Abstract:

We show endogenous information production can cause agent motivation and potential outcomes to vary across open RCTs and real-world settings with discretionary choice. Before treatment/control assignment/choice, agents can generate costly signals about (treatment-related) effort returns. Since information improves treatment/control choice, incentives for information production are generally higher under discretion. Information has three effects: higher mean outcomes conditional upon effort; an ambiguous effect on mean effort as positive/negative signals increase/decrease effort returns; and higher effort to facilitate learning about signal quality. Each effect can generate SUTVA violations. Since signals improve effort decisions, the return to information increases with RCT treatment probability. If information is sufficiently unimportant for discretionary choice, there exists a unique RCT treatment probability achieving external validity. At this treatment probability, expressed preference for treatment equals the observational data treatment probability, a simple feasible diagnostic. Finally, we consider implications for: randomised micro-credit; draft lotteries; and open medical trials.

General information:

Attendance to this seminar is possible by invitation only. Please send an e-mail to finsec-abs@uva.nl if your are interested in attending this seminar.