Hi,Fi Call Girl In Marathahalli - 7001305949 with real photos and phone numbers
Levels of analysis and levels of inference in ecological study
1. Presented by:
Kamal Bahadur Budha
Roll No: 11
MPH (HPE) 2017
School of Health and Allied Sciences
Faculty of Health Science
Pokhara University
1
2. Ecologic Studies
• “A study in which units of analysis are populations or
groups of people than individuals.” –[medical Dictionary, 2012]
• “An ecologic or aggregate study focuses on the
comparison of groups rather than individuals” –
Morgenstern, Modern Epi, 2008
• Units of observation are a population rather than
individual
– Country, region
– Group of population
• Mean values of hypothesized risk factor and outcome
are generated from each observation unit.
– Often plotted together to assess their relationship
2
3. Key issues with ecologic Studies
• Explores correlations between aggregate (group level)
exposure and outcomes
• Unit of analysis: usually not individual, but clusters
(e.g. countries, counties, schools)
• Useful for generating hypothesis
• Prone to “ecological fallacy”
• Cannot adjust well for confounding due to lack of
comparability (due to lack of data on all potential
covariates)
• Missing data is another concern
• Incomplete study(Kleinbaum et. al. 1982)
3
4. Levels of Measurement in Ecologic Studies
• Aggregate measures:
– Summarized the characteristics of individuals within a group as means or
proportions in groups, derived from individuals in each groups (e.g.
Prevalence of SHARC, Proportion of smoking, proportion of smokers and
Median family income)
• Environmental measures:
– Physical characteristics of the place in which members of each group works.
– Environmental measure has an analog at the individual level, but not easy to
measure
– E.g. air pollution level in a country or global
• Global measures:
– Attributes for groups or places for which there is not analog to individual
– E.g. population density, type of healthcare system, political system in the
country
4
5. Ecologic study of lung cancer risk factors in the US and
Japan, with special reference to smoking and diet
Example of Aggregate measures
characteristics of individuals
within a group as means or proportions in groups,
5
6. annual energy-related CO2 emissions /
population in different countries
5.13
16.9
1.37
10.8
8.6
9.2
7.3
15.4
0.12
6.66
16.22
1.56
10.2
9.35 8.93
7.12
16.61
0.21
0
2
4
6
8
10
12
14
16
18
China United states India Russian
federation
Japan Germany Islamic
republic of
Iran
Canada Nepal
CO2emissionsintonnes
tonnes CO 2 / capita
2009
2014
International Energy Agency (IEA) organization CO2 Emissions From Fuel Combustion
Example of Aggregate measures:
characteristics of individuals
within a group as means or proportions in groups,
6
7. Eg: Air pollution and case fatality of SARS in the
People's Republic of China: an ecologic study
Cui et al. Air pollution and case fatality of SARS in China
Example of Environmental measures
Physical characteristics of the place
Environmental measure characteristics of group
7
8. Mortality by inequality (Robin Hood index) in United States
(abbreviations are for each state).
Bruce P Kennedy et al. BMJ 1996;312:1004-1007
Example of Global measures:
Attributes for groups or places for which
there is not analog to individual
8
9. Levels of Analysis in Ecologic Studies
• Analysis is the process of breaking a complex topic or
substance into simple in order to gain a better
understanding of it.
Levels of Analysis deal with the
origin of that cause
•Individual
•Country/ national level
•International /systematic level
9
10. Levels of Analysis in Ecologic Studies
• The unit of analysis is the common level for which
the data on all variable are reduced and analyzed.
• Ecologic analysis can best be presented by dividing
ecologic study into
– Exploratory study (observe geographic differences in the
disease rate among several regions . For eg, For oral
cancers, one studies found a striking difference in
geographic patterns by sex: among men.)
