Questions: Ecological Analysis and the Ecological Fallacy

Question 1

A researcher finds a strong positive correlation between average per-capita alcohol consumption and liver cancer mortality rates across 50 countries. She concludes that individuals who drink more alcohol face higher liver cancer risk. What is the most serious methodological problem?

• A. The correlation coefficient may not reach statistical significance with only 50 countries
• B. Liver cancer mortality may be underreported in some countries, biasing the correlation
• C. Country-level correlations cannot validly establish that heavy-drinking individuals have higher cancer risk — drawing individual-level conclusions from group-level data is the ecological fallacy
• D. The study should have included more countries to increase statistical power

Answer: (C)

The core error is the ecological fallacy: inferring individual-level associations from group-level data. Countries that drink more may differ in dozens of other ways (income, healthcare access, diet, screening rates) that explain the mortality difference. We cannot know from the country-level correlation whether the individuals within those countries who drink the most are the ones getting liver cancer — that requires individual-level data.

Question 2

A researcher adds area-level poverty rates as a covariate to her ecological model of area-level alcohol use and liver disease. This adjustment...

• A. Fully resolves the ecological fallacy by controlling for the key confounder
• B. Addresses between-area poverty differences but cannot remove within-area individual-level confounding
• C. Makes the ecological fallacy worse by introducing additional ecological-level variables
• D. Is unnecessary if the original correlation was statistically significant at p < 0.05

Answer: (B)

Adding area-level covariates controls for differences between areas, but if poorer individuals within areas are both more likely to drink and more likely to develop the outcome, that within-area confounding remains completely unaddressed. The ecological fallacy cannot be resolved by adding more group-level variables — it requires individual-level data to partition exposure, outcome, and confounders at the person level.

Question 3

A strong positive ecological correlation between area-level smoking rates and lung cancer mortality provides reliable evidence that individual smokers face higher lung cancer risk.

• T. True
• F. False

Answer: False

This is a textbook ecological fallacy. Even a perfect group-level correlation does not establish individual-level causation — we would need individual-level data confirming that the smokers within high-smoking areas are the ones developing lung cancer. In practice, smoking and lung cancer do have a well-established individual-level causal relationship, but that conclusion comes from cohort and case-control studies, not ecological analysis.

Question 4

Ecological studies retain genuine scientific value for generating hypotheses and for studying exposures that are inherently area-level, such as policies or environmental pollutants.

• T. True
• F. False

Answer: True

Ecological analysis is not worthless — it is efficient when individual data are unavailable, appropriate when the exposure is truly contextual (air pollution, policy interventions), and useful for hypothesis generation. The discipline is interpretive: ecological associations describe places, not people, and should be used to motivate but not replace individual-level investigation.

Question 5

Why does adding area-level covariates to an ecological model fail to eliminate the ecological fallacy?

Model answer: Area-level covariates only adjust for variation between groups. If exposure and a confounder vary within areas — if, for example, poorer individuals within a region both drink more and have higher disease rates — then the within-area individual-level confounding is invisible to the model and completely unaddressed. Only individual-level data can disentangle who has the exposure, who has the confounder, and who gets the outcome.

The ecological fallacy arises because the within-group distribution of exposure, outcome, and confounders is hidden when only group averages are observed. A model that adjusts for the group average of a confounder removes the between-group effect of that confounder, but leaves within-group heterogeneity untouched. Multilevel models that include both individual and area-level data are required to properly partition this variance.