Difference-in-differences Designs for Controlled Direct Effects: An Application to Reducing Intergroup Prejudice


(with Adam Glynn, Hanno Hilbig, and Connor Halloran Phillips)

Recent experimental studies in the social sciences have demonstrated that short, perspective-taking conversations are effective at reducing support for discriminatory public policies, but it remains unclear if these effects occur even if subjective feelings about the minority group are unchanged. Unfortunately, the identification and estimation of the controlled direct effect—the natural causal quantity of interest for this question—has required strong selection-on-observables assumptions for any mediator. Given that this assumption is too strong for many social science settings, in this paper we show how to identify and estimate controlled direct effects under a difference-in-differences design where we have measurements of the outcome and the mediator before and after treatment is assigned. This design allows us to weaken the identification assumptions to allow for linear and time-constant unmeasured confounding between the mediator and the outcome. Furthermore, we develop a semiparametric efficient and multiply robust estimator for these quantities. We find that there is a robust controlled direct effect of perspective-taking conversations when subjective feelings are neutral but not positive or negative. An open-source software package implements the approach with a variety of flexible, machine-learning algorithms for nuisance functions estimation.