The method of causal effect decomposition is intended to partition observed associations among variables into meaningful components based on path analyses according to the hypothesized causal mechanisms. While the causal effect decomposition method was first introduced to the field of sociology in the 1970s, it is largely unknown to epidemiologists. This paper describes the method and uses an empirical example to demonstrate that how an observed association among study variables can be partitioned according to their causal relationships. The findings suggest that when causal paths are specified, one can estimate the total effects (direct and indirect) of a factor on the outcome. Moreover, a causal decomposition approach also helps to differentiate and quantify different confounding effects. This demonstration underscores the importance of specifying and testing causal mechanisms in epidemiological studies.

causal decomposition, mediation, epidemiology method.