Conditional likelihood inference in generalized linear mixed models

Consider a generalized linear model with a canonical link function, containing both fixed and random effects. In this paper, we consider inference about the fixed effects based on a conditional likelihood function. It is shown that this conditional likelihood function is valid for any distribution of the random effects and, hence, the resulting inferences about the fixed effects are insensitive to mis- specification of the random effects distribution. Inferences based on the conditional likelihood are compared to those based on the likelihood function of the mixed effects model.