Objective

Conduct a multi-site, retrospective investigation of predictive bias and disparate impact post-implementation of the Indiana Risk Assessment System-Pretrial Assessment Tool (IRAS-PAT).

Methods

Black and White defendants who received IRAS-PAT assessments (n = 2,570) were matched to two comparison conditions (n = 1,527 and n = 3,107) of defendants who did not receive assessments. Area under the curve statistics and multivariable logistic regression models tested for predictive bias. Weighted, multivariable mixed-effects models examined effects of assessments on release decisions by race.

Results

IRAS-PAT assessments produced lower levels of predictive validity for Black defendants relative to White defendants. Although there were disparities in pretrial release rates, bond amount, and days in detention, there was no evidence that effects of pretrial risk assessments differed by race.

Conclusions

Even a tool producing biased assessments of risk can improve pretrial release outcomes for defendants irrespective of race and relative to practice as usual.

中文翻译：

---

实践中审前风险评估的差异预测和不同影响：多地点评估

客观的

对印第安纳州风险评估系统-审前评估工具 (IRAS-PAT) 实施后的预测偏差和不同影响进行多地点回顾性调查。

方法

接受 IRAS-PAT 评估 ( n = 2,570) 的黑人和白人被告与 未接受评估的被告 的两个比较条件 ( n  = 1,527 和n = 3,107) 相匹配。曲线下面积统计和多变量逻辑回归模型测试预测偏差。加权、多变量混合效应模型检查了评估对种族释放决定的影响。

结果

与白人被告相比，IRAS-PAT 评估对黑人被告的预测有效性水平较低。尽管审前释放率、保证金金额和拘留天数存在差异，但没有证据表明审前风险评估的影响因种族而异。

结论

即使是产生有偏见的风险评估的工具也可以改善被告的审前释放结果，无论其种族如何，也与往常一样。