– Multiple-group comparison study (ecologic association
by regressing Y on X)
– Time trend study (change in the average exposure level
and the change in the disease rate for a single
population)
– Mixed study (change in the average exposure level and
the change in the disease rate among several groups:
can calculate the RR, R, EF) Source: Morgenstern 198210
11. Levels of Analysis in Ecologic Studies
1. Individual-level analysis:
– Measurements are available for each individual in
the study
– Each variable is assigned to every subject in the
study
– Eg: average pollution level of each country might be
assigned to every subject
11
12. Levels of Analysis in Ecologic Studies
2. Completely ecologic analysis:
– All variables (exposure, outcome, covariates) are
ecologic measures, so unit of analysis is the
group(School, region, community)
– We do not know the joint distribution of variable
at individual level (frequency of exposed cases,
unexposed cases, exposed non case and
unexposed none cases)
– All we know the marginal distribution of each
variable. (proportion exposed and disease rate)
12
13. Levels of Analysis in Ecologic Studies
3. Partially ecologic analysis
– Information on certain joint distribution (M,N or
A/B frequency)
– but we still do not know the full joint distribution
of variables with in each group. (‘?’ cell are
messing)
– Eg.: ecological study of cancer incidence, joint
distribution of age and disease status then
estimation of age specific cancer rates.
– Outcome: cancer
– Covariate: Age
– Exposure: No 13
14. Joint distribution in each group of a simple ecological analysis
exposure status (X=1vs 0)
disease status (y = 1 vs 0)
Covariate status (z =1 vs 0)
T frequencies are the only data available in a completely ecological
analysis of all three variables,
M frequencies require additional data on the joint distribution of z and y
with each group.
N frequencies required additional data on the joint distribution of x and z
with each group,
A and B frequencies require additional data on the joint distribution of x
and y within each group,
? Cells are always missing ecological analysis
Z = 1
X = 1 X= 0
Y =1 ? ? M11
Y =0 ? ? M01
N11 N01 T0
Z = 0
X = 1 X= 0
? ? M10
? ? M00
N10 N00 T0
Total
X = 1 X= 0
A1+ A0+ T+1
B1+ B0+ T+1
T1+ T0+ T++
14
15. Levels of Analysis in Ecologic Studies
4. Multi-level analysis
– Combines data collected at two or more levels
– Contextual analysis
– Eg.: Individual level analysis might be conducted in
each group followed by ecological analysis of all
group using result from individual level
15
16. Ecologic study of lung cancer risk factors in the US and
Japan, with special reference to smoking and diet
Example of Completely ecologic
analysis
Additional information on certain joint distribution (M,N or A/B
frequency)
16
17. Annual energy-related CO2 emissions /
population in different countries
5.13
16.9
1.37
10.8
8.6
9.2
7.3
15.4
0.12
6.66
16.22
1.56
10.2
9.35 8.93
7.12
16.61
0.21
0
2
4
6
8
10
12
14
16
18
China United states India Russian
federation
Japan Germany Islamic
republic of
Iran
Canada Nepal
CO2emissionsintonnes
tonnes CO 2 / capita
2009
2014
International Energy Agency (IEA) organization CO2 Emissions From Fuel Combustion
Example of Individual-level
analysis
Each variable is assigned to every subject in the study
17
18. Eg: Air pollution and case fatality of SARS in the
People's Republic of China: an ecologic study
Cui et al. Air pollution and case fatality of SARS in China
Example of Partially ecologic
analysis
Additional information on certain joint distribution (M,N or A/B
frequency) but we still do not know the full joint distribution of
variables
18
19. Mortality by inequality (Robin Hood index) in United States
(abbreviations are for each state).
Bruce P Kennedy et al. BMJ 1996;312:1004-1007
Example of Partially ecologic
analysis
Additional information on certain joint distribution (M,N or A/B
frequency) but we still do not know the full joint distribution of
variables
19
20. Ecological and individual level analysis of risk factors for HIV infection
in four urban populations in sub-Saharan Africa with different levels of
HIV infection
Example of Multi-level analysis
Combines data collected at two or more levels
Source: Auvert et.al. 2001 20
21. Levels of Inference in Ecologic Studies
• Inference is the process of evolving from
observations and axioms to generalizations.
usually with calculated degrees of uncertainty
• Inference is the generalization of results from a
sample to the population from which the sample
came.
