDS 2007, Vol 21 (suppl 7) Table 2. HIV prevalence at baseline (1998–2001) and follow-up (2001–2003) in wealth groups and change in prevalence in the open cohort. Sex Wealth index tercile Baseline prevalence (N) Follow-up prevalence (N) Change Pa Men Low 0.21 (1061) 0.19 (700) À11% 0.33 Med 0.18 (1650) 0.16 (797) À13% 0.076 High 0.20 (1635) 0.16 (1292) À25% 0.004 Women Low 0.28 (1573) 0.23 (1460) À18% < 0.001 Med 0.25 (2058) 0.22 (1362) À12% 0.020 High 0.24 (1569) 0.19 (1893) À21% 0.004 a Wald-test P value from logistic regression models controlling for age in 5-year groups and site type. There was a significant trend of decreasing incidence by group (incidence rate ratio 0.69; 95% CI 0.50–0.95; wealth index tercile after controlling for site type and age P ¼ 0.02, Poisson regression). [incidence rate ratio 0.73; 95% confidence interval (CI) 0.56–0.93; P ¼ 0.03, Poisson regression]. This trend was No clear significant or monotonic trends in incidence by even more marked in young men 17–24 years of age, in wealth index were observed in women of all ages or whom rates in the highest wealth index group were 8.3 young women (Fig. 1d–f). Controlling for education or per 1000 person-years and 23.3 in the lowest wealth index mobility (living outside the village in the past year) did not (a) Men: All agesa (b) Men: 25–54 year olds (c) Men: 17–24 year oldsa 50 50 50 40 40 40 30 30 30 20 20 20 10 10 10 0 0 0 Low Med High Low Med High Low Med High (d) Women: All ages (e) Women: 15–24 year olds (f) Women: 25–44 year olds 50 50 50 40 40 40 30 30 30 20 20 20 10 10 10 0 0 0 Low Med High Low Med High Low Med High Fig. 1. HIV incidence by wealth tercile in Manicaland Zimbabwe, 1998–2003. Incidence is the number of new HIV infections per 1000 person-years. Person-years at risk are contributed by participants uninfected (HIV-negative) at baseline and followed up at round 2 of the survey. Points and whiskers show the observed cumulative incidence and 95% confidence intervals. Lines and shaded area illustrate fitted Poisson model and 95% interval controlling for age and site type. a Significant linear trend (likelihood ratio test P < 0.05) Poisson regression model, controlling for age and site type.
- 66. HIV incidence andpoverty in Manicaland, Zimbabwe Lopman et al. S61 significantly improve the models for men or women index women between the ages of 15 and 34 years (Wald test P > 0.35 for all tests). Mobility was not (Poisson regression Wald P ¼ 0.024). In women over 35 associated with the wealth index tercile for sex, in which years of age there was no apparent mortality trend by education and wealth index tercile were positively wealth index. Controlling for education level or mobility associated for men (chi-squared P < 0.0001) and women did not significantly improve the models for men or (chi-squared P < 0.0001). women (Wald test P > 0.45 for all tests). Mortality Sexual behaviour Overall, 300 HIV-positive deaths were observed in the Considering the whole of the male study population, cohort from 1998 to 2003 [4]. Mortality rates decreased men of higher wealth index were more likely to have from 25 to 20 to 15 deaths per 1000 person-years in casual sexual partners and to have multiple partners in the increasing wealth index terciles for men, a trend 3-year follow-up period, but were also more likely to statistically significant after controlling for site type and report consistent condom use in their casual relationships age (Poisson regression P ¼ 0.024; Fig. 2a–c). Although (all controlling for site type and age in logistic regression not significant when split into young and older adulthood models; Table 3, model 1). In towns, however, men of (35 years of age), the same trend of decreasing mortality higher wealth index did not report greater numbers of was observed. Mortality was also lower in higher wealth partnerships but did report higher condom usage in casual (a) Men: All agesa (b) Men: 17–34 year olds (c) Men: 35–54 year olds 50 50 50 40 40 40 30 30 30 20 20 20 10 10 10 0 0 0 Low Med High Low Med High Low Med High (d) Women: All agesa (e) Women: 15–34 year oldsb (f) Women: 35–44 year olds 50 50 50 40 40 40 30 30 30 20 20 20 10 10 10 0 0 0 Low Med High Low Med High Low Med High Fig. 2. Mortality rates of HIV-positive individuals by wealth tercile in Manicaland Zimbabwe, 1998–2003. Mortality rate is the number of deaths among HIV-positive individuals per 1000 person-years in the cohort excluding HIV-negative participants who died. Deaths among HIV-infected represents individuals leaving the population of infected and therefore is directly comparable to the incidence of new infections. Points and whiskers show observed cumulative incidence and 95% confidence intervals. Lines and shaded areas illustrate the fitted Poisson model and 95% interval controlling for age and site type. a Significant linear trend (likelihood ratio test P < 0.10) Poisson regression model, controlling for age and site type. b Significant linear trend (likelihood ratio test P < 0.05) Poisson regression model, controlling for age and site type.
- 67. S62 AIDS 2007, Vol 21 (suppl 7) Table 3. Association of wealth and sexual behaviour: logistic regression models, men. Model 1a Model 2b % (n) b 95% CI P b 95% CI P Started sex (< 25 year olds) 56 (1456) 0.0 (À0.8; 0.8) 0.95 0.0 (À0.9; 0.8) 0.92 Towns 66 (130) À0.9 (À2.8; À1.0) 0.40 0.0 (À2.0; 2.0) 0.99 Estates 63 (454) À0.4 (À1.8; 1.1) 0.63 À0.4 (À1.9; 1.1) 0.56 Rural areas 47 (806) 0.5 (À0.6; 1.6) 0.31 0.5 (À0.7; 1.6) 0.42 Had casual partnership 59 (3263) 0.7 (0.2; 1.1) 0.007 0.6 (0.1; 1.1) 0.025 Towns 69 (549) À0.2 (À1.2; 0.9) 0.73 À0.3 (À1.3; 0.9) 0.69 Estates 61 (1300) 1.0 (0.2; 1.9) 0.015 0.9 (0.1; 1.7) 0.034 Rural areas 54 (1414) 1.0 (0.3; 1.8) 0.007 1.0 (0.2; 1.7) 0.012 Consistent condom use with casual partners 32 (1976) 1.0 (0.4; 1.7) 0.001 0.7 (0.1; 1.4) 0.27 Towns 40 (373) 1.3 (0.1; 2.4) 0.030 0.8 (À0.4; 2.0) 0.19 Estates 26 (783) 1.1 (0.0; 2.4) 0.050 0.9 (À0.4; 2.1) 0.17 Rural areas 34 (820) 1.1 (0.2; 2.1) 0.022 1.0 (0.0; 2.0) 0.050 Multiple partnerships 39 (3333) 0.4 (0.0; 0.8) 0.078 0.3 (À0.1; 0.8) 0.18 Towns 46 (547) À0.8 (À1.7; 0.1) 0.089 À0.9 (À1.8; 0.1) 0.074 Estates 41 (1295) 0.4 (À0.4; 1.1) 0.29 0.3 (À0.5; 1.1) 0.41 Rural areas 34 (1491) 1.4 (0.7; 2.1) < 0.001 1.3 (0.6; 2.0) < 0.001 Transactional sex 6 (1662) À0.7 (À2.0; 0.6) 0.29 À0.7 (À2.1; 0.7) 0.34 Towns 9 (332) À0.4 (À2.5; À1.7) 0.70 À0.7 (À3.0; À1.6) 0.54 Estates 6 (778) À2.5 (À4.9; À0.2) 0.035 À2.4 (À4.8; 0.0) 0.052 Rural areas 4 (552) 0.9 (À1.7; 3.5) 0.47 1.3 (À1.3; 4.0) 0.33 CI, Confidence interval. a Logistic regression models adjusting for age and site type (when models include participants from all sites). b Logistic regression models adjusting for age, education level and site type (when models include participants from all sites). All P values are from the Wald test. partnerships. In estates, relatively wealthier men were than one partner in 3 years of follow-up, or engage in more likely to have casual partners but were more likely to transactional sex (all controlling for site type and age in use condoms and not engage in transactional sex. logistic regression models; Table 4, model 1). These differentials were most pronounced in towns, with all Women of higher wealth index were less likely to begin remaining significant when restricting analyses to urban sex (under 25 year olds) have casual partners, have more women. Higher wealth index women in estates were less Table 4. Association of wealth and sexual behaviour: logistic regression models, women. Model 1a Model 2b % (n) b 95% CI P b 95% CI P Started sex (< 25 year olds) 45 (1911) À1.1 (À1.9; À0.3) 0.008 À0.5 (À1.3; 0.3) 0.22 Towns 52 (253) À4.5 (À6.6; À2.4) < 0.001 À4.0 (À6.2; À1.7) < 0.001 Estates 45 (543) À1.0 (À2.3; 0.3) 1.4 À0.4 (À1.7; 1.0) 0.56 Rural areas 38 (1115) 0.0 (À1.1; 1.1) 0.95 0.5 (À0.6; 1.7) 0.36 Had casual partnership 15 (4703) À1.1 (À1.7; À0.5) < 0.001 À1.0 (À1.7; À0.5) < 0.001 Towns 24 (614) À2.8 (À3.9; À1.7) < 0.001 À2.8 (À3.9; À1.5) < 0.001 Estates 18 (1273) À0.2 (À1.1; 0.8) 0.73 À0.2 (À1.3; 0.8) 0.69 Rural areas 12 (2816) À0.7 (À1.5; 0.1) 0.096 À0.8 (À1.5; 0.1) 0.078 Consistent condom use with casual partners 21 (718) 0.1 (À1.2; 1.5) 0.83 0.2 (À1.2; 1.5) 0.78 Towns 38 (146) 0.1 (À2.0; 2.3) 0.91 0.1 (À2.2; 2.5) 0.91 Estates 15 (210) 0.3 (À2.4; 3.1) 0.81 0.1 (À2.7; 3.0) 0.94 Rural areas 17 (362) À0.1 (À2.1; 2.0) 0.93 0.1 (À2.0; 2.2) 0.91 Multiple partnerships 8 (4794) À1.0 (À1.7; À0.3) 0.004 À1.0 (À1.8; À0.3) 0.006 Towns 17 (607) À1.6 (À2.7; À0.4) 0.010 À1.4 (À2.7; À0.1) 0.032 Estates 11 (1270) À1.1 (À2.3; 0.1) 0.080 À1.1 (À2.3; 0.2) 0.10 Rural areas 5 (2917) À0.4 (À1.6; 0.9) 0.56 À0.5 (À1.7; 0.8) 0.42 Transactional sex 7 (2334) À2.1 (À3.2; À1.0) < 0.001 À2.1 (À3.3; À1.0) < 0.001 Towns 14 (353) À2.7 (À4.4; 0.9) 0.003 À2.6 (À4.5; 0.7) 0.008 Estates 9 (667) À2.7 (À4.7; À0.7) 0.007 À2.8 (À4.9; À0.7) 0.009 Rural areas 4 (1314) À0.6 (À2.6; 1.5) 0.58 À0.8 (À2.9; 1.4) 0.47 CI, Confidence interval. a Logistic regression models adjusting for age and site type (when models include participants from all sites). b Logistic regression models adjusting for age, education level and site type (when models include participants from all sites). All P values are from the Wald test.
- 68. HIV incidence andpoverty in Manicaland, Zimbabwe Lopman et al. S63 likely to engage in transactional sex. In rural areas in Sexual mixing patterns particular there were no significant (P < 0.05) associations Sexual behaviour data suggest that higher wealth index between sexual behaviour and the wealth index. Condom men may be engaging in riskier sexual behaviours, at least use was not associated with the wealth index in women in for certain indicators such as having casual partners. any setting. Patterns of mixing will, however, predict the probability of engaging with an infected partner. Controlling for completed secondary education had no substantive effects on the estimates of the association Both men and women in higher wealth index groups of wealth index and the five sexual behaviours in men were more likely to have attended secondary or higher (Table 3 and Table 4, model 2). For women, secondary education (men 61, 74, and 76%; women 44, 52, and education was a stronger determinant of starting sex 63%; x2 < 0.001 for both sexes). The proportions who (for under 25 years olds) than the wealth index, but had achieved secondary/higher education were much controlling for education levels had little effect on higher in those under the age of 30 years compared with wealth index coefficients for other indicators of sexual those aged 30 years and over (Fig. 3a,b). In higher wealth behaviour. index groups young individuals (< 30 years) were more (a) Education and mixing by wealth index (< 30 years) (b) Education and mixing by wealth index (< 30 + years) 1 1 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 Low Med High Low Med High Low Med High Low Med High Men Women Men Women (c) HIV prevalence by education (< 30 years) (d) HIV prevalence by education (30+ years) 0.3 0.5 0.4 0.2 0.3 0.2 0.1 0.1 0 0 Men Women Men Women Fig. 3. Uneven risk resulting from sexual mixing patters: Mixing patterns and education by wealth index and HIV prevalence by education level. (a) and (b) Education and mixing by wealth index. Secondary/higher educated; – –^– – Q [degree of assortative (like-with-like) mixing]. (c) and (d) HIV prevalence by education. & Primary/none; & secondary/higher.
- 69. S64 AIDS 2007, Vol 21 (suppl 7) likely to have secondary education, and mixing was poorer men. This is for two reasons. First, men of higher increasingly assortative. This is important because in both wealth index in all sites (including towns where they do sexes, HIV prevalence was lower among individuals with not have more partners) were more likely to use condoms secondary education, although the difference was much in their casual partnerships. In effect, men reduce the greater in women (secondary/higher 24.8%; none/ transmission probability if encountering an infected primary 12.5%; chi-squared P < 0.001) than men woman. In addition, for each partnership that is formed, (secondary/higher 11.8%; none/primary 7.6%; chi- there may be a lower probability of serodiscordance in squared P ¼ 0.017). In other words, young men and higher wealth index groups; if partnerships are assortative women in higher wealth index groups were more likely to (made between members of the same wealth index) and be educated and to have an educated partner, and that HIV prevalence is lower in higher wealth index groups, partner was less likely to be infected, with this pattern these partnerships will tend to be less risky. Given the more pronounced in men. limitations of the present Manicaland data, we cannot measure directly the degree to which mixing is assortative Some 61% of women without any secondary education by wealth index. Participant reports on the level of reported that their last partner was of the same education of their most recent partners, however, suggest education level and 80% of women with secondary that higher wealth index men and women are markedly education reported that their last partner had the same more likely to form partnerships with individuals with level. Assortativeness of mixing and proportion with secondary education, and in turn, young people with secondary education also increased with wealth index in secondary/higher education have substantially lower HIV older participants (30þ years, Fig. 3b and d), but in this prevalence. Therefore, men and women of higher wealth older age group men and women with secondary index are less likely to form partnerships with infected education had a higher prevalence of HIV. Therefore, individuals. This crude measure of the sexual network men and women in higher wealth index groups would be requires substantial refinement in two ways. First, the more likely to contact an infected individual. In summary, level of education is only one dimension of HIV patterns of mixing appear to confer an increased risk for prevalence. Education as a function of age, as discussed the higher wealth index groups in the older ages but a briefly here, is another. As noted in a number of other lower risk for the young. studies in sub-Saharan Africa, the relationship between education and HIV vulnerability seems to be reversing, with education becoming protective [10,18,19]. Second, and preferably, the serostatus of each individual in a Discussion partnership would be known to understand the degree to which HIV has penetrated certain wealth index groups HIV incidence was associated with poverty in men, and to what degree serosorting is occurring in new especially young men, from 1998 to 2003 in Manicaland, partnerships. The association between HIVand education Zimbabwe. No such trend was observed in women. reversed completely, with education being protective in Lower HIV incidence in men of higher wealth index is young people and a risk in older groups. partly explained and supported by other observations from this cohort. The study was undertaken during a Analysis of mortality is one way to understand historical period of general decline of HIV prevalence, but, overall, trends in incidence because there is approximately a 10- the biggest decreases in prevalence occurred in higher year period between infection and death [20,21]. As with wealth index groups. By our ‘summed score’ measure of incidence, mortality was lower in higher wealth index the wealth index, towns were the ‘wealthiest’ of the site groups in both young men and young women, suggesting types, but they also had the greatest variance in their that the patterns of incidence have not changed markedly wealth index. This finding, alongside the generally higher since the estimated 10-year period when the groups prevalence in towns, supports the suggestion that HIV currently suffering mortality became infected. Modelling transmission may be enhanced by heterogeneity when studies of the HIV epidemic in Zimbabwe suggest that different social or economic groups mix [17]. behaviour change began in about 1992. (Hallett et al. unpublished information) and data from the Demo- The relationship between reported sexual behaviour and graphic and Health Survey from as early as 1994 show HIV incidence was not always straightforward. Men of women of higher wealth index delaying sexual debut as higher wealth index reported having more partners and well as the more frequent use of condoms by both men were more likely to have a casual partner. This is the same and women (Lopman et al., unpublished information). pattern observed early in the African HIV epidemic, This suggests that behaviour change have been underway which was used to explain the higher prevalence in the approximately 10 years before this study, with the more mobile and relatively well off [1]. The evidence resulting impact on infection only now becoming from the present study suggests that although higher apparent. An alternative explanation is that survival rates wealth index men may be having more partners, they may are lower in poorer groups. If malnutrition leads to faster be lower-risk relationships than those entered into by disease progression, as some research has suggested
- 70. HIV incidence andpoverty in Manicaland, Zimbabwe Lopman et al. S65 [22,23], and poorer groups are more malnourished, dynamic that may not be adequately represented by radio higher mortality rates could be caused by reduced ownership, floor type, etc., or any combination of these survival, rather than different levels of infection. variables. Finally, the summed score measure is to some extent a marker of urban residence, as evidenced by The present analyses have focused on incidence in order the higher mean wealth index score in towns. Levels to understand the direction of causation between the of follow-up were comparable to other major cohort wealth index and HIV as well as to reflect contemporary studies in Africa [6], but the wealth index was not patterns of infection. Previous analyses have examined independently associated with a probability of follow-up, poverty and prevalent infection, and provide an inter- so it is unlikely that these results are biased with respect esting comparison to the incidence findings. Seropreva- to the wealth analysis. Having secondary or higher lence was not associated with wealth index among men, education and being more mobile at baseline were, whereas poorer women were more likely to be infected in however, associated with lower follow-up rates. If survival the baseline survey of this cohort [24]. This is in contrast or incidence rates differed in the lost-to-follow-up groups to lower incidence in higher wealth index men and no the analysis may be biased with respect to mobility and association with incidence in women. This suggests a education. For example, if more educated groups left general shift away from risk in higher wealth index the Manicaland study sites to find employment in large groups, perhaps with the shift lagging behind in women. cities, they may have been at increased risk because of At baseline, poor women from rural areas were more the higher prevalence in cities and the possibility of likely to have started sex, whereas poor women from meeting new sexual partners after relocation. The group towns were more likely to engage in transactional sex. By of migrants that were followed up, however, did not the follow-up survey poorer women in towns were still have different levels of incidence or sexual behaviour, more likely to engage in transactional sex, but were also but this was a small group of the total migrant population more likely to have multiple and casual partners and to [27]. start sex younger. It may be expected that the first group of women to be motivated and able to change behaviour Despite these limitations of the current data, we have are relatively wealthy women in towns, and this is precisely observed a decreased risk of HIV incidence in higher what was observed. wealth index men. Although such a trend was not observed in women, the finding that lower wealth index HIV risk has reduced substantially among teenagers; women engage in riskier behaviour combined with their however, the girls who are still becoming infected have an tendency for having less-educated male partners suggests identifiable vulnerability such as being orphaned or that future trends may follow the emerging pattern in having experienced the death of another household men. At this advanced stage of the epidemic, a number of member [25]. Orphaned girls or girls with an HIV- factors may contribute to infection and risk behaviour. infected parent are more likely to drop out of school and HIV prevention activities in Zimbabwe have included begin sex, leading to pregnancy, poor reproductive health the treatment of sexually transmitted infections, social and HIV. So, despite not observing a general trend of marketing of condoms, voluntary counselling and testing, wealth index and incidence in women, there is a clear education through mass media and the activities of the causal pathway from vulnerability to leaving school, National AIDS Trust Fund (which is supported by ultimately leading to HIV infection in young women in income tax). These initiatives, as well as fear of AIDS this population. It has previously been observed that mortality, may have disproportionately affected those of households experiencing a death, and particularly an higher wealth index. Risk reduction behaviour, ushered AIDS death, disproportionably suffered the loss of the in by the relatively well off is a hopeful trend, but, in the household head, increased healthcare expenditure, and frail Zimbabwean economy, where the poor are an were more likely to dissolve [26]. increasing demographic, the clustering of HIV in lower wealth strata is cause for concern. This highlights the fact that our measure of the wealth index may be limited in a number of ways. When grouped into terciles, it becomes a relative measure, with individuals categorized on the basis of the asset ownership Appendix: Justification of ‘summed score’ of their household, compared with the asset ownership of other households. Therefore, the secular decrease in There is a high level of correlation between all binary and wealth index likely to be occurring because of AIDS ordinal wealth variables. Therefore, exploratory analyses mortality and the collapse of the Zimbabwean economy were undertaken to reduce the 10 assets to a simplified [11] has not been expressed in this measure. Furthermore, measure of the wealth index [24]. A simplified measure simplified as a relative measure, asset ownership may be a was created using multidimensional scaling analysis crude indicator of how the wealth index is a determinant (MDS), a statistical technique for exploring similarities of sexual behaviour. For example, falling below a certain and differences in data [28]. Starting from a correlation poverty threshold may drive a woman to sex work; a matrix between variables, MDS is used to assign a score to
