What does exchangeability mean in observational economic data?

• A. The assumption that groups are identical in all measured characteristics.
• B. The assumption that groups are comparable after adjustment, enabling causal inference about differences.
• C. The assumption that outcomes are independent of exposure.
• D. The assumption that time-to-event is constant.

Answer: (B)

Exchangeability means that after you adjust for the observed characteristics, the groups defined by exposure would have had the same distribution of outcomes if no one had been exposed. In other words, once you control for those covariates, any differences you see in outcomes between exposed and unexposed groups can be attributed to the exposure itself, not to other differences between the groups. This is the condition that makes causal inference possible in observational data, because it mimics the balance you’d get from random assignment. In practice we aim for conditional exchangeability: E[Y(0) | X] = E[Y(1) | X], where X represents the covariates you adjust for, using methods like regression, matching, or propensity scores. The other options describe stronger or unrelated ideas: requiring groups to be identical in all measured characteristics is rarely true; independence of outcomes from exposure without conditioning ignores confounding; and time-to-event being constant is not about exchangeability.