Research map / Probabilistic and scientific modeling / Causal discovery
Evaluating Pre-trial Programs Using Interpretable Machine Learning Matching Algorithms for Causal Inference
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Applies interpretable matching (FLAME-DAME) and other observational causal inference methods to evaluate a Durham, NC pre-trial program, finding no significant effect on new criminal charges.
Problem
Evaluating whether pre-trial community supervision programs reduce new criminal charges, where randomized trials are infeasible and prior observational methods lack interpretability or accuracy.
Approach
Uses FLAME-DAME, an interpretable machine learning matching algorithm that automatically selects matching covariates and forms interpretable matched groups, as a rigorous observational causal inference pipeline for program evaluation, compared against propensity score and other black-box methods.
Key results
- FLAME-DAME estimated an average treatment effect of 1.8% for new criminal charges with 95% CI including zero.
- Propensity score matching gave ATE of -0.0375 with 95% CI [-0.517, 0.440], also null.
- FLAME-DAME produced more cohesive matched groups than propensity score matching configurations, matching on more expert-identified covariates.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| participate in the CJRC pre-trial program rather than receive pre-trial release without supervision | People recommended for the CJRC program in Durham · New criminal charges within one year after case disposition | ATE: 0.0198 (null) | No clear effect |
| participate in the CJRC pre-trial program rather than receive pre-trial release without supervision | People recommended for the CJRC program in Durham · New criminal charges within one year after case disposition | ATE: -0.0375 (null) | No clear effect |
| participate in the CJRC pre-trial program rather than receive pre-trial release without supervision | People recommended for the CJRC program in Durham · New criminal charges within one year after case disposition | ATE: -0.0322 (null) | No clear effect |
| participate in the CJRC pre-trial program rather than receive pre-trial release without supervision | People recommended for the CJRC program in Durham · New criminal charges within one year after case disposition | ATE: -0.0315 (null) | No clear effect |
| participate in the CJRC pre-trial program rather than receive pre-trial release without supervision | People recommended for the CJRC program in Durham · New criminal charges within one year after case disposition | ATE: 0.0143 (null) | No clear effect |
Surprising findings
- Program showed no significant effect on new criminal charges, suggesting pre-trial supervision may be ineffective for low-risk individuals.
- Most individuals' charges were ultimately dismissed (82%), raising questions about whether supervision was warranted.
Limitations
- Observational design without randomization; unobserved confounding possible.
- Single program in one jurisdiction; generalizability limited.
- Smaller matched groups less reliable; DAME is computationally expensive.
Experimental setup
Datasets: CJRC program data Durham County jail records NC AOC ACIS criminal records
Metrics: average treatment effect 95% confidence interval matched group cohesiveness (number of matching covariates)
Baselines: Propensity Score Matching Mahalanobis distance matching GenMatch Causal BART
Builds on: FLAME DAME propensity score matching genetic matching causal BART
Closest papers by meaning
- Combining Machine Learning and Queueing Theory for Data-Driven Incarceration-Diversion Program Management (AAAI 2024 · similarity 0.84)
- On the Importance of Application-Grounded Experimental Design for Evaluating Explainable ML Methods (AAAI 2024 · similarity 0.82)
- Position: Causal Machine Learning Requires Rigorous Synthetic Experiments for Broader Adoption (ICML 2025 · similarity 0.80)
- Providing Fair Recourse over Plausible Groups (AAAI 2024 · similarity 0.80)
- Evaluating Index-based Treatment Allocation in Underresourced Communities (AAAI 2025 · similarity 0.80)
- Causal Fairness under Unobserved Confounding: A Neural Sensitivity Framework (ICLR 2024 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.