- 71. S66 AIDS 2007, Vol 21 (suppl 7) each individual using fewer dimensions coded in a 12. Boerma JT, Weir SS. Integrating demographic and epidemio- logical approaches to research on HIV/AIDS: the proximate- reduced number of variables. The first dimension of determinants framework. J Infect Dis 2005; 191 (Suppl. 1):S61– MDS was compared with a summed score of all assets. For S67. the summed score the binary and ordinal measures were 13. Gregson S, Garnett GP, Nyamukapa CA, Hallett TB, Lewis JJ, Mason PR, et al. Supporting online material: HIV decline each transformed to lie between 0 and 1. The 10 variables associated with behavior change in eastern Zimbabwe. Science were added and expressed as a percentage. A high degree 2006; 311:1–25. of correlation was found between the first dimension of 14. Gregson S, Mushati P, White PJ, Mlilo M, Mundandi C, Nyamukapa C. Informal confidential voting interview methods the MDS and the summed score in subsistence farming and temporal changes in reported sexual risk behaviour for areas (R2 ¼ 0.96), roadside business centres (R2 ¼ 0.97), HIV transmission in sub-Saharan Africa. Sex Transm Infect 2004; 80 (Suppl. 2):ii36–ii42. commercial estates (R2 ¼ 0.94) and towns (R2 ¼ 0.95). 15. Ray CS, Mason PR, Smith H, Rogers L, Tobaiwa O, Katzenstein The summed score was thus considered equivalent to the DA. An evaluation of dipstick-dot immunoassay in the detec- first dimension of the MDS and a general indicator of tion of antibodies to HIV-1 and 2 in Zimbabwe. Trop Med Int Health 1997; 2:83–88. poverty. Given the reproducibility and more intuitive 16. Ghani AC, Garnett GP. Measuring sexual partner networks for interpretation of the summed score, it was used for transmission of sexually transmitted diseases. J R Stat Soc Series all analysis. A – Stat in Soc 1998; 161:227–238. 17. Anderson RM, May RM, Boily MC, Garnett GP, Rowley JT. The spread of HIV-1 in Africa: sexual contact patterns and the Conflicts of interest: None. predicted demographic impact of AIDS. Nature 1991; 352: 581–589. 18. Michelo C, Sandoy IF, Fylkesnes K. Marked HIV prevalence References declines in higher educated young people: evidence from population-based surveys (1995–2003) in Zambia. AIDS 1. Shelton JD, Cassell MM, Adetunji J. Is poverty or wealth at the 2006; 20:1031–1038. root of HIV? Lancet 2005; 366:1057–1058. 19. Hargreaves JR, Boler T. Girl power: the impact of girls’ educa- 2. United Nations. UNDP poverty report 2000. New York: United tion on HIV and sexual behaviour. Johannesburg, South Africa: Nations; 2000. Action Aid; 2006. 3. UNAIDS. Evidence for HIV decline in Zimbabwe: a compre- 20. Morgan D, Mahe C, Mayanja B, Okongo JM, Lubega R, hensive review of the epidemiological data. Geneva: UNAIDS; Whitworth JA. HIV-1 infection in rural Africa: is there a 2005. difference in median time to AIDS and survival compared with 4. Lopman BA, Barnabas R, Hallett TB, Nyamukapa C, that in industrialized countries? AIDS 2002; 16:597–603. Mundandi C, Mushati P, et al. Assessing adult mortality in 21. Collaborative Group on AIDS Incubation and HIV Survival. HIV-1-afflicted Zimbabwe (1998–2003). Bull WHO 2006; Time from HIV-1 seroconversion to AIDS and death before 84:189–197. widespread use of highly-active antiretroviral therapy: a 5. Hallett TB, Aberle-Grasse J, Bello G, Boulos LM, Cayemittes MP, collaborative re-analysis. Lancet 2000; 355:1131–1137. Cheluget B, et al. Declines in HIV prevalence can be associated 22. Anabwani G, Navario P. Nutrition and HIV/AIDS in sub- with changing sexual behaviour in Uganda, urban Kenya, Saharan Africa: an overview. Nutrition 2005; 21:96–99. Zimbabwe, and urban Haiti. Sex Transm Infect 2006; 82 23. Vorster HH, Kruger A, Margetts BM, Venter CS, Kruger HS, (Suppl. 1):i1–i8. Veldman FJ, Macintyre UE, et al. The nutritional status of 6. Gregson S, Garnett GP, Nyamukapa CA, Hallett TB, Lewis JJ, asymptomatic HIV-infected Africans: directions for dietary Mason PR, et al. HIV decline associated with behavior change intervention? Public Health Nutr 2004; 7:1055–1064. in eastern Zimbabwe. Science 2006; 311:664–666. 24. Lewis JJ. Behavioural, demographic and social risk factors for HIV 7. Halperin DT, Epstein H. Concurrent sexual partnerships help to infection in rural Zimbabwe. London: Imperial College; 2006. explain Africa’s high HIV prevalence: implications for preven- 25. Gregson S, Nyamukapa CA, Garnett GP, Wambe M, Lewis JJ, tion. Lancet 2004; 364:4–6. Mason PR, et al. HIV infection and reproductive health in 8. Piot P, Bartos M, Ghys PD, Walker N, Schwartlander B. The teenage women orphaned and made vulnerable by AIDS in global impact of HIV/AIDS. Nature 2001; 410:968–973. Zimbabwe. AIDS Care 2005; 17:785–794. 9. Gregson S, Terceira N, Mushati P, Nyamukapa C, Campbell C. 26. Gregson S, Mushati P, Nyamukapa C. Adult mortality and Community group participation: can it help young women to erosion of household viability in AIDS-afflicted towns, estates, avoid HIV? An exploratory study of social capital and school and villages in eastern Zimbabwe. J Acquir Immune Defic Syndr education in rural Zimbabwe. Soc Sci Med 2004; 58:2119– 2007; 44:188–195. 2132. 27. Mundandi C, Vissers D, Voeten H, Habbema D, Gregson S. No 10. Gregson S, Waddell H, Chandiwana S. School education and difference in HIV incidence and sexual behaviour between out- HIV control in sub-Saharan Africa: from discord to harmony? migrants and residents in rural Manicaland, Zimbabwe. Trop J Dev 2001; 13:467–485. Med Int Health 2006; 11:705–711. 11. World Bank. Zimbabwe: country brief. Washington, DC, USA: 28. Borg I, Groenen P. Modern multidimensional scaling: Theory World Bank; 2006. and applications. New York: Springer; 1996.
- 72. The economic impactsof premature adult mortality: panel data evidence from KwaZulu-Natal, South Africa Michael R. Cartera, Julian Mayb, Jorge Agueroc and ¨ Sonya Ravindranatha Measuring the household level economic impacts of AIDS-related deaths is of particular salience in South Africa, a country struggling with a legacy of poverty and economic inequality in the midst of an HIV epidemic. Household panel data that span more than a decade permit us to resolve many of the statistical problems that make it difficult to determine these impacts. After allowing for the impact of demographic adjustments and other coping strategies, we found evidence that these impacts are quite different across different types of households, and that the largest and most persistent effects were in the middle ranges of the South African income distribution, that is, households just above the poverty line. Households below that level seem less severely affected, whereas those above it seem to recover more quickly. All these results need to be treated with caution because their statistical precision is weak. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins AIDS 2007, 21 (suppl 7):S67–S73 Keywords: Africa, economics Introduction would have experienced in the absence of the death. Second, the impacts may be heterogeneous across the A number of recent studies have measured the economic different households that experience an AIDS-related impacts of AIDS-related deaths on the economic death. Third, and finally, the impacts may show wellbeing of affected households [1–3]. Measuring these differential persistence over time for households on the impacts is of particular salience in South Africa, a country basis of their ability to cope with the death and the stress it struggling with a legacy of poverty and economic places on household resources. inequality in the midst of an HIV epidemic. Among adults aged 15–49 years, the HIV prevalence rate is In an effort to deal with these difficulties, this study estimated to be 21.5%. As the epidemic moves from employs three waves of panel data on a sample of infection into impact, premature adult mortality rates are households from the KwaZulu-Natal province of South increasing rapidly, with an estimated 370 000 South Africa. These data, which span the 1993–2004 period, Africans dying of AIDS-related illness in 2003, making effectively permit this study to use each household’s the disease the leading cause of death in almost all South trajectory in the period preceding the onset of AIDS- African provinces [4]. It is estimated that the majority of related deaths to estimate what the household’s counter- AIDS-related deaths have still to occur [4]. Three factual economic status would have been without such difficulties, however, confront efforts to measure these deaths. Although the approach here is somewhat impacts reliably. First, measuring the economic impacts of distinctive, several earlier studies have utilized panel data an AIDS-related death requires an estimate of the to examine the dynamics of poverty status and the impact counterfactual level of wellbeing that the household that premature deaths have on pathways into or out of From the aProfessor, University of Wisconsin-Madison (Agricultural and Applied Economics), the bAssociate Professor, University of KwaZulu-Natal (Development Studies), and the cAssistant Professor, University of California-Riverside (Economics), and Graduate student, University of Wisconsin-Madison (Agricultural and Applied Economics). Correspondence to Julian May, Associate Professor and Head of School, School of Development Studies, Memorial Tower Building, University of KwaZulu-Natal, King George V Ave, Durban. Tel: +27 31 2602841; fax: +27 31 2602359; e-mail: mayj@ukzn.ac.za ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S67
- 73. S68 AIDS 2007, Vol 21 (suppl 7) poverty [5–8]. In addition, the analysis here explores the of an AIDS-related death, or whether they become heterogeneous impact of premature adult mortality by trapped in a permanently lower standard of living. permitting its estimated impacts to vary according to the households’ specific initial conditions. Although the immediate economic effects of an AIDS- The KwaZulu-Natal Income Dynamics related death are of course important, its long-term Study impacts on poverty depend on whether households recover from the death economically, or whether they fall Covering the 1993–2004 period, the KwaZulu-Natal into a long-term poverty trap. Both the economic theory Income Dynamics Study (KIDS) allows an analysis of of poverty traps [9] and evidence on natural disasters [10] household wellbeing both before and after the onset suggest that severe shocks can indeed push households of AIDS-related deaths in South Africa. The first round of below a critical level from which they cannot recover what became the KIDS data was part of the nationally economically. Similar insights emerge from the anthro- representative survey conducted by the Project for pological literature that shows how a poverty trap Statistics on Living Standards and Development [19]. situation can emerge for resource-poor households. The key decision makers (or ‘core’ members) of the 1354 The living standards of vulnerable households that face African and Indian households visited by the Project for multiple shocks over time may ratchet down over time to Statistics on Living Standards and Development in the the point at which they eventually become trapped in a KwaZulu-Natal province became the basis for the follow- situation of structural poverty [11,12]. up KIDS surveys undertaken in 1998 and 2004 [20,21]. When subsequent survey rounds found that co-resident In the context of AIDS-related deaths, these ideas of core members had split into separate residences, each core repeated shocks and poverty traps have an important was interviewed along with their corresponding new resonance. In countries in which there has been a high household. In the 2004 survey round, adult children of prevalence of HIV, death is preceded by comparatively the core members were also interviewed. In total, lengthy episodes of illness (and corresponding episodes of information was obtained in all three periods from at care; Hornbrook et al. [13] have suggested framing least one eligible respondent for 74% of the original analysis of protracted illness in terms of the costs of 1354 Project for Statistics on Living Standards and episodes of illness and care) that present households with a Development KwaZulu-Natal households. Ethical protracted series of shocks as the illness progresses. In the approval for study was obtained through the appropriate case of AID-related deaths in Tanzania, the average length committees at the University of Natal and the University of debilitating illness preceding death was 12 months of Wisconsin-Madison. Informed consent was obtained [14]. Another Tanzanian study showed that on average, before interviews were undertaken. Although similar to an adult experiences 17 different episodes of illness that found in other panel studies, the attrition rate in the before dying [15]. Each of these episodes is likely to be KIDS sample needs to be kept in mind when considering accompanied by episodes of care, which become more the results reported below. Maluccio [22,23] analysed the costly as death approaches. Moreover, in many cultures, pattern of attrition over the 1993 to 1998 period, and the death itself does not signal the end of the episode noted that it came from both the upper and lower tails of because funeral celebrations are required that necessitate the livelihood distribution. Although there has not yet further expenditure and possible indebtedness [16]. Even been a similar analysis of attrition over the 1998–2004 these expenditures may extend over several years if there period, the age-specific mortality patterns in the KIDS are annual celebrations or customs that families observe. data are similar to those found in another study of South Some of these changes are not unique to the HIV Africa [24]. Whereas these observations suggest that the epidemic. Discussing the impact of malaria, Sachs and KIDS data accurately reflect the reality of AIDS-related Malaney [17] noted the costs associated with changes in deaths, it is possible that sample attrition over the 1998– the behavior of household members concerning decisions 2004 period disproportionately reflected households such as schooling, child-bearing, savings and work- most severely affected by AIDS-related deaths. If that seeking are often overlooked when measuring the indeed happened, then the results here will probably economic impact of disease. Such changes have been understate the true impacts of those deaths. documented in the case of AIDS-related illness or death in Tanzania, where it has been shown that children may The KIDS data show that at ages 20–50 years, the marry earlier, drop out of school to help support the proportion of individuals dying between the second and family, and take on informal labor schemes [18]. These third waves was nearly three times the proportion dying observations all suggest that the economic effects of an between the first two waves [20]. In total, 309 members of AIDS-related death may be long lived, if not permanent. KIDS households between the ages of 20 and 50 years Following the lead of a study of Tanzania [8], the analysis died between 1998 and 2004. Of these deaths, 74 were here will exploit the available panel data to see whether the result of injury or accident. As we are interested in the households recover over time from the economic impacts impact of death associated with illness, we excluded this
- 74. Economic impacts ofadult mortality Carter et al. S69 ÀyitÀ1 group and designated the remaining 235 deaths git ¼ yityitÀ1 . Under our data structure, we observe git premature adult mortality (PAM). Premature mortality twice: once for t ¼ 1998, measuring the growth (positive refers to death occurring before some standard age. We or negative) in wellbeing between 1993 and 1998, and have used South African life expectancy in 2004 (51.4 once in 2004 (measuring the growth since 1998). years) as a guideline for this standard [25]. Other studies have used a slightly younger age group (15–50 years) on Consider the following fixed effects regression model for the grounds that this age range is the group most at risk of this growth in economic wellbeing measure: HIV infection [26]. git ¼ b1 hit þ d1 Sit þ d2 ½lnðyitÀ1 ÞŠ þ yi þ li þ eit (1) where hit is a binary indicator variable that takes a value of 1 Methodology and results when family i experienced PAM between times t–1 and t, and is 0 otherwise. We here treat the 25 households with Evaluating the impacts of PAM on the economic status of more than one PAM as identical to those with only one a family is difficult because we cannot observe what the PAM. Efforts to identify different effects for additional PAM family’s status would have been counterfactually in the failed statistically, presumably because of small sample sizes. absence of the death. The economic status of families The variable Sit signifies other unfavorable shocks that struck unaffected by PAM may be a very bad proxy for this the household between times t–1 and t, including crop loss, counterfactual status, especially in the case of the HIV theft, spousal abandonment, and death of an elderly house- epidemic in which specific behaviors and situations are hold member. The terms yi, li, d and b are all parameters to known to make infection and death more likely. be estimated, and eit is a random error term that we assume is unrelated to the included variables. Other studies have approached this statistical problem in several ways. One approach is to use propensity score Consistently estimating the coefficient b1, which gives methods to match affected with unaffected households, the impact of an adult death on the growth in wellbeing, effectively using the latter as the counterfactual for PAM is of course our primary interest. Note that this households [1]. Propensity score methods, however, only specification assumes that the impact of a premature control for observable differences between affected and adult death on wellbeing is the same for all households. unaffected households. Alternatively, with panel data it is This ‘homogenous effect’ regression model thus says that possible to use fixed effects estimation methods that can growth in household wellbeing over time depends on a also control for any unobserved differences between household-specific growth factor that does not change households that do not vary over time [3,6]. over time (yi), as well as on a time-specific intercept (li, t ¼ 98 or 04) that is assumed to be the same for The approach here is similar to this fixed effect approach all households. except that the three periods of KIDS data permit us to work in rates of growth in wellbeing rather in levels of Our ability to use fixed effects panel data methods to wellbeing. In particular, the KIDS data allow us to control for the household-specific effect is key to our observe a family’s economic trajectory (their growth in effort to identify the impact of a prime age adult mortality wellbeing) before the onset of the epidemic. Using this on economic wellbeing. Note that yi will capture time information, and a few modest statistical assumptions, we invariant observable and unobservable factors that can use fixed effects methods to predict reliably what the influence the growth in household wellbeing. It is affected family’s economic status would have been in the precisely these unobservable differences between house- absence of PAM. Effectively, this procedure allows each holds that make it difficult to estimate the impact of family’s past experience to inform the counterfactual that premature adult death. Once we control for the fact that is used to judge the impacts of PAM. households with adult deaths are likely to grow more slowly (or perhaps more rapidly) than the typical The KIDS data described above contain measures of household, we can be more confident in our estimate household economic wellbeing at three points in time, of b1. More formally, failing to control for the household- 1993, 1998 and 2004. We denote the economic wellbeing specific fixed effect would tend to exaggerate the impact of household i in time period t as, yit. Economic wellbeing of a premature death if households that suffer such deaths is measured as total household expenditures per capita. tend on average to experience lower growth even in the Expenditures include the imputed value of home- absence of the death. Given that the performance of the produced food, owner-occupied housing, etc. Although South African economy improved over the 1998–2004 in principle this measure should be scaled for the period, a change that is reflected in the profile of poverty demographic composition of the household, we have of the KIDS sample, we would expect l04 > l98. Note, not done so here in order to maintain comparability with however, that our methodology does not account for the the de facto per capita standard used to define poverty spillover effects of premature death. An analysis of Zambia in South Africa. In turn, we define the growth rate of estimated that local economic growth was negatively household wellbeing between period t–1 and t as: influenced by high concentrations of AIDS-related
- 75. S70 AIDS 2007, Vol 21 (suppl 7) deaths [7]. Although such macro effects may occur in different depending on the household’s initial level of South Africa, the urbanized and well-integrated nature of wellbeing (measured as the natural logarithm of the the South African economy makes it less likely that these household’s level of wellbeing at the beginning of the effects can be picked up at the local community level. period, yit–1). Conventional economic theory predicts Similar to Grimm [6], we controlled for other shocks that that d < 0, indicating that initially less well-off households potentially affect the growth rate of household economic experience more rapid growth. Other theory suggests the wellbeing. As measures of these shocks, we employed opposite [9]. For the purposes of this study, we are simply binary indicator variables as to whether the household concerned to control for the impact of initial levels of experienced the shock. The study by Grimm [6] also wellbeing on subsequent changes. controlled for changes in household demographic composition. We chose explicitly not to control for Table 1 displays the fixed effects estimates for the demographic changes as we suspected that such changes homogenous effects model. As the underlying data were are themselves coping strategies employed by families that collected through cluster design, robust standard errors suffer PAM. Statistically, demographic changes would be were calculated that allow for intracluster correlation in directly related to the error term eit in (1), and including it the regression errors. For the key variables of interest, the would yield biased estimates of the effect of PAM. P value (the level of statistical significance at which it is Although it would be possible to employ simultaneous possible to reject the hypothesis that the reported equation methods to address the statistical endogeneity of coefficient is zero) is reported in square brackets. The demographic changes, we prefer here to estimate reduced estimated coefficient of the PAM variable is negative, but, form models such as (1). The parameter estimates we surprisingly, it is not statistically significant. Its value obtained thus give us the full or bottom line effect of a (À0.21) means that PAM would be expected to lower a PAM on household wellbeing after the household has household’s 5-year growth rate by 21%, controlling for utilized available coping strategies (including demo- the unobserved time-invariant factors that influence each graphic changes). household’s growth rate (yi) and other variables. Note also that model 1 does not condition on the charac- The estimated coefficient of the initial level of wellbeing teristics of the adult who has died (as in Yamano and Jayne is statistically significant and signals a convergent process, [3]). Although we have no doubt that these characteristics with initially less well-off households estimated to grow matter, we are here interested in identifying the average or faster than others. None of the shock variables are typical effect of PAM in our South Africa data. statistically significant, although most are negative. Their insignificance may signal that most of these shocks are of a Finally, the basic regression model includes a term that short-term nature, and that whatever their short-term allows the expected growth in economic wellbeing to be effects on consumption, households had largely recovered Table 1. Fixed effects estimates of the impact of premature adult mortality on the growth rate of economic wellbeing. Explanatory variables Homogenous effects model Heterogeneous effects model Impact persistence model PAM impact Common effect, b1 À0.21 [25%] 1.8 [22%] 5.4 [27%] Differential effect, b2 – À0.370 [14%] À0.96 [25%] Persistence of PAM effect Common effect, b3 – – À0.10 [33%] Differential effect, b4 – – 0.02 [35%] Convergence, d À1.8MM À1.8MM À1.8MM Time effects 1998 intercept, l98 10.8MM 10.4MM 10.4MM 2004 intercept, l04 11.3MM 10.9MM 10.9MM Other shocks Illness 0.15 0.16 0.16 Job loss À0.05 À0.05 À0.05 Lose remittances À0.12 À0.12 À0.17 Lose grant À0.02 À0.02 À0.01 Abandonment À0.34 À0.27 À0.24 Theft 0.02 0.01 0.02 Crop loss À0.27 À0.26 À0.25 Elderly death 0.33 0.32 0.32 Household fixed effects, yi Included Included Included R2 within 0.34 0.34 0.35 PAM, Premature adult mortality. Reported significance based on robust standard errors corrected for clustering. Figures in square brackets are P values. M Statistically significant at the 10% level. MM Statistically significant at the 5% level.
