Development and Validation of the Oregon Pretrial Release Scales (OPRS)

Problem

The Problem Addressed: 

The report addresses limitations in existing pretrial risk assessment tools in Oregon. The two most commonly used tools, the Public Safety Checklist (PSC) and the Virginia Pretrial Risk Assessment Instrument (VPRAI), lacked accuracy and local applicability for predicting pretrial misconduct, including failures to appear (FTA) and new criminal offenses.

General Impact: 

Inaccurate risk assessments can result in either unnecessary detention or risks to public safety if high-risk individuals are released. Enhancing these tools could improve the equity, accuracy, and efficiency of pretrial release decisions.
 

Research Questions Answered:

1. How effectively can a new tool predict pretrial misconduct, including FTA and arrests for new offenses?
2. Can predictive accuracy be achieved across demographic subgroups without bias?
3. Can the tool be automated using existing criminal justice databases?
   
    

Method and Analysis

Program Evaluated: 

The study developed and tested the Oregon Pretrial Release Scales (OPRS), designed specifically for Oregon’s Circuit Courts, addressing gaps in the existing PSC and VPRAI tools.
 

Data and Sample Size: 

The study used data on 24,438 criminal cases filed in Oregon’s Circuit Courts in 2018, down from an initial dataset of 84,692 cases. Selection criteria ensured high confidence in the relevance of cases to pretrial risk assessment.
 

Analysis Used:

• LASSO regression: For selecting and weighting the most predictive risk factors.
• Bootstrapped, k-fold cross-validation: To ensure reliability and minimize bias.
• AUC (Area Under the Curve): To measure the accuracy of predictions, with thresholds to classify prediction strength (e.g., "moderately strong" if AUC > 0.639).
   

Outcome

Key Findings:

• The OPRS outperformed the PSC and VPRAI in predicting pretrial misconduct.
  • FTA: OPRS AUC = 0.730 (vs. 0.671 for PSC and 0.624 for VPRAI).
  • Any rearrest: OPRS AUC = 0.741 (vs. 0.675 for PSC and 0.614 for VPRAI).
  • Person rearrest: OPRS AUC = 0.663 (vs. 0.553 for PSC and 0.555 for VPRAI).
• The OPRS demonstrated minimal bias across sex and racial/ethnic subgroups, though slightly lower accuracy for FTA among Black and Hispanic defendants and higher accuracy for person-crime predictions among Native Americans.
   

Implications/Recommendations: 

• Implementation of OPRS: Recommended for state-wide use in Oregon to improve the accuracy and equity of pretrial decisions.
• Automation: Utilize existing criminal justice databases to streamline integration.
• Further research: Investigate ongoing performance and potential adjustments for demographic subgroup differences.

This report offers a strong foundation for practitioners and policymakers aiming to improve pretrial processes through evidence-based tools.