Sentencing algorithms like the one under development in Pa. are ‘deeply flawed,’ researchers say

It’s been nearly a decade since state lawmakers ordered the Pennsylvania Commission on Sentencing to create a mathematical equation to calculate a defendant’s risk for violence. 

Pennsylvania judges were supposed to use tool to dole out fairer sentences. But a growing body of evidence suggests such tools are deeply flawed.

This week, 27 leading criminal justice researchers signed a letter denouncing the use of sentencing risk algorithms in courtrooms across the country. They say the tools’ reliance on distorted data does little to reduce jail populations and unfairly penalizes minority defendants.

Sentencing algorithms like the one being developed in Pennsylvania use an array of data — including age, gender, and criminal history — to predict a defendant’s likelihood of recidivism, or committing another crime.

Some tools are designed for use in pre-trial sentencing — the stage at which a judge decides whether or not a defendant should be released on bail while they await trial. 

Cash bail, explained: How it works and why criminal justice reformers want to get rid of it

But criminal justice experts say that pre-trial violence is so rare that it’s hard to statistically predict it with any accuracy.

They also say that risk assessments of all types perpetuate bias, in part because black and hispanic defendants are more heavily policed, and therefore more likely to have criminal records. 

“Risk assessments that incorporate this distorted data will produce distorted results,” the letter reads. “These problems cannot be resolved with technical fixes. We strongly recommend turning to other reforms.”

The researchers sent the letter to letter sent to lawmakers in California and Missouri statehouses and Los Angeles County — three jurisdictions currently considering the use of pre-trial sentencing algorithms.

In Pennsylvania, lawmakers and criminal justice reform advocates have called on the General Assembly to repeal the mandate, first passed in 2010, requiring the sentencing commission to create a pre-sentencing risk assessment tool. 

The aim of the initial legislation was to divert non-violent offenders out of prison and minimize sentencing disparities, according to state Sen. Sharif Street, D-Philadelphia, who serves on the commission.

But in the years since, the Sentencing Commission has struggled to develop a sentencing tool that passes public muster. 

A draft algorithm they released in 2018 drew near-unanimous criticism from criminal justice reform advocates at public comment sessions in Philadelphia, according to WHYY.

Legislation that Street introduced this year would repeal the 2010 mandate once and for all. In a memo seeking his colleagues’ support for the bill, Street said that creating an unbiased, reliable sentencing tool may be an impossible task.

“Since the passage of the [2010] mandate, the Commission has worked hard to create an automated tool that is statistically predictive of risk and does not show bias against any protected class,” Street wrote in a memo seeking his colleagues’ support for the bill. “After reviewing more than eight years of thorough research and development conducted by the Commission, many, including members of the Commission itself, have serious concerns that such an automated tool is possible.”