Use the technique discussed in Section9.4 to develop a GEE approach for zeroinflated Poisson model for count responses in longitudinal studies.

Section9.4:

Transcribed Image Text:

9.4 Marginal Models for Longitudinal Data Instead of explicitly modeling the correlation, we may deal with the correlation using robust inference. In marginal models, we only assume the marginal distribution for the outcome at each time point. For example, we may simply assume a linear regression for a continuous response at each time point: Yit ~N (xB, 02), or Yit =x+ B+ it, it ~N (0, 0). Such models are called marginal models because the distributional assumption for it does not involve the responses at other time points. GLMs such as logistic models may also be applied for categorical responses. Unlike linear mixed-effects models, inference for marginal models relies only on the mean response, which we will discuss shortly. For a longitudinal study with n subjects and m assessment times, let it be a response and xit a set of explanatory variables of interest from the ith subject at time t as before. A general marginal model may be specified for the response yit at each time t by the mean response: Yit | Xiti.d.f (it), 9 (pit)=x, 1