The Effect of an Ageing Population on British Diets
1. The Effect of an Ageing Population
on British Diets
Alexander Lewzey
A dissertation submitted to The University of Manchester for the degree
of Master of Science in the Faculty of Economics
2016
7969814
School of Social Science
2. Abstract
This paper analyses the channels through which population ageing affects British diets.
Household consumption of food and nutrients varies over the life-cycle. It is likely the
population ageing we are currently witnessing in the UK will have a significant impact on
the diets households choose, and the health outcomes associated with such diets.
Using a OLS specification that controls for cohort and time effects I analyse 6 years
of the Living Costs and Food Survey. I produce a series of age curves with corresponding
semi-elasticities that map the non-linear relationship between age and expenditure, as well
as the relationship between age and nutrient intake. Finally, I use fitted values from the
aforementioned estimations in combination with ONS population projections to forecast the
possible impact age effects will have in the event of the predicted population ageing.
The results indicate that there are significant age effects influencing British diets. The
impact of the age effects evolves over the life-cycle. These effects in combination with the
population ageing observed in Britain will have a significant effect on the food products
households choose to purchase, and the nutrients they consume.
Keywords: Ageing population, Food and nutrient demand, Pooled household survey
data, Population projections
2
3. Declaration
• I declare this paper is my own original work unless referenced clearly to the contrary,
and no portion of the work referred to in the dissertation has been submitted in support
of an application for another degree or qualification of this or any other university or
other institute of learning
Intellectual Property Statement
1. The author of this dissertation (including any appendices and/or schedules to this
dissertation) owns certain copyright or related rights in it (the “Copyright”) and he
has given The University of Manchester certain rights to use such Copyright, including
for administrative purposes.
2. Copies of this dissertation, either in full or in extracts and whether in hard or electronic
copy, may be made only in accordance with the Copyright, Designs and Patents Act
1988 (as amended) and regulations issued under it or, where appropriate, in accordance
with licensing agreements which the University has entered into. This page must form
part of any such copies made.
3. The ownership of certain Copyright, patents, designs, trademarks and other intellectual
property (the “Intellectual Property”) and any reproductions of copyright works in the
dissertation, for example graphs and tables (“Reproductions”), which may be described
in this dissertation, may not be owned by the author and may be owned by third parties.
Such Intellectual Property and Reproductions cannot and must not be made available
for use without the prior written permission of the owner(s) of the relevant Intellectual
Property and/or Reproductions.
4. Further information on the conditions under which disclosure, publication and commerc
ialisation of this dissertation, the Copyright and any Intellectual Property and/or
Reproductions described in it may take place is available in the University IP Policy, in
any relevant Dissertation restriction declarations deposited in the University Library,
and The University Library’s regulations.
3
7. 1 Introduction
This report investigates the mechanisms through which population ageing effects the diets of
British consumers. Many factors influence food purchasing behaviour, economists typically
tend to focus on prices and income. However household consumer spending on different
foods changes markedly over the life-cycle. This is in part due to a person’s age as well as
many other factors that change with age, such as household size, income and the presence
of children. The population ageing we are currently witnessing in the UK is likely to have
a substantial impact on the food products households choose to buy, and the nutrients
that they consume. The need to better understand ageing and the implications it has for
social security sustainability and growth, has attracted a substantial amount of attention
(Steptoe et al., 2012; Bloom et al., 2010). Yet the impact population ageing will have
on the composition of diets and the potential health outcomes associated with such diets
has not received the attention it warrants. The prevalence of obesity has increased by
almost 400% in the last 25 years, and will soon surpass smoking as the number one cause
of premature loss of life (HOC, 2004). Treating obesity and its consequences is currently
estimated to cost the NHS £5.1 billion pounds a year. Obesity is also one of the risk factors
for type 2 diabetes which accounts for spending of £8.8 billion a year, which is around 9%
of the NHS budget. The wider costs of obesity to society such as loss of productivity and
self-esteem are estimated to be 3 times that (HOC, 2015). While activity has great benefit,
increasing activity alone cannot address the current obesity crisis. There is clear evidence
that measures to improve food environment and reduce calorie intake lie at the heart of a
successful strategy. Given that the growing trends of obesity and other nutrition related
issues show no signs of stopping, it is of vital importance to policy makers and public health
that the effect age has on diet is better understood. For example, is demand for food mainly
influenced by age specific tastes or by age-specific income differences? Is it the change in
household composition resulting in smaller household size and older parents which is driving
demand? It is of great importance that these questions are investigated in order to gain
insight into future demand structures. This will enable policy makers to develop strategies
that effectively target different population groups at different stages in the life-cycle.
In this paper I use the Living Costs and Food Survey (LCF) for the years 2008 to 2013.
To this data I apply a model specification that controls for the confounding covariates of
cohort and time effects, as well as the socio-economic and demographic characteristics that
influence household diets. This allows me to estimate the effect age has on the composition
of diets. In order to capture the nonlinear effects of age it is modelled with the use of a
7
8. quadratic term. The model specification is first implemented to measure the responsiveness
of expenditure on 11 different composite food goods, and then the specification is applied to
map the responsiveness of the macronutrients contained within the foods. The analysis of
macronutrients is of great interest, as it is the consumption of macronutrients that results in
the health outcomes associated with an individual’s diet. In the final stage of the empirical
analysis the aforementioned estimates are applied to ONS population projections, in order
to predict how age effects are likely to influence consumption per person in the future.
In this paper I find several notable results that have important implications for policy
and public health:
• I find that the relationship between age and expenditure or nutrient intake is typically
characterised by a sharp initial increase during adolescence, which becomes less steep
and eventually levels out around the age of 60. Beyond the age of 60 consumption
either continues to level off, or in some instances begins to decrease. This result is
of great importance as it helps to explain the persistent increase in weight that is
commonly observed among middle aged individuals.
• The age estimates in this paper indicate that expenditure values are more responsive to
changes in age than nutrient intake. This has important implications for policy design,
as policies aimed at single food goods are unlikely to yield significant improvements to
the nutritional quality of British diets.
• We find that there are asymmetries in the responsiveness of different goods to changes
in age, for example fruit consumption is highly influenced by age and is likely to
respond positively to population ageing. Vegetable consumption on the other hand
is less responsive and may lag behind fruit consumption if policies to promote the
consumption of vegetables are not implemented.
• Sugar intake displays marginal increases in calories consumed with age even in the latter
stages of the life-cycle. This is most likely the result of the large marginal increases
in expenditure allocated to sweets, cereal and soft drinks observed in the latter stages
of the age distribution. This implies that policies which focus on soft drinks alone
may not be as effective at reducing sugar intake as policies that also target cereals and
sweets.
• Cohort effects have a significant impact on the amount of expenditure allocated to
soft drinks, with more recent cohorts exhibiting increased expenditure on soft drinks.
8
9. Attention needs to be paid to how these cohort effects continue to develop. If the trend
continues in its current fashion, soft drinks will only become increasingly popular with
future cohorts
• There is a strong association between takeaways purchased for consumption in the
home, and a reduction in expenditure on fruit and vegetables. Takeaway consumption
is also associated with an increase in total calorie intake, with a disproportionately
large amount of those calories coming from sugar and saturated fat. These trends
combined with a strong negative relationship between age and the propensity to order
a takeaway raises concerns of younger generation’s preferences for processed food, and
their possible inability to produce home cooked food.
This paper is organised as follows: section 2 gives a brief review of the relevant literature.
Section 3 gives a description of the data used in the analysis, section 4 gives the intuition
behind the model specification used in the subsequent analysis. Section 5 contains the
analysis of expenditure allocated to composite food goods, section 6 contains the analysis of
macronutrients. Section 7 contains the projections of how the estimated age effects are likely
to interact with population projections. Section 8 discusses the results and the implications
they have for policy makers and wider society, and section 9 concludes the paper.
2 Literature
My work takes place at the cross over of a number of literatures. The two it is most closely
relates to are, the literature on “consumer demand for food and nutrients in the UK” and
“the impact of population ageing on consumer demand.”
