Inferring Beliefs from Selected Samples: Evidence from Civil Litigation

Abstract

In many applications it is standard to assume that economic agents have complete knowledge about the distributions of uncertain outcomes. Often, the most likely source of information that agents will use to infer their beliefs about a particular distribution is the set of observed past realizations. In the case of civil litigation, litigants may base expectations of outcomes at trial on observed jury verdict awards in similar cases. However, economic models of settlement predict that the set of cases going to trial is nonrandom, so using these cases to estimate distributional parameters may cause litigants to make systematic mistakes. I show that when litigants base expectations about the variability of jury awards on observations of completed trials they systematically underestimate the true variability of awards. The model predicts that when litigants settle cases based on biased expectations, the observed variability in cases resolved at trial will be negatively autocorrelated from year to year; otherwise, it follows a random walk. I confirm the autoregressive pattern in award variability empirically using historical data on jury verdicts. These results suggest that economic agents use available information to formulate beliefs even if the information is faulty, and as a result likely make systematic errors in formulating beliefs.

Keywords: litigation, settlement, information

JEL Classification: K41, D83