- 76. Economic impacts ofadult mortality Carter et al. S71 their expected level of economics by the time of the positive effects of PAM for poorest households and survey. The death of an elderly household member, negative effects as households become better off. The however, is not a short-term event like an illness. The estimated coefficients are, however, not statistically positive, but statistically insignificant, coefficient on the different from zero at conventional probability levels. elderly death variable may seem surprising in the context We re-estimated the reported models using a lower age of South Africa, where the death of an older person nearly cutoff to define PAM (40 and 45 years instead of 50 years). always results in the loss of significant pension income. Its Lowering the cutoff had little effect on the magnitude of insignificant effect may reflect the fact that the households the estimated coefficients, but did improve their statistical were prepared (economically) for the death. One such significance. Increasing the PAM age cutoff towards 60 coping strategy may be through the shedding of the years also left the point estimates of the PAM coefficients household members who other studies have shown tend stable, but made them even less precise. This pattern is to immigrate into households when an elderly person consistent with the notion that the earning power of becomes of pensionable age. adults begin to fall off as they enter their 50s. In addition, households presumably become better prepared for an Whereas it is common to think of a premature adult death adult death as that death becomes (statistically) more reducing household economic wellbeing, of course it likely. This insignificance reflects the heavy demands put need not be so, especially when large numbers of adults on the data by fixed effects procedures, as well as the are unemployed or underemployed. Negative effects clustering of the underlying data that further reduces could also be muted if other family members involved in the precision obtainable from a sample of the size of the care-giving were also unemployed or underemployed at KIDS. When the impact of clustering on the standard the time of the onset of an AIDS-related illness. In this errors is ignored, the estimated coefficients are significant circumstance, an adult death may actually increase the at conventional levels. The estimated impacts of the other living standards of those remaining alive in the household shock variables are qualitatively identical to those in the as there are now fewer needs to meet from the family’s homogenous effect model. modest resources [27]. It should be stressed that the analysis here ignores other benefits (even those that Given that the estimated impact of PAM now depends on are solely economic) that an individual may bring to the the household’s initial level of wellbeing, we used the household, including support for children, their socia- estimated coefficients from Table 1 to calculate the impact lization and education [28]. The opposite could of course of PAM on the livelihood trajectories for three typical be the case for somewhat better-off households, in which households: one that began in the 20th percentile of the the premature death of an adult results in a net reduction initial wellbeing distribution, another at the 50th and a of the goods available for others. third at the 80th percentile. For each of these typical households, we took the average of the fixed effect terms From a statistical perspective, these observations suggest (the yi) for economically similar households. For example, that our basic regression model (1) mixes together two for the 20th percentile household, we took the average different regimes, one in which the immediate livelihood fixed effect estimates for all households between the 15th effects of PAM are negative, and another in which they and the 25th percentile. A similar band was used for the are positive. The average effect estimated in Table 1 other two household estimates. Using this estimate, plus would, in this case, be a data-weighted average of the two the household’s initial level of wellbeing (yi93) we then underlying regimes or regression relationships. From this calculated the predicted growth that would be expected perspective, we see that this data-weighted average effect for such a household over the 1993–1998 period, the is negative, but not surprisingly, it is insignificant. period before the onset of significant AIDS-related deaths. Using this predicted growth, we then calculated In an effort to pull these two regimes apart, and allow for the household’s predicted standard of living for 1998. To heterogeneous PAM effects, we modified the basic fixed make this value more easily interpretable, we have divided effect regression equation as follows: it by the poverty line such that a standard of living of 1 would imply a living standard exactly equal to the poverty git ¼ b1 hit þ b2 ½hit lnðyitÀ1 ÞŠ þ d1 Sit þ d2 ½lnðyitÀ1 ÞŠ line, 2 a living standard double the poverty line, and so þ yi þ lt þ eit ; (2) forth. For this analysis, we used the de facto official South African poverty line of R322 per month per person where the new coefficient b2 allows the PAM impact to (in year 2000 prices) [29]. Figure 1 shows these pre-PAM change with the household’s level of initial economic well- estimates for each of the three typical households. As being. As discussed above, we might expect b2 0 and can be seen, all but the least well-off households b1 ! 0. experienced negative growth in wellbeing over the 1993–1998 period. The second column in Table 1 shows the results of this expanded, heterogeneous effects model. The impact In order to assess the predicted impact of PAM, we then coefficients have the anticipated signs, indicating the performed the same exercise for the 1998–2004 period
- 77. S72 AIDS 2007, Vol 21 (suppl 7) where the new variable pit measures the passage of time (in Without PAM months) between the PAM death and the survey. Note that 200 this model permits the rate of recovery from PAM to vary by Percentage of poverty line 80th Percentile household income level. If households tend eventually to recover after 150 PAM, we would expect the effect of time-since-death to be positive, as found in an analysis of Tanzania [8]. In contrast, a Poverty line study of Kenya [3] did not find evidence that PAM effects 100 dissipate over time (although their period of observation was 50th Percentile household shorter than that in Beegle et al. [8]). 50 20th Percentile household The estimates (reported in the third column of Table 1) are again not statistically significant at conventional levels. 0 1992 1994 1996 1998 2000 2002 2004 Although this evidence is thus a bit weak, the estimated Year coefficients imply that a household that began in the 80th percentile of the wellbeing distribution would have Fig. 1. Impact of premature adult mortality on livelihood begun to recover its growth rate 5 years after the PAM. trajectories. PAM, Premature adult mortality. Better-off households would be estimated to recover more quickly. Less well-off households (for whom the effects are less pronounced) are estimated to recover for the three typical households. The starting point for less quickly. each household was their predicted level of wellbeing for 1998, as described above. For each household a growth In conclusion, this paper has explored the ability of panel rate (and resulting living standard level) was calculated data to permit more reliable inferences on the impact of both with and without a premature adult death. As can be AIDS-related deaths on household economic wellbeing. seen, the predicted impact of PAM on the household that The results obtained, which allow us to compare the began at the 20th percentile is slightly negative, but economic wellbeing of AIDS-affected households with imperceptibly so. Households at the 50th percentile in what their wellbeing would have been in the absence of 1993 (whose wellbeing levels had collapsed over the mid- AIDS, are less strong statistically than they might be. They 1990s) show a similar pattern. In contrast, households at do, however, suggest a consistent story in which the the 80th percentile show a large predicted drop in immediate impacts of an AIDS-related death are most wellbeing. Without PAM, the household would have severe for somewhat better-off households. It should be grown to a living standard in excess of 225% of the stressed that the analysis here gives a ‘bottom line’ impact poverty line. With PAM, the household’s wellbeing is estimate that reflects households’ adaptation to the shock, only 175% of the poverty line. This 50% drop is correctly including demographic adjustments in which severely interpreted as the impact of PAM on initially better- affected households may send dependent members to off households. better-off friends and relatives. Demographic adjustments of this sort would improve the economic wellbeing of the Finally, our data permit us to explore whether these remaining members, but would also hide some of the estimated PAM impacts tend to dissipate over time. To do effects of the AIDS shock. this, we modified the model by including an additional variable that indicates the number of months between the Although somewhat at odds with the conventional premature death and the date of the survey. This wisdom that the HIV/AIDS crisis most severely affects specification implies a linear relationship between time the poorest households, our findings are consistent since death and effects. We also estimated a more general with the observation that premature death may actually specification (with different dummy variables for different reduce poverty as we measure it [27]. These findings periods since death) and obtained qualitatively similar also suggest that an analysis of the impacts of AIDS- results. We also tried a specification in which we looked related deaths needs to allow for the possibility that for effects based on the period of time when the these impacts could be quite variable across household individual who eventually died first became unable to types. perform his ordinary economic activity. These specifica- tions proved to be uninformative. This same variable was Finally, the analysis here has explored the question of also interacted with initial expenditures to give the whether households are able to recover over time from following ‘impact persistence model’: the immediate impacts of an AIDS-related death. The git ¼ b1 hit þ b2 ½hit lnðyitÀ1 ÞŠ þ b3 pit results are again statistically weak, but they suggest that better-off households do manage to recover their rate of þ b4 ½pit lnðyitÀ1 ÞŠ þ d1 Sit þ d2 ½lnðyitÀ1 ÞŠ economic progress eventually. Although further investi- þ yi þ li þ eit ð3Þ gation is needed, the results do suggest that households
- 78. Economic impacts ofadult mortality Carter et al. S73 in the middle range of the South African income 10. Carter MR, Little P, Mogues T, Negatu W. Poverty traps and the long-term consequences of natural disasters in Ethiopia and distribution are more vulnerable to experience an AIDS- Honduras. World Dev 2006; 35:835–856. related death as a permanent setback in household 11. Moser C. The asset vulnerability framework: reassessing urban wellbeing. poverty reduction strategies. World Dev 1998; 26:1–19. 12. Davies S. Adaptable livelihoods: coping with food insecurity in the Malian Sahel. London: MacMillan Press; 1996. The work reported here is an outcome of a collabora- 13. Hornbrook MC, Murtado RV, Johnson RE. Health care episodes: tive project between the University of KwaZulu-Natal, definition, measurement, and use. Med Care Res Rev 1985; 42:163–218. the University of Wisconsin-Madison, the London 14. Beegle K. Labor effects of adult mortality in Tanzanian house- School of Hygiene & Tropical Medicine, UK Economic holds. Policy research working paper 3062. World Bank; 2003. Research Council, the International Food Policy 15. Bollinger L, Stover J, Riwa P. The economic impact of AIDS in Research Institute (IFPRI) and the South African Tanzania. Policy working paper. 1999. Available at: http:// Department of Social Development. Further financial www.policyproject.com/pubs/SEImpact/tanzania.pdf. Acces- sed: 22 May 2006. support was provided by: Department for International 16. Stover J, Bollinger L. The economic impact of AIDS. Washington, Development (DFID); the United States Agency for DC: Futures Group; 1999 . International Development (under agreement No. 17. Sachs J, Malaney P. Insight review article: the economic and LAG-A-00-96-90016-00 through the BASIS Collabora- social burden of malaria. Nature 2002; 415:7. 18. Ainsworth M, Semali I. The impact of adult deaths on children’s tive Research Support Program); the Mellon Founda- health in North-Western Tanzania. Policy research working tion; and a project grant (RES-167-25-0076) from the paper 2266. World Bank; 2000. UK Economic and Research Council. The authors have 19. Project for Statistics on Living Standards and Development. no conflicts of interest, including financial, consultant, South Africans rich and poor: baseline household statistics. South African Labour and Development Research Unit, Uni- institutional and other relationships that might lead to versity of Cape Town; 1994. bias or a conflict of interest. 20. May J, Aguero J, Carter MR, Timaeus I. The KwaZulu-Natal ¨ Income Dynamics Study (KIDS) 3rd wave: methods, first find- ings and an agenda for future research. Dev Southern Afr 2007; References in press. 21. May J, Carter MR, Haddad L, Maluccio J. KwaZulu-Natal 1. Gertler P, Levine D, Ames M. Schooling and parental death. Rev Income Dynamics Study (KIDS) 1993–1998: a longitudinal Econ Stats 2004; 86:211–225. household data set for South African policy analysis. Dev 2. Naidu V, Harris G. The impact of HIV/AIDS morbidity and Southern Afr 2000; 17:567–581. mortality on households–a review of household studies. S Afr J 22. Maluccio JA. Using quality of interview information to assess Econ 2005; 73:533–544. nonrandom attrition bias in developing-country panel data. 3. Yamano T, Jayne T. Measuring the impact of working age adult Rev Dev Econ 2004; 8:91–109. mortality on small-scale farm households in Kenya. World Dev 23. Maluccio JA. Attrition in the KwaZulu-Natal Income Dynamics 2004; 32:91–119. Study 1993–1998. Food Consumption and Nutrition Division 4. Medical Research Council. South African National Burden of discussion paper no. 95. International Food Policy Research Disease Study 2000: provincial estimates of mortality. Medical Institute; 2000. Research Council; 2004. 24. Hosegood V, Vanneste A-M, Timaeus I. Levels and causes of 5. Chapoto A, Jayne TS. Socio-economic characteristics of adult mortality in rural South Africa. AIDS 2004; 18:663–671. individuals afflicted by AIDS-related prime-age mortality in 25. Statistics South Africa. Mid-year population estimates, statistical Zambia. In: IFPRI/Renewal Conference on HIV/AIDS and release, P302. Pretoria: Statistics South Africa; 2006. Food and Nutrition Security. Durban, South Africa, 14–16 April 26. Ainsworth M, Dayton J. The impact of the AIDS epidemic on the 2005. health of older persons in northwestern Tanzania. World Dev 6. Grimm M. Mortality and survivors’ consumption. Discussion 2003; 31:131–148. paper 611. German Institute for Economic Research; 2006. 27. Kanbur R. Conceptual challenges in poverty and inequality: one 7. Jayne TS, Chapoto A, Byron E, Ndiyoi M, Hamazakaza P, development economist’s perspective. 2002. Available at: Kadiyala S, Gillespie S. Community-level impacts of AIDS- http://www.chronicpoverty.org/pdfs/conferencepapers/kanbur. related mortality: panel survey evidence from Zambia. Rev pdf. Accessed: 12 May 2006. Agr Econ 2006; 28:440–457. 28. Bell C, Shantayanan D, Gersbach H. The long-run economic 8. Beegle K, De Weerdt J, Dercon S. Adult mortality and economic costs of AIDS: theory and an application to South Africa. Work- growth in the age of HIV/AIDS. Working paper series. World ing paper. Heidelberg, Germany: University of Heidelberg; Bank; 2006. 2003. 9. Carter MR, Barrett CB. The economics of poverty traps and ¨ 29. Hoogeveen JG, Ozler B. Not separate, not equal: poverty and persistent poverty: an asset-based approach. J Dev Studies inequality in post-apartheid South Africa. Working paper no. 2006; 42:178–199. 739. William Davidson Institute; 2005.