2.1 UK consumer demand for food
The majority of studies on UK consumer demand focus on aggregated bundles of consumer
goods including food consumption as one of the featured bundles (Blundell et al., 1993, 1994;
Blundell and Stoker, 2005). However, over the last 20 years there have been a number of
studies that have disaggregated consumption down to the level of individual food groups in
an attempt to see what drives consumers to allocate expenditure to different foods (Chesher,
1997, 1998; Blundell and Robin, 2000; Lechene, 2000; Chesher and Lechene, 2002; Agostini,
2014; Tiffin et al., 2011; Tiffin and Arnoult, 2010; Griffith et al., 2015). The primary focus
of most of these studies is to measure the responsiveness of a consumer’s expenditure, to
9
10. changes in the prices of goods and an individual’s income. Up until 1998 the National Food
Survey (NFS) provided estimates of price and income elasticities of demand for a number of
food groups, but by this time it was felt that the methodology used was dated empirically
and theoretically and was dropped in favour of new methods. For example, Lechene (2000)
used an Almost Ideal Demand System (AIDS) to measure the responsiveness of expenditure
allocation to changes in price and income over the period of 1988 to 2000. The paper found
that most price elasticities of demand are well below 1 in value, and that income elasticities
are typically small and positive indicating food goods are normal goods1
. The methodology
applied in analysis has continued to evolve, the most recent study of UK consumer food
demand in relation to price and income comes from Tiffin et al., (2011). Tiffin introduces a
dynamic component that allows for the modelling of short term and long term elasticities,
this allows him to capture habit formation in the analysis. They find that if the prices of
certain goods increase consumers continue buying these products out of habit despite higher
prices, and only in the long run look for cheaper alternatives. The income effects estimated
in the study also exhibited dynamic behaviour.
2.2 The demand for nutrients
One important evolution in the analysis of food demand is the development of nutrient
elasticities. Nutrient elasticities measure the impact that a change in the price of a good
or an individual’s income has on the consumption of the macronutrients that make up an
individual’s diet. The macronutrients that are typically analysed are carbohydrates, fats and
sugar, which all together sum to the total calories an individual consumes. This approach is
borne out of the fact that ultimately it is the consumption of particular nutrients contained
in the foods that leads to poor health outcomes, not the foods themselves. This method
has been implemented in a hand full of studies (Behrman, 1988; Subramanian and Deaton,
1996; Behrman and Deolalikar, 1990; Strauss and Thomas, 1990; Pitt, 1983; Huang, 1996,
1999; Huang and Lin, 2000). Applications of this method in the UK by Agostini (2014)
and the aforementioned work by Tiffin find that nutrients are less responsive to changes in
price and income than expenditure on specific goods tends to be. There are some possible
explanations for the relatively inelastic demand for nutrients. Individuals can increase their
time shopping seeking better deals and switch between non nutritional characteristics i.e.
branded to non-branded products (Griffith, 2015). This indicates that taxes levied on foods
1
A normal good is an item with an income elasticity of demand between 0 and 1 i.e. consumption responds
less than proportionally to changes in income.
10
11. may be somewhat ineffective as a means to reduce the prevalence of obesity when used in
isolation. In two studies that use nutrient price elasticities to measure the impact of indirect
taxes,2
it was found that the implementation of taxes did not bring consumption in line with
recommended UK guidelines (Briggs et al., 2013; Tiffin and Arnoult, 2011). In an article by
Tiffin3
he concludes there needs to be a better understanding of the other socio-economic
and demographic factors driving the demand for food if we are going to be able to design
and implement effective policies capable of averting bad dietary choices.
2.3 The impacts of population ageing on consumer demand
The ageing populations of the developed world are quickly becoming one of the more salient
demographic shifts, yet little research has been done into the effect an ageing population has
on disaggregated consumer demand. Most research on the impact an ageing population
has on the economy focuses on the macroeconomic implications, much of it reinforcing
Modigliani’s life-cycle hypothesis (1963) that predicts per capita consumption to decline
with population ageing (Hamermesh, 1984; Fair and Dominguez, 1991; Banks et al., 1998;
Bernheim et al., 2001; Hurd and Rohwedder, 2003). There are few papers that analyse
the effect of population ageing on consumer demand in the UK at a microeconomic level.
One notable exception, Luhrmann (2008) evaluates the effect population ageing has on the
demand for composite bundles of consumer goods. Luhrmann finds that there are age specific
demand curves that result in increased overall demand, as well as increased expenditure on
health related goods, household services, leisure goods and spending on food consumed away
from the home. The result in this paper is contingent upon a redistribution of income towards
older generations due to labour reforms that require individuals to extend working lives.
Luhrmann concludes that the changes in demand are not mainly caused by age-specific tastes,
but to a large extent by differences in spending power between different age groups. Drescher
and Roosen (2013) carried out a study analysing 25 years of German consumption data, and
obtained similar results also finding that an ageing population reduced expenditure on food
consumed at home. They determined the relationship between age and food consumed in
the home is a quadratic one, and that expenditure allocated to food decreases with age at
an increasing rate.
2
One a 20% tax levied on sweetened fizzy drinks and another a tax levied on fat.
3
Tiffin. (2014, January 23rd).Will a Soft Drinks Tax Solve Britain’s Obesity Crisis? Fat Chance[Article].
Retrieved fromhttp://www.huffingtonpost.co.uk/professor-richard-tiffin/soft-drink-tax˙b˙4191154.html?
11
12. 2.4 The impacts of population ageing on the demand for food
The intercept of the two aforementioned literatures is defined by the micro economic analysis
of the impact an ageing population has on the demand for food. To date only a handful of
papers have touched on this subject. Internationally there is a series of papers by Hiroshi
Mori (Mori and Clason, 2004; Mori et al., 2006; Mori and Saegusa 2010; Mori et al., 2015)
which explore the impact Japan’s ageing population has on the diets of Japanese households.
The external validity of these studies is somewhat limited due to the substantial differences
in culture, consumer preferences and diet between Japan and the western world. However,
the studies are relevant in the sense that Mori observes the phenomena of pure age effects
influencing food choices in Japan, and his work stresses the importance of being able to
isolate the impact of age from income, time and cohort effects.
In the UK there have been a string of papers analysing this relationship between age and
diet. The first contribution was from Chesher, (1997) who observed nutrition trends over
the age distribution while controlling for income and region. Chesher estimated energy-age
curves to measure the impact of age on consumption of calories and fat. The paper uses
non-linear least squares4
to capture the non-linearity’s in the relationship between age and
consumption. Chesher finds that energy and fat consumption increases sharply until the
teen years, then it decreases in the mid 20s. Consumption then increases steadily peaking
around 50 to 60 years of age, after which there is a steady decline. Chesher accredits some of
the differences across age to biological factors associated with body size, growth, metabolic
rate and age dependent activity levels5
. However, Chesher struggles to explain the increase
in calories consumed beyond the age of 30. The study’s analysis is somewhat limited in the
fact it only includes age curves for calories and fats; analysis of protein or carbohydrate is
not included. The estimates are also likely to suffer from bias induced by omitted variables,
and the data used in the study6
by construction induces a number of biases such as wastage7
,
eating out and visitor impacts. Sweets, chocolates and soft drinks were not recorded in the
survey at that point in time, which excludes some of the most pertinent foods groups in
terms of induced health outcomes. Most of the criticisms of this paper are due to the use of
data and methods that are somewhat dated, however the paper is an important contribution
and a good starting place.
There have been two subsequent continuations of this paper (Chesher, 1998; Agostini,
4
A non-parametric method of estimation.
5
Manual labour vs service oriented jobs and the employed vs the retired.
6
The National Food Survey (NFS)
7
Calories purchased but not consumed.
12
13. 2005) which have expanded on the original analysis and addressed some of the short comings.
The subsequent paper by Chesher introduces a dynamic aspect by constructing a pseudo
panel8
to observe how the relationship differs over time. Chesher finds there is significant
time variation in nutrient intake, for example a reduction in calories consumed at any age is
observed in sample period under study9
. This result is consistent with more contemporary
studies of calorie intake in the UK (Griffith et al., 2013). Chesher concludes the age profiles
of nutrient intake are complex and may vary through time in a complex fashion. Agostini
(2005) further builds on the research by including additional nutrient dependent variables10
,
controlling for additional covariates11
and analysing a more recent time period 1975-2000
and finds similar results.
These studies provide great insight into this relationship between age and diet, however
they are all limited by their use of non-parametric estimation. Non-parametric methods are
good for modelling the noisy data but lacks power against their parametric counterparts, and
they do not allow for the accurate quantification of age related impacts. Another continuation
of the work yet to be touched upon is the issue of distinguishing age and cohort variation
from calendar time driven changes. This is highlighted by Chesher and Agostini as a prime
candidate for future research in this area. These studies also include a parsimonious selection
of demographic variables, for example they fail to control for working status, working hours,
and the quadratic nature of income effects. In this paper I expand upon the work of Chesher
and Agostini and contribute to the literature in several ways:
1. I disentangle the effect of age from time and cohort effects, by constructing a model
that is capable of controlling for both covariates.
