Building data science muscle to improve probation services
To take advantage of evidence-based approaches to supervising offenders on parole, one federal office wants to teach interested staff about data science.
The Administrative Office of the United States Courts’ Probation and Pretrial Services Office (PPSO) is looking for help training selected staff members in data science.
To take advantage of evidence-based approaches to community corrections supervision of offenders after they’ve released from federal prison, particularly for predicting recidivism and evaluating programs, PPSO wants to offer interested staff an “immersive course on data science,” it said in a recent solicitation.
PPSO said it currently uses risk assessment tools to assess released offenders' likelihood of committing new crimes, but it want to see whether data science tools could improve those predictions.
Additionally, PPSO produces large quantities of unstructured data that it has been unable to analyze, it said in a Q&A document. It said it expects to learn how data science applications could be applied to examine these unstructured data so it can be used for program evaluations and other potential research projects.
The four- to six-month overview would cover various machine-learning data science applications that would improve staff members’ capacity to predict key outcomes, such as recidivism, for individuals under federal supervision. It would address violent or sexual offenses, revocations from supervision, failed drug tests, successful/unsuccessful program participation, likelihood of successful officer contacts, attaining long-term employment, etc.
The course would introduce the foundations of data science and the data science toolkit. Participants would learn how to conduct exploratory data analysis, use traditional statistical modeling techniques and apply machine learning techniques. Additionally, PPSO wants an overview of data science concepts for senior management covering the major concepts, topical areas, capacities and challenges inherent in data science.
The use of algorithms for predicting recidivism has been hotly debated as some studies have shown that some applications are no better at predicting recidivism than untrained volunteers and others inject or boost racial bias.
A recent study by researchers at Stanford University and the University of California at Berkeley, however, found that risk assessment tools are significantly better at interpreting the complexity of the criminal justice system and making more accurate decisions than humans. When a large number of factors are considered, the algorithms performed far better than humans. In some tests, they were nearly 90% accurate in predicting which defendants might be rearrested; humans were on target only about 60% of the time.