- 80. The financial impactof HIV/AIDS on poor households in South Africa Daryl L. Collinsa,b and Murray Leibbrandtb Background: Rising mortality rates caused by HIV/AIDS in South Africa have sub- stantial and lingering impacts on poor households. Methods: This is a descriptive paper using a new dataset of daily income, expenditure and financial transactions collected over a year from a total of 181 poor households in South African rural and urban areas. One of the key pathways through which HIV/AIDS impacts on household wellbeing is through the socioeconomic impacts of death, which this dataset is especially useful in quantifying. Results: The key impacts of death on households are funerals and the loss of income. Funerals often cost up to 7 months of income. Nearly all households in the sample attempt to cover such costs by holding a portfolio of funeral insurance. Despite these efforts to insure against funeral costs, 61% of households are underinsured against the cost of a funeral. Nearly half the sample households are dependent on a regular wage earner, and another quarter are dependent on a grant recipient. Eighty per cent of these households would lose over half of their monthly income should the highest income recipient in the household die. Even by selling liquid assets, only one third of the sample households would be able to maintain their pre-death living standards for a year or more. Conclusion: Death poses substantial and lingering burdens from the funerals that surviving household members need to finance and the ongoing loss of income once brought into the household by the deceased. These costs pose so great a threat to households that they dominate household saving and insurance behavior. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins AIDS 2007, 21 (suppl 7):S75–S81 Keywords: South Africa, AIDS-related mortality, household finance, funerals, insurance Introduction greater destitution is revealed in household-level studies [3–7]. The severe impact of HIV/AIDS has led to a dramatic increase in the probability of death in South Africa’s This paper focuses on the financial impact of death at the population. The latest forecasts from the Actuarial Society household level, using a new dataset called the Financial of South Africa show the likelihood of death among adult Diaries, which tracks household-level cash flows over one men jumping from 36% in 1990 to a forecasted 61% in year. The dataset used in this paper is not geared 2008, whereas the likelihood of death among adult specifically towards capturing information about HIV/ women increases from 21% in 1990 to a forecasted 53% in AIDS. As households were observed over an entire year, 2008 [1]. Given high unemployment rates in South however, the impact of illness and death revealed more Africa, macroeconomic estimates of the impact of this details about the impacts of death on finances than other increasing death rate are somewhat benign [2], but far survey data. These are underappreciated factors in From the aNew York University, New York, New York, USA, and the bSouthern Africa Labour and Development Research Unit, University of Cape Town, Cape Town, South Africa. Correspondence and requests for reprints to Daryl Collins, Wagner School of Public Policy, New York University, 295 Lafayette Street, New York, NY 10012, USA. Tel: +1 914 433 9014; e-mail: dlc300@nyu.edu ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S75
- 81. S76 AIDS 2007, Vol 21 (suppl 7) understanding the impacts of HIV/AIDS on household 70% wellbeing. Although the Financial Diaries did record 60% situations revealing the burdens of caretaking and medical expense, a key finding of our research is that these costs 50% were dwarfed by the overwhelming financial impacts of death [9]. The impact of death was felt the strongest with 40% the high cost of funerals and with the loss of income from a breadwinner. These impacts are known to households, 30% and to a large extent they dominate the saving and insurance decision-making of households. This paper 20% provides a descriptive analysis that substantiates these points and details the extent to which sample households 10% were prepared for these financial burdens. 0% Regular wages Grants Remittances Business Casual wages Rental income Pension UIF Agriculture profits Fig. 1. Source of income, by percentage of average monthly Methods household income. Langa; Lugangeni; Diepsloot. This paper investigates the financial impact of HIV/AIDS on vulnerable households living in three urban and rural areas of South Africa. The Financial Diaries dataset was period. Households earn their income from a number of collected to examine a variety of questions about financial different sources. Figure 1 demonstrates this by reflecting management in poor South African households. The data the average household income in each of the three areas of can be accessed at http://www.datafirst.uct.ac.za/data_f- the study, and the percentage of average income earned diaries.html. A sample of 181 black households was from different activities. In the urban areas, most income selected in three of South Africa’s low income areas: comes from regular wages, from one or more people in Langa, Cape Town (urban); Diepsloot, Johannesburg the household. Government grant income is, however, (peri-urban); and Lugangeni, Eastern Cape (rural). A also important. There are several types of monthly stratified selection criteria based on relative household government grants in South Africa. At the time of the wealth was used to select households from each study, old age grants for older people above the age of 65 neighborhood in these areas. Between July 2003 and years were US$114 per month, disability grants were December 2004, the households were interviewed on a US$114 per month, child support grants for children fortnightly basis by a team of six field researchers. Detailed below the age of 7 years were US$26 per month and daily income, expenditure and financial transactions, as foster care grants were US$83 per month. In the entire well as open-ended qualitative data, were captured sample, 27% of households depend on a grant for the during this period using a specially built and conceived majority of their income and in rural Lugangeni, grants database. An attrition rate of 19% was experienced over account for close to 50% of average monthly household the course of data collection, primarily in relatively income. Other sources of income were recorded but were higher income households. The analysis presented in this less important. paper is based on the 152 households for which there is continuous data for the entire year. Full details regarding There is a fairly strong literature looking at the impact of the dataset, including survey instruments, can be found the death of a working age individual on a variety of on www.financialdiaries.com and in a preliminary household outcomes, and even some research on the descriptive overview [8]. burden of caring for those who become ill as a result of HIV/AIDS [5–7]. This literature has not, however, The General Household Survey, a national South African explored the large costs of the funerals associated with sample survey, reports national mean household expen- these deaths. Given that these funerals are frequent diture and national mean per capita expenditure to be occurrences within poor communities, they represent US$3287 and US$2060, respectively. The comparable daunting claims on household resources. This paper takes figures for the black population group are US$1288 and a close descriptive look at the extent of the financial US$814. Even allowing for comparability issues between burden that these funerals impose on households as well these datasets, it is clear that the sample households used the behavior of households in response to these burdens. in this paper are poor compared with average South The adequacy of funeral expense coverage is established African households, with a mean household income of by measuring the costs and sources of funding for funerals US$432 and mean per capita income of US$155. All and the sufficiency of funeral insurance against these costs. dollar amounts are converted from South African Rands Although funeral insurance does dampen the financial at a market rate of ZAR6.5 per US$, the average rate shock of death, the results show that 61% of sample during the period of the Financial Diaries data collection households are inadequately insured for potential funeral
- 82. Financial impact ofHIV/AIDS Collins and Leibbrandt S77 costs. Within the poorest of the sample, 84% are with interest and without, and by using money remaining inadequately insured for potential funerals in their from the two grants. She managed to pay for the funeral, households. but was left with a significant debt that she struggled to repay for the remainder of the year. The financial impact after the funeral is assessed by estimating the amount by which these households would The funerals observed in the Financial Diaries agree with see a decline in income per capita with the loss of the main similar estimates of funeral costs from a less detailed but income earner. The results show that 80% of the sample broader survey on funerals [10]. Most funerals appear to would lose more than half their per capita income with cost approximately US$1500. Compared with an average the death of the highest income earner, suggesting a household income of between US$155 and US$308 per lingering and debilitating shock of death. This analysis is month, households can easily spend an amount com- extended by assessing the amount of time households parable to approximately 7 months of income on one would be able to maintain their living standards after a funeral. Roth [10] found evidence that funerals in the death, given current liquid asset levels. With the sale of Grahamstown township cost approximately 15 times the liquid assets, only one third of the sample households average monthly household income. It is no surprise, would be able to maintain their pre-death living standards then, that the funeral industry in South Africa is for a year or more. substantial. An estimated US$770 million is spent on funerals each year, with 3000–5000 funeral parlors to facilitate them [11]. Results Such a large one-off cost cannot be managed out of monthly income, and some sort of financial instrument Data from the Financial Diaries [9] recorded five funerals must be used to manage the costs. Savings instruments, of household family members within 152 sample although helpful, are not feasible in meeting the entire households over the study year. Although five funerals cost of the funeral, as it would take many years for most are too few to generalize about how households manage households to save this amount. Borrowing would put their finances to pay for funerals, they do provide a very households in severe debt, even if they were able to find strong basis for understanding the financial impact of someone willing to lend them such a large amount these funerals. of money. An example of the expenses and funding sources It is therefore easy to understand why funeral insurance demonstrates how much funerals cost and how house- dominates the financial portfolios of the poor in South holds pay for them. Thembi (names of respondents have Africa. It is also, however, a strategy chosen to insure been changed to protect their identity) is one of the urban against a worst case scenario rather than an optimal use of respondents, a 50-year-old woman who lives with her 47- scarce savings to facilitate a movement out of poverty. year-old brother. The major source of income for the There is no doubt that this behavior is driven by the fact household was the disability grants of US$114 per month that, in the age of HIV/AIDS, working-age deaths are an that each received, plus a part-time job that Thembi held. immediate reality in the households and communities of Thembi belonged to a burial society but when her the poor [1]. Ten million people in South Africa have brother died, reportedly of tuberculosis, in June 2004, she funeral insurance [12]. Of these 10 million, 8 million was left scrambling for resources to pay for his funeral. A people belong to an informal burial society. There are an set of consolidated accounts for the funeral is shown in estimated 80 000–100 000 burial societies in South Africa Table 1. Of the sources of funds, only 11% came from [11]. Thembi’s burial society. The majority of the costs (54%) were paid for through relative’s contributions. Thembi This paper distinguishes between three different forms of was able to scrap together a bit more by borrowing, both funeral insurance: formal funeral plans with an insurance Table 1. Sources and uses of funds for Thembi’s brother’s funeral. Sources of funds US$ Uses of funds US$ Cash contribution from relatives 538 Undertaker 538 In-kind contribution from relatives 225 Tent 91 Burial insurance payout 154 Pots 35 Borrowed from aunt’s burial society (no interest) 154 Food 750 Borrowed from cousin’s savings club (30% per month) 92 Borrowed from cousin (no interest) 108 Leftover money from grant 92 Leftover money from brother’s grant 50 Total 1413 Total 1414
- 83. S78 AIDS 2007, Vol 21 (suppl 7) company; burial societies, which are informal funeral required should every person in the household need a plans administered through a group of friends, relatives funeral. or neighbors; and funeral parlor funeral plans. Within X the Financial Diaries sample, funeral instruments make INSUREi ¼ FUNERAL COSTS INSUREDi = X up between 10 and 20% of the total number of FUNERAL COSTS REQUIREDi ð1Þ instruments in the average household portfolio [10]. Most households have more than one burial society or formal funeral plan, a feature that is also observed in other An example of INSURE is calculated below for a rural South African research [11,13], as well as in Ethiopia and household. This household is headed by Mzwamadoda Tanzania [14]. (names of respondents have been changed to protect their identity), an older man who lives in Lugangeni with his Despite the prioritization of funeral insurance in house- wife, Tembisa, one child and six grandchildren. Despite a hold portfolios, however, this funeral cover rarely covers modest income (US$228 per month at the time of the the entire cost of a funeral. The example above showed study), this household had recently joined five out of their that the bulk of funds for the funeral come from seven plans in the past 4 years alone. All of the plans cover contributions from extended family members. This is a Mzwamadoda, Tembisa and their two daughters. A third consistent theme in the Financial Diaries data. Over the daughter is covered in all but three of the plans. If either study year, 81% of sample households contributed to the Mzwamadoda or his wife were to die, US$7708 would be funeral of a relative outside the immediate household at paid out for the funeral from all of the plans. By any least once [9]. These contributions are substantial, often standards, this is a lot of money for a funeral. In total, the costing up to 20% of the monthly income in some cases, sum of benefits insured for their family under all their requiring households to borrow or dip into savings. In a plans is US$23 614, whereas the amount they would situation in which the rate of death is increasing, as in the require to have adequate funerals for the family is only AIDS epidemic in South Africa, relatives may become less US$9230. Therefore, the value for the INSURE variable motivated or able to continue to contribute at the same for this household would be 2.558 (Table 2). rate they have in the past. INSURE is calculated household by household for the How much insurance would it take for households to entire sample of 152 households. The results show that 61% insure themselves against the death of a household of households are inadequately insured (with INSURE member, without requiring help from relatives? A funeral below 1) versus 39% that are adequately insured (with adequacy ratio is calculated from information in the INSURE above 1). The ratio of inadequately insured Financial Diaries dataset. This calculation uses an households to adequately insured households is particularly estimated cost of a funeral of US$1500 for an adult high in rural Lugangeni, which also has much poorer and US$770 for a child, as well as several other estimates households than the two urban areas. Is this a choice within based on funerals observed in the dataset. There are households, or are households too cash-constrained to further assumptions that needed to be made for this insure themselves adequately? estimate and we based these assumptions on what was recorded in the Financial Diaries data and reported in This question is investigated by splitting the sample in other papers [15]. Often benefits from burial societies or each area into three tiers based on income per capita per funeral parlors will be in kind. As burial society costs are month, taking into account not only the number of usually food, the value of this in-kind benefit is assumed individuals supported in the household but also income to be US$308. Funeral parlor benefits are usually a coffin, relative to others in the area. As Figure 2 below shows, transport and burial fees, the cost of which is estimated at only 16% of relatively low income households are US$770. The variable INSURE is calculated as the sum of adequately insured, compared with 38% in medium funeral costs insured over the sum of funeral costs income households and 62% in high income households. Table 2. Mzwamadoda’s portfolio of funeral cover. Type Member Description Monthly premium (US$) Benefits (US$) Burial society Tembisa Pay in cash when someone dies 9.25 each time 2000 Burial society Tembisa Pay in kind when someone dies 7.70 each time 923 Burial society Mzwamadoda Pay monthly 15.40 1538 Burial society Tembisa Pay monthly 9.25 2230 Funeral plan Tembisa Pay monthly; well-known retail bank 4.60 3692 Funeral plan Tembisa Pay monthly; well-known retail bank 5.85 7385 Funeral plan Tembisa Pay monthly; unknown company 15.40 5846 Total benefits insured US$23 614 Total benefits required US$9230 INSURE 2.558
- 84. Financial impact ofHIV/AIDS Collins and Leibbrandt S79 90% 40% 80% 35% 70% 30% 60% 25% 50% 20% 40% 15% 30% 10% 20% 5% 10% 0% 100% 75%–100% 50%–75% < 50% 0% Low income Medium income High income Fig. 3. Percentage of householdsa that would lose income Fig. 2. Percentage of adequately and inadequately insured from the death of the highest income earner, arranged by households in each relative income group. Relative income is proportion of income per capita lost. aDoes not include calculated in each area by dividing the sample into three households that are alone or are entirely dependent on groups on the basis of per capita income. In Langa, low remittances. income is defined as less than R577 per month, medium income is defined as between R577 and R981, and high income is greater than R981. In Lugangeni, low income is These observations can be generalized to quantify the loss defined as less than R233, medium income is between R233 of income that might happen should the main income and R639, and high income is greater than R639. In Diep- recipient die. Within the Financial Diaries sample, there sloot, low income is defined as less than R534, medium are 128 income-generating families, 22 households that income is between R534 and R1152, and high income is consist of one person living alone, and five households greater than R1152. Adequately insured; inadequately that are entirely dependent on outside remittances. Only insured. the 128 income-generating families would need to sustain themselves should the main income earner die. Following Bernheim et al. [16], the percentage difference between The average level of INSURE is 2.26 in the high income per capita income with and without the main income tier, 1.03 in the medium income tier and only 0.71 in the earner is calcuated. Figure 3 separates these households low income tier. Therefore, households in the high or into different tiers based on the percentage of income lost medium income levels in this sample are adequately from the death. insured or even overinsured, whereas households in the low income tier are underinsured. Moreover, within the The result is sobering. Over one third of these households low income tier, only 24% have no insurance at all, so would lose 100% of their per capita household income. In the majority (76%) of them does have insurance of some other words, the person who died was the sole income kind, but just not enough. provider for the household and they would be left with no income if this person died. Twenty-three per cent more The financial impact of death does not end with the costs households would lose between 75 and 100% of their per of the funeral. A crucial component of understanding capita household income, and another 22% would lose the impact of death is measuring the forgone income between 50 and 75% of their per capita household associated with the cessation of income activity as a result income. Therefore, 80% of the households in this sample of death. As Figure 1 earlier in the paper showed, this would lose more than half of their per capita income with income activity is not only wage earning, but also grant the death of their highest income earner. receiving. In some situations, the household left behind is cash-flow neutral after the death. For example, in In addition to this direct impact on household members, Thembi’s household, discussed earlier, after her brother the Financial Diaries data show that many households died, she was left without his income, but there was also support relatives who live outside their immediate only one in the household now rather than two. In household. Nearly all wage earners in the sample give another household, the main income before the death an average of 15% of their monthly income to someone was a monthly grant of US$114 per month, split among outside the household. This amounts to contributions of five people. After the death of this grant recipient, there between US$19 and US$35 worth of remittances every was one less person to feed, but there was also the loss of a month. To outside households that are dependent on main source of income. The remaining family of four had these remittances, this can be a substantial loss of income. to get by on casual work done by the oldest daughter. Their living standard deteriorated steadily through the How would a stock of assets or savings change this remainder of the year, from an income per capita of picture? Only 10 households in the entire sample have life US$70 to US$23. insurance. More common are provident or pension funds,
- 85. S80 AIDS 2007, Vol 21 (suppl 7) a quarter of households have these, so the household death of a prime age working adult as the key proxy might receive the benefit of the deceased’s savings, less his variable for an HIV death. or her liabilities. More generally, households do have some amount of financial savings and assets to support There are detailed implications from our study for surviving household members. A net assets figure for each understanding whether forms of insurance could provide household can be calculated by adding financial assets to better support for poor households than they currently physical assets and subtracting financial liabilities. The do. This analysis suggests that households are, on the majority of household net worth is, however, tied up in whole, inadequately insured or resourced for the funeral the value of the home in which they live. In this exercise, and ongoing costs attendant on the death of a household the value of the house is excluded, not only because it is member. It further suggests that innovative financial often a highly illiquid asset, but also because selling the instruments could be useful in addressing this inadequacy, family home would leave the family homeless. Other but we know very little about how a household would physical assets, particularly livestock, do not have the behave if new financial instruments were brought to the same restrictions. Households would commonly report market. Would people drop some of their funeral selling livestock for emergencies but rarely any other insurance products? Would they worry less about paying movable assets. for a funeral, or spend less time in making those payments? Testing behavioral change as insurance The amount of net assets (excluding housing) divided by increases and uncertainty decreases would require more pre-death per capita household income provides the information than the Financial Diaries dataset can number of months that the household could use the sale provide. Understanding the existing levels of under and of assets to maintain pre-death living standards. We find overinsurance in poor households is helpful to begin to that only one third of the sample households would be think about how needed innovative insurance and long- able to use assets to maintain living standards for one year term savings products are. To capture the entire or more, but the other two thirds would be left in a more complexity of decision-making and insurance, however, dire condition. Twenty-two per cent of the households a new field experiment would be the most robust method would be left with negative net assets, owing more than of inquiry. they own. Forty-five per cent have positive net assets, but only enough to sustain themselves for a year or less. A deeper problem is that selling assets would ultimately set households back by the number of years that it took to Acknowledgements obtain the asset, and potentially the income it can The authors are grateful for helpful comments from provide. participants in the UNAIDS/HEARD Research Sym- posium as well as to the post-symposium referees for their careful review. Discussion Sponsorship: The Financial Diaries project was sup- ported by the Ford Foundation, FinMark Trust and the The contribution of this paper has been to show that MicroFinance Regulatory Council of South Africa. death represents serious negative income shocks to poor Murray Leibbrandt acknowledges funding from the households and poor communities. Households try to National Institute of Child Health and Development cope with this through a sustained commitment to funeral and the National Institute of Aging (grant R01 insurance. Despite this, households are inadequately HD045581-01). insured and funerals impose huge costs on the household, the extended family and the community. In contrast, Conflicts of interest: None. health insurance is conspicuous by its absence. It is disconcerting that financial provisioning for medical References treatment and care seem to take second place to coping with the costs of death. This funeral insurance seems to 1. Dorrington RE, Johnson LF, Bradshaw D, Daniel T. The demo- graphic impact of HIV/AIDS in South Africa. National and crowd out other savings and insurance provisions. provincial indicators for 2006. Cape Town: Centre for Actuarial Research, South African Medical Research Council and Actuarial Society of South Africa; 2006. There is also a strong dependency on a single income 2. Bureau for Economic Research. The macroeconomic impact of provider in most of the households. The relevance of this HIV/AIDS under alternative intervention scenarios (with specific finding to a discussion of the impact of HIV in particular reference to ART) on the South African economy. Cape Town: University of Stellenbosch; 2006. turns on the vulnerability of such providers to illness and 3. Booysen F. Income and poverty dynamics in HIV/AIDS-af- death. It is beyond dispute that HIV-related deaths have fected households in the Free State province of South Africa. changed the shape of the distribution of South Africa’s S Afr J Econ 2004; 72:522–545. 4. Samson M. HIV/AIDS and poverty in households with children prime age working population. Most studies of the suffering from malnutrition: the role of social security in Mount impact of HIV on household wellbeing have used the Frere. S Afr J Econ 2002; 70:1148–1172.
