Talk Title: Excludability is a Matter of Research Design: How to Conduct Credible Indirect Manipulations
Abstract: Since the credibility revolution, scholars have increasingly used experiments to assess whether a variable of interest affects an outcome. Because many putative causes cannot be directly manipulated, scholars often manipulate such variables using encouragements. When using instrumental variables regression to estimate the average effect of the encouraged variable on the outcome, scholars must assume that the encouragement affects the outcome only through the mediator of interest (the exclusion restriction). And yet few papers address the possibility that another mediator may transmit the encouragement's influence to the dependent variable. Here, we propose a three-step design-based approach for making the exclusion restriction more defensible. This approach uses theoretical arguments to design encouragements that are then tested iteratively using a combination of manipulation checks and estimation of ``first-stage'' effects of the encouragements on posited mediators. Although no method can ensure that that exclusion restrictions are satisfied, our proposed method leverages researchers' theoretical intuitions about causal pathways, multiple encouragements, and assorted manipulation checks to make core assumptions as credible as possible.
LOCATION:CGIS South S040 STATUS:CONFIRMED DTSTART:20250214T170000Z DTEND:20250214T183000Z END:VEVENT END:VCALENDAR