2. I use a parametric specification which has higher power then the previously used
non-parametric techniques, and allows for the quantification of the impact ageing has
on the consumption of composite goods and nutrients.
3. I reduce the amount of omitted variable bias present in the previous papers by including
a more comprehensive collection of demographic and socio-economic controls.
4. I expand the knowledge of how nutrient intake evolves over the life-cycle by including
analysis of sugar, which has not featured in the aforementioned work by Chesher and
8
A method of constructing panel data out of repeated cross sections first implemented by Deaton (1985)
9
1975-94
10
Protein and saturated fat.
11
Presence of children and region.
13
14. Agostini. I also carry out a more extensive analysis of saturated fat which was only
covered briefly in Chesher (1997).
5. I continue to map the changes in age specific demand by studying a more recent time
period 2008-2013. To my knowledge this is the first paper to use the Living Costs and
Food survey to analyse the relationship between age and diet.
6. This is the first study which uses estimates in combinations with population projections,
to attain a forecast of the impact age effects may potentially have on British diets.
3 Data
3.1 Living Costs and Food Survey
The analysis in this paper is based on the UK Living Costs and Food Survey (LCF),
sponsored by the Office for National Statistics (ONS) and the Department for Environment,
Food and Rural Affairs (DEFRA). The LCF covers the period 2008 to 2013 and was
previously known as the Expenditure and Food Survey (EFS). The LCF is the successor
of the National Food Survey (NFS) which is the survey used in the majority of the previous
literature (Chesher, 1997, 1998; Agostini, 2005, 2014). The LCF is a cross sectional survey of
households in the UK, that collects information at a household and personal level. Data on
purchases are collected over two weeks in a personal diary. In this diary households record
expenditure on goods and the quantities purchased. The recorded values are averaged over
the two weeks in order to obtain values in terms of expenditure per week. Individuals then
take part in an interview where additional socio-economic and demographic information
is collected. Around 12,000 households are selected per year and the rate of attrition is
typically 50 to 60%; the survey is carried out all year round. 256 food and drink categories
are included, for the purposes of this paper I only include food consumed in the home as this
represents the majority of food and drink consumption, around 70% (Griffith et al, 2013).
I partition the food items into 11 composite food goods, of which details can be found in
appendix D.
3.2 Food Prices
One short coming of the LCF is that it does not include data on food prices, however there
are two commonly used methods to obtain price data. One is to use an Aggregated price
14
15. index (Chester and Lechene, 2002) such as the CPI or RPI. This method imposes some
limitations on the analysis in that it assumes the same price is observed nationally, and it
puts restrictions on the goods you can bundle together12
. The more commonly implemented
alternative is to obtain a derived price by dividing the total expenditure on a good, by the
quantity of the good purchased for each household (Huang,1999; Lechene 2000; Agostini,
2014; Meng et al., 2014). This method is not perfect as it exacerbates any measurement
error in the expenditure and quantity data, further it results in an unrealistic amount of
heterogeneity in prices. However, it has the advantage of producing a derived price for any
good purchased in the data set, and it also captures the geographical differences in prices.
I average prices by year and government region to reduce the impact of any measurement
error (Agostini, 2014).
3.3 Nutrient data
It is well established that in order to be healthy, one has to include the right balance of
nutrients in one’s diet. It is of paramount importance to study nutrient intakes as well
as consumption of composite food goods if we want to prevent the poor health outcomes
associated with bad dietary choices. In this report I include an additional set of dependent
variables: several different macronutrients measured in terms of the energy you gain from
eating them, more commonly referred to as calories (kcal). The LCF includes quantities
purchased in millilitres and grams of different food products, nutritional quantities can be
obtained by applying the nutrient conversion factors provided by DEFRA. The conversion
factors assume a 10% wastage, that is 10% of nutrients brought into the home are thrown
away or not consumed. DEFRA provides conversion factors for 47 macro and micro nutrients13
,
however for the purposes of this report I am only concerned with 5 of them. The 3 primary
macronutrients carbohydrates, fats and proteins are the main feature as they they sum
to total calories consumed, and therefore determine the nutritional aspect of weight gain.
Additional to these I have included 2 sub components, sugar a type of carbohydrate and
saturated fat a type of fat. Both of which have strong associations with a range of poor health
outcomes when consumed in excess (Mozaffarian et al., 2010; Burt and Pai, 2001; Malik et
al., 2010). Once the quantities of nutrients are obtained in grams, you can calculate the
12
And in application the low levels of variation lead to multicollinearity issues.
13
Macronutrients are the energy providing component of nutrition, which includes Carbohydrates, Fats
and Proteins. Micronutrients consist of vitamins, minerals and trace elements which play an important
role in health and well-being, however in this study we are concerned with calories and hence only concern
ourselves macronutrients.
15
16. energy content by multiplying the quantities of carbohydrates, fats and proteins by their
respective calorie per gram 3.75, 9 and 4 kcal’s per gram (DEFRA, 2000).
3.4 Data cleaning
All nominal food expenditures and prices have been deflated to 2008 prices using the
corresponding food and drink components of the CPI, provided by the ONS. When preparing
the data for analysis there were a number prices which were clearly the result of measurement
error and the implicit nature of the derived calculation, to address these outliers around 30
observations were dropped. The nutritional quantities were derived from the raw diary entries
and hence subject to larger amounts of measurement error, in the form of unrealistically
extreme values. To deal with these outliers I implement a more parsimonious version of
the approach taken by (Meng et al., 2014), and systematically drop observations above the
99.9th or below the 0.1th percentile of the distribution for each macronutrient, this comes
to around 400 observations.
3.5 Limitations
When using pooled cross sectional data of this nature, it is important to recognise its
limitations. There are several sources of measurement error in the data. Personal diary
entries will be subject to under reporting, as well as interviewer and respondent errors
(Agostini, 2014). Zero purchases, also known as censoring are difficult to interpret given
that they could occur for a number of reasons. It could be that at a given price and level
of income, an individual finds a good unattractive (Tiffin et al., 2011) in which case the
zero purchase represents a valid consumer preference. Alternatively, zero purchases can be
a result of infrequency of purchase. The expenditure diary takes place over a two-week
period so for items that are infrequently purchased for example, sugar, rice, pickles etc.
there is a possibility purchases will not fall within the diary window (Tiffin et al., 2011).
The nutritional data assumes a flat rate of wastage across all households. However, I would
hypothesise that those who earn more would have a higher propensity to waste food. Food
costs take up a much larger proportion of a low income familie’s budget, therefore the
opportunity cost of wasting food would also be greater for a low income family. This could
induce an upward bias in the nutritional values for higher income households. The data
also suffers from visitor effects, as guests may consume part of a household’s food stock,
or a household may consume food they did not purchase (Chester, 1997; Agostini, 2005).
16
17. However, the data is highly suited for this analysis and has a number of advantages over other
sources. Intake data where individuals record quantities of goods and nutrients, typically
have much higher rates of under reporting than expenditure surveys (Chester, 1997). The
fact the data is publicly available means research is open to verification (Tiffin et al., 2011).
4 Demand model selection
4.1 Model Specification
The application of consumer demand theory in empirical work over the last few decades
has fostered a vast literature on consumer demand theory, and the econometric techniques
used in its applications. There have been a number of models used to estimate consumer
responsiveness. By far the most commonly used models are the Almost Ideal Demand
System (AIDS), and the Quadratic Almost Ideal Demand System (QUAIDS) (Deaton and
Muellbauer, 1980; Banks et al., 1997). These models are championed as they are widely
applicable, consistent with the theory of consumer behaviour and have a functional form
consistent with known household budget data14
. However, in this paper given the variable
of interest is age, my specification does not have to be subjected to so many theoretically
imposed restrictions. Instead I opt for a single equation15
ordinary least squares (OLS)
approach, and use a similar specification to those used in the following papers (Meng et al.,
2014; Luhrmann, 2008). A single equation OLS approach allows for easier application of
socio-economic controls, yet still allows for the quantification of age effects. Equation (1) is
estimated for each composite good j and each nutrient n.
Expenditureij/Caloriesin = α + β1Agei + β2Age2
i + β3Incomei + β4Income2
i +
j
βjPriceij
+ βRegionDummiesi + βHHCompositionDummiesi + βQuarterDummiesi + β5HHSizei
+ β6Child0to4i + β7Child5to17i + β8Takeawayi + β9Workersi + β10Hoursi
+ βY earDummiesi + βCohortDummiesi + i
(1)
14
For a more complete over view of the literature on the empirical application in consumer food demand
I recommend (Tiffin et al., 2011).
