However, if HPV16 infection were assessed repeatedly over several visits it would not be appropriate to use model (1.1) to incorporate all the available data: model (1.1) would naively treat repeated observations of each individual over time as if they are independent (i.e., as if they were from different subjects), in essence amplifying the sample size. This can result in underestimation of the variation in the estimate of β and its P-value, as well as a confidence interval that is too narrow(10). To avoid these problems, a generalized estimating equation (GEE)(11) can be incorporated in the logistic regression model.
