Researchers are often interested in analyzing conditional treatment effects.One variant of this is "causal moderation," which implies that interventionupon a third (moderator) variable would alter the treatment effect. In thispaper, I ask: What are the conditions and assumptions under which causalmoderation effects can be identified, and how can they be properly estimated? Ipresent a generalized, non-parametric framework for estimating causalmoderation effects given randomized treatments and non-randomized moderatorsthat achieves a number of goals. First, it highlights how conventionalapproaches in the literature do not constitute unbiased or consistentestimators of causal moderation effects. Second, it offers researchers asimple, transparent approach for the estimation of causal moderation effectsand lays out the assumptions under which this can be performed consistentlyand/or without bias. Third, as part of the estimation process, it allowsresearchers to implement their preferred method of covariate adjustment,including both parametric and non-parametric methods, or alternativeidentification strategies of their choosing. Fourth, it provides a set-upwhereby sensitivity analyses designed for the average-treatment-effect contextcan be extended to the moderation context.
展开▼
