Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rethinking recidivism through a causal lens

Vik Shirvaikar, Choudur Lakshminarayan

TL;DR

This study reframes recidivism analysis from prediction to causality, treating incarceration as a treatment and applying two causal methods—DAG-based adjustment and double/debiased machine learning (DML)—to a well-known North Carolina dataset. Across both approaches and multiple subgroups, the authors find that additional prison time causally increases the likelihood of future recidivism, with the effect ranging from roughly 3–5 percentage points overall and stronger for individuals whose most recent offense was a felony. The work combines causal discovery with robust adjustment and a modern ML-based debiasing framework, augmented by a sensitivity analysis for unobserved confounding, to demonstrate that incarceration can have a detrimental causal impact within the studied context. While results are not intended to generalize beyond the dataset, the paper showcases a practical pipeline for applying causal inference to criminal justice questions and highlights how such methods can inform policy evaluation and reform efforts.

Abstract

Predictive modeling of criminal recidivism, or whether people will re-offend in the future, has a long and contentious history. Modern causal inference methods allow us to move beyond prediction and target the "treatment effect" of a specific intervention on an outcome in an observational dataset. In this paper, we look specifically at the effect of incarceration (prison time) on recidivism, using a well-known dataset from North Carolina. Two popular causal methods for addressing confounding bias are explained and demonstrated: directed acyclic graph (DAG) adjustment and double machine learning (DML), including a sensitivity analysis for unobserved confounders. We find that incarceration has a detrimental effect on recidivism, i.e., longer prison sentences make it more likely that individuals will re-offend after release, although this conclusion should not be generalized beyond the scope of our data. We hope that this case study can inform future applications of causal inference to criminal justice analysis.

Rethinking recidivism through a causal lens

TL;DR

This study reframes recidivism analysis from prediction to causality, treating incarceration as a treatment and applying two causal methods—DAG-based adjustment and double/debiased machine learning (DML)—to a well-known North Carolina dataset. Across both approaches and multiple subgroups, the authors find that additional prison time causally increases the likelihood of future recidivism, with the effect ranging from roughly 3–5 percentage points overall and stronger for individuals whose most recent offense was a felony. The work combines causal discovery with robust adjustment and a modern ML-based debiasing framework, augmented by a sensitivity analysis for unobserved confounding, to demonstrate that incarceration can have a detrimental causal impact within the studied context. While results are not intended to generalize beyond the dataset, the paper showcases a practical pipeline for applying causal inference to criminal justice questions and highlights how such methods can inform policy evaluation and reform efforts.

Abstract

Predictive modeling of criminal recidivism, or whether people will re-offend in the future, has a long and contentious history. Modern causal inference methods allow us to move beyond prediction and target the "treatment effect" of a specific intervention on an outcome in an observational dataset. In this paper, we look specifically at the effect of incarceration (prison time) on recidivism, using a well-known dataset from North Carolina. Two popular causal methods for addressing confounding bias are explained and demonstrated: directed acyclic graph (DAG) adjustment and double machine learning (DML), including a sensitivity analysis for unobserved confounders. We find that incarceration has a detrimental effect on recidivism, i.e., longer prison sentences make it more likely that individuals will re-offend after release, although this conclusion should not be generalized beyond the scope of our data. We hope that this case study can inform future applications of causal inference to criminal justice analysis.

Paper Structure

This paper contains 13 sections, 5 equations, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Predictive Modeling
  4. Survival Analysis
  5. From Prediction to Causality
  6. Methods
  7. Causal Discovery and Adjustment
  8. Double/Debiased Machine Learning
  9. Results
  10. Data and Limitations
  11. Causal Discovery and Adjustment
  12. Double/Debiased Machine Learning
  13. Conclusion

Figures (3)

  • Figure 1: Example of directed acyclic graph (DAG)
  • Figure 2: Directed acyclic graph for North Carolina recidivism data, elicited using PC algorithm with mixed-data conditional independence testing
  • Figure 3: Contour plot for double machine learning sensitivity analysis, showing additional explanatory power required for unobserved confounders to affect main causal effect