- 86. Financial impact ofHIV/AIDS Collins and Leibbrandt S81 5. Chapoto A, Jayne T. Impact of HIV/AIDS-related mortality on 10. Roth J. Informal micro-finance schemes: the case of funeral rural farm households in Zambia: implications for poverty insurance in South Africa. ILO working paper no. 22. Geneva: reduction strategies. In: International Union for the Scientific International Labour Office; 1999. Study of Population (IUSSP) Seminar on Interactions between 11. Genesis Analytics. A regulatory review of informal and formal Poverty and HIV/AIDS. December 12–14, 2005. Cape Town, funeral insurance markets in South Africa. Johannesburg: South Africa; 2005. Genesis Analytics; 2005. 6. Linnemayr S. Awareness, morbidity, mortality: when does the 12. Leach J. Presentation to Portfolio Committee on Finance: hear- economic impact of HIV/AIDS at the household level com- ings on funeral benefits schemes. Johannesburg: FinMark Trust; mence? In: International Union for the Scientific Study of 2005. Population (IUSSP) Seminar on Interactions between Poverty 13. Thomson R, Posel D. Burial societies in South Africa: risk, trust and HIV/AIDS. December 12–14, 2005. Cape Town, South and commercialization. Cape Town: Actuarial Society of South Africa; 2005. Africa Convention; 2001. 7. Case A, Ardington C. The impact of parental death on school 14. Dercon S, Bold T, De Weerdt J, Pankhurst A. Group-based enrollment and achievement: Longitudinal evidence from funeral insurance in Ethiopia and Tanzania. The Centre for the South Africa. Demography 2006; 43:410–420. Study of African Economies working paper series 227. Oxford, UK: Oxford University; 2004. 8. Collins D. Financial instruments of the poor: initial findings 15. Collins D. Financial decisions and funeral costs. Southern Africa from the Financial Diaries study. Dev Southern Africa 2005; Labour and Development Research Unit working paper. 22:735–746. University of Cape Town, Cape Town; 2006. 9. Collins D. Funerals and finance: events in the lives of Financial 16. Bernheim D, Forni L, Gokhale J, Kotlikoff L. The mismatch Diaries households. Southern Africa Labour and Development between life insurance holdings and financial vulnerabilities: Research Unit working paper. University of Cape Town, Cape evidence from the Health and Retirement Study. Am Econ Rev Town; 2005. 2003; 93:354–365.
- 88. Father figures: theprogress at school of orphans in South Africa Ian M. Timaeusa and Tania Bolerb Objective: To examine the progress in their schooling of maternal and paternal orphans in a province of South Africa with high AIDS mortality and contrast it with that of both children who lived in different households from their parents and children who resided with their parents. Methods: The KwaZulu-Natal Income Dynamics Study is a panel of households first interviewed in 1993. The 1998 and 2004 waves of fieldwork collected 5477 reports on children aged 8–20 years. We studied the determinants of the proportion of these children who had completed 2þ grades fewer than expected for their year of birth using both household fixed-effects models and difference-in-difference models fitted to children reported on twice. Results: Co-residence with a well-educated mother benefited children’s schooling, but the fixed-effects models provide no evidence that maternal orphanhood or living apart from their mother adversely affected children’s schooling. In contrast, both paternal orphanhood and belonging to a different household from one’s father resulted in slower progress at school. Although absence of the father was associated with household poverty, this was not why it was associated with falling behind at school. Discussion: Both the substantial benefits of living with their fathers for children’s schooling and the limited importance of maternal orphanhood conflict with the results of most studies in this issue, including those of other research in the same part of South Africa. These findings caution against drawing general conclusions about the impact of the AIDS epidemic from a few studies of geographically localized populations. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins AIDS 2007, 21 (suppl 7):S83–S93 Keywords: AIDS orphans, cohort study, educational achievement, South Africa Introduction compare orphans with children who live in different households from their mothers and fathers and children One consequence of the AIDS epidemic in Africa has who reside with their parents. been a dramatic increase in the number of children who are orphans. The exact scale of the problem remains the In KwaZulu-Natal, HIV seroprevalence among antenatal subject of debate, but UNAIDS estimates that were 12 clinic attendees rose from approximately 1% at the million orphans aged less than 18 years in Africa in 2005 beginning of the 1990s to 41% in 2004 [3]. The mortality who had lost one or both of their parents to AIDS [1]. of young adults from AIDS is already very high [4,5]. By The prevalence of orphanhood rises rapidly with age. the time of the 2001 census, 6% of young people aged 5– Most orphaned children are of school age and more than 19 years in the province had a dead mother and 18% a half of all orphaned children are aged 12–17 years [2]. dead father (10% sample data). This paper uses data from a household panel study to examine the progress in their schooling of maternal and Some early studies of orphans in AIDS-affected African paternal orphans in KwaZulu-Natal, South Africa, and to countries found rather equivocal evidence as to whether From the aCentre for Population Studies, London School of Hygiene and Tropical Medicine, London, UK, and the bSection on HIV and AIDS, UNESCO, Paris, France. Correspondence to Ian Timaeus, Centre for Population Studies, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK. Tel: +44 20 7299 4689; fax: +44 20 7299 4637; e-mail: ian.timaeus@lshtm.ac.uk ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S83
- 89. S84 AIDS 2007, Vol 21 (suppl 7) orphans were educationally disadvantaged [6,7]. Three impact of paternal orphanhood on schooling outcomes comparative studies of Demographic and Health Survey remains statistically significant after controlling for other (DHS) and other household survey data have, however, variables, the adverse impact of death of the mother is found that orphans are either less likely to be enrolled in four to five times greater. The authors did not investigate school [8,9] or are less likely to be at the correct grade for whether progress at school differs between children who their age [10] than other children. A fourth such study live with their parents and children who have living suggested that the effects of orphanhood are inconsistent parents but do not live in the same household as them. and small [11]. The most sophisticated of the four studies used household fixed-effects models to control for other determinants of educational participation [8]. It found that orphans are less likely to attend school than other Methods children living in the same households. It concludes, therefore, that their educational disadvantage cannot KwaZulu-Natal Income Dynamics Study result solely from them being poorer than other children. The 1993 Project for Statistics on Living Standards and Development was a nationally representative survey of Cross-sectional surveys such as the DHS have serious households in South Africa with a design based on that of limitations for assessment of the impact of orphanhood on the World Bank’s Living Standards Measurement Studies children’s education. Few such surveys (and no DHS) [17,18]. The KwaZulu-Natal Income Dynamics Study have measured money-metric poverty. In addition, if (KIDS) re-interviewed the African and Indian house- children move between households after the death of holds from the Project for Statistics on Living Standards their parent(s), information is seldom collected about the and Development in the province of KwaZulu-Natal in household from which they come. It is likely that children 1998 to establish a panel study [19]. A third wave of field are fostered into relatively advantaged households. Lower work took place in 2004 [20]. educational participation rates among orphans than other children in the same households may thus predate the Accounts of the design and implementation of KIDS illness and death of their parents. have been published elsewhere [19–21]. The question- naire included a roster of household members and their In the past few years several studies have been published characteristics and detailed modules on household that use longitudinal designs to try and circumvent the income and expenditure. Each wave of the study limitations of previous research on orphanhood and underwent ethical review and the fieldwork in 2004 education in Africa. The results of these investigations are was approved by the ethics committees of all three rather mixed. Some find that the death of parents and universities involved in the project. other adults has a large impact on primary school enrolment or completion [12,13]; others have found In 1998, KIDS sought to interview all households rather small effects [14]. Most of those studies, together containing ‘core’ members from 1993, that is the 1993 with some cross-sectional studies [10,15], suggest that the household head, his or her spouse and other adults who educational disadvantages of maternal orphans are far had children in 1993. In order to refresh the panel and greater than those of paternal orphans. improve the follow-up of children, the 2004 wave also sought to interview ‘next-generation’ households, that is The previous study of most relevance to the present one households established by the children of core members examined the impact of parental death on schooling from 1993 who by 2004 were adults and had children of outcomes using data from the Africa Centre for Health their own. It also administered a shortened questionnaire and Population’s demographic information system to households fostering children aged less than 18 years (ACDIS) in the Hlabisa subdistrict of northern Kwa- of core individuals. Zulu-Natal [16]. It analysed data on nearly 20 000 children aged 6–16 years. Whereas paternal orphans in The 2004 wave of KIDS interviewed 1377 households Hlabisa tend to live in poorer households than non- [20]. They arose from 793 of the 1354 households orphans, no evidence exists that this results from interviewed in 1993. The fieldworkers located and orphanhood. Moreover, paternal orphans’ low grades- interviewed 319 next-generation households (68% of for-age are explained by their relative poverty, and they those identified) and 193 households containing foster are no more likely than other children to drop out of children (41% of those identified). The high rate of school. Maternal orphans, however, have significantly attrition of foster children reflects their mobility; many of lower grades-for-age and higher dropout rates than non- them had already moved on by the time an interviewer orphans living in the same households. The paper also visited the household identified as their home. In total, presents cross-sectional models fitted to the 10% sample data were collected on at least one 2004 household for data from the 2001 census for both KwaZulu-Natal and 62% of the baseline households interviewed in 1993. South Africa as a whole [16]. The results are similar to Although the panel has suffered considerable attrition, the those for Hlabisa. Although in the census dataset the age distribution of the resident members of the core and
- 90. Progress at schoolof orphans Timaeus and Boler S85 next-generation households considered together still Africa remain at school until the age of 19 or 20 years. approximates to that of the African and Indian population Therefore, we included these age groups in our analysis. of KwaZulu-Natal [20]. Grade progression can be measured as either a continuous The analysis presented in this paper was based on the 1998 variable, mean number of years behind at school, or a outcomes of 2609 children born in 1977–1989 and the categorical one, the proportion of children who are 2004 outcomes of 2868 children born in 1984–1995, behind at school. We adopted the categorical formulation giving a total of 5477 reports on children. for several reasons. Some children start school early and are ahead of their birth cohort. It is unclear whether they Definition of measures should be recoded to zero years behind at school or KIDS distinguished between resident household mem- allowed to exert a compensating effect on the mean. bers, who had slept there for at least 15 days in the past Second, more than half the school-age children in KIDS month, and other members, who might have spent as had completed at least one fewer grade than they should, little as 2 weeks in the past year residing with the often because they started school late. We are uncon- household but nevertheless pooled resources with it when vinced that it is appropriate to consider all these children they were present [17,19,20]. We examined the progress educationally disadvantaged. Third, the categorical at school of both the resident and non-resident children indicator is less vulnerable to measurement errors than who were members of panel households. the continuous indicator. Such errors would offset, at least partly, the gains in statistical power that would be Each wave of the study identified the parents of each obtained from working with a continuous outcome. In household member, if they were resident or non-resident practice, modelling the two measures of educational members of the same household as the index person, and progress led to very similar conclusions. The results established whether they were alive or dead, if they were presented in the tables are all for the proportion of not. Linking the responses to these questions about children who were 2þ grades behind in their schooling. parents across the three waves reveals that sometimes, When appropriate, mention is made in the text of when the person reported to be a parent in an earlier wave estimates for the continuous outcome. had died, someone else was reported as the parent in a later wave. This suggests that orphanhood is under- The economic status of households was measured using a reported in cross-section. New orphans among those detailed series of questions on household expenditures. In children contacted more than once were identified, particular, we used a poverty score calculated as the ratio therefore, using the information collected in later waves of household expenditure per resident member (adjusted on the survival of the first person reported to be the child’s for inflation) to a poverty line of R322 per month in 2000 mother or father. (US$47) [20,23]. In 2004, KIDS included detailed questions about Complete information is available on all of the children’s schooling and administered numeracy and characteristics of children who were analysed, except literacy tests to children aged 7–9 years. The analysis of their mothers’ education. The latter information is only these data is reported elsewhere [22]. Less information available for waves of KIDS in which a child’s mother was was collected on schooling in 1998, however, and only a member of the same household. Therefore, we the question asked in successive household rosters on the modelled only whether the educational benefits of living highest grade at school that each individual had with mothers were conditional on maternal schooling, completed is available as an outcome if one wishes to not whether the educational outcomes of foster children analyse the study in a way that makes full use of its and orphans were also affected by their mothers’ longitudinal nature. schooling. The educational background of the mothers of 16 of the children who were co-resident is unknown The South African school year coincides with the and those analyses that used this characteristic are based calendar year. Children should start school at the on 5461 reports, not 5477. beginning of the year in which they have their 7th birthday and progress to the next higher grade each Statistical methods subsequent January. Ideally, they matriculate from grade We first modelled the impact of orphanhood and 12, aged 18 years. By comparing actual progress with this separation from parents on grade progression by fitting ideal, we established whether each child was in the correct a household random-effects model to the data [24]. Such grade for his or her year of birth, or had fallen behind at models allow for the possibility that households may differ school as a result of enrolling late, repeating grades or from one another in ways that affect children’s progress at dropping out altogether. If children had completed fewer school but were not picked up by KIDS. They assume, grades than is ideal, they are described in this paper as however, that these unmeasured characteristics of house- ‘behind in their schooling’ or ‘behind at school’. Because holds are unassociated with those that are included in the they have fallen behind, many young people in South model explicitly.
- 91. S86 AIDS 2007, Vol 21 (suppl 7) The second model presented is a household fixed-effects The proportion of children who were behind at school model [24]. The model is estimated by comparing the dropped between 1998 and 2004. Although this seems outcomes of children within the same household who encouraging, it may not accurately portray trends in had differing characteristics or whose characteristics KwaZulu-Natal. Over time, the panel has probably changed between waves. Such models control for all fixed become less representative of the African and Indian characteristics of the children’s households, including population of the province. Moreover, the 2004 wave of sources of heterogeneity in educational outcomes that KIDS followed up children in next-generation house- KIDS did not measure explicitly. They have the holds and fostered children who would have been disadvantage that they are less statistically efficient than excluded by design from the 1998 data. random-effects models because no use is made of the data on households that contained only solitary or hom- Using information on whether the first person identified ogenous groups of children. to be the parent had died raises the estimate of the proportion of the children who were maternal orphans in Third, we present an individual fixed-effects model of the 2004 to 13%, compared with a cross-sectional estimate of determinants of grade progression between 1998 and 11%, and the estimate of paternal orphanhood to 26%, 2004 among the 925 children born in 1984–1989 who compared with 22% initially. In 1998, 21% of the children were observed in both waves. They include 286 children and in 2004 33% of them had lost at least one parent. Only who switched between categories of the outcome just over half the children were living with their mothers variable between the waves. on a day-to-day basis and only a minority of them were with their fathers. In 1998, 67% of children and in 2004 62% of them were Results living in poor households. The rise in median household income between the waves only benefited those children Table 1 presents some descriptive information on the who ended up in next-generation and foster households. school-age children and young people in KIDS house- The proportion of urban households that were poor was holds in 1998 and 2004. Over 80% of them were enrolled half that of rural households. Only half the children who in school and 97% of the children aged less than 17 years lived with their father belonged to poor households, were enrolled. Many of the children were 2þ years compared with approximately 70% of other children. behind in their schooling. Some of these children had Separation from or death of the mother was not associated started school late or dropped out, but many of them had with child poverty. fallen behind because they repeated grades. By the time that they should have been in the top three grades of Table 2 presents the main results of the analysis. Looking secondary school, the majority of children were 2þ years first at the random-effects model, Indian children had behind in their schooling. only one quarter the odds of being behind at school in Table 1. Characteristics of children and young people of school age who were members of KIDS households in the 1998 and 2004 wavesa. Characteristic Coding 1998 Wave 2004 Wave % of children 2þ years behind at school by expected grade for year of birth Expected grade 2 to 5 17 12 Expected grade 6 to 9 35 25 Expected grade 10 to 12 55 51 % Distribution according to mother’s residence and survival Resident household member 65 56 Non-resident member 10 4 Not household member 19 27 Dead 6 13 % Distribution according to father’s residence and survival Resident household member 37 30 Non-resident member 13 5 Not household member 33 40 Dead 17 26 % Distribution according to residence Rural 72 76 Urban 17 15 Metropolitan 11 9 % Indian 7 5 Household expenditure per head as % of the poverty line (R322 in 2000) 10th Percentile 28 26 Median 67 75 90th Percentile 206 292 Mean number of children per household 3.9 3.7 Number of children 2609 2868 a Children and young people born in 1977–1989 in 1998 and born in 1984–1995 in 2004, that is those who are aged 8–20 years approximately. See text for explanation.
