Compensation Without Distortion; Stochastic Termination and Incentive Compatibility in Preference Elicitation

High quality data is necessary across empirical fields. Often, this means eliciting preferences and beliefs from individuals in a variety of contexts. To ensure accuracy, individuals are compensated based their responses. However, when multiple questions are linked to payment, previous decisions can distort future behavior and bias results. I theoretically propose a stochastic mechanism, the Random Stopping Procedure (RSP), that mitigates many sources of preference “contamination” present in currently-used protocols. After every question, the RSP introduces some probability the experiment ends and the subject is paid based on their most recent response. Once a subject is asked another question, they know that their last answers are no longer “payoff relevant,” mechanically isolating decisions and avoiding contamination. Moreover, subjects are incentivized not to consider possible future questions, since those are only seen if the current decision is not paid. To test these theoretical predictions, I construct an experiment in a setting previous literature has suggested is heavily impacted by preference distortions. Results from this context demonstrate the existence of bias with current best practices, and the RSP can attenuate, if not eliminate, these data inaccuracies