There are, however, important differences in the estimates obtained from marginal and mixed effects models. Our paper examined the major current forms of these models relevant to the analysis of HPV data. The marginal models were GEE for the analysis of HPV prevalence, and WLW for the incident detection or clearance of HPV. Marginal models estimate the average relative difference in risk between two groups of subjects with different levels of risk factor exposure. This is often referred to as the “population-averaged effect”. The mixed effects models examined were frailty models for analyzing the incident detection (or clearance) of HPV, and mixed effects models for analyzing HPV prevalence. Mixed effects models estimate the change in risk that would occur within a person were risk factor exposure to change. This is often referred to as the “subject-specific effect”.
