Mediation Analysis Based on Propensity Score Method

時間：2016年6月6日（一）9:25-12:30
地點：中央研究院經濟研究所慕寰廳
主辦單位：中央研究院經濟研究所

主講人：Prof. Martin Huber (Chair of Applied Econometrics, University of Fribourg)
 

This lecture discusses mediation analysis with a binary treatment based on propensity score methods, in particular inverse probability weighting and matching. The aim of mediation analysis is disentangling a (total) average treatment effect into an average indirect effect, which operates through an intermediate variable (or mediator) that is situated on the causal path between the treatment and the outcome, as well as the (unmediated) direct effect. In the first part of the lecture, a (sequential) selection on observables assumption is assumed which implies that all confounders jointly associated with the treatment, mediator, and/or the outcome are observed. In this case, matching on or weighting by two different treatment propensity scores can be used to identify the effects of interest. The approach is illustrated by an application to Swiss labor market data in order separate the direct effect of caseworker’s cooperativeness (or counselling style) on the employment of jobseekers from the indirect effect operating via the assignment of active labor market programs. The second part of the lecture discusses identification based on two distinct instruments for the (endogenous) treatment and the mediator in the presence of unobservable confounders. It is shown under which conditions local average direct and indirect effects can be identified for treatment compliers (whose treatment state reacts on its instrument) under a discrete or continuously distributed mediator. The approach is illustrated by an application to the US Job Corps program to disentangle the earnings effect of program participation, which may affect earnings indirectly via working longer hours and directly via higher wages per hour.