21
22. Levels of Inference in Ecologic Studies
Eg: When we draw inferences from normally
distributed data, base on the relationships of
mean and SD to the normal curve (illustrated in
Figure) when we draw inferences from data and
appears normal, we assume that the population
of sample data came to the normally distributed.
then assume that if we had all possible
observations from that population, we would find
that 68.3%, 95.5%, and 99.7% of the population
would lie between the mean and ___, __, __SD
Also assume that 95%CI
__%, ___ %, and ___ %
22
23. Areas under the normal curve that lie between 1, 2,
and 3 standard deviations on each side of the mean
23
24. Levels of Inference in Ecologic Studies
1. Biologic (Individual) inferences
About effects on individual risks
Eg. counselling: people's tobacco smoke is a cause of lung cancer.
2. Ecologic inferences
Ecological inferences about effect on group rate
Magnitude of ecologic effect depend not only one biologic
effect but also on degree and pattern of compliance with
polices, lows act in groups.
Validity of ecologic effect estimate depends on our ability to
control for differences among groups in joint distribution of
confounders including individual level variables
Eg.harmful effect i.e 90% coverege of negative effect on
Source: Morgenstern 1982
Connor et al. 1984
Valkanen 1966
24
25. Levels of Inference in Ecologic Studies
• Cross-level inferences
– Here, ecologic inference doesn’t always match the level of analysis
– It are often made ecological effects are interpreted as individual effects and
this is vulnerable to bias
– Eg. Explicit or implicit objective of ecological analysis may be to male a
individual inference about the effect of a specific exposure on individual
disease risk.
• Contextual effect
– Ecologic exposure on individual risk is also biologic inference
– If the ecologic exposure is an aggregate measure, we would generally want to separate
its effect for the effect of individual level analog.
– Eg. We might estimate the contextual effect of living in a poor area on risk of disease,
controlling for individual poverty level.
Source: Rothman
Humphreys 199125
27. Example: ecologic effect
E.g. Do rates of motorcycle-related mortality of riders vary
across different states that have different helmet laws in place
27
28. Conclusion
• The correlation at the group level was valid. It
was only invalid as a statement of individual
causal effect.
• Ecologic exposure on individual risk is also
biologic inference
• Despite these advantages, ecologic analysis
poses problems of interpretation when
making inferences at the individual level.
28
29. References
1. Ecologic study. (n.d.) Medical Dictionary for the Health Professions and
Nursing. (2012). Retrieved May 2 2017 from http://medical-
dictionary.thefreedictionary.com/ecologic+study
2. Rothman KJ, Greenland S, Lash TL. Modern epidemiology (3rd ed.). Wolterd
Kluwer Pvt: New Dilhi; 2009.
3. Morgenstern H. Ecologic Studies in Epidemiology: Concepts, Principles, and
Methods. Annual Review of Public Health 1995; Vol. 16: 61-81.
4. Morgenstern H. Ecological studies. In: Modern Epidemiology. 3rd Edition.
Editors: Rothman, Greenland, Lash. Lippincott Williams and Wilkins, 2008.
6. Ouellet JV, Kasantikul V. Motorcycle helmet effect on a per-crash basis in
Thailand and the United States. Traffic injury prevention. 2006 Mar 1;7(1):49-
54. Cui Y, Zhang ZF, Froines J, Zhao J, Wang H, Yu SZ, Detels R. Air pollution
and case fatality of SARS in the People's Republic of China: an ecologic study.
Environmental Health. 2003 Nov 20;2(1):15.
7. Nau, Henry R. Perspectives on International Relations, 2ndEdition, CQ Press,
2009.
8. Michael Beaney . "Analysis". The Stanford Encyclopedia of Philosophy.
Michael Beaney. Retrieved 23 May 2012
29
30. References
9. Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic research: principles and
quantitative methods. John Wiley & Sons; 1982 May 15.
10. Morgenstern H. Uses of ecologic analysis in epidemiologic research. American
journal of public health. 1982 Dec;72(12):1336-44.
11. Connor MJ, Gillings D. An empiric study of ecological inference. American journal of
public health. 1984 Jun;74(6):555-9.
12. Valkonen T. Individual and structural effects in ecological research. Helsingin
yliopiston Sosiologian laitoksen; 1966.
13. Humphreys K, Carr-Hill R. Area variations in health outcomes: artefact or ecology.
International Journal of Epidemiology. 1991 Mar 1;20(1):251-8.
