False Confessions: An Experimental Study of the Innocence Problem

Abstract

The innocence problem, which occurs when an innocent person is falsely accused or convicted of a crime, is impossible to study with empirical data, because “true” innocence and guilt are unobservable in the “real world.” In this study, we replicate the criminal justice system in the laboratory with real salient crimes and subjects acting in the roles of defendants, prosecutors, and jurors in order to study the innocence problem. In a controlled environment, we identify individuals who are falsely or accurately accused of a crime and track them through the plea-bargaining system. This allows us to explore how being falsely accused of a crime affects plea bargaining decisions. We find evidence for a substantial innocence problem, reflected by a high willingness of truly innocent defendants to accept plea bargains. However, they do so at a lower rate than the truly guilty suggesting preferences for truth telling, an irrelevant factor in most economic theory. We also find evidence that individual preferences over uncertainty influence plea decisions, as predicted by economic theory. Overall, we find that loss aversion has a significant positive influence on plea decisions and that the reduced propensity of the truly innocent to accept plea bargains is driven by an interaction between their preferences to avoid lying and their preferences over uncertainty.