Summary
Examines ecological and group-level designs, focusing on the gap between group-level data and individual-level inference that produces the ecological fallacy as Robinson documented with race and literacy in 1950. Walks through aggregate, environmental, and global variable types, three sources of ecologic bias including within-group misclassification and effect modification, and the modifiable areal unit problem in which boundary choices change correlations. Closes by invoking Geoffrey Rose's distinction between the etiology of a case and the etiology of incidence, defending group-level inferences for genuinely population-level questions like herd immunity while citing Dufault and Klar's evidence that most published ecologic studies fail to justify their unit choice.