Table 3:

Mean Risk and Disparate Impact Estimates

	Linear Extrapolation	Quadratic Extrapolation	Local Linear Extrapolation
Panel A: Mean Risk by Race	(1)	(2)	(3)
White Defendants	0.338 (0.007)	0.319 (0.021)	0.346 (0.014)
Black Defendants	0.400 (0.006)	0.394 (0.021)	0.436 (0.016)
Panel B: System-Wide Disparate Impact
Mean Across Cases	0.054 (0.002)	0.054 (0.007)	0.042 (0.006)
Panel C: Judge-Level Disparate Impact
Mean Across Judges	0.054 (0.003)	0.054 (0.007)	0.042 (0.006)
Std. Dev. Across Judges	0.038 (0.003)	0.037 (0.003)	0.037 (0.003)
Fraction Positive	0.929 (0.016)	0.931 (0.036)	0.873 (0.036)
Judges	268	268	268

Notes. This table summarizes estimates of mean risk and disparate impact from different extrapolations of the variation in Figure 2. Panel A reports estimates of race-specific average misconduct risk, Panel B reports estimates of system-wide (case-weighted) disparate impact, and Panel C reports empirical Bayes estimates of summary statistics for the judge-level disparate impact prior distribution. To estimate mean risk, column 1 uses a linear extrapolation of the variation in Figure 2, while column 2 uses a quadratic extrapolation and column 3 uses a local linear extrapolation with a Gaussian kernel and a rule-of-thumb bandwidth. Robust standard errors, two-way clustered at the individual and judge level, are obtained by a bootstrapping procedure and appear in parentheses.