DP11343 Persuasion Bias in Science: Can Economics Help?
The widespread adoption of randomized controlled experiments owes much to their ability to curtail researchers' conflicts of interest. This paper casts data collection and analysis in a game-theoretic framework. A researcher aims at persuading an evaluator that the causal effect of a treatment outweighs its cost, so as to justify acceptance. The researcher uses private information to (1) sample subjects based on their treatment effect (challenging external validity), (2) assign subjects to treatment based on their baseline outcome (challenging internal validity), or (3) selectively report experimental outcomes (challenging both external and internal validity). The resulting biases have different welfare implications: for sufficiently high acceptance cost, the evaluator loses in cases (1) and (3), but benefits from the researcher's information in case (2).
