Bayesian Analysis for Probability-Matching

Probability-matching is a phenomenon where people, given a repeated choice between alternatives with different payoff rates, select each alternative proportional to its payoff rate. We attempt to isolate the strategy responsible for this generally irrational behavior using Bayesian data analysis to score three models over three conditions. While two strategies were broadly utilized, one causing significant differences in subjects’ performances over varied conditions, and the other causing few changes at all, the corresponding models did not predict subjects’ responses more accurately than their competitors, indicating that we have uncovered two persistent behavioral trends in probability-matching tasks, but not its general cause.