Regression models for pretest/posttest data in blocks

Abstract

We consider regression models with no intercepts to analyse pretest/posttest data from a dental study conducted under an experimental design involving a blocked factorial structure with two within individual factors. The proposed models accommodate block effects, heteroscedasticity, nonlinear relations between pretest and posttest measures and repeated measures. We compare multiplicative lognormal and gamma models to additive normal models fitted via generalized linear models methodology for repeated measures. Alternatively, we consider standard linear mixed models methodology to fit lognormal models, an option that facilitates modelling the within subjects covariance structure.