- 92. Table 2. Oddsof being 2R grades behind at school, controlling for year of birtha. Household random-effects Household fixed-effects Individual fixed-effects model model model Variable Coding Odds ratio 95% CI Odds ratio 95% CI Odds ratio 95% CI Race (reference category: African) Indian 0.25 (0.13–0.48) Residence in 1993 (reference category: rural) Urban 0.53 (0.38–0.73) Metropolitan 0.52 (0.35–0.77) Sex of child (reference category: boys) Girls 0.44 (0.38–0.51) 0.44 (0.37–0.53) Household expenditure per head ln (poverty score) 0.64 (0.57–0.73) 0.84 (0.71–1.00) 0.65 (0.38–1.11) (relative to poverty line) Mother’s education, residence and survival No schooling 1.98 (1.40–2.81) 1.43 (0.88–2.33) 7.87 (0.46–135) (reference category: resident with primary schooling) Incomplete secondary 0.75 (0.57–0.99) 0.97 (0.66–1.41) 0.89 (0.05–14.4) Complete secondary 0.45 (0.29–0.69) 0.73 (0.42–1.27) 0.26 (0.02–4.23) Not resident in household 1.05 (0.81–1.36) 1.11 (0.80–1.53) 1.95 (0.19–20.1) Mother dead 1.15 (0.84–1.59) 1.19 (0.81–1.77) 3.00 (0.16–57.1) Progress at school of orphans Timaeus and Boler Father’s residence and survival (reference category: resident) Non-resident member 1.30 (0.94–1.79) 1.24 (0.83–1.87) 1.68 (0.44–6.49) Not a household member 1.57 (1.26–1.96) 1.72 (1.31–2.27) 2.03 (0.56–7.28) Father dead 1.57 (1.24–1.99) 1.71 (1.27–2.32) 4.00 (0.81–19.7) Wave of KIDS (reference category: 1998) 2004 0.63 (0.54–0.74) 0.61 (0.51–0.73) 12.22a (6.76–22.1) Between household variance (mi) 1.06 (0.93–1.20) Reports on children (households) contributing information 5461 (1002) 4190 (580) 572 CI, Confidence interval. a The first two models include a series of indicators that control for year of birth relative to year of fieldwork (expected grade). The final model compares the same children at the two waves and the relative odds for 2004 reflect the ageing of the children since 1998 as well as the passage of time. S87
- 93. S88 AIDS 2007, Vol 21 (suppl 7) 1998 or 2004 of African children. The odds of children household being 2þ years behind at school were more whose families were residing in urban areas in 1993 being than two-thirds greater than those of other children in the behind were only half those of children from rural areas. same households. These odds ratios hardly change at all if The odds of girls being behind in their schooling were the poverty index is excluded from the model (model not less than half those of boys, and the odds of being shown). 2þ grades behind fell by more than a third between 1998 and 2004. The coefficient on the poverty score suggests The last two columns of Table 2 present the difference-in- that the probability that children were 2þ grades behind difference model fitted to the data on children who were at school shrank by 15% with each doubling of household reported on both in 1998 and 2004. Once again it expenditure per head. provides no evidence that death of, or separation from, their mothers affected children’s grade progression. Whether living with their mothers benefits children’s Children whose fathers died between 1998 and 2004, progress at school depends entirely on the mother’s own however, completed on average 0.8 fewer grades than level of education. Therefore, compared with children other children (P ¼ 0.00) during the interval between with co-resident mothers who went to primary school, the waves (model not shown). These children’s odds of children whose mother was co-resident but uneducated dropping 2þ grades behind in their schooling were four had twice the odds of being behind in their schooling, times those of children with co-resident fathers whereas having a co-resident mother who completed (P ¼ 0.09). No evidence exists that children whose secondary school halved the odds of children being fathers moved out were affected similarly. behind. The progress of orphans and other children whose mother was not a member of their household It is possible that children whose fathers died between the was no worse than that of children with co-resident waves shared some other characteristic that affected their mothers with only moderate levels of schooling, and was progress at school. Using data from the 1998 wave, significantly better than that of children with uneducated however, one can investigate whether the children who co-resident mothers. went on to be orphaned or became separated from their fathers were already making slower progress at school This study provides no evidence that the educational level than other unorphaned children. No evidence exists of co-resident fathers was important for their children’s that they were (model not shown). If anything, after progress at school. Moreover, children whose fathers controlling for the other factors included in the random- belonged to the same household, but were not currently effects model, children whose father went on to die were living there, did no worse than children with resident making somewhat better progress at school in 1998 than fathers. Children whose father was either dead or children who remained unorphaned in 2004 (odds ratio unlinked to their household, however, had 1.57 times 0.7, P ¼ 0.20). the odds of being 2þ grades behind in their schooling. Both boys and girls were affected (model not shown). Table 3 assesses whether these findings are likely to be sensitive to the misreporting of orphanhood. It presents The effect of mothers’ education on children’s grade two household fixed-effects models that differ only in that progression was attenuated greatly in the fixed-effects the first uses our preferred measure of orphanhood, model that compares foster children with children in the whereas the other uses only the cross-sectional infor- same households whose mother was present. Therefore, mation from each wave of KIDS in isolation to identify most of the apparent effect of education among co- orphans. The two sets of results are similar. They both resident mothers in the previous model stemmed from indicate that children whose fathers were members of confounding with other determinants of educational their households were less likely to be behind in their progress. Children with co-resident mothers who had schooling than children whose fathers were dead or were completed secondary school remained significantly less not household members. The analysis based on cross- likely (P ¼ 0.04) to be behind at school than children with sectional reports of orphanhood, however, suggests that a co-resident mother who never attended school. children who were separated from their mothers were Aggregating across levels of maternal schooling does more likely to be behind at school than children whose not provide any evidence that, as a whole, children with mothers were co-resident. The analysis using our resident mothers were making better progress at school preferred method of identifying orphans does not. than maternal orphans or other foster children living in the same households as them (model not shown). A more serious source of bias than the inaccurate measurement of orphanhood might be attrition of the In contrast to the effects for mothers, the estimated panel. Table 4 examines risk factors for attrition by 2004 impact of the absence or death of the father on their of the children in panel households in 1998. Considerable children’s progress at school was stronger in the fixed- attrition of the panel has occurred; only 68% of 1998 effects models than the initial model. The odds of orphans household members from the 1984–1989 birth cohorts and other children whose fathers did not belong to their were found in households interviewed in 2004.
- 94. Progress at schoolof orphans Timaeus and Boler S89 Table 3. Odds of being 2R grades behind at school for all orphans and orphans reported on the roster (household fixed-effects model, controlling for year of birth, n U 4197 children in 580 households). All orphans Roster orphans Variable Coding Odds ratio 95% CI Odds ratio 95% CI Household expenditure per head ln (poverty score) 0.84 (0.71–0.99) 0.84 (0.71–1.00) (relative to poverty line) Mother’s residence in household and Non-resident member 1.08 (0.74–1.57) 1.31 (0.95–1.81) survival (reference category: resident) Not a member 1.13 (0.89–1.42) 1.20 (0.94–1.52) Mother dead 1.21 (0.88–1.67) 1.08 (0.78–1.50) Father’s residence in household and Non-resident member 1.23 (0.82–1.85) 1.18 (0.81–1.72) survival (reference category: resident) Not a member 1.67 (1.27–2.20) 1.58 (1.21–2.06) Father dead 1.70 (1.26–2.30) 1.64 (1.23–2.19) Sex of child (reference category: boys) Girls 0.44 (0.37–0.53) 0.45 (0.38–0.53) Wave of KIDS (reference category: 1998) 2004 0.59 (0.49–0.71) 0.63 (0.53–0.76) CI, Confidence interval. Unsurprisingly, the oldest children, particularly older manifest some years later when the affected household boys, were least likely to remain in panel households by finally exhausts its savings and external sources of 2004. Attrition was also high among the few children assistance. Endogeneity is also a major issue. Children who were not resident in the household that reported on may become poor because they are orphaned or be them in 1998. No evidence exists of greater attrition of orphaned because their families are poor. In addition, orphans, of the poor, or of urban residents. Children who adult death is a risk factor for household dissolution [25]. belonged to a different household from their mother in 1998, however, suffered higher attrition than other Such complexities suggest that only panel studies can children. No doubt many of them left the panel hope to assess the household-level impact of the AIDS household to move in with their mothers. epidemic, identify the determinants of vulnerability and evaluate interventions intended to mitigate impact. Unfortunately, panel studies have their own limitations. They are prone to differential attrition of the poorest and Discussion most vulnerable households. Moreover, household-based studies alone are a poor source of data on the quality of the One methodological issue in the study of the impact of services available to households (e.g. schools). AIDS and other adult deaths on other household members, including children, is that the impact of having The results presented here are based on a panel study in an AIDS case in the household may be felt before the KwaZulu-Natal. We found that rural children were death, when the person first becomes ill, or only become approximately twice as likely as urban children to be badly Table 4. Risk factors for attrition between 1998 and 2004, children born in 1984–1989 (n U 1350). Variable Coding Odds ratio 95% CI Race (reference category: African) Indian 1.09 (0.54–2.20) Residence in 1993 (reference category: rural) Urban 1.20 (0.76–1.90) Metropolitan 1.09 (0.57–2.09) Household expenditure per head in 1998 (relative to poverty line) ln (poverty score) 0.98 (0.77–1.24) Mother’s residence and survival in 1998 (reference category: resident) Non-resident member 1.11 (0.66–1.87) Not a household member 1.78 (1.25–2.53) Mother dead 1.47 (0.81–2.68) Father’s residence and survival in 1998 (reference category: resident) Non-resident member 0.75 (0.45–1.25) Not a household member 1.00 (0.69–1.45) Father dead 1.47 (0.92–2.33) Year of birth (reference category: 1989) 1984 1.33 (0.90–1.95) 1985 1.32 (0.93–1.86) 1986 0.65 (0.42–1.01) 1987 0.88 (0.59–1.30) 1988 1.00 (0.67–1.50) Sex of child (reference category: boys) Girls 0.85 (0.66–1.08) Residence in 1998 (reference category: resident) Non-resident 2.14 (1.36–3.36) CI, Confidence interval.
- 95. S90 AIDS 2007, Vol 21 (suppl 7) behind at school, but that household poverty only had a way fully through the child population. Therefore, the small impact on children’s progress at school once number of orphans in South Africa will continue to grow residence and other confounders were allowed for. This rapidly for many years [29]. Our results suggest that this finding suggests that the educational disadvantage of rural trend represents a serious threat to efforts to improve the children is something that, in principle, the government levels of education of young South Africans. could aspire to correct. The data collected in KIDS provide only hints as to the We also found that girls were only approximately half as mechanism by which co-resident fathers promote their likely to be behind at school as boys. This pattern is children’s schooling. Their children were better off than consistent with the findings of previous research in other children. The household fixed-effects model, southern Africa [26]. It should not be taken to imply that however, suggests that changes in households’ incomes schools in South Africa are free of attitudes and practices only exerted a small influence on whether the children in that discriminate against girls [27]. The superior them dropped behind at school. The economic benefits educational achievement of girls has not, however, always of co-residence with fathers thus explain little of its received the recognition that it should in the literature on beneficial impact on children’s progress at school. In gender and schooling in this region. addition, although co-residence with fathers was more common in urban households, the estimated benefits of Our findings concerning the impact of orphanhood and having a father in the household persist in the household separation from parents on children’s schooling differ fixed-effects model. It seems unlikely, therefore, that from those of most other studies in Africa. On the one school quality and journey times are important con- hand, KIDS indicates that co-residence with fathers founders of the co-residence–schooling relationship. greatly benefited children’s progress at school. On the other, it provides no evidence that co-residence with Some of the children with fathers who were alive, but not mothers improved children’s progress at school unless members of the same household, will have lost touch with the mother had been to secondary school herself. The them altogether. The fact that not only orphans but impacts of separation from and death of the father were children with absent fathers did less well at school, similar, except that the panel analysis suggests that although many of them had never lived with their fathers, children whose fathers died between 1998 and 2004 may suggests that, although witnessing the illness and death of have completed fewer grades during these years than their father is obviously distressing for any child, it is the children who became separated from their father. This lack of a father–child relationship that is more serious in could reflect the short-term impact of the shock the longer run. As the amount of schooling completed by of bereavement. co-resident fathers themselves did not significantly affect whether their children were behind at school, it seems KIDS data on children in core and next-generation unlikely that men who live with their children are households remain broadly representative of KwaZulu- unusually strong advocates for education. It is probably Natal [20]. They indicate that, by 2004, approximately more generic supportive and directive aspects of fathering 25% of children of school age had dead fathers. Half as that benefit children’s schooling [30–32]. many children again, 37%, were living in households to which their father did not belong. Unfortunately, both The results presented here appear to conflict with those of domestic situations became more common between 1998 Case and Ardington [16], who conducted a similar and 2004. Annual deaths of adults who were aged 20–44 analysis of data from a largely rural area of KwaZulu- years in 1998 increased fivefold during the 6 years Natal. KIDS and ACDIS are two of the most detailed and between these two waves of KIDS [20]. Although the well-conducted longitudinal studies from anywhere in study did not collect data on HIV status or precise causes Africa that can be used to study the impact of orphanhood of death, this rise undoubtedly reflects the rapid growth in on children’s educational outcomes. The fact that the two AIDS mortality in South Africa [4,28]. Only 32% of datasets yield different results is sobering. deaths of men aged 20–44 years in 1998–2003 and 6% of those of women resulted from injuries, and there was The apparent contradiction between the findings of the no clear trend in the number of injury deaths across this two studies concerning mothers probably arises because period. KIDS is too small a study to pick up the moderately sized effects of maternal orphanhood on schooling that Case The rate of increase in the prevalence of HIV infection in and Ardington [16] identified with much larger datasets. South Africa has slowed since 2004 [3]. In addition, a Using a household fixed-effects model, they estimated growing number of AIDS patients in the country now from ACDIS that maternal orphans have completed 0.12 receive antiretroviral therapy and this may be moderating years less schooling than other children, with a standard the upward trend in mortality. Although these are error of 0.05 (see Table 5 in Case and Ardington [16]), and welcome developments, it takes 18 years for the effect on from the 2001 census sample that maternal orphans have orphanhood of any change in adult mortality to work its 0.22 years less schooling than other children, with a
- 96. Progress at schoolof orphans Timaeus and Boler S91 standard error of 0.03 (see Table 9 in Case and Ardington should have been concentrated among those orphans [16]). Fitting an equivalent model to KIDS suggests that who were making poor progress at school. If anything, maternal orphans have 0.15 years less schooling than one might expect those paternal orphans who had other children, but with a standard error of 0.09. The acquired a new ‘father’ to make better progress at school effect is of similar magnitude but statistically insignificant. than other paternal orphans. We thus remain open- minded as to whether underreporting of paternal Moreover, whereas the design of the two analyses is orphanhood is the main explanation of the inconsisten- similar, it is not identical. Case and Ardington [16] can cies between our results and those of Case and Ardington control only for previous asset poverty, not previous [16]. income poverty. Our results did not, however, suggest that previous poverty confounds the relationship between Another plausible explanation of why our findings about paternal orphanhood and progress at school. Perhaps fathers differ from those obtained from the ACDIS data is more importantly, they analysed data on children aged not bias in either set of results but that paternal 6–16 years who had been followed up in ACDIS for orphanhood has less impact on children’s schooling in 2.5 years on average. Our analysis included all children the Hlabisa subdistrict than in KwaZulu-Natal as a whole. and young people aged 9–20 years and has a longitudinal Of course, this cannot explain why Case and Ardington component that involves the follow-up of children aged [16] obtained much smaller estimates of the impact of 9–14 years for 6 years. Children’s schooling may suffer paternal orphanhood in the province from the 2001 most when their parent is ill and recover after census data than we do from KIDS. Some insight into the orphanhood [14]. No evidence of this exists for maternal importance of place of residence can be obtained from orphans in Hlabisa [16]. Nevertheless, if the impact of a KIDS, however, by examining how residence interacts father’s death in KwaZulu-Natal emerges only in the with the father variable in the household fixed-effects longer run, it is possible that the difference between the model in Table 2. Fitting this more complex model follow-up periods considered by the two studies could reveals that, although having one’s father living in the explain their contrasting findings. same household was of significant benefit to children’s progress at school in both rural and urban areas, the One problem that might affect the validity of the KIDS adverse impact of paternal orphanhood (but not results is attrition of the panel. Those orphans with the separation) on children’s grade attainment was smaller most disrupted family lives and children from the most in rural households than in urban households (odds ratio impoverished households are probably more prone to loss 0.5, P ¼ 0.04; model not shown). Extrapolating from this to follow-up than other children. Attrition may be less of finding, the educational benefit to a child of co-residence a problem in ACDIS. Children who move within the area with his or her father probably varies between different covered by ACDIS should be picked up in their new rural areas within KwaZulu-Natal. In some disadvantaged household even though children who move outside the areas, perhaps including Hlabisa, fathers may be unable to surveillance area will still be lost to follow-up. The do anything effective to promote their children’s analysis of attrition in KIDS presented in Table 4 is not schooling. In such areas, the death of their fathers might definitive. A subset of maternal orphans who make poor have little impact on children’s progress at school. progress at school could exist that it is so difficult to track that they are largely unrepresented in Table 4. This seems unlikely. It also seems unlikely that KIDS has suffered Conclusion disproportionate attrition of those paternal orphans who are progressing well at school. Therefore, we doubt that The analysis of KIDS shows that fathers matter for the findings from KIDS are invalidated by attrition bias. children’s schooling. Children whose fathers were members of their household were considerably less likely One bias that is probably more severe in both ACDIS and than other children to be behind at school. This is not the 2001 census than in KIDS is the underreporting of because the households of orphans and other children orphanhood. The former data sources lack the historical who were not living with their fathers were poorer than depth that enabled us to identify unreported orphans in those of children whose fathers belonged to the same the KIDS data. According to the 2004 wave of KIDS, household. It is the relationship that children have with 10% of children aged 6–16 years in core and next- their fathers that seems to be important. Our results are generation households had dead mothers, compared with consistent with those of previous research that found that 9% in 2003–2004 in ACDIS, and 21% had dead fathers, maternal orphanhood has some adverse impact on compared with only 15% in ACDIS [16]. Although it is children’s progress at school in KwaZulu-Natal, but do impossible to determine exactly how many children not provide positive evidence in support of this finding. should have been reported as orphans in each study, it is likely that ACDIS has failed to detect a substantial A considerable body of research from other parts of the minority of paternal orphans. Nevertheless, it is not world suggests that fathers (or at least father figures) obvious why the underreporting of paternal orphanhood benefit children as well as mothers [32,33]. Although little
- 97. S92 AIDS 2007, Vol 21 (suppl 7) evidence exists as to whether the contribution of fathering the UK Economic and Social Research Council by is either specific or unique, it plays an important role in means of a PhD studentship awarded to T.B. and a determining the wellbeing, development and educational project grant awarded to I.M.T. (RES-167-25-0076). outcomes of children. More effort thus needs to be made to Conflicts of interest: None. relate other issues pertaining to child wellbeing to that of fatherhood. Men should not be marginalized in research, policy debates and interventions pertaining to the care, socialization and schooling of children. References 1. UNAIDS. 2006 Report on the global HIV/AIDS epidemic. Research into the impact of the AIDS epidemic in Africa Geneva: United Nations Joint Programme on HIV/AIDS; 2006. still lags behind research into other aspects of HIV and 2. UNICEF, UNAIDS, USAID. Children on the brink 2004. AIDS. More is known now than a few years ago about the New York: United Nations Childrens Fund; 2004. 3. South Africa. National HIV and Syphilis Prevalence Survey, impact of orphanhood on children’s education. A handful South Africa, 2005. Pretoria: National Department of Health; of methodologically adequate studies, however, provide a 2006. poor basis for the development of global policies to 4. Dorrington R, Moultrie TA, Timaeus IM. Estimation of mortality using the South African census 2001 data. Monograph 11. Cape benefit orphans and other vulnerable children. Our Town: Centre for Actuarial Research, University of Cape Town; results may not be relevant to other populations. South 2004. Africa differs from most of the rest of Africa in many 5. Hosegood V, Vanneste A-M, Timaeus IM. Levels and causes of adult mortality in rural South Africa: the impact of AIDS. AIDS respects, including near universal school enrolment up to 2004; 18:663–671. the age of 16 years. Our findings caution against drawing 6. Kamali A, Seeley JA, Nunn AJ, Kengeya-Kayondo JF, Ruberantwari A, Mulder DW. The orphan problem: experience general conclusions about the impact of the AIDS of a sub-Saharan Africa rural population in the AIDS epidemic. epidemic from investigations in a few geographically AIDS Care 1996; 8:509–516. localized populations. If two studies in the same province 7. Urassa M, Boerma J, Isingo R, Ngalula J, Ng’weshemi J. The impact of HIV/AIDS on mortality and household mobility in of South Africa disagree as to whether the death of the rural Tanzania. AIDS 2001; 15:2017–2023. father, as well as the mother, impacts directly on the 8. Case A, Paxson C, Ableidinger J. Orphans in Africa: parental progress children make in their schooling, we lack the death, poverty and school enrollment. Demography 2004; 41:483–508. knowledge needed to prescribe solutions for Africa as 9. Monasch R, Boerma JT. Orphanhood and childcare patterns in a whole. sub-Saharan Africa: an analysis of national surveys from 40 countries. AIDS 2004; 18 (suppl 1):S55–S65. 10. Bicego G, Rutstein S, Johnson K. Dimensions of the emerging orphan crisis in sub-Saharan Africa. Soc Sci Med 2003; Acknowledgements 56:1235–1247. 11. Ainsworth M, Filmer D. Poverty, AIDS and children’s schooling: a targeting dilemma. World Bank Policy Research Paper. The authors are grateful to the individuals and Washington, DC: The World Bank; 2002. communities who agreed to be interviewed in the 12. Evans DK, Miguel E. Orphans and schooling in Africa: a long- KwaZulu-Natal Income Dynamics Study (KIDS). The itudinal analysis. Demography 2007; 44:35–57. 13. Yamano T, Jayne TS. Working-age adult mortality and primary paper has benefited from the comments of Julian May, school attendance in rural Kenya. Econ Dev Cult Change 2005; Stuart Gillespie and other participants at the HEARD/ 53:619–654. UNAIDS research symposium for investigating the 14. Ainsworth M, Beegle K, Koda G. The impact of adult mortality and parental deaths on primary schooling in northwestern empirical evidence for understanding vulnerability and Tanzania. J Dev Stud 2005; 41:412–439. the associations between poverty, HIV infection and 15. Nyamukapa C, Gregson S. Extended family’s and women’s AIDS impact, Durban, South Africa, 16–17 October, roles in safeguarding orphans’ education in AIDS-afflicted rural Zimbabwe. Soc Sci Med 2005; 60:2155–2167. 2006, and from discussions with Cally Ardington and 16. Case A, Ardington C. The impact of parental death on school Anne Case. The analysis remains the responsibility of the outcomes: longitudinal evidence from South Africa. Demogra- authors alone and the views expressed in the paper are not phy 2006; 43:401–420. 17. South African Labour Development Research Unit (SALDRU). necessarily those of the individuals and organizations South Africans rich and poor: baseline household statistics. Cape mentioned already or of UNESCO. Town: SALDRU, University of Cape Town; 1994. 18. Grosh M, Munoz J. A manual for planning and implementing the LSMS survey. Washington, DC: The World Bank; 1996. Sponsorship: The 2004 wave of KIDS was a collabora- 19. May J, Carter MR, Haddad L, Maluccio J. A KwaZulu-Natal tive project between researchers at the University of Income Dynamics Study (KIDS): 1993–1998 a longitudinal KwaZulu-Natal, the University of Wisconsin, the household data set for South African policy analysis. Dev S London School of Hygiene and Tropical Medicine, Afr 2000; 17:567–581. the International Food Policy Research Institute and 20. May JD, Aguero J, Carter MR, Timaeus IM. The KwaZulu-Natal Income Dynamics Study (KIDS) third wave: methods, first the South African Department of Social Development. findings and an agenda for future research. Dev S Afr 2007; In addition to the contributions of these institutions, 24:629–648. the following provided financial support: the UK 21. Maluccio J. Attrition in the KwaZulu-Natal Income Dynamics Department for International Development, USAID, Study 1993–1998. Washington, DC: International Food Policy Research Institute; 2000. the Mellon Foundation, and the National Research 22. Boler T. Facing the consequences of AIDS: orphans, educational Foundation/Norwegian Research Council. The analysis outcomes and cash grants in South Africa (PhD). London: of KIDS data reported here was partly supported by University of London; 2007.
