A Modified Model-Selection Criteria in a Generalised Estimating Equation for Latent Class Regression Models

Abstract

In recent years, generalised estimating equations (GEEs) have played an important role in many fields of research, such as biomedicine. In this paper, we use GEEs for latent class regression (LCR) with covariate effects on underlying and measured variables. However, there are only a few model-selection criteria in GEEs. The widely known Akaike information criterion (AIC) cannot be used directly, since AIC is a full likelihood-based model, whereas GEEs are nonlikelihood based. Hence, we propose a modification to AIC in GEEs for (LCR) models, where the likelihood is replaced by quasi-likelihood, and a proper adjustment is made by giving a penalty term. The data of the modified hospital elder life program (mHELP) project are used to illustrate our method.