Identification and Estimation of Causal Mechanisms and Net Effects of a Treatment∗

Estimation of the effect of a treatment or intervention on a given outcome is an important topic in economics and many other sciences. Sometimes researches want to go a step further and try to understand the mechanisms through which the given treat-ment affects the outcome. Although some recent applications in economics have looked at this question, their parameters of interest are usually not clearly defined and the as-sumptions needed for a causal interpretation are rarely stated explicitly. In this paper we fill this void by considering identification and estimation of causal mechanisms of a treatment and causal effects of a treatment net of such mechanisms. We provide pre-cise definitions of our parameters of interest using the potential outcomes framework and motivate their usefulness for policy. Our parameters result in an intuitive decom-position of the total effect of a treatment on an outcome into the causal mechanism and the causal effect net of that mechanism. We then investigate conditions under which these causal effects are identified and can be estimated. We start by assuming a randomly assigned treatment, and then consider the case in which selection into the treatment is assumed to be random conditional on a set of covariates. We also state conditions under which the common approach of directly controlling for the observed value of the post-treatment or mechanism variable can be interpreted as estimating a causal average net treatment effect. We close with two empirical applications that illustrate the concepts and methods discussed in this paper. Key words and phrases: causal inference, post-treatment variables, principal strat-ification