- 98. Progress at schoolof orphans Timaeus and Boler S93 23. Hoogeveen J, Ozler B. Not separate, not equal: poverty and 28. Dorrington R, Bourne D, Bradshaw D, Laubscher R, Timaeus inequality in post-apartheid South Africa. William Davidson IM. The impact of HIV/AIDS on adult mortality in South Africa. Institute working paper no. 739. Ann Arbor: University of Burden of Disease Unit Technical Paper. Tygerberg: South Michigan; 2005. African Medical Research Council; 2001. 29. Dorrington RE, Johnson LF, Bradshaw D, Daniel T. The demo- 24. Hu FB, Goldberg J, Hedeker D, Flay BR, Pentz MA. Comparison graphic impact of HIV/AIDS in South Africa: national and of population-averaged and subject-specific approaches for provincial indicators for 2006. Cape Town: Centre for Actuarial analyzing repeated binary outcomes. Am J Epidemiol 1998; Research, South African Medical Research Council and Actuar- 147:694–703. ial Society of Southern Africa; 2006. 25. Hosegood V, McGrath N, Herbst K, Timaeus IM. The impact of 30. Engle P, Breaux C. Fathers’ involvement with children: per- adult mortality on household dissolution and migration in rural spectives from developing countries. Society for Research in South Africa. AIDS 2004; 18:1585–1590. Child Development Social Policy Report 1998; 12:1–23. 31. Hunter M. Fathers without amandla: Zulu-speaking men and 26. Facing the future together: report of the Secretary General’s task fatherhood. In: Richter L, Morrell R, editors. Baba: men and force on women, girls and HIV/AIDS in Southern Africa. Johan- fatherhood in South Africa. Cape Town: HSRC Press; 2006. nesburg: UNAIDS Regional Support Team, East and Southern 32. Richter L. The importance of fathering for children. In: Richter L, Africa; 2004. Morrell R, editors. Baba: men and fatherhood in South Africa. 27. Unterhalter E. Gender equality and education in South Africa: Cape Town: HSRC Press; 2006. measurements, scores and strategies. In: Chisholm L, September 33. Tamis-LeMonda CS, Cabrera N, editors. Handbook of father J, editors. Gender equity in South African education 1994– involvement: multidisciplinary perspectives. Mahwah, NJ: 2004. Cape Town: HSRC Press; 2005. Lawrence Erlbaum Associates; 2002.
- 100. Exploring the Cinderellamyth: intrahousehold differences in child wellbeing between orphans and non-orphans in Amajuba District, South Africa Anokhi Parikha, Mary Bachman DeSilvab, Mandisa Cakwea, Tim Quinlana, Jonathon L. Simonb, Anne Skalickyb and Tom Zhuwaua Objective: To determine whether differences in wellbeing (defined by a variety of education and health outcomes) exist between recent school-aged orphans and non- orphans who live in the same household in a context of high HIV/AIDS mortality in KwaZulu Natal, South Africa. Design: The data come from the first 2 years (2004–2006) of an ongoing 3-year longitudinal cohort study in a district in KwaZulu-Natal, the Amajuba Child Health and Well-being Research Project. Using stratified cluster sampling based on school and age, we constructed a cohort of 197 recent orphans and 528 non-orphans aged 9–16 years and their households and caregivers. Household heads, caregivers, and children were interviewed regarding five domains of child wellbeing: demographic, economic, educational, health/nutrition/lifestyle, and psychosocial status. Methods: The analytical sample consists of 174 children (87 orphans and 87 com- parable non-orphans who live together) at baseline and 124 children in round 2. We estimated a linear regression model using household fixed effects for continuous outcomes (grade adjusted for age, annual expenditure on schooling and body mass index) and a logit model using household fixed effects for categorical variables (malnutrition) to compare co-resident orphans and non-orphans. Results: We found no statistically significant differences in most education, health and labour outcomes between orphans and the non-orphans with whom they live. Paternal orphans are more likely to be behind in school, and recent mobility has a positive effect on schooling outcomes. ß 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins AIDS 2007, 21 (suppl 7):S95–S103 Keywords: caregivers, children, HIV, intrahousehold, mortality, orphans, South Africa Introduction currently 1.2 million AIDS orphans in South Africa [2], and this number is expected to peak at a staggering 2.3 With an antenatal seroprevalence of 40.7% in 2005, million in 2015 [3]. As such, it is no surprise that the issue KwaZulu-Natal has the highest HIV prevalence of any of orphaning has attracted significant attention. province in South Africa, a country with 5.1 million individuals infected [1,2]. AIDS-related mortality is high, Much of this attention has characterized orphans as and the consequent impact on orphaning is likely to children who are growing up without the care and be dramatic in the years to come, irrespective of the support of their families, who have poorer learning and expansion of the antiretroviral programme. There are knowledge levels, and who are suffering from the From the aHealth Economics and AIDS Research Division (HEARD), University of KwaZulu Natal, Durban, South Africa, and the b Boston University School of Public Health, Center for International Health and Development, Boston, Massachusetts, USA. Correspondence to Anokhi Parikh, HEARD, Private Bag X 54001, Durban, 4000, South Africa. E-mail: anokhip@gmail.com ISSN 0269-9370 Q 2007 Wolters Kluwer Health | Lippincott Williams & Wilkins S95
- 101. S96 AIDS 2007, Vol 21 (suppl 7) ‘absence of adults in their socialization’ [4]. The data national 2003 General Household Survey, approximately reveal that most orphans, defined in the AIDS literature as two-thirds of all orphans in KwaZulu-Natal live in such having one or more parents dead, in sub-Saharan Africa households. These households are of particular interest have at least one parent living and either live with their also because the co-resident non-orphans make a natural surviving parent or are absorbed into other families where comparison group for orphans as they both share they have some adult supervision. household characteristics (e.g. household income, literacy of caregiver etc.). Nevertheless, the death of a parent (and the potentially long illness preceding it if the parent has died of AIDS) In one of the few studies to make intrahousehold may have various impacts on a child’s wellbeing. comparisons, Case and Ardington [15] showed that in Numerous studies have accordingly examined the impact Hlabisa district in KwaZulu-Natal, maternal orphans of orphaning on children, but have focused primarily on were more likely to be behind in school and had less spent educational outcomes (commonly defined by school on their schooling, but were equally likely to be enrolled enrolment), with few studies looking at any other aspect as the non-orphans with whom they lived. Implicit in the of wellbeing such as health or labour outcomes. finding that orphans are worse off when compared with other children is the notion that caregivers may prioritize Although the empirical evidence is mixed and depends their own children over the fostered child, a Cinderella on the location, data sources, and methods used, two approach, so to speak. For example, when facing income broad themes emerge in the literature: that the impact of constraints, a caregiver might spend less on the fostered orphaning depends often on which parent dies and that child than on his/her own child (although perhaps with income is often a greater predictor of outcomes than less malice than Cinderella’s stepmother). Gender bias in orphan status. Case et al. [5] used the Demographic and educational expenditures has been well documented in Health Survey data from 19 countries in Africa and found Asia; a similar bias could also apply to orphans, but very systematic differences in school attendance between little research has examined the intrahousehold aspect of orphans and non-orphans. The finding of difference in this question to date. attendance and enrolment between orphans and non- orphans has been supported by other studies in rural In summary, the research to date is limited in two ways. Kenya through longitudinal studies [6,7], and in cross- First, few studies explicitly assess intrahousehold differ- sectional studies elsewhere in sub-Saharan Africa [8]. ences between orphans and non-orphans. Second, both Beegle et al. [9], using a panel that follows children for intra and interhousehold comparisons have used limited 13 years, showed that maternal orphanhood was indicators such as school enrolment, grade progression, associated with lower educational attainment and health and height, and thus do not provide a comprehensive (as measured by height) in the long term, and paternal picture of other important aspects of child wellbeing. orphanhood was associated with lower educational attainment for certain groups only. Using Demographic To respond to these shortcomings, this paper compares and Health Survey data and other nationally representa- the differences in wellbeing between orphans and non- tive household surveys from 51 countries in Africa, orphans who live with each other using longitudinal data Ainsworth and Filmer [10] highlighted the variation in from 2004–2006 from Amajuba District in KwaZulu- orphan/non-orphan differentials across countries and Natal. This paper addresses the question ‘do orphans and found that income plays a greater role in determining non-orphans living within the same household fare school enrolment than orphaning. differently in terms of wellbeing?’ We define ‘wellbeing’ to include education, health, and labour outcomes. We A lack of statistically significant differences in enrolment conclude with potential explanations for the observed have been found by other researchers conducting results and trends. longitudinal studies in East Africa [11–13]. Chatterji et al. [12], using longitudinal data from Rwanda, showed no differences in school enrolment and food intake between orphans and non-orphans. Adato et al. [14] Methods found no statistically significant differences in schooling indicators but qualitatively did find some cases of Study area discrimination towards orphans within the household. The Amajuba district was chosen because it included a broad cross-section of urban, peri-urban and rural areas. Many of these studies consider orphans and non-orphans The district has approximately 470 000 inhabitants and is in general, but do not distinguish between orphans who poor. The economy used to be driven by the coal mining live with other orphans and orphans who live with non- industry, but the closure of coal mines has led to high orphans in mixed households (with some exceptions) unemployment in the region and consequently high rates [6,7,15]. Moreover, they are an important type of of migration. Additional details are provided elsewhere household, as according to the Statistics South Africa’s (Bachman DeSilva et al., manuscript in preparation).
- 102. Wellbeing of co-residentorphans and non-orphans Parikh et al. S97 Data and sample selection a result of a change of caregiver, return to living with The data come from the first 2 years of an ongoing 3-year parents, or move for family reasons. The movement of cohort study that commenced in 2004. The study was one child led to the exclusion of the comparison child designed, and the data collected by the Amajuba Child from the analysis. Health and Well-being Research Project, a joint initiative between the Health Economics and HIV/AIDS Analytical methods Research Division (HEARD) at the University of This paper examines the education, health and labour KwaZulu-Natal and the Center for International Health outcomes of orphans compared with the non-orphans and Development at the Boston University School of with whom they live. For education, we assessed two Public Health. The Institutional Review Board of Boston indicators: grade normalized for age, and annual University Medical Center and the Ethics Committee of expenditures on schooling for the child. For physical the University of KwaZulu-Natal provided ethical health, we examined body mass index (BMI), which was approval for the study. calculated then translated into z-scores and percentiles using the United States Centers for Disease Control and The annual survey has four components: a household and Prevention age and sex-specific reference curves. As demographic information questionnaire administered to malnutrition often manifests itself as obesity, analysing the household; a questionnaire for the primary caregiver BMI as a continuous variable can be misleading. We of the study child; and two questionnaires administered therefore looked at malnourishment as a categorical to the study child, one on general wellbeing including variable in which malnourishment was defined, in self-reported health, educational attainment, and the use accordance with Centers for Disease Control and of time, and a second that assesses the self-reported Prevention definitions, as being in the bottom 5 psychosocial wellbeing of children. percentile or the top 5 percentile of the age and sex- specific BMI distribution. For labour outcomes we Sample selection took place using randomized stratified examined the categorical variables: whether the child had cluster sampling from 60 of 252 schools in the district. worked outside the house in the last week and whether The study population were predominantly Zulu-speak- the child had done chores within the house in the ing children aged 9–16 years, resident in the district and last week. attending school at the time of sampling. Only ‘recent orphans’, defined as children who had lost one or both Bivariate relationships between child type and demo- parents to any cause during a 6-month period between graphic variables were assessed using Mantel–Haenszel March and August 2004 were included. This was done in chi-square tests for categorical variables and t-tests for order to capture the incidence of orphaning and measure continuous outcomes. For the multivariate analysis, we the exposure period of parent death. Three comparison estimated a linear regression model using household fixed non-orphan children were randomly selected from the effects. The household fixed effect allows for the same school, grade, and age as index-orphan children. In comparison of children (with different characteristics) households in which there were both orphan and non- within households by controlling for all observed and orphan children, a secondary comparison child was unobserved child invariant household characteristics such selected in the same age range as that household’s primary as income, assets, household size, distance to school, etc. study child. A cohort of secondary comparison children As co-resident orphans and non-orphans will have the was thus also constructed in order to investigate same household characteristics, this method allows us to intrahousehold differences in children’s wellbeing. identify the within-household differences between orphans and non-orphans. Study households were classified into three groups: orphan-only, non-orphan-only, and mixed households. For continuous variables, we estimated the following The overall baseline sample includes 50 orphans from the linear model: 50 orphan-only households, 377 non-orphans from 377 non-orphan-only households, and 298 children from the Yijt ¼ b1 maleij þ b2 ageijt þ b3 mobilityijt 210 mixed households (a total of 725 children). Of the 210 mixed households, 87 had non-orphans of com- þ b4 orphan typeijt þ b5 resident parentijt parable age. The analytical sample used for the current þ Hj þ eijt analysis thus consists of 174 children at baseline, 87 orphans and 87 non-orphans who live together in mixed households; and 124 children, 62 orphans and 62 non- where Yijt represents the educational attainment/BMI for orphans, in the second round. Of the 25 pairs that child i from household j at time t; male is an indicator dropped out in round 2, six dropped out as the variable for whether child is male or not; age is the age of comparison non-orphans were orphaned (therefore the the child; mobility is an array of categorical variables household no longer remained mixed); the remaining 19 reflecting when the child moved into the house; orphan pairs had at least one child moved to another household as type is the type of orphan (maternal orphan, paternal
- 103. S98 AIDS 2007, Vol 21 (suppl 7) orphan, double orphan); resident parent is whether the mately half the sample had repeated a grade once. Average surviving parent of the orphan or the parent of the non- body mass was within normal range and was comparable orphan is living at home (father lives at home, mother for orphans and non-orphans. Approximately 10% of the lives at home). Hj is the child invariant household fixed children work outside the house and 91% report assisting effect; and eijt is the error term. Several models were with chores in the household (with no differences estimated using the different independent variables in between orphans and non-orphans). Although mobility different combinations, but results from only one such of the sample was high, it was equally high for orphans model are presented in this paper. and non-orphans, with approximately 30% of the sample having moved at least once and approximately 12% in the For categorical variables, we estimated a logit model: past 2 years. Living arrangements were different, however, with a parent being the primary caregiver for only 39% of PðYijt ¼ 1Þ the non-orphans and 12% of the orphans. Grandparents ¼ Fðb1 maleij þ b2 ageijt þ b3 mobilityijt were primary caregivers for 56% of the orphans and 43% of non-orphans, although this difference was not þ b4 orphan typeijt þ b5 resident parentijt statistically significant. These findings of non-difference remained unchanged in round 2 (results not shown). þ Hj þ eijt Þ Table 2 shows the demographic characteristics of the where Yij represents whether child i from household j at time sample that was lost in round 2. None of the t is malnourished or not/works at home or not/does chores characteristics listed were significant predictors of at home or not; and the remaining variables are as defined attrition (results not shown). above. Of the orphans at baseline, 13 were maternal-only The decision to include the variables on mobility and orphans, 30 were paternal-only orphans, 26 were double resident parent was informed by the literature. This orphans, and 19 had missing information on which parent literature shows that fostering and mobility are high even had died. This changed to nine maternal-only orphans, for all kinds of children in South Africa because of high 21 paternal-only orphans, 21 double orphans, and 12 levels of adult migration and children born out of with missing data in the following year. Table 3 wedlock [14,16,17]. We thus feel it is important not only summarizes the demographic and socioeconomic charac- to introduce these variables as controls. teristics of children by orphan type. Examining children who are co-resident using household As mentioned in the methodology section, several fixed effects allows us to control for common household models, each controlling for a different combination of characteristics. It is impossible to know, however, with independent variables, were estimated to examine the these data, whether the children being compared were orphan/non-orphan differentials across different out- indeed comparable before the death of the parent because comes. Table 4 shows results from two such models. The we do not have information on the orphan’s character- results were consistent across specifications: paternal istics before being orphaned (which may have themselves orphans are more likely to be behind in school than non- been affected by orphaning). We are also unable to orphans with whom they live, they are on average a third consider the fixed and unobserved characteristics of the of a year behind in their grade. Maternal orphans are on child him/herself, even though this is longitudinal data, as average half a year behind in schooling, but this effect is most of the variables of interest have remained constant statistically insignificant. Mobility within the past 2 years over time. As a result of this limitation of the data, this is seen to have a positive effect on grade progression. paper does not try and isolate the impact of parental death Orphanhood does not seem to have any effect on on children. Rather, it compares orphans and non- expenditures on schooling. Recent mobility is associated orphans on a range of indicators and tries to identify some with a substantial negative effect on schooling expendi- causal pathways for the results. ture. The impact of parents being present at home is insignificant (results not shown). Bivariate analysis demonstrates no significant differences Results in nutrition, health proxy, and labour outcome indicators such as going to bed hungry the previous night, being sick At baseline, bivariate analysis of sociodemographic in the past 6 months, and working both within and characteristics shows few differences between the 87 outside the house (see Table 1). Analysis of BMI, orphans and 87 non-orphans in the analytical sample presented in Table 4, shows that BMI is lower for orphans. (Table 1). There are no statistically significant differences This is also robust when controlling for mobility. As between the demographic, educational, health, or labour malnutrition also manifests itself as obesity, however, outcomes between orphans and co-resident non- lower BMI is not necessarily informative, especially in orphans. Whereas attendance was near 100%, approxi- adolescents. Logistic analysis (Table 5) shows that
- 104. Wellbeing of co-residentorphans and non-orphans Parikh et al. S99 Table 1. Demographic characteristics of orphans and non-orphans at baseline (2004–2005). Non-orphans (N ¼ 87) Orphans (N ¼ 87) Characteristic % (N) Mean (SD) % (N) Mean (SD) Sex (male) 55.2 (48) 55.2 (48) Age (years) 12.3 (1.9) 12.2 (1.8) Education % of children attending school 98.8 (86) 100 (87) Mean current grade 6.3 (2.0) 6.0 (2.0) % of children who have repeated a grade at least once 53.5 (47) 51.7 (45) Mean number of times grade has been repeated 1.1 (1.1) 1.3 (1.3) Expenditures on schooling (SA Rand, per year) 392.7 (307.8) 357.5 (206.5) Health Mean body mass index 18.7 (3.6) 18.0 (3.1) % of children who were sick in the past 6 months 28.7 (25) 32.2 (28) % of children who ate breakfast this morning 86.2 (75) 81.6 (71) Labour % of children who worked outside the house last week 9.2 (8) 10.3 (9) Mean number of hours worked last week 1.9 (0.6) 1.4 (0.5) % of children who did chores in the house last week 93.1 (81) 89.5 (78) Mean number of hours of chores last week 1.6 (0.5) 1.6 (0.5) Mobility % of children who have never changed residence 71.3 (62) 70.1 (61) % of children who changed residence in the past 2 years 11.5 (10) 14.9 (13) % of children who changed residence 2–5 years ago 6.9 (6) 4.6 (4) % of children who changed residence over 5 years ago 10.3 (9) 10.3 (9) Living arrangements % of children whose primary caregiver is their parenta 39.0 (34) 12.6 (11) % of children whose primary caregiver is their grandparent 43.7 (38) 56.3 (49) % of children whose mother lives at homeb 67.5 (59) 50 (44) % of children whose father lives at home 30.4 (26) 27.8 (24) SA, South African. a Significant at 1%. b Significant at 5%. Table 2. Characteristics of the 25 children lost to attrition in round 2. Characteristics % (N) Mean (SD) Sex (male) 62.5 (15) Age (years) 12.9 (1.7) Orphan status Non-orphan 40 (10) Maternal-only orphan 4 (1) Paternal-only orphan 20 (5) Double orphan 16 (4) Orphans with missing parent death data 16 (4) Education % of children attending school 95.8 (23) Mean current grade 6.7 (2.0) % of children who have repeated a grade at least once 62.5 (15) Mean number of times grade has been repeated 1.2 (0.4) Expenditures on schooling (SA Rand, per year) 339.5 (174.0) Health Mean body mass index 18.4 (2.5) Mobility and living arrangements % of children who have never changed residence 58.3 (14) % of children who changed residence in the past 2 years 25 (6) % of children who changed residence 2–5 years ago 8.33 (2) % of children who changed residence over 5 years ago 8.33 (2) % of children whose primary caregiver is their parent 20.8 (5) % of children whose primary caregiver is their grandparent 54.2 (13) % of children whose mother lives at home 53.3 (13) % of children whose father lives at home 27.3 (7) SA, South African.
