In criminal cases judges evaluate and combine probabilistic evidence to reach verdicts. Unavoidably, errors are made, resulting in unwarranted conviction or acquittal of defendants. This paper addresses the questions (1) whether hearing cases by teams of three persons leads to less error than hearing cases alone; (2) whether deliberation leads to better decisions than mechanical aggregation of individual opinions; and (3) whether participating in deliberations improves future individual decisions. We find that having more than one judge consider cases reduces error effectively. This does not mean that it is necessary to deliberate about all cases. In simple cases many errors can be avoided by mechanical aggregation of independent opinions, and deliberation has no added value. In difficult cases discussion leads to less error. The advantage of deliberation goes beyond the case at hand: although we provide no feedback about the quality of verdicts, it improves individual decisions in subsequent cases.
