Table 5:
Disparate Impact Decompositions
| Baseline | No Racial Bias | Equal Signal Quality | Both | |
|---|---|---|---|---|
|
|
||||
| Panel A: Change Black Parameters | (1) | (2) | (3) | (4) |
|
|
||||
| Disparate Impact | 0.047 | −0.042 | 0.095 | 0.039 |
| Release Rates (W/B) | 0.768 / 0.703 | 0.768 / 0.795 | 0.768 / 0.652 | 0.768 / 0.709 |
| Racial Bias | 0.074 | 0.000 | 0.074 | 0.000 |
| Marginal Outcomes (W/B) | 0.650 / 0.577 | 0.650 / 0.650 | 0.650 / 0.577 | 0.650 / 0.650 |
| Signal Quality (W/B) | 1.386 / 0.970 | 1.386 / 0.970 | 1.386 / 1.386 | 1.386 / 1.386 |
| Panel B: Change White Parameters | ||||
| Disparate Impact | −0.006 | 0.136 | 0.062 | |
| Release Rates (W/B) | 0.716 / 0.703 | 0.853 / 0.703 | 0.781 / 0.703 | |
| Racial Bias | 0.000 | 0.074 | 0.000 | |
| Marginal Outcomes (W/B) | 0.577 / 0.577 | 0.650 / 0.577 | 0.577 / 0.577 | |
| Signal Quality (W/B) | 1.386 / 0.970 | 0.970 / 0.970 | 0.970 / 0.970 | |
|
| ||||
| Judges | 268 | 268 | 268 | 268 |
Notes. Column 1 of this table reports average disparate impact and racial bias across judges and 250 simulations of the hierarchical MTE model, along with average release rates, marginal released outcomes, and signal quality of Black and white defendants. Simulations are based on the estimates from columns 2 and 4 of Appendix Table A18. Column 2 recomputes the statistics for a counterfactual in which Black (Panel A) or white (Panel B) release rates are set to eliminate racial bias, while column 3 adjusts Black (Panel A) or white (Panel B) signal quality to equalize signal quality across race. Column 4 applies both counterfactuals simultaneously.