Estimating Mediated Effects in Health and Epidemiogical Research

A central objective in health and epidemiological research is to examine the processes by which explanatory variables yield certain outcomes. For example, two overarching, policy-relevant issues are (1) the pathways through which various socio-demographic, attitudinal, and behavioural risk factors lead to particular health outcomes (e.g., death, disease), and (2) the mechanisms through which intervention and treatment programs produce their desired effects (e.g., increasing screening rates for serious illness, as well as reducing drug, alcohol, and tobacco use). Such issues can be addressed empirically via estimation of mediated (i.e., indirect) effects; however, the required statistical procedures are complex and not yet widely applied in the fields of health and epidemiology. The purpose of this paper is to provide an overview of the major theoretical and methodological concepts involved in mediated effects analysis, as well as present some illustrative applications using Structural Equation Modeling (SEM) and data from the 2003 Canadian Health Services Access Survey (HSAS).