ABSTRACT

A multivariate model is a mathematical model which describes how a number of variables and their parameters specify a stochastic variable. Effect modification means that the magnitude of the effect varies when a third variable varies. It is easy to test models with various kinds of effect modification when multivariate models are being constructed. Multivariate models are less sensitive than stratified data for sparse data. This chapter discusses various kinds of epidemiological data. A model which is linked to this and frequently used is the multiplicative Poisson model. Multivariate logistical regression has become the leading multivariate model for use with case-control studies. The chapter demonstrates how the logistic model can be specified so that it performs the same analysis as a stratified analysis, and how general programs for multivariate models can be used to calculate maximum-likelihood estimates in stratified material.