- 105. S100 AIDS 2007, Vol 21 (suppl 7) Table 3. Demographic characteristics of maternal, paternal and double orphans at baseline (2004–2005). Maternal-only orphans (N ¼ 13) Paternal-only orphans (N ¼ 36) Double orphans (N ¼ 26) % (N) Mean (SD) % (N) Mean (SD) % (N) Mean (SD) Sex (male) 53.9 (7) 70 (25) 42.3 (11) Age (years) 12.7 (2.0) 12.2 (1.7) 12.2 (1.9) Education % of children attending school 100 (13) 100 (36) 100 (26) Mean current grade 5.7 (2.1) 6.2 (1.8) 6.1 (2.3) % of children who have repeated a grade at least once 41.7 (5) 51.8 (19) 33.3 (9) Expenditures on schooling (SA Rand, per year) 336.3 (127.8) 363.4 (158.7) 344.3 (309.2) Health Mean body mass index 19.3 (4.9) 17.7 (2.2) 17.8 (3.4) % of children who were sick in the past 6 months 38.5 (5) 23.3 (8) 38.5 (10) % of children who ate breakfast this morning 84.6 (11) 90 (32) 76.9 (20) Labour % of children who worked outside the house last week 7.7 (1) 3.3 (1) 7.7 (2) % of children who did chores in the house last week 100 (13) 86.2 (31) 76.9 (20) Mean number of hours of chores last week 1.5 (0.5) 1.6 (0.6) 1.6 (0.5) Mobility and living arrangements % of children who have never changed residence 61.5 (8) 86.7 (31) 69.2 (1.8) % of children who changed residence in the past 2 years 7.7 (1) 5.5 (2) 26.9 (7) % of children who changed residence over 2 years ago 23.1 (3) 5.5 (2) 3.8 (1) % of children whose primary caregiver is their parent 0 (0) 1.7 (6) % of children whose primary caregiver is their grandparent 69.2 (9) 43.3 (16) 65.4 (17) % of children whose mother lives at home 50 (18) % of children whose father lives at home 23.0 (3) SA, South African.
- 106. Wellbeing of co-residentorphans and non-orphans Parikh et al. S101 Table 4. Educational and health outcomes for orphans and co-resident non-orphans (household fixed effects). Grade normalized Grade normalized Annual expenditure Annual expenditure Body mass for age for age on school on school index Age (in years) À0.06b À0.07b À2.18 À1.59 0.77b (0.01) (0.01) (38.25) (37.40) (0.10) Sex (male) À0.02b À0.02b 23.19b 25.75b À3.11b (0.00) (0.00) (8.74) (8.46) (0.45) Changed residence in past 2 years 0.05b À170.11b (0.02) (55.93) Changed residence more than 2 years ago À0.03 106.43a (0.02) (48.28) Maternal-only orphan À0.04 À0.03 À11.14 À54.29 1.5a (0.02) (0.02) (60.97) (60.74) (0.73) Paternal-only orphan À0.04a À0.04b À5.15 6.53 À1.12a (0.01) (0.01) (40.88) (40.25) (0.48) Double orphan 0 À0.01 À61.42 À27.76 À1.71b (0.01) (0.01) (42.74) (42.27) (0.50) Constant 1.24b 1.25b 120.89 84.97 11.29b (0.04) (0.04) (113.95) (110.35) (1.32) Observations 296 296 298 298 298 Number of field code 87 87 87 87 87 R-squared 0.28 0.32 0.04 0.12 0.42 a Significant at 5%. b Significant at 1%. Standard errors in brackets. Variable with missing orphan type included but not shown. maternal and double orphans are at greater odds of being coefficient on the father being a resident within the malnourished than non-orphans but this is not statistically household is insignificant (results not shown), suggesting significant. Maternal and paternal orphans are at greater that the impact of paternal orphanhood may be caused by odds of doing chores within the house and at lower the fact that the death of a father could be an economic odds of working outside the house compared with co- shock that may have, at some point, resulted in children resident non-orphans, but again these differences are not dropping out of school. We, unfortunately, have no way statistically significant. to test this hypothesis and can only offer it as a potential explanation. What could explain the lack of overall differences Discussion between orphans and non-orphans? First, temporal issues may be driving the result. It is important to remember The results show some statistical differences in edu- that we are merely looking at incident (recent) orphans, cational outcomes and no differences for health and i.e. children that have lost at least one of their parents in labour outcomes between orphans and non-orphans who the 6 months before the survey and in the same year the live in the same households. survey was conducted. This may not be sufficient time to see a large effect on children, or the households may have Case and Ardington [15], in their intrahousehold analysis effective short-term coping mechanisms to mitigate the also set in KwaZulu Natal (albeit in a poorer district), effect on children [9]. On the other hand, one can argue found that maternal orphans are ‘on average, 0.12 of a that the critical period for an orphan child is the terminal year behind in their schooling and have 7% less spent on illness period as a result of the trauma of seeing a parent their education’ compared with the non-orphans with wasting away and sometimes having to miss school in whom they live. Although the differences in expenditure order to attend to sick parent(s), and that the impacts may are moderate, the magnitude of the difference between diminish over time. orphans and non-orphans in terms of schooling is small: 0.12 of a year behind equates to orphans being behind by Second, Case and Ardington [15] have argued that a little over a month. They found no difference for differences in outcomes between orphans and non- paternal orphans. Our results, on the other hand, show orphans are driven by the tendency to live with distant or that paternal orphans are behind in school. unrelated caregivers. All the orphans in the sample are either living with close relatives (aunt or grandmother) or Given the fact that a large proportion of South African their surviving parent, as in a study from the same fathers are absent or not linked to the household [16], the province by Adato et al. [14]. Therefore, they are likely to significant effect of paternal orphanhood is curious. The receive the same support as the non-orphans with whom
- 107. S102 AIDS 2007, Vol 21 (suppl 7) Table 5. Health and labour outcomes for orphans and co-resident non-orphans (logistic model with household fixed effects). Child is malnourished Child works outside the house Child does household chores Odds ratio 95% CI Odds ratio 95% CI Odds ratio 95% CI Age (in years) À0.29 (À0.81–0.23) 0.01 (À0.31–0.32) À0.14 (À0.46–0.18) Sex (male) 0.01 (À1.50–1.52) 0.11 (À1.37–1.59) À1.54 (À3.33–0.24) Maternal-only orphan 0.48 (À1.88–2.83) À0.24 (À2.28–1.81) 0.69 (À1.77–3.15) Paternal-only orphan À0.18 (À2.29–1.92) À0.03 (À1.41–1.35) 0.25 (À1.22–1.72) Double orphan 0.56 (À1.13–2.24) À0.72 (À2.41–0.96) À0.95 (À2.61–0.72) Observationsa 54 86 96 Number of groups 16 24 26 CI, Confidence interval. aMultiple positive/negative outcomes were encountered within groups and thus the groups were dropped from the regression resulting in fewer numbers of observations. Standard errors in brackets. Variable with missing orphan type included but not shown. they live. With time, destination households may become moved houses in the preceding year, 13 orphans and 10 oversaturated and could struggle to absorb more children non-orphans. Moreover, in round 2, sample attrition was and this may change. South Africa’s extensive social grants equally high for orphans and non-orphans. Therefore, system potentially mitigates against this phenomenon and although there may be endogeneity in placement assist families in coping. decisions, we do not believe it is disproportionately so for orphans when compared with non-orphans. Child Third, in Amajuba District’s context of high adult migration is a historical/cultural phenomenon, and migration, having a parent alive does not equate to the fostering literature shows how children have lived away presence of a parent at home, thus orphanhood itself may from their ‘nuclear’ families (although this may be not be associated with lower educational or health exacerbated by AIDS mortality) [17]. outcomes. The majority of both orphans and non- orphans live without parents present at home. Table 1 Fifth, it is possible that the indicators used may not be indicates that only 38.64% of non-orphans have parents as sensitive to differences, particularly because the orphans primary caregivers, and a large proportion of both were so recently orphaned. There may be some orphans and non-orphans live with their grandparents. limitations of BMI, but in general it is difficult to Even single parent orphans tend not to live with their identify good measures of health of older/school-aged surviving parent. Migration for employment was the most children because this age group is generally very healthy frequently cited reason for parents’ not living at home. (self-reported or otherwise). In terms of schooling, there Furthermore, approximately a third of the fathers were maybe differences in performance within a grade that are not living at home because they were not married to the not captured by these instruments. Whether an orphan mother. This figure also calls into question the role of child is truly learning, as opposed to progressing, like biological parents (especially fathers) in caregiving, and non-orphaned children, may not be fully captured by supports other literature that shows that the absence of the data. fathers is high in South Africa, with 55% of fathers being absent in rural South Africa in 2002 [16]. Current The study has some limitations worth mentioning that definitions of orphan inaccurately privilege the biological may bias the results towards the null. First, the tests have parent in a context in which even non-orphans do not low power as a result of the relatively small sample size, live with their parents. This calls into question our and this may contribute to not finding statistically thinking on the category of orphan in South Africa. significant effects. The attrition in round 2 only further reduced the sample. Comparisons with much larger Fourth, when orphans have moved from their original studies should be made with this in mind. households (i.e. they were fostered into the survey household, which has non-orphans), there may be Second, the study sample was drawn from a random endogeneity in placement decisions, in that orphans sample of schools in the district. Using schools as our sole are strategically moved to better-off households and this recruitment source for study participants was both may bias the results towards the null. The positive and methodological and practical. Drawing a sample of significant coefficient on recent mobility (within the past school-going ‘recent’ orphans and non-orphans intro- 2 years) supports the idea that children are often moved duces a sampling bias that potentially biases the for schooling. It is important to note, however, that intrahousehold results towards the null as worse-off mobility is equally high for both orphans and non- orphans may have been excluded from the sample. What orphans. Table 1 shows how there is no statistically is important to note is that school enrolment rates in significant difference between orphans’ and non-orphans’ KwaZulu-Natal are extremely high. The national 2003 mobility. Of the 174 children at baseline, 23 children General Household Survey conducted by Statistics South
- 108. Wellbeing of co-residentorphans and non-orphans Parikh et al. S103 Africa showed school enrolment rates to be 97.5% for United States National Institutes of Health under its non-orphans and 95% for orphans. Upon further analysis African Partnerships programme (grant R29 of this survey, we found orphans who are not enrolled HD43629). neither worse off than the non-orphans with whom they Conflicts of interest: None. live nor are they are worse off compared with enrolled orphans in terms of health and labour outcomes. Their schooling outcomes do differ, however, and this may bias References the results towards the null. 1. Department of Health. National HIV and Syphilis Antenatal Sero-Prevalence Survey in South Africa 2005. Pretoria: National Finally, Evans and Miguel [6] argued that studies that do Department of Health, South Africa; 2006. not take into consideration child fixed effects are likely to 2. UNAIDS. 2006 Report on the global AIDS pandemic. Geneva: UNAIDS; 2006. be biased to the null because they are unable to account 3. Dorrington R, Bradshaw D, Johnson L, Budlender D. The for omitted variable bias as well as endogeneity because demographic impact of HIV/AIDS in South Africa: national we do not have information on the child’s characteristics indicators for 2004. Cape Town: Centre for Actuarial Studies, South Africa Medical Research Council, Actuarial Society of before orphaning. The results therefore serve as a lower South Africa; 2004. bound on the differences between orphans and non- 4. Meintjes H, Giese S. Spinning the epidemic: the making of orphans. mythologies of orphanhood in the context of AIDS. Childhood 2006; 13:407–430. 5. Case A, Paxson C, Ableidinger J. Orphans in Africa: parental death, poverty and school enrolment. Demography 2004; 41:483–508. 6. Evans D, Miguel E. Orphans and schooling in Africa: a long- Conclusion itudinal analysis. Demography 2007; 44:35–57. 7. Yamano T, Jayne TS. Working-age adult mortality and primary We have shown a lack of systematic difference in school attendance in rural Kenya. Econ Dev Cultural Change 2005; 53:619–653. education and health outcomes between orphans and co- 8. Bicego G, Rutstein S, Johnson K. Dimensions of the emerging resident non-orphans, with a few exceptions. Policy orphan crisis in sub-Saharan Africa. Soc Sci Med 2003; responses (and literature) do not always distinguish 56:1235–1247. 9. Beegle K, Weerdt JD, Dercon S. Orphanhood and the long-run between orphans living with other orphans or orphans impact on children. Oxford: Centre for Studies of African living in mixed households, even though they may live in Economies; 2006. completely different circumstances. Comparing the 10. Ainsworth M, Filmer D. Inequalities in children’s schooling: AIDS, orphanhood, poverty and gender. World Dev 2006; results presented in this paper with the results in the 34:1099–1128. literature reveals the heterogeneity of the category of 11. Lloyd CB, Blanc AK. Children’s schooling in sub-Saharan orphans. The contradiction of the results of Case and Africa: the role of fathers, mothers and others. Popul Dev Rev 1996; 22:265–298. Ardington [15] highlights the need to refrain from 12. Chatterji M, Dougherty L, Ventimiglia T, Mulenga Y, Jones A, generalizing to the country based on district or even Mukaneza A, et al. The well-being of children affected by HIV/ province level results. The paper thus illustrates the need AIDS in Lusaka, Zambia and Gitarama Province, Rwanda: findings from a study. Working paper no. 2. Washington, DC: for context-specific approaches that pay attention to Community REACH; 2005. definitions, as opposed to sweeping global responses to 13. Crampin A, Floyd S, Clynn J, Madise N, Nyondo A, Khondwe M, the crisis. The remarkable similarity of orphans’ and et al. The long-term impact of HIV and orphanhood on the mortality and physical well-being of children in rural Malawi. non-orphans’ living arrangement calls into question the AIDS 2003; 17:389–397. category of orphan in South Africa. Evidence from 14. Adato M, Kadiyala S, Roopnaraine T, Biermayr-Jenzano P, Norman A. Children in the shadow of AIDS: studies of vulner- the study thus far challenges the Cinderella assertion in able children and orphans in three provinces in South Africa. the case of orphans in South Africa. Unstable households Washington, DC: IFPRI; 2005. with absent fathers are common and mobility is high, as 15. Case A, Ardington C. The impact of parental death on school outcomes: longitudinal evidence from South Africa. Demogra- also suggested by other literature. Reframing the phy 2006; 43:401–420. discussion around orphanhood to be appropriate to 16. Posel D, Devey R. The demographics of fathers in South Africa: South Africa’s social context is warranted. an analysis of survey data, 1993–2002. In: Richter L, Morrell R, editors. Baba: men and fatherhood in South Africa. Cape Town: HSRC Press; 2005. pp. 38–52. Sponsorship: This project was funded by the National 17. Madhavan S. Fosterage patterns in the age of AIDS: continuity Institute of Child Health and Development of the and change. Soc Sci Med 2004; 58:1443–1454.
- 109. List of contributors Agüero J., S67 Hargreaves J. R., S39 Parikh A., S95 Assche A.V., S17 Hong R., S17 Phetla G., S39 Assche, S.B-V.,S17 Hosegood V., S29 Porter J.D.H., S39 Pronyk P.M.,S39 Bärnighausen T., S29 Kadiyala S., S5 Boerma J. T., S17 Khan S., S17 Quinlan T., S95 Boler T.,S83 Kim J.C., S39 Bonell C. P., S39 Ravindranath S., S67 Rutstein S., S17 Lam D., S49 Cakwe M.,S95 Leibbrandt M., S49,S75 Simon J.L., S95 Carter M.R., S67 Lewis J., S57 Skalicky A., S95 Collins D.L., S75 Lopman B., S57 Chandiwana S., S57 Timaeus I. M. S29,S83 DeSilva M.B., S95 May J., S67 Vaessen M., S17 Dinkelman T., S49 Mishra V., S17 Morison L. A., S39 Watts C., S39 Ghys P. D., S17 Mushati P., S57 Whiteside A., S1 Gillespie S., S1,S5 Whitworth J., S1 Gregson S., S57 Newell M.-L., S29 Greener R., S1,S5,S17, Nyamukapa C., S57 Zhuwau T., S95 S104