Fixed Effects Regression Methods for Longitudinal Data Using SAS

4.7. Summary

Fixed effects models for count data can be estimated using either PROC GENMOD or PROC NLMIXED. When there are only two observations per individual, conditional maximum likelihood estimation of a fixed effects Poisson model can be implemented by converting the Poisson model into a logistic regression model for grouped data, with difference scores as independent variables. When there are more than two observations per individual, conditional maximum likelihood estimation of the Poisson model can be accomplished with PROC NLMIXED. Unconditional maximum likelihood can be done with PROC GENMOD using dummy variables to estimate the fixed effects. Fortunately, conditional and unconditional estimation of the Poisson model yield identical ...