Assessing Risk Among Correctional Community Probation Populations: Predicting Reoffense With Mobile Neurocognitive Assessment

The study introduces the first mobile, self-scoring risk assessment software, using neurocognitive testing to predict reoffense among probation populations. This software, administered on a tablet, evaluates cognitive traits such as impulsivity, empathy, and aggression, aiming to predict recidivism and aid in directing offenders towards appropriate rehabilitative programs.

BACKGROUND

Traditional risk assessments have relied on static factors (e.g., criminal history) and subjective measures, which can introduce biases and fail to capture dynamic behavioral changes. The NeuroCognitive Risk Assessment (NCRA) diverges from these approaches by employing interactive, gamified neurocognitive tests. Using machine learning, the NCRA offers an innovative, unbiased alternative to predict recidivism more accurately while reducing administrative burdens.

KEY FINDINGS

  1. New Assessment Tool: A mobile, neurocognitive assessment tool was developed to predict reoffense among probationers, measuring traits like impulsivity and aggression using machine learning.
  2. Comparable Accuracy: The tool predicted recidivism with an AUC of 0.70, comparable to traditional risk assessments, while reducing subjectivity and potential bias.
  3. Efficiency and Objectivity: This self-scoring tool eliminates the need for lengthy interviews, reduces administrative tasks, and requires minimal training.
  4. Data and Analysis: Data from 730 probationers in Houston were analyzed using machine learning to link cognitive traits with reoffending.
  5. Future Applications: The tool aims for broader use across different crime types and justice system stages, such as pre-trial and re-entry programs, enhancing individualized rehabilitation strategies.

RECOMMENDATIONS

– Wider Implementation: Expand the use of NCRA at various stages of the criminal justice process, including pre-trial, probation assessment, and post-release supervision, to refine its predictive capabilities across diverse populations.

– Enhanced Data Collection: Increase sample size and incorporate more diverse crime categories to fine-tune the predictive model, particularly for specific offense types like violent or non-violent crimes.

– Bias Mitigation: Further refine the NCRA to eliminate dependence on demographic factors entirely, focusing solely on cognitive measures to enhance fairness and objectivity in risk predictions.

CONCLUSION

The NCRA provides a novel, efficient, and unbiased approach to recidivism prediction through mobile neurocognitive assessments. Its gamified, self-administered format, combined with advanced machine learning analytics, positions it as a valuable tool in optimizing offender management and rehabilitation strategies. Continued development and broader application of this technology could significantly improve criminal justice outcomes and public safety by more accurately assessing reoffense risks and tailoring interventions accordingly.

Sample and Demographics

The study involved 730 probationers from Harris County, Texas, assessed between 2017 and 2019. The sample included 75.3% males and 24.7% females, with offenses ranging from misdemeanors to felonies. Of the participants, 17.3% recidivated within two years.

Predictive Validity

The NCRA achieved a recidivism prediction accuracy with an Area Under the Curve (AUC) of 0.70, comparable to existing risk assessments. The AUC measures the model’s ability to distinguish between those who will recidivate and those who will not, with values ranging from 0.5 (no predictability) to 1 (perfect predictability).

Cognitive Measures

The NCRA evaluates seven neurocognitive tests, including the Balloon Analog Risk Task (measuring risk-taking) and the Go/No-Go task (measuring impulsivity), providing insights into behavioral traits linked to criminal activity.

Machine Learning Application

The study utilized machine learning techniques, such as Generalized Linear Models and Linear Discriminant Analysis, to enhance predictive accuracy by optimizing feature selection and accounting for data non-linearities. The Recursive Feature Elimination (RFE) method improved the predictive performance by 0.02 on the AUC scale.

Comparative Performance

The NCRA, using cognitive data alone, achieved an AUC of 0.66, which increased to 0.70 when demographic data (age, gender, current offense) were included. This demonstrates the potential of the NCRA to outperform or match traditional assessments using purely neurocognitive measures.