A Unified Model of How Jurors Update and Re-Evaluate Evidence to Reach a Verdict.

摘要

Prominent researchers have suggested that criminal prosecutions should require more than the testimony of a single eyewitness. The present study investigated the effects of how additional evidence, beyond a single eyewitness, affected mock jury decision making. The results are compared to predictions from Bayes Theorem, which provides a normative/rational/statistical model of how mock jurors should update judgments given new information. Participants (N = 440, data collected via MTurk) read a summary of testimony from the victim of a home invasion, and then at Time 1 (T1), they judged the Likelihood of the defendant's Guilt, rated their Trust in the Victim's testimony, and rendered a Verdict. At Time 2 (T2), participants read one of six additional types of evidence: strong or weak forensic, strong or weak eyewitness, or strong or weak confession evidence, and answered the same questions as at Time 1. All participants were then instructed to disregard the additional T2 evidence, and at Time 3 (T3), they answered the questions for a third time. Three independent evaluations of each piece of additional evidence supported a "true" value of that information for Bayesian predictions. However using these values, Bayesian predictions for Likelihood of Guilt at T2 and T3 always over-adjusted the change in Likelihood of Guilt. At T2, Bayes over-predicted guilt, whereas at T3, Bayes underpredicted guilt. Contrary to Bayes Theorem, the value of prior information, such as the eyewitness's testimony, changed as new evidence was added or removed. All participants other than the weak forensic group increased in Likelihood, Trust, and Verdict at Time 2, and declined at Time 3 to levels similar to Time 1. The weak forensic evidence group declined on all measures at Time 2, which replicated Martire et al. (2013), but surprisingly they rebounded on Trust and Verdict at Time 3 to levels significantly higher than at Time 1. In conclusion, Bayes theorem does not describe how mock jurors use more than one piece of evidence in making decisions. The value of any piece of evidence depends in part on the value of other evidence. Multiple regression predicted mock juror decision-making better than Bayes theorem.