15
Single equation method implies that for each good and each nutrient, equations are estimated
independently.
17
18. 4.2 Dependent Variables
In this study there are 2 sets of dependent variables, 1 expenditure based and 1 nutrition
based. The food items that make up the expenditure variables have been partitioned into
the following food groups “Sweets”, “Cereals”, “Meat”, “Fish”, “Dairy”, “Fats”, “Fruit”,
“Vegetables”, “Condiments”, “Beverages” and “Soft drinks” and are measured in (£) expenditure16
.
The other set of dependant variables consists of nutritional values, which features “Total
Calories”, “Carbohydrates”, “Sugar”, “Fat”, “Saturated Fat” and “Protein” all measured in
calories (kcal). “Total Calories” is the sum of Carbohydrates, Fat and Protein.
4.3 Independent Variables
The primary variable of interest in this study is age. Chesher (1997) uses non-parametric
estimation techniques to capture the non-linearities in the relationship between age and
consumption, this paper like other studies of age effects (Blundell, 1994; Luhrmann, 2008;
Drescher, 2013) includes a quadric term to capture the relationship. Lyssiotou et al. (2002)
carries out a non-parametric analysis of consumer demand using the UK Family expenditure
survey. They find that in order to capture all the age effects in a parametric specification,
linear and quadratic terms alone are not adequate. This is due to the markedly different
behaviour of the under 30s. Lyssiotou recommends that in addition to the quadratic term,
dummies for individuals under the age of 30 should be included. However, they also find
that the radical behaviour of the under 30s is not observed for food, hence a quadratic
relationship will be sufficient to model the variation. In studies using household level data it
is usual practice to use the age of the household representative person (HRP)17
as a proxy
for “age of the household” (Jorgenson, 1997; Blundell, 1994; Luhrmann, 2008). This is not
a particularly problematic assumption as for households of one person the age of HRP is
the age of the household exactly. For households with multiple members the average age
difference between household members is only 1.39 years of age, with a standard deviation
of 5.31 (appendix A). Such a small difference poses little threat to the validity of the results;
note this statistic excludes children who are controlled for separately.
In order to obtain an unbiased estimate or to least minimise the bias in the age coefficients,
we need a specification that controls for the covariates that are correlated with age which
also have an impact on expenditure and nutrition. In the absence of adequate controls,
age coefficients will be highly correlated with the error term. This is particularly hazardous
16
Details of these groups can be found in appendix D.
17
This is simply the biggest economic contributor in a household and normally the oldest.
18
19. given the variable of interest “age” is strongly correlated with a large number of confounding
factors. I have included following controls to minimise the omitted variable bias.
Numerous studies have found there are significant cohort and time effects that influence
diets (Agostini, 2005; Mori, 2015; Mori, 2006). Most studies fail to control for both of these
effects. Luhrmann (2008) assumes cohort effects to be zero and only allows variation by time,
a needlessly strong assumption. To control for time effects I include dummy variables for each
year of the survey sample 2008-13. To control for cohort effects I include dummy variables
for each 5-year birth cohort in the period 1930 to 1995. These cohorts are constructed using
the year of birth of the HRP, given the small variations in age within household this is not
a problematic assumption.
Figure 1: Average gross household income by age, 2008-2013
Mori (2006) stresses the importance being able to control for all the income effects when
trying to estimate age effects. This is because income is highly correlated with age and
expenditure, see figure 1. In order to control for all the income effects, I include a quadratic
term as a multiplicative relationship between income and consumption has been established
in prior research (Luhrmann, 2008; Drescher, 2013). As is typical in demand analysis I
include the prices of the 11 composite food goods (Lechene, 2000; Griffith, 2015). It is
well established that the absolute and relative prices of goods have an effect on quantity
demanded.
It is also important to put in place certain demographic controls. Given the analysis
is based on data aggregated at household level, it is important to control for the different
household compositions included in the sample. This is because different types of household
have different dietary needs. I control for household composition by including the following
19
20. dummy variables: single, lone parent, couple with no children and couple with children
(Agostini, 2014).
Figure 2: Confounding covariates of household composition by age, 2008-2013
(a) Average household size (b) Average number of children
I include a control for household size, as households with more members will clearly
consume larger quantities. Household size is also strongly correlated with age. As individuals
get older and move in with a partner and have children, household size increases. Beyond
the age of 40 households tend to get smaller as children leave the nest, see figure 2a. Another
important aspect of household composition is the presence of children. Children of different
ages have specific dietary needs, to account for this I include two variables. One that indicates
the number of children aged 0 to 4, and another that indicates the number of children aged
5 to 17. It can be seen in figure 2b that the number of children present in a household is
strongly associated with age.
Figure 3: Confounding covariates of work-life balance by age, 2008-2013
(a) Average weekly working hours (b) Share of age group who purchased a takeaway
Labour force participation is believed to have an effect on diet. Hectic work lives increase
the need for convenience. As people work longer hours the opportunity cost of their time
20
21. increases, resulting in less home cooked food and an increased demand for convenience food,
which is typically energy dense and high in saturated fat and added sugar (Monsivais, 2014).
To control for this, I have included a variable that controls for number of workers in a
household. Another factor that drives the need for convenience food is female labour force
participation. Women labour participation increases the cost of their time, and reduces the
amount of time spent cooking healthy meals at home (Agostini, 2005). To control for this,
I have included the combined working hours of the couple18
.
To capture a household’s natural propensity to purchase convenience food, I have included
a dummy variable that indicates if a household recorded any expenditure on takeaways.
Additional to these controls I have included dummy variables for different regions, of which
there are four. Due to the survey being carried out throughout the year, I have included
dummy variables for each quarter. It is well documented that food consumption exhibits
seasonal trends (Tiffin et al., 2011). Summary statistics of all the aforementioned variables
are included in appendix A.
4.4 Strengths and Weaknesses
Single equation models fail to recognise that decisions of budget allocation are made contempo
raneously and therefore are interdependent. The single equation approach I am using does
not capture this interdependence of consumer choices. However, the analysis in this paper is
dealing with absolute marginal values, as opposed to expenditure shares. A single equation
specification is adequate to model these marginal changes. There is also the issue that a
single equation model cannot include households which have zero purchases for a particular
composite good, as the dependent variable would be undefined. This is problematic if
observations are omitted in a non random fashion which is contingent upon the dependent
variable. However, this is not a major problem as there a very few zero observations for
the more commonly purchased food groups, and there are no zero observations at all when
nutrient values are the dependent variable. This is due to the fact that it is rare for foods
to not contain at least some quantity of all macronutrients.
One limitation of the data is that it does not include values of consumption for individuals.
It only includes quantities of goods and expenditures aggregated at household level. Treating
households as a single entity forces us to assume that preferences are homogenous within
households. This assumption is unlikely to hold up in real life (Chiappori, 1988, 1992).
Treating households as single entities makes it difficult to control for gender which is not
18
Couple consists of the HRP and their partner if they have one.
21
22. ideal given gender has a significant effect on consumption. However, gender is not correlated
with age so is unlikely to bias our variable of interest.
One advantage of not using a typical budget share model is that the specification used
in this paper does not have to adhere to “adding up”. When using an AIDS specification,
total income is set equal to total expenditure19
which makes the model endogenous by
construction. In my specification income is measured as “total gross household income”,
which is exogenous. The specification used in this paper also allows for an easy and
intuitive interpretation of the dynamic marginal effects age has, at different points in the
age distribution.
19
This is because under the restriction of adding up, all expenditure shares must add up to one.
22
23. 5 Expenditure Analysis
5.1 Descriptive Statistics
One of the main ways an older population is likely to affect diets, is through a change in
consumption structure. In this section I use descriptive statistics to identify how expenditure
patterns vary across the age distribution.
Figure 4: Expenditure (£) by age, 2008-2013
Figure 4 depicts average weekly household expenditure on 11 composite food goods20
by
age. Total expenditure increases sharply at the beginning of the age distribution, with total
expenditure peaking at ages 40 to 45. Beyond the peak total expenditure decreases steadily
until 70 years of age where the decline becomes sharper. The increases and decreases in
expenditure are common across all food groups. Figure 4 does not enable us to determine
the cause of the variation. However most of the variation in expenditure is likely due to the
concurrent trends of household size and household income, which also peak in the late 40s
(Figures 1 and 2a).
