Explaining Judge-Jury Differences. Table 9 suggests more rigorously than heretofore established the greater conviction proneness of judges, at least for our sample. It does not reveal the cases in which judges tend to disagree with juries or explain the cases in which they disagree. Models of conviction, standing alone, cannot reveal the pattern of disagreement. Yet it is the pattern of disagreement that most interested Xxxxxx and Xxxxxx. They limited much of their book to discussing cases in which judges and juries disagreed. After noting the generally high rates of judge-jury agreement, they all but excluded the agreement cases from their analysis. They classified disagreements as relating to guilt, charge, or hung-jury status, and then stated: “The precise quest of this study then is to explain what caused the disagreements in these 1063 instances, constituting 30 percent of all trials.”57 57Kalven & Zeisel, supra note 1, at 110. Table 7, Panel B, reports summary statistics for jury characteristics across the possible combination of judge-jury agreement and disagreement. To further explore disagreement, we divide the cases into three categories— those in which the judge and jury agreed, those in which the judge would have convicted but the jury acquitted, and those in which the jury convicted but the judge would have acquitted. These categories, whose occurrence rates are reported in Table 1, Panel A, are used to construct a categorical dependent variable in multinomial logit models.58 The models employ the same sets of explanatory variables as are used in Table 9. Table 10 reports the results. Models 1–3 account for the stratified struc- ture of the sample by site. Models 4–6 use dummy variables to control for locale effects. Table 10 suggests that education, a factor helpful in explaining jurors’ conviction proneness in Table 9, also helps explain judge- jury disagreement. In all models, the coefficient for “Education level” for the jury-convicts/judge-acquits outcome, compared to judge-jury agree outcome, is negative. This suggests that increased education is associated with the jury not convicting compared to agreeing with the judge. The coef- ficient for education for the judge-convicts outcome is consistently positive. This suggests that greater jury education is associated with the judge con- victing when the jury would not. In all models other than Model 6, the two education coefficients are statistically significantly different from each other (p < 0.05) or nearly so (p < 0.10),...
