Joint Modelling of Cluster Size and Binary and Continuous Outcomes

摘要

In clustered designs often multiple outcome variables are collected for each individual. Some of the dependent variables may be measured at the individual level while others (for example cluster size) may be measured at the cluster level. It is both important and challenging to model all variables jointly taking into account the correlation between the variables. In this paper we consider a data example with a binary and continuous individual-level outcomes and an ordinal cluster-level variable, define a multivariate random effects model and obtain maximum likelihood estimates using standard software. We also compare bias in dose effect estimates when misspecifying the correlation structure of the random effects and when ignoring cluster size using a simulation study.