20
Details on these groups can be found in appendix D
23
24. Figure 5: Average expenditure shares by age, 2008-2013
It is not very clear from figure 4 how the relative proportions of expenditure vary with
age. Hence included in figure 5 are the expenditures on the 11 food groups as a share of
total expenditure on food purchased for consumption in the home. This gives us a clearer
image of how consumption structure changes with age. It is apparent that the majority of
expenditure at any age is allocated to cereals, meat, dairy and vegetables, making up around
70% of expenditure. This is likely where the majority of nutrients in a household’s diet comes
from. Expenditure shares of dairy, vegetables and sweets are fairly persistent across all ages,
with vegetable exhibiting a mild peak mid life-cycle. Meat has a large peak latter on in
the life-cycle, increasing from a 18.6% share in the late teens to a 22.4% share in the late
60s. The reverse trend is observed for cereal as it peaks (22.4%) in the late teens, and hits
a low (16.7%) in the early 60s. Then there are the food groups that exhibit monotonically
increasing or decreasing shares with age. Fats exhibits a 1.6% increase in share, and beverages
a 1.9% decrease in share across the distribution. Particularly striking are the relatively large
increases and decreases in the shares of fruit and soft drinks respectively, both exhibit a
change of around 6 to 7%. As previously mentioned it is not possible to extrapolate how
much of the variation in figure 5 is due to age effects, as it does not control for the confounding
cohort and time effects. For example, it is possible that the demand for soft drinks is mostly
influenced by the cohort you belong to; the product might have become more popular with
later cohorts as opposed to the variation in consumption being the result of a persistent age
dictated preference. However, as a descriptive statistic it helps identify the overarching age
24
25. related trends, that might be attributable to age when all other covariates are controlled for.
5.2 Results - Expenditure
Table 3 contains the results of the expenditure model estimates that will be discussed in
the following section. Due to age and income being included in the model with quadratic
terms, the marginal effects of these variables change in direction and magnitude across the
distribution, hence it is not immediately apparent from the parameter estimates what the
marginal effect of each of these variables is. To make for a more intuitive interpretation I
have included age expenditure curves, (figure 6) to map the estimated relationship between
age and expenditure. Included in table 1 are the semi-elasticities that correspond to the age
curves in figure 6. The semi-elasticities measure the percentage change in expenditure that
results from a unit change in age at different points in the age distribution. Engel curves and
elasticities have also been provided for the corresponding income parameters in table 3. The
figures in table 2 represent the income elasticities of demand, which can be interpreted as the
percentage change in expenditure that results from a percentage change in an individual’s
income at different points in the income distribution.
5.2.1 Expenditure age curves
In figure 6 are the age curves for each composite good. These curves map out the estimated
quadratic relationship between age and expenditure. For 9 out of 11 composite goods the
linear and the quadratic age terms are highly significant, indicating that if the model is
constructed properly there are significant age effects for the majority of food expenditures. It
can be seen that the majority of the curves exhibit a hump shape that is steeper in the earlier
stages, and proceeds to level off around the age of 60. Some of the age curves include a portion
that exhibits marginal decreases in expenditure with age. Despite the superficially similar
appearance of the curves, the estimated impacts vary considerably between expenditure
groups. Table 1 gives a clearer image of how the curves vary quantitatively.
25
27. For example, expenditure on sweets, fruits and beverages are considerably more responsive
to age when evaluated at younger ages, with semi-elasticities of 6.25, 12.75 and 8.15 respective
ly. For a 20-year-old a unit change in age would lead to a 12.75% increase in expenditure
on fruit. The age curves for these three goods level off around the age of 60, and display
marginal decreases in expenditure at the end of the age distribution. For example, fruit
has a semi elasticity of demand of -0.71 when evaluated at 80 years of age, this means that
a unit change in age for someone who is 80 years old will result in them reducing their
expenditure on fruit by 0.71%. There are goods that exhibit a similar hump shape but are
not as responsive at younger ages. Goods such as cereals, meat and dairy, which typically
have semi-elasticities of 3 to 4% when evaluated at 20 years of age. These curves also display
diminishing increases in expenditure with increases in age.
Table 1: The marginal effect of unit change in age on expenditure (%)
Age Sweets Cereals Meat Fish Dairy Fats Fruit Veg Cond Bev Soft
20 6.25 3.57 3.64 1.27 2.19 4.35 12.75 3.17 -0.35 8.15 0.78
30 3.26 2.19 2.14 1.13 1.49 2.26 4.73 1.84 -0.30 3.88 0.62
40 1.96 1.41 1.29 1.02 1.01 1.14 2.49 1.04 -0.23 2.29 0.48
50 1.19 0.87 0.70 0.93 0.65 0.34 1.36 0.44 -0.16 1.42 0.36
60 0.64 0.46 0.22 0.85 0.36 -0.38 0.59 -0.08 -0.09 0.83 0.26
70 0.19 0.10 -0.23 0.79 0.09 -1.19 -0.06 -0.61 -0.01 0.37 0.16
80 -0.24 -0.25 -0.71 0.73 -0.16 -2.34 -0.71 -1.25 0.06 -0.04 0.06
The semi-elasticities in this table correspond to the curves in figure 6
Another trend that seems to characterise some of the expenditure age curves, is a strong
negative segment of the curve. This negative portion can be observed from 60 years of age and
onwards, and features in the curves corresponding to vegetables and fats. This implies that
the marginal effect of an increase in age for someone who is above the age of 60, is to reduce
their expenditure on fats and vegetables. The remaining 3 curves differ more considerably
in shape to the curves we have covered. Fish and soft drinks display more of a linear
relationship, and condiments displays marginal decreases in expenditure across the whole age
distribution. It can be seen from the wide confidence intervals corresponding to the age curves
for fish, soft drinks and condiments, that they lack precision. The linear and the quadratic
terms are both highly insignificant in all 3 of these expenditure equations. However, for all
other 8 composite goods, the coefficients are highly significant. The confidence intervals for
the curves that correspond to significant coefficients indicate that estimates around ages 40 to
60 tend to be considerably more precise than estimates at either end of the age distribution.
The amount of variation in expenditure explained by the model differs between goods. A
27
28. larger amount of the variation in cereals, dairy and vegetables is explained by the model, with
R-squared values of 30 plus. Meat, fruit and soft drinks have R-squared values of around
20. The remaining equations explain even less of the variation in the dependent variable.
5.2.2 Engel curves
Also featured in the results are the Engel curves (figure 7) and income elasticities of demand
(table 2) that correspond to the income parameters in table 3. The Engel curves, like the
age curves tend to have a characteristic shape that is fairly consistent across goods. The
characteristic shape on the Engel curves differs from the prevailing shape of the age curves.
The marginal increases in expenditure become greater with larger values of household income,
this trend is common across all eleven composite goods. All goods seem to be relatively
unresponsive to changes income at lower values of household income. For a household that
earns £250 a week, a 1% increase in income will typically yield a 0.02 to 0.06% increase in
expenditure on most food items. Expenditure becomes more responsive to changes in income
for households that earn more. We start to observe marked differences in income effects
between different composite goods at higher levels of income. Certain goods become a lot
more responsive at higher income levels for example, when evaluated at an income of £2000
a week a 1% percent increase in income will result in a 0.86% increase in expenditure on fish.
The remaining elasticities lie in the range 0.25 to 0.82 when evaluated at an income of £2000,
indicating that all goods are normal goods21
, that is expenditure on goods increases less than
proportionally with the increase in income. The impact that a change in income differs with
a household’s income, and between different goods. There is less variation between goods
in the shape of the Engel curve, than there is between goods in the shape of age curves.
The Engel curves are estimated with more precision; the 95% confidence intervals hug the
fitted values closely and only becomes less precise at very high levels of income. The age
curves are generally not as precise and become increasingly vague at either end of the age
distribution. Nearly all income coefficients are significant or verging on significance apart
from fat which is highly insignificant in both linear and quadratic terms. There is some
evidence of linearity in sweets, dairy and beverages. This can be seen in the flatter curves
and from the insignificant quadratic terms in the corresponding equations.
21
A normal good is a good that has an income elasticity of demand between 0 and 1.
