Understanding Marginal and Conditional Effects from GEEs and GLMMs

One tricky thing in correlated and longitudinal data is understanding the difference between “marginal” (population-average) effects from generalized estimating equations (GEEs) and “conditional” (subject-specific) effects from generalized linear mixed models (GLMMs). This presentation tries to explain why in relatively simple terms, borrowing liberally from Fitzmaurice, Laird, and Ware.

In terms of level, it should be appropriate for anyone who understands logistic regression. Enjoy! Comments/error identifications welcome.