Pseudo-likelihood inference for clustered binary data

Abstract

Molenberghs and Ryan (1996) proposed a likelihood-based model for clustered binary data, based on a multivariate exponential family model (Cox, 1972). The model benefits from the elegance and simplicity of exponential family theory and is flexible in terms of allowing response rates to depend on cluster size. A main problem however, particularly with large clusters is the evaluation of the normalizing constant. In this paper, pseudo-likelihood is explored as an alternative mode of inference. The pseudo-likelihood equations are derived, the model is applied to data from a developmental toxicity study, and an asymptotic and small sample relative efficiency study is performed.