28
30. Table 2: Income elasticity of demand (%)
Income(£) Sweets Cereals Meat Fish Dairy Fats Fruit Veg Cond Bev Soft
250 0.03 0.02 0.03 0.11 0.06 0.01 -0.12 0.05 0.02 0.05 0.03
500 0.06 0.05 0.07 0.23 0.11 0.02 0.22 0.12 0.08 0.10 0.08
750 0.09 0.09 0.12 0.35 0.15 0.05 0.32 0.19 0.17 0.15 0.15
1000 0.13 0.13 0.17 0.47 0.18 0.09 0.40 0.27 0.29 0.21 0.22
1250 0.16 0.19 0.23 0.58 0.22 0.13 0.47 0.35 0.42 0.26 0.30
1500 0.19 0.24 0.30 0.68 0.24 0.18 0.54 0.42 0.56 0.32 0.38
1750 0.22 0.30 0.36 0.77 0.27 0.24 0.60 0.50 0.69 0.38 0.47
2000 0.25 0.36 0.43 0.86 0.29 0.30 0.65 0.58 0.82 0.44 0.55
The elasticities in this table correspond to the curves in figure 7
5.2.3 Other determinants of expenditure
This section briefly covers the estimated coefficients for the socio-economic and demographic
variables, for more details see table 3. The effect of region on expenditure varies considerably
between regions and goods. The effects are typically significant, with certain regions having
higher expenditures on some goods and lower expenditure on others. Households in the
North, Midlands and London typically spend less on sweets, cereals, meat and soft drinks
relative to the base category. Household composition dummies included in the specification
exhibit varying levels of significance. Unsurprisingly expenditure on sweets, cereals and
meats is much higher during the festive season whereas expenditure on fruit, vegetables
and soft drinks is much higher in the summer months. The coefficients for household size
are highly significant, positive and large in magnitude for the majority food groups. The
coefficients for the number of children present in a household indicate that having children
of the age 0 to 4 in the household will reduce the amount spent on cereals, meats and
soft drinks, but will increase the amount spent on dairy. Having children of the age 5 to
17 will increase the amount spent on sweets and cereals, yet decrease the amount spent
on condiments. The dummy variable that indicates whether a household has purchased a
takeaway is significant and negative in nearly all equations apart from meat and soft drinks,
indicating those who order takeaways typically spend less on store bought foods particularly
fish, fruit and vegetables. The coefficients for number of workers and the number of hours
worked are typically insignificant or very small. There are a few significant patterns in the
cohort effects, for example in the cohorts 1931 to 1941 there is a significant increase in
expenditure on meat and dairy. Those born in cohorts 1951 to 1980 would typically spend
less on fat relative to the base category. The cohort effects associated with expenditure on
soft drinks are significant in nearly all cohorts included in the sample, and indicate that
households who belong to younger cohorts typically consume more soft drinks.
30
36. 6 Nutritional Analysis
As previously mentioned it is important to be able to understand what drives consumers to
allocate expenditure to different food products. This is because expenditure has important
implications for policies such as indirect taxes and product specific information campaigns.
However, it is just as important to gain an understanding of what influences household’s
consumption of macronutrients. A poor choice of nutrient intake is what ultimately leads to
poor health outcomes. Due to the fact that all foods contain at least some amount of any
given macronutrient, the effect of changes in expenditure allocated to different composite
goods can have an ambiguous effect on nutrient intake. For example, many dairy products
are high in fat but also high in protein, often there are also low fat alternatives which contain
little fat but high levels of added sugar. All of these products would be bundled together
in the same composite group, hence a change in the expenditure allocated to such a group
would have an ambiguous effect on nutrient intake. The following section continues the
analysis with a focus on nutrients as opposed to expenditure.
6.1 Descriptive Statistics
Firstly, we identify the overarching trends in the nutrient data before proceeding into the
demand model analysis.
Figure 8: Average calories consumed (kcal) by age, 2008-2013
Included in figure 8 are the average calories consumed of each macronutrient by age.
Figure 8 exhibits a similar shape to figure 4 with total household calorie intake peaking at
36
37. ages 40 to 45. However, the decline is less linear for total calories consumed than for total
expenditure and seems to be split into 3 separate stages; a steep rate of decline from ages
45 to 60, followed by a much less pronounced decline from ages 60 to 70, beyond the age
of 70 calories consumed declines rapidly. This pattern is common across all macronutrients.
Similarly to figure 4 the variation in calories consumed across the age distribution cannot be
interpreted as the effect of age on calories consumed as it does not isolate the confounding
factors of household size, cohort and income, which will no doubt have an impact on calories
consumed.
Figure 9: Share in total calories by age, 2008-2013
Included in figure 9 are the average macronutrient intakes represented as a share of total
calories consumed. It can be seen that the majority of calories come from carbohydrates,
which make up roughly 50% of calories consumed. The remaining calories are shared between
fats and protein, which make up around 38% and 13% of calories consumed respectively. By
far the most striking aspect of this figure is the persistence of macronutrient shares across
the age distribution. We observe less variation in the macronutrient shares than we do
in the expenditure shares in figure 5, however this is to be expected as small changes in
macronutrient intake have a large impact on weight and health over an extended period of
time. While the variation in nutrient shares is superficially less substantial than the variation
observed in expenditure shares, there are some particularly salient trends in nutrient intake
over the life-cycle. Fat as a share of total calories has a peak in latter life, increasing from
37.2% initially reaching a maximum of 38.9% in the late 50s and throughout the 60s, then
declining back down to 37.9% for ages 80 and older. Protein also exhibits similar behaviour
starting at a 12.2% share and peaking at a 13.8% share in the early 60s. To offset the
37
38. increased shares of fat and protein, the complete opposite trend is observed for carbohydrates.
Carbohydrate consumption exhibits peaks at either end of the age distribution. The peaks
are just under a 50% share; carbohydrate reaches a low of 47.4% throughout the 60s.
Figure 10: Sugar and Saturated Fat Shares
This report includes analysis of the macronutrient subgroups, sugar and saturated fat.
Sugar and saturated fat make up proportions of carbohydrate and fat respectively. These
nutrients are of interest due to their links with numerous fatal health conditions. Included in
figure 10 is sugar consumption as a share of carbohydrate consumption by age, and saturated
fat consumption as a share of fat consumption by age. Both charts exhibit a persistent
increase in share with age, however the increase in sugar is more pronounced particularly
at latter ages. Initially for saturated fat as a share of fat we observe a value of 37% which
increases to 41%, a 4% difference. Whereas sugar as a share of carbohydrates starts as 42%
and increases to 49% across the age distribution, a 7% increase. These trends combined
with the aforementioned trends indicate that as households get older, they consume less
carbohydrates of which a growing proportion is sugar. Additionally, as households get older
they consume more fat of which a growing proportion is saturated. If these trends are
associated with age specific effects, the predicted shift towards an older population could
lead to an increase in average saturated fat consumed per household.
38
39. 6.2 Results - Nutrients
In order to make any claims about the effect of age we need to control for the confounding
covariates. Included in table 4 are the results of the nutrient model estimations. It is the
same model specification used in the expenditure equations in table 3, only with a new set of
explanatory variables “Total Calories”, “Carbohydrates”, “Sugar”, “Fat”, “Saturated fat”
and “Protein”, all measured in terms of calories (kcal). For reasons of space the remainder
of the socio-economic and demographic variables not included in table 4 are included in
appendix B table 9. Age curves and the corresponding elasticities are included in figure 11
and table 5 respectively. Nutrient Engel curves and nutrient elasticities are included in figure
12 and table 6.
Table 4: Nutrient regression results
TotalCals Carb Sugar Fat SatFat Protein
Income 4.1*** 2.2*** 1.4*** 1.3*** 0.5*** 0.7***
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Income2
-0.0 -0.0** -0.0*** -0.0 -0.0 -0.0**
(0.11) (0.02) (0.00) (0.87) (0.43) (0.01)
Age 611.1*** 231.4*** 91.1** 278.3*** 99.0*** 101.4***
(0.00) (0.00) (0.01) (0.00) (0.00) (0.00)
Age2
-4.6*** -1.5*** -0.5 -2.3*** -0.7*** -0.8***
(0.00) (0.01) (0.11) (0.00) (0.00) (0.00)
R2
0.51 0.49 0.41 0.43 0.42 0.45
F-stat 685 622 442 508 488 582
# Obs. 31321 31321 31321 31321 31321 31321
P values are in parentheses *,**,*** denote significance at 10, 5, 1%. All
estimations are carried out using robust standard errors. For reasons of
space the remaining regressions results are included in appendix B table 9.
6.2.1 Nutrient age curves
We shall first analyse the age curves in figure 11, all the nutrient age curves display a very
similar shape. The curves increase substantially at first then level off around the age of
60, then proceed to decline at latter stages of the age distribution apart from sugar which
seems to be characterised by a more linear relationship. It can be seen in table 5 that the
39
40. quantifiable effects of these curves are similar for each nutrient, however there are a few key
differences worth noting.
