Estimating Causal Moderation Effects with Randomized Treatments and Non-Randomized Moderators

TL;DR

This paper develops a generalized framework for estimating causal moderation effects in experiments with randomized treatments and non-randomized moderators, allowing flexible covariate adjustment and sensitivity analysis.

Contribution

It introduces a new method for identifying and estimating causal moderation effects with flexible adjustment strategies and extends sensitivity analysis to this context.

Findings

Applied framework to US welfare terminology and European attitudes towards asylum seekers.

Demonstrated the method's flexibility with different covariate adjustment techniques.

Provided insights into how moderators influence treatment effects in real-world scenarios.

Abstract

Researchers are often interested in analyzing conditional treatment effects. One variant of this is "causal moderation," which implies that intervention upon a third (moderator) variable would alter the treatment effect. This study considers the conditions under which causal moderation can be identified and presents a generalized framework for estimating causal moderation effects given randomized treatments and non-randomized moderators. As part of the estimation process, it allows researchers to implement their preferred method of covariate adjustment, including parametric and non-parametric methods, or alternative identification strategies of their choosing. In addition, it provides a set-up whereby sensitivity analysis designed for the average-treatment-effect context can be extended to the moderation context. To illustrate the methods, the study presents two applications: one dealing with the effect of using the term "welfare" to describe public assistance in the United States, and one dealing with the effect of asylum seekers' religion on European attitudes toward asylum seekers.