ABSTRACT
In Chapter 6, the pseudo-likelihood estimation procedure was proposed as a viable and attractive alternative to maximum likelihood estimation in the case of clustered (multivariate) binary data. In practice, one will often want to perform a flexible model selection. Therefore, one needs extensions of the Wald, score, or likelihood ratio test statistics to the pseudo-likelihood framework. Rotnitzky and Jewell (1990) examined the asymptotic distributions of generalized Wald and score tests, as well as likelihood ratio tests, for regression coefficients obtained by generalized estimating equations for a class of marginal generalized linear models for correlated data. Using similar ideas, we derive different test statistics, as well as their asymptotic distributions for the pseudo-likelihood framework. Liang and Self (1996) have considered a test statistic, for one specific type of pseudo-likelihood function, which is similar in form to one of the tests we will present below.