Figure 11: Age nutrient curves (with 95% confidence intervals)
At younger ages fat, saturated fat and protein are the most responsive to age, with
elasticities ranging from 1.85 to 2.09. Sugar and carbohydrates are the least responsive with
elasticities of 1.22 to 1.31, total calories lies somewhere in between with an elasticity of
1.61. This indicates that when evaluated at the average, someone who is 20 years of age will
increase calorie consumption by 1.61% in response to a 1 year increase in age. At 50 years
of age just before the age effects level off, there is much less variation in the quantifiable
effects of age. At this stage of the age distribution all nutrients exhibit elasticities in the
range 0.38 to 0.56. Later on in the life-cycle, 60 years of age and older the age effects enter
a negative phase where a unit change in age leads to a decrease in calories consumed, apart
40
41. from sugar which remains positive. The negative age effect is stronger for fat and protein;
displaying elasticities of -0.63 and -0.69 respectively. The effects are not as large for calories
and saturated fat. Carbohydrates and sugar seem to be the least responsive to changes in
age. It can be observed in table 4 that all the age coefficients are highly significant apart
from the quadratic age term in the sugar equation, which gives further evidence of a linear
relationship between age and sugar consumption. The R-squared values for the nutrition
equations are large and do not vary considerably between equations, with values in the
range 0.41 to 0.51. This indicates the models explain a substantial amount of the variation
in nutrient intake. Similarly to the expenditure age curves the estimates around the middle
of the distribution are more precise, and the confidence intervals for the estimated elasticities
gets wider at both ends of the age distribution.
Table 5: The marginal effect of a unit change in age on nutrients consumed (%)
Age Calories Carb Fat Protein Sugar Sat-fat
20 1.61 1.31 1.85 2.09 1.22 1.93
30 1.11 0.96 1.21 1.34 0.95 1.31
40 0.73 0.69 0.75 0.82 0.73 0.87
50 0.43 0.47 0.38 0.41 0.56 0.54
60 0.16 0.28 0.05 0.05 0.41 0.25
70 -0.09 0.11 -0.28 -0.30 0.28 -0.02
80 -0.36 -0.07 -0.63 -0.69 0.15 -0.29
The semi-elasticities in this table correspond to the curves in
figure 11
6.2.2 Nutrient Engel curves
We now look at the nutrient Engel curves in figure 12 and the corresponding elasticities in
table 6. The nutrient Engel curves excluding carbohydrate and sugar tend to be more linear
in shape, with only carbohydrates and sugar exhibiting a clear quadratic curve towards the
tail end of the distribution. In table 6 it can be observed that the quantifiable effects are
relatively small, with elasticity values rarely exceeding 0.1. The elasticities are all positive
indicating nutrients are normal goods, apart from sugar which becomes an inferior good22
at high levels of income, £1750 a week and greater. Carbohydrate has a similar quantifiable
effect, but does not slip into negative values at the end of the distribution.
22
An inferior good is a good for which you reduce consumption in response to an increase in income i.e.
has a negative income elasticity of demand.
41
42. Figure 12: Nutrient Engel curves (with 95% confidence intervals)
Calories and protein exhibit a mild quadratic trend in the quantifiable effects, initially
starting off with minimal values of around 0.03 at low incomes. Calories and protein go on to
reach a peak of around 0.09 at a higher income values of around £1500, and begin to exhibit
decreasing marginal effects for greater values of income. Fat and saturated fat display a
quadratic relationship characterised by persistent increasing marginal income effects. That
is the percentage increase in nutrient intake that results from a 1% increase in income, is
bigger the further along the income distribution to go. Both fat and saturated fat start
off with elasticities of around 0.03 at lower incomes, the elasticities increase monotonically
across the income distribution peaking at values of around 0.15. That is a 1% increase in
income for someone who earns £2000 a week, will result in them increasing their fat intake
by 0.15%. The nutrient Engel curve estimates exhibit more precision than the corresponding
age curve estimates; the confidence intervals only begin to flare out at very large values of
42
43. income.
Table 6: Income elasticities of nutrients (%)
Income(£) Calories Carb Fat Protein Sugar Sat-fat
250 0.03 0.03 0.02 0.04 0.04 0.03
500 0.05 0.05 0.04 0.06 0.07 0.05
750 0.07 0.06 0.06 0.08 0.08 0.07
1000 0.08 0.07 0.08 0.09 0.07 0.08
1250 0.08 0.07 0.10 0.09 0.06 0.09
1500 0.09 0.06 0.12 0.08 0.03 0.10
1750 0.09 0.05 0.13 0.07 -0.01 0.11
2000 0.08 0.03 0.15 0.06 -0.06 0.12
The elasticities in this table correspond to the curves in figure 12
7 Projected age effects
In this is section, estimates from tables 3 and 4 are used to project the possible age effects
that could arise from the UK’s ageing population. ONS projections predict the average age
of the UK population will increase significantly over the next 50 years. If food choices are
subject to age specific preferences then this could lead to a substantial increase or decrease in
consumption per household, this change in the structure of British diets could be beneficial
or detrimental to the health of the British population. As seen in the previous section the
way age effects expenditure on food items and nutrient intake is complicated as it varies
through the age distribution and between products. These results indicate that a change in
the composition of age will likely shift the consumption structure. To test whether this is
the case, ONS population projections are combined with the fitted values from the estimated
models in order to generate an out-of-sample prediction.
7.1 Projection methodology
This forecast is in no way a point estimate of where consumption per household will be in
50 years time as it is subject to very strong assumptions. However, it is a projection of the
possible impacts the estimated age effects could have if the population evolves in accordance
with ONS projections. The projections are carried out with a similar methodology to the
one used in by Lefebvre, (2006). Using ONS population predictions and fitted values from
prior estimations, Ct and ∆Ct% are calculated at every five-year interval until 2059, using
43
44. 2014 as the baseline year. Ct is the average consumption per household in a given year, and
is calculated using equation (2).
Ct = a PatXa
a Pat
(2)
Equation (2) consists of multiplying the predicted population P of each age group a in
a certain year t, by the fitted value from the estimated model Xa when evaluated at the
corresponding age a and all other variables are evaluated at the mean. This is done for each
age group and summed. It is then divided by the total projected population in that given
year a Pat. To work out the percentage change in consumption per person in a certain year
relative to the base year, equation (3) is used.
∆Ct% =
Ct
C2014
− 1 (3)
Equation (3) simply divides the estimated consumption per person in that year by the
average consumption per person in the base year (2014), and subtracts one. These estimates
have been carried out for all expenditure groups and nutrients and are plotted in figures
13 and 14. These projections are subject to some limitations in that they assume a ceteris
paribus situation where population characteristics and food technology do not vary over
time, however this does allow us to isolate the long term impact of age effects.
7.2 Projection results
In figure 13 you can see the projected changes in expenditure per person for all 11 composite
goods, plus total expenditure on food purchased for consumption in the home. It can be
seen that the age effects are likely to induce an increase in expenditure per person in most
food items, by varying magnitudes. The projections predict small decreases of up to 1% in
condiments and fats, and little effect on vegetable consumption per person. Larger increases
of 1 to 2.5% are predicted for soft drinks, meat, dairy, cereal and total food expenditure
over the forecast horizon. There are composite goods for which expenditure per person
rises significantly over the forecast horizon; consumption per person in sweets, fish and fruit
increase by almost 4%, and beverages increases by almost 5% over the forecast horizon.
By simultaneously analysing figure 6 it can be seen that the future predictions seem to be
contingent upon the dynamic of the age curve, from 60 years of age and greater. This is
mostly likely due to the fact the biggest growth in population over the next 50 years is
44
45. predicted to be in the over 60s. Vegetables and fats exhibit strong marginal decreases past
the age of 60, hence their projections for future changes in expenditure per person are very
small or negative. Fish and fruit display marginal increases with age even at latter stages in
the life-cycle, hence fish and fruit display much bigger increases in expenditure per person
over the forecast horizon.
Figure 13: Percentage change in expdenditure per person
All percentage changes are with respect to the 2014 base year. Estimated using ONS
population projections.
Included in figure 14 are the projections of calories per person for all the macronutrient
groups. It can be seen that the population ageing is likely to increase calorie intake per person
in all nutrient groups over the forecast period. Like the coefficient estimates, the effects on
nutrients tend to be smaller than effects on expenditure. Fat and protein consumption
per person increases by 0.5%, total calories per person increases by just under 1% over the
forecast horizon. The increase is more substantial for the remaining nutrients, carbohydrates
and saturated fat increase by just under 1.5%, and sugar intake increases by around 2.0%.
