Regression Methods for Epidemiological Analysis

Basic tabular and graphical methods are an essential component of epidemiological analysis and are often sufficient, especially when one considers only a few variables at a time. They are, however, limited in the number of variables that they can examine simultaneously and in the detail that they can analyze continuous variables. Regression analysis encompasses a vast array of techniques designed to overcome the numerical limitations of simpler methods. This advantage is purchased at a cost of stronger assumptions, which are compactly represented by a regression model. Such models and the assumptions they represent have the advantage of being mathematically explicit; a disadvantage is that the assumptions may not be well understood by the intended audience or even the user. Regression models should thus be tailored by the analyst to suit the topic at hand; the latter process is sometimes called model specification. This process is part of the broader task of regression modeling. To ensure that the assumptions underlying the regression analysis are reasonable approximations to reality, it is essential that the modeling process be actively guided by the scientists involved in the research, rather than be left solely to mechanical algorithms. Such active guidance requires familiarity with the variety and interpretation of models. Hence, the present chapter will focus primarily on forms of models and their interpretation, rather than technicalities such as fitting methods.