14. Wynder EL, Taioli E, Fujita Y. Ecologic study of lung cancer risk factors in the US and
Japan, with special reference to smoking and diet. Japanese journal of cancer
research. 1992 May 1;83(5):418-23.
15. Auvert B, Buvé A, Ferry B, Caraël M, Morison L, Lagarde E, Robinson NJ, Kahindo M,
Chege J, Rutenberg N, Musonda R. Ecological and individual level analysis of risk
factors for HIV infection in four urban populations in sub-Saharan Africa with
different levels of HIV infection. Aids. 2001 Aug 1;15:S15-30.
30
Spatial epidemiology is a subfield of health geography focused on the study of the spatial distribution of health outcomes.
Specifically, spatial epidemiology is concerned with the description and examination of disease and its geographic variations. This is done in consideration of “demographic, environmental, behavioral, socioeconomic, genetic, and infections risk factors.
Example:
Ecological fallacy arises from thinking that relationships observed for groups necessarily hold for individuals: if provinces with more Protestants tend to have higher suicide rates, then Protestants must be more likely to commit suicide; if countries with more fat in the diet have higher rates of breast cancer, then women who eat fatty foods must be more likely to get breast cancer.
•Such inferences made using group-level data may not always be correct at the individual level.
•Ecological bias can be interpreted as the failure of associations seen at one level of grouping to correspond to effect measures at the grouping level of interest.
•For example, associations seen using country-level data may not correlate with associations that exist at the individual or neighborhood-level.
Ecological fallacy arises from thinking that relationships observed for groups necessarily hold for individuals: if provinces with more Protestants tend to have higher suicide rates, then Protestants must be more likely to commit suicide; if countries with more fat in the diet have higher rates of breast cancer, then women who eat fatty foods must be more likely to get breast cancer.
•Such inferences made using group-level data may not always be correct at the individual level.
•Ecological bias can be interpreted as the failure of associations seen at one level of grouping to correspond to effect measures at the grouping level of interest.
•For example, associations seen using country-level data may not correlate with associations that exist at the individual or neighborhood-level.
Incomplete designs ecologic is frequently used whe data are not readily available for conductiong another type of study. It is often relatively inexpensiive on convenient to make use of secoandary data sources to test or generete hypothesis with these design before spending considerablly more time and money on primary data collection.
Missing : ecological design are studies in which information is missing on one or more relevant factors. Since we can never know whether all distorting influences.
More information : International Energy Agency (IEA) organization CO2 Emissions From Fuel Combustion:
http://www.iea.org/
Reference: Mortality by inequality (Robin Hood index) in United States (abbreviations are for each state)
The Robin Hood Index is conceptually one of the simplest measures of inequality used in econometrics. It is equal to the portion of the total community income that would have to be redistributed (taken from the richer half of the population and given to the poorer half) for the society to live in perfect equality.
The Robin Hood index is based on the Lorenz Curve and is closely tied to the better known inequality measure the Gini coefficient which is also based on the Lorenz curve.
In layman words, the Robin Hood index is the proportion of money needed to be transferred from the rich to the poor to achieve equality.
The Lorenz Curve is a graphical representation of the proportionality of a distribution. It represents a probability distribution of statistical values, and is often associated with income distribution calculations and commonly used in the analysis of inequality.
The Gini Coefficient is a way to measure equity and is derived from the Lorenz curve.
The Gini coefficient is defined graphically as a ratio of two surfaces involving the summation of all vertical deviations between the Lorenz curve and the perfect equality line (A) divided by the difference between the perfect equality and perfect inequality lines (A+B).
The Gini coefficient is defined as a ratio with values between 0 and 1.
How the Robin Hood Index is calculated
The Robin Hood index is equivalent to the maximum vertical distance between the Lorenz curve, or the cumulative portion of the total income held below a certain income percentile, and the Perfect Equality Line, that is the 45 degree line of equal incomes.
The value of the index approximates the share of total income that needs to be transferred from households above the mean to those below the mean to achieve equality in the distribution of incomes.
The Robin Hood index values
The Robin Hood index is a measure of income inequality ranging from 0 (complete equality) to 100 (complete inequality).
