Generalized Nonlinear Mixed-Effects Models

Although fairly general, the nonlinear mixed-effects model (7.4.6) of

Chapter 7 is somewhat restrictive in that both the S S random effects and the

within-subject errors are assumed to be normally distributed. Even in the

semi-nonparametric case where we relax our assumption regarding the

normali ty of the random-effects, the conditional model is s t i l l restricted by our

underlying normali ty assumptions on the within-subject errors, e-. Since many

applications dealing wi th repeated measurements involve discrete rather than

continuous outcomes (e.g., binary data, count data and ordinal data),

alternative methods have been proposed for these situations. In this chapter,

we consider a class of generalized nonlinear mixed-effects models which include

both population-averaged ( P A ) and subject-specific (SS) generalized linear

models. These models and the accompanying estimation methods allow us to

analyze repeated measurements data arising from a wide range of applications.