Numerically, compared to β in the GEE model, δ in the mixed effects model is in general larger in magnitude(20), but because the standard error for the estimate of δ is also larger their significance levels (p-value) are similar. The interpretations of these estimators also differ, and mixed effects models may be of greater clinical relevance. For example, if a doctor wants to describe to an individual patient how much her risk of HPV16 infection will change if her CD4+ count changes, the subject-specific estimator δ more directly addresses this. The population-averaged estimator given by the GEE is of greater interest from a public health perspective because it describes on average how two groups of women with different CD4+ levels will differ in prevalence of HPV infection. For a more detailed discussion on the comparison of β and δ, see Hu(22).