Analysis begins with describing the characteristics of the subjects and progresses to calculating rates, creating comparative tables (e.g., two-by-two tables), and computing measures of association (e.g., risk ratios and odds ratios), tests of statistical significance (e.g., chi-square), confidence intervals, and the like.
The fundamentals of ecologic analysis can best be presented by dividing ecologic study designs into four types.
The simplest of these is the exploratory study in which we observe geographic differences in the disease rate among several regions (groups). The objective is to search for spatial patterns that might suggest an environmental etiology or a special etiologic hypothesis. No exposures are measured and, generally, no formal data analysis is used. For example, the National Cancer Institute (NCI) mapped the age-adjusted cancer mortality rates in the US by county for the period 1950-69
For oral cancers, they found a striking difference in geographic patterns by sex: among men.
In the multiple-group comparison study, we observe the association between the average exposure level (X) and the disease rate (Y) among several groups. We can quantify and test this ecologic association by regressing Y on X, i.e., by fitting the data to a mathematical model5, such as Y = Bo + BIX (4)
where Y is the predicted value of Y for any given value of X, B1 is the estimated slope, and Bo is the estimated Yintercept.
In the time trend study (or time series study), we observe the relationship between the change in the average
exposure level (or intervention) and the change in the disease rate for a single population. t With time trend studies involving a sudden change in exposure, such as the start of an intervention program, we compare the slope in the disease trend before and after the intervention. For example, in Figure 2, the age-adjusted mortality rates for two infectious diseases are graphed for the period 1900-1973.
In the mixed study, we observe the relationship between the change in the average exposure level and the change in the disease rate among several groups. The analysis and interpretation of such data are similar to that of the comparison study; the only difference is that in the mixed design both variables are measured as absolute changes between the same two times (or periods).7 Thus, we can estimate the risk ratio (RR), the correlation coefficient (R), and the etiologic fraction (EF) from the ecologic regression coefficients, using expressions 5, 6, and 3. Frequently, it is more convenient or informative to categorize the exposure variable and to compare the changes in disease rate among groups having different mean exposure levels. For example,
Crawford, et al,'3 observed the absolute changes in the average annual cardiovascular disease (CVD) mortality rate
The fundamentals of ecologic analysis can best be presented by dividing ecologic study designs into four types.
The simplest of these is the exploratory study in which we observe geographic differences in the disease rate among several regions (groups). The objective is to search for spatial patterns that might suggest an environmental etiology or a special etiologic hypothesis. No exposures are measured and, generally, no formal data analysis is used. For example, the National Cancer Institute (NCI) mapped the age-adjusted cancer mortality rates in the US by county for the period 1950.69.2 For oral cancers, they found a striking difference in geographic patterns by sex: among men.
In the multiple-group comparison study, we observe the association between the average exposure level (X) and the disease rate (Y) among several groups. We can quantify and test this ecologic association by regressing Y on X, i.e., by fitting the data to a mathematical model5, such as Y = Bo + BIX (4)
where Y is the predicted value of Y for any given value of X, B1 is the estimated slope, and Bo is the estimated Yintercept.
In the time trend study (or time series study), we observe the relationship between the change in the average
exposure level (or intervention) and the change in the disease rate for a single population. t With time trend studies involving a sudden change in exposure, such as the start of an intervention program, we compare the slope in the disease trend before and after the intervention. For example, in Figure 2, the age-adjusted mortality rates for two infectious diseases are graphed for the period 1900-1973.
In the mixed study, we observe the relationship between the change in the average exposure level and the change in the disease rate among several groups. The analysis and interpretation of such data are similar to that of the comparison study; the only difference is that in the mixed design both variables are measured as absolute changes between the same two times (or periods).7 Thus, we can estimate the risk ratio (RR), the correlation coefficient (R), and the etiologic fraction (EF) from the ecologic regression coefficients, using expressions 5, 6, and 3. Frequently, it is more convenient or informative to categorize the exposure variable and to compare the changes in disease rate among groups having different mean exposure levels. For example,
Crawford, et al,'3 observed the absolute changes in the average annual cardiovascular disease (CVD) mortality rate
More information : International Energy Agency (IEA) organization CO2 Emissions From Fuel Combustion:
http://www.iea.org/
Reference: Mortality by inequality (Robin Hood index) in United States (abbreviations are for each state)
The Robin Hood Index is conceptually one of the simplest measures of inequality used in econometrics. It is equal to the portion of the total community income that would have to be redistributed (taken from the richer half of the population and given to the poorer half) for the society to live in perfect equality.