These impacts may seem small but a 1% in increase in calories consumed by the average
household is equivalent to an additional 350 calories per week. 3500 calories are required to
create a pound of body fat, assuming the extra calories are not offset with physical activity
45
46. it would only take 10 weeks for a household to put on an extra pound. Given that most
weight gain takes place over an extended period of time, it is the marginal increases and
decreases that one has to be concerned with. It is also important to recognise that this is
solely the impact of the estimated age effects, all the other variables that are also likely to
be influenced by an population ageing are held constant. Many of these variables such as
income and household size would be influenced by an ageing population in a way that would
increase intake per person.
Figure 14: Percentage change in calories per person
All percentage changes are with respect to the 2014 base year. Estimated using ONS
population projections.
8 Discussion
The very broad title of this paper is “The Effect of an Ageing Population on British Diets”,
more specifically the ultimate goal of this paper is to determine if there are age specific effects
that influence diets, if there are how are they influencing diets, and what consequences could
these effects have for the UK population in the event of population ageing?
This report contains a range of empirical analysis on the effect age has on expenditure
and nutrient intake. Based on the body of evidence in this paper you would conclude that
there are significant age effects that play an important role in determining British diets. I
estimate the relationship between age and consumption while controlling for cohort effects,
time effects and all other covariates that could potentially confound the estimates for age.
The estimates display highly significant age effects for the majority of goods and nutrients.
Further the variation in most instances is characterised by a quadratic relationship. The
results in this paper are consistent with established contributions of Chesher and Agostini.
46
47. With reasonable confidence that age effects are present, the question of how these effects
manifest themselves becomes the point of interest. There are a number of key empirical
findings in this paper, all of which have their own implications for policy makers and
wider society. In this next section I shall discuss the main findings from this paper and
the possible implications they may have.
8.1 Age effects
The expenditure and nutrient age curves featured in this report tend to display a particular
shape, which is characterised by a sharp initial increase23
which proceeds to level off around
the age of 60. Beyond the age of 60 it either continues to level off, or in some instances
begins to decrease. Some of the variation in consumption particularly the sharp initial
increase during adolescence and the sharp decline at the end of life-cycle, will be due to
biological factors that are contingent upon age, such as body size, growth, metabolic rate
and age dependent activity levels (Chesher, 1997). However there is no apparent reason for
the persistent increase in consumption observed from 30 years of age through till 60 years of
age. No major biological transformations that effect calorie requirements take place during
this period in the life-cycle, and there is no corresponding increase in physical activity, in fact
this is typically a period in the life-cycle when most people become increasingly sedative.
Like Chester I struggle to come to a conclusion on the reason for this result. However
regardless of being able to explain the result it still has important implications for policy
makers. In section 7 the estimates are applied to ONS population projections where we
find that age effects are likely to increase average calories per person. This indicates that
in the future age effects are likely to contribute by tipping the balance of calories in favour
of weight gain. Policies targeted at individuals of ages 30 to 60 need to be designed to
reduce the calorie intake over this period of the life-cycle, either by reducing consumption
or offsetting it with physical activity. This is of great importance as 30 years of age and
onwards is when individuals typically put on the excess weight (He et al., 2004) that results
in the poor health outcomes encountered later on in life.
One notable trend observed throughout the age related results, is that expenditure
values are more responsive to changes in age than the nutritional values are. This trend
becomes more evident when you compare the elasticities in tables 1 and 5. Early on in
the age distribution you can observe expenditure semi-elasticities of magnitudes that are
typically greater than 3, with some elasticities reaching values of 12.57. For nutrients the
23
This is also found in Chesher’s work
47
48. semi-elasticities of an individual who is 20 years of age typically fall in the range 1 to 2. For
expenditure you observe more extreme negative values at the end of the age distribution,
with values as extreme as -2.24, whereas for nutrients values tend to be less extreme -0.69.
The projected age effects in section 7 also tend to be smaller for nutrients. The phenomena of
nutrients being less responsive than expenditure has been observed in work that studies the
impact of prices and income. Studies of this nature find that expenditure reacts to changes
in prices and income more substantially than nutrient intakes do (Agostini, 2014; Tiffin et
al., 2011). This can be explained by the fact that expenditure does not equate to nutrient
intake; as households get older they may switch the non-nutrient characteristics of foods,
and increase purchases of branded items that are typically more expensive but have the same
nutritional content (Griffith et al, 2015). The negative semi-elasticities observed for some
goods towards the end of the age distribution could be explained by the increase in leisure
time that becomes available upon entering retirement. The increase in freely allocatable
time could be used to spend more time looking for lower prices (Stigler 1961), or households
may allocate more time to home production of food (Becker 1965). This result adds to the
growing consensus that taxes alone may not be able to solve the issues surrounding poor
nutrition (Tiffin and Arnoult, 2011; Briggs et al., 2013). Because households are able to
maintain an excessive consumption of calories despite significant changes to the consumption
structure of goods, taxes on specific goods may be somewhat ineffective at curbing excessive
nutrient consumption. Tiffin and Briggs suggest that taxes would be most effective when
implemented in conjunction with complementary policies such as information campaigns and
product reformation.
Despite the presence of an overarching trend, there is a great deal of heterogeneity in the
effect age has on different composite goods and nutrients. It is important to note the marked
differences between the age effects observed for different goods. This level of granularity can
illuminate some of the more specific policy opportunities. One issue that is high on the policy
agenda is promoting the consumption of fruit and vegetables, as high fruit and vegetable
intake is associated with a significant reduction in the risk of chronic diseases such as coronary
heart disease and cancer (Hung et al., 2004; Steinmetz and Potter, 1996; Joshipura et al.,
2001). The elasticity values in table 1 indicate that fruit expenditure is highly responsive
to age, for example an individual who is 30 years of age would increase their expenditure
on fruit by 4.73% in response to an additional year of age. Vegetable expenditure it is
much less responsive, for the same individual an additional year of age would only yield
an 1.84% increase in expenditure on vegetables. In the projections featured in section 7
48
49. this leads to large disparities in the projected changes in expenditure per person for the
two goods. Age effects are projected to increase fruit expenditure per person by 4% over
the forecast horizon, whereas age effects leave vegetable consumption unchanged (figure 13).
This indicates that in order to reach the goal of increased fruit and vegetable consumption,
considerable measures may need to be taken to increase vegetable consumption or it may
lag behind. Given vegetables make up the larger proportion of expenditure (Figure 5), it
would be particularly beneficial to implement policies that increase demand for vegetables
in order to offset the weak age effects associated with vegetable consumption.
In terms of expenditure on specific goods, age effects are likely to have a positive effect
on one aspect of British diets by considerably increasing the consumption of fish. The age
curve for fish in figure 6 exhibits persistent marginal increases in expenditure, even beyond
the age of 60 a characteristic not observed for any other of the composite goods. This trend
culminates in a large predicted age effect for fish, resulting in an increase in fish expenditure
per person which can be seen in figure 13. Superficially this is good news for British diets,
as fish is typically a good source of protein and bolsters diets with healthy fats.
A particularly notable result is the impact age effects have on calorie intake from sugar,
in the latter stages of the age distribution. The age curve for sugar differs from the age curves
corresponding to the other nutrients at the end of the age distribution. All the nutrients
exhibit marginal decreases in calorie intake past the age of 7024
, apart from sugar which
continues to display marginal increases in calorie intake with age. It has already been noted
in this report that the behaviour of age effects beyond the age of 60 plays a big role in
determining the projected age effects that result from an ageing population. It can be seen
in figure 14 that the age effects for sugar are likely to contribute by increasing sugar intake
per person by almost 2% over the forecast period. This is bad news given excessive sugar
consumption is already high on the policy agenda. This is a particularly salient finding as a
majority of the increased consumption per person will be among older individuals, who are
at much greater risk from the health outcomes associated with excess sugar consumption,
such as dental caries and type 2 diabetes (Burt et al., 2001; Malik et al., 2010). Looking
at figure 6 it can be deduced that the persistent marginal increases in sugar intake at the
latter end of the age distribution, are most likely the culmination of the persistent age effects
observed for sweets, cereals, fruit and soft drinks for individuals who are 60 years of age and
greater. This has meaningful implications for policy makers, currently a disproportionate
amount of effort is focused on reducing expenditure allocated to soft drinks, as a means to
24
Carbohydrates not so much but this is likely due to the proportion of carbs that consists of sugar.
49
