Comparing the performance of two estimation procedures, the weighted least squares and the generalized estimating equations, in analyzing repeated observations with time nestin

Public health researchers are increasingly encountering repeated categorical observations. Analyzing this type of data requires techniques that account for their inherent properties. Two estimation procedures are used in the analysis of repeated observations: the Weighted Least Squares (WLS) and the Generalized Estimating Equations (GEE). Using a longitudinal model with time nesting effects, we studied the performance of these two procedures under different conditions. We found disparity in estimation bias between the two procedures. On the other hand, both techniques were comparable in terms of efficiency and precision. We looked at the relation between the bias and the study factors and found few associations. We also compared estimates from the GEE using four different working correlation structures with estimates from the logistic regression and found no significant differences. Simulation technique and results are also presented.