The Robin Hood index is based on the Lorenz Curve and is closely tied to the better known inequality measure the Gini coefficient which is also based on the Lorenz curve.
In layman words, the Robin Hood index is the proportion of money needed to be transferred from the rich to the poor to achieve equality.
The Lorenz Curve is a graphical representation of the proportionality of a distribution. It represents a probability distribution of statistical values, and is often associated with income distribution calculations and commonly used in the analysis of inequality.
The Gini Coefficient is a way to measure equity and is derived from the Lorenz curve.
The Gini coefficient is defined graphically as a ratio of two surfaces involving the summation of all vertical deviations between the Lorenz curve and the perfect equality line (A) divided by the difference between the perfect equality and perfect inequality lines (A+B).
The Gini coefficient is defined as a ratio with values between 0 and 1.
How the Robin Hood Index is calculated
The Robin Hood index is equivalent to the maximum vertical distance between the Lorenz curve, or the cumulative portion of the total income held below a certain income percentile, and the Perfect Equality Line, that is the 45 degree line of equal incomes.
The value of the index approximates the share of total income that needs to be transferred from households above the mean to those below the mean to achieve equality in the distribution of incomes.
The Robin Hood index values
The Robin Hood index is a measure of income inequality ranging from 0 (complete equality) to 100 (complete inequality).
inference The process of evolving from observations and axioms to generalizations. In statistics, the development of generalization from sample data, usually with calculated degrees of uncertainty. Causal inference from observational data is a key task of epidemiology and other sciences as sociology, education, behavioral sciences, demography, economics, or health services research; these disciplines share methodological frameworks for causal inference
Sometimes we calculate measures of location and dispersion to describe a particular set of data. At other times, when the data represent a sample from a larger population, we might want to generalize from our sample to the larger population that the data came from—or, said another way, we want to draw inferences from the data. A large body of statistical methods is available to allow us to do this. In this section, we will look at some of the methods for drawing inferences from data that are normally distributed.
When we draw inferences from normally distributed data, we base our conclusions on the relationships of the standard deviation and the mean to the normal curve. We use these relationships, which were illustrated in Figure 3.9, when we draw inferences from data. When the graph of a frequency distribution appears normal, we assume that the population of data our sample came from is normally distributed. We then assume that if we had all possible observations from that population of data, we would find that 68.3%, 95.5%, and 99.7% of the population would lie between the mean and ―SD
inference The process of evolving from observations and axioms to generalizations. In statistics, the development of generalization from sample data, usually with calculated degrees of uncertainty. Causal inference from observational data is a key task of epidemiology and other sciences as sociology, education, behavioral sciences, demography, economics, or health services research; these disciplines share methodological frameworks for causal inference
Area Variations in Health Outcomes: Artefact or Ecology
KEITH HUMPHREYS & ROY CARR-HILL
It is a long-standing belief that the size of the difference between the poor (lower social groups) and the rich (higher social groups) in health outcomes will vary according to the characteristics of the area. However typical approaches to analyses of this kind of question violate standard statistical assumption.
The basic problem is how to estimate the size of a ‘ward effect’ (the disadvantage of living in a ‘poor’ ward over and above effects associated with individual or household circumstances). This is complicated by the hypothesized existence of intra-ward correlated errors; the only way to avoid this bias is to explicitly model the different variance components using multi-level modelling techniques. The purpose of this paper is to illustrate this technique.
Analysis using several of the health outcomes in the Health and Lifestyle Survey data, suggests that the ward effect is quite substantial, and remains after ‘controlling for’ age, gender and several other socio-demographic variables. This ‘ward effect’ appears to be best represented by the proportions without access to a car and the preponderance of working class members (RGSC IV and V) in the population.
Whilst the verdict on the original hypothesis remains ‘not proven’, the hypothesis has been shown to be more complex than its simplistic statement suggests. The analyses have shown how to unpack these complexities and, more generally, have illustrated the power of the multi-level modelling technique.