Statistical modelling

This chapter provides a brief introduction to the role of statistical modelling in epidemiology, focusing on the use of generalized linear regression models. It also covers analysis of clustered data in more detail. The chapter discusses epidemiological analyses, which frequently involve consideration of categorical input variables (often called factors). All of the methods presented can be used both for data exploration and, with caution, for confirmatory analysis. Selection of the 'best' model depends on the problem being addressed. Model selection is not a simple task and, importantly, it is not a purely statistical task; it requires knowledge about the system being investigated. Model checking for mixedâeffects models is conducted to test the model assumptions. For linear models it is useful to examine the residuals, recalling that the total residual error is partitioned.