Measures of Association in Epidemiological Studies: How Best to Compare Discrete and Continuous Variables?

Abstract

The strength of the association in an observational study, often measured by a relative risk estimate, is an important factor in determining whether an exposure is likely to be causally related to the outcome. One Bradford-Hill criterion is that strength of association shifts the weight of evidence toward causation [1]. Assessing the relative strength of association between variables in a multivariate model is thus an important part of epidemiological data analysis. Today many variables may find their way into regression analyses. They may include both discrete (categorical) and continuous variables and be measured on non-commensurate scales. Direct comparison of beta-coefficients or measures of association derived from them, is often