Which Cases Go to Trial?: An Empirical Study of Predictors of Failure to Settle

Abstract

Although most scholars agree that litigated cases are an unrepresentative sample of legal disputes, scholars disagree on what causes trials, and there is little empirical evidence to illuminate the question. In order to understand the extent to which published cases reveal only a distorted view of peoples' behavior in response to legal rules, we need to determine the mechanisms by which cases are selected for trial.

Empirical tests of formal models of suit and settlement have been limited by the strong assumptions behind the models, assumptions that generally do not hold true in actual cases. Another difficulty in studying empirically what causes trials is the general lack of data on settlements. However, unlike most courts, the United States Tax Court (Tax Court), retains records on cases that settle after being docketed. Tax Court cases therefore provide an unusual opportunity to compare cases that went to trial with those that settled prior to trial. My study of these cases found that trials are nonrandom, and that identifiable features of the case, the judge, and the taxpayer are predictive of whether or not a case will go to trial. This is important new information that illuminates theoretical modeling of which cases go to trial.

My data set consisted of random sample of about 400 recent cases docketed in Tax Court, of which approximately half settled and the other half went to trial. In order to avoid the problem of limiting assumptions that are not true of actual cases, and the problem of subjective judgments about the merits of a case, I tested for nonrandomness by comparing the attributes of tried cases in the sample with the attributes of settled cases in the sample. More specifically, I tested whether any of nine characteristics of a case were statistically significant predictors of the type of outcome, that is, trial or settlement. Each of the nine variables used in the study, STAKES, APPEALS, JUDGETYPE, DECADE, BACKGROUND, PARTY, TAXPAYER, REGION, and COUNSEL, reflects a characteristic of either the case itself, the taxpayer involved in the case, or the judge assigned to the case. I used multiple regression analysis to test whether any of the nine independent variables, controlling for the other variables, were predictive of the dependent variable, TRIAL, which would indicate an effect on whether a case would go to trial or not.

Analysis of the data revealed that five of the nine independent variables, APPEALS, STAKES, JUDGETYPE, DECADE, and BACKGROUND, were statistically significant in predicting an increased likelihood that a case will go to trial. Thus, importantly, the data demonstrated a nonrandom selection of cases for trial: Tried cases were statistically different from settled ones. The specific results are even more interesting in that they provide insights into the mechanisms at work in Tax Court, which in turn illuminates the applicability of the various theoretical models to actual cases.

JEL Classification: K34, K41