What You Need to Know About Pennsylvania’s Criminal Sentencing Algorithm
Could a computer program determine your fate after a criminal conviction? If the Pennsylvania sentencing commission has its way, computer-assisted sentencing could be just around the corner. While this technology is alluring, there are definite issues the level of foreknowledge this next generation of sentencing professes. Critics with the American Civil Liberties Union and the Defender Association of Philadelphia claim that the proposed Pennsylvania criminal sentencing algorithm cannot accurately predict the risk of recidivism in criminal offenders, and even worse expresses the same racial biases as some human judges.
How Does the Risk Assessment Instrument Work?
In 2010, the Pennsylvania legislature passed a law mandating the commonwealth’s sentencing commission to develop a risk assessment instrument “as an aid in evaluating the relative risk that an offender will re-offend and be a threat to public safety,” and that could identify which offenders should be offered alternatives to incarceration.
The risk assessment instrument is a complex algorithm that predicts an offender’s future behavior by taking into account factors like their age and criminal history. This information is fed into a computer program, which produces an estimate of a person’s risk of reoffending. Judges will consider the results from the risk assessment instrument in deciding how to sentence offenders in their courtrooms.
Many states, from Wisconsin to Florida, use these algorithms already. However, most are developed by private corporations such as Northpointe, whose product is called Correctional Offender Management Profiling for Alternative Sanctions, or COMPAS. Northpointe refuses to release to the public how its algorithm works, which raises important due process concerns.
When a Wisconsin offender tried to appeal his sentence on this basis, the state supreme court ruled that the use of COMPAS did not violate due process since the offender would have received a similar sentence with or without the use of the algorithm. Therefore, the court saw no reason to compel Northpointe to release the details of its risk assessment instrument.
Pennsylvania, on the other hand, is developing its own algorithm.
Pennsylvania Is Developing Its Algorithm in Full Transparency
Since it started work in 2010, the Pennsylvania Sentencing Commission has held 11 public hearings on its risk assessment tool and produced over 15 reports to explain changes to the algorithm in response to feedback. For example, in 2017 the Defender Association of Philadelphia successfully convinced the commission to use conviction records instead of arrest records. Because people of color are arrested at a much higher rate than whites, had the algorithm used arrest records, it would have predicted that people of color are much more likely to re-offend–an unacceptable result that reflects biases within the criminal justice system.
The sentencing commission has also been open about the inaccuracy of its algorithm in predicting violent recidivism. At best, the risk assessment tool can only predict with 18 percent accuracy the risk that someone will be arrested again for a violent offense. To increase the accuracy of the prediction, the commission has redefined recidivism as any arrest for a parole violation, misdemeanor, or felony within a three year period. But critics correctly point out that it doesn’t make sense to bundle parole violations, which could simply be missed curfews, together with serious criminal conduct.
The Sentencing Algorithm Will Have Limited Use
The commission is well aware of the algorithm’s potential to unfairly classify some people as high risk (or others as low risk). For this reason, they propose that judges should not use the algorithm’s results when it shows a defendant to be either very high risk or very low risk. In such case, judges will order a presentence investigation report that will delve into that particular offender’s circumstances and produce an individualized risk assessment.
According to the commission, the algorithm will never be used to give a person a harsher or more lenient sentence than usual. It will only be used to identify outlying cases that require an in-depth, individualized risk assessment. It’s good that the sentencing commission recognizes the limits of its risk assessment tool. However, it’s unclear why the commission has had to spend so much time and money to reach this conclusion.
The limits of sentencing algorithms have been known for some time. Academic research has consistently proven that algorithms cannot predict recidivism beyond 70 percent accuracy. In one study from Dartmouth University, random people selected from the internet were consistently better at predicting the likelihood of reoffending when compared to an algorithm developed by the researchers, which has the same 65 percent accuracy rate as COMPAS.
Perhaps the most troubling research was performed at George Mason University by Megan Stevenson, whose study suggests that judges often deviate from the recommendations of their risk assessment tools. This calls into question whether even the narrow use of the risk assessment tool recommended by the Pennsylvania sentencing commission is appropriate or practical.
In July, the commission pushed back the vote on the adoption if its risk assessment tool by six months in response to all the criticism it has received. In the meantime, they will publish another report on their algorithm, and the Urban Institute will conduct its own assessment of the program. Another round of public hearings will take place in December.
Our Pittsburgh Criminal Defense Lawyers Are Here to Help
At Worgul, Sarna & Ness, Criminal Defense Attorneys, we are closely following these developments as they may significantly impact the outcome of our clients’ cases. If you or a loved one are facing a criminal charge, don’t take unnecessary risk. We are highly experienced in the Pennsylvania sentencing process and are just a phone call away. Call us at (412) 281-2146 for a free, initial consultation.