Primary Submission Category: Mediation Analysis, Mechanisms
Post-treatment problems: What can we say about the effect of a treatment among sub-groups who (would) respond in some way?
Authors: Tanvi Shinkre, Chad Hazlett, Nina McMurry,
Presenting Author: Tanvi Shinkre*
Investigators are often interested in how a treatment affects an outcome for units responding to treatment in a certain way. We may wish to know the effect among units that, for example, meaningfully implemented the intervention, passed an attention check, or survived to the endpoint. Simply conditioning on the observed value of the relevant post-treatment variable introduces problematic biases. Further, assumptions such as “no unobserved confounding” (of the post-treatment mediator and the outcome) or of “no direct effect” (of treatment on outcome) required of several existing strategies are typically indefensible. We propose the Treatment Reactive Average Causal Effect (TRACE), which we define as the total effect of the treatment in the group that, if treated, would realize a particular value of the relevant post-treatment variable. Given the total effect of treatment, and by reasoning about the treatment effect among the “non-reactive” group, we can identify and estimate the range of plausible values for the TRACE. We discuss this approach and its connection to existing estimands and identification strategies, then demonstrate its use with two applications: (i) a community-policing intervention in Liberia, among locations where the project was meaningfully implemented, and (ii) a field experiment studying how in-person canvassing affects support for transgender rights, among participants whose feelings towards transgender people become more positive.