Logistic Regression Using SAS®: Theory and Application

8.8. Comparison of Methods

Table 8.4 summarizes differences among the four methods for handling clustered data. All four methods produce consistent estimates of the standard errors of the coefficients in the presence of clustering. But GEE goes further than robust variance estimation because it also produces coefficient estimates that have less sampling variability. On the other hand, GEE produces estimates of population-averaged coefficients rather than subject-specific coefficients. Population-averaged coefficients are subject to heterogeneity shrinkage—attenuation toward 0 in the presence of heterogeneity in the population. Heterogeneity shrinkage can be corrected by either conditional logit analysis or mixed models. Conditional logit analysis ...