The U.S. Department of Justice now encourages the use of risk assessments based on predictive modeling algorithms at all stages of the criminal justice process. Criminal sentencing has long been based on the crime and the defendant’s past criminal record. Judges can now consider a new dimension: the future.

Predicting Dangerous Criminals

Criminologists have long tried to predict which criminals are more dangerous before deciding whether they should be released. Today computers crunch data—arrests, type of crime committed, and demographic information—and a risk rating is generated.

Risk assessments have existed in various forms for years, a century in fact. But during the past 20 years, they have gained favor in the justice system, driven by advances in social science. The tools try to predict recidivism. These statistical probabilities are based on factors like age, employment history, and prior criminal record.

Risk assessment is used at some stage of the criminal justice process in nearly every state — guiding decisions about which prisoners to offer parole and helping to set bail for inmates awaiting trial.

Modern risk tools were originally designed to help judges determine what types of treatment an individual would most profit from — from drug treatment to mental health counseling.

But being given a high risk score (and therefore ineligible for alternative treatments) usually meant incarceration. Defendants rarely had the opportunity to challenge their assessments.

New Laws Create New Challenges

In the 1980s, new laws established many mandatory sentences and in many cases, abolished parole. The result made exercising discretion much more difficult for judges and parole boards.

But as prison populations and costs soared, the idea to create a guide less likely to be subject to unconscious biases, the mood of a judge, or other human shortcomings began to win favor.

Similar assessment tools have long been used to decide where to place inmates in prison, who to approve for parole, determine how closely to supervise parolees, and predict whether certain individuals arrested for nonviolent, non felony crimes will re-offend.

New Challenges, New Concerns

Critics contend risk algorithms are:

  • too mysterious to be used in court
  • punish people for the crimes of others
  • can hide and aggravate old prejudices
  • hold demographics against defendants

Risk Assessments and Criminal Sentences

Risk scores, generated by algorithms, are an increasingly common factor in sentencing. Several states, notably Pennsylvania and Virginia, now use risk assessments results in determining the criminal sentences of nonviolent felony offenders, arguing that data about a person’s past help judges make better sentencing determinations.

According to state records, Virginia judges sent nearly half of defendants to alternatives to prison in 2014 using the predictive modeling tool. As a result, the growth rate in state prisons has slowed to 5 percent — from a rate of 31% during the previous decade.

How Risk Assessment Models Work

Effective risk assessment models require tens of thousands of profiles to be input into a computer. Profile includes data on those arrested: age when first arrested, the
neighborhood they’re from, how long the arrestee spent in jail, etc. The data also contain information about individuals that are rearrested. The computer identifies patterns. Those patterns are the basis for predictions about who is likely to reoffend.

Machine learning doesn’t concern itself with understanding what causes someone to be violent. Risk assessment tools don’t try to develop any philosophies on the origins of criminal proclivity. However this theory comes under scrutiny each time a supposedly neutral algorithm produces a non-neutral result.

Risk assessment tools can be as simple as questionnaires filled out by a jail staff member, probation officer or psychologist. Computerized risk assessment tools are all built in essentially the same way.

Hundreds of facts about former prisoners are examined. Their lives are then followed over several years to see who is involved with further criminal activity. Factors have been identified that appear to be linked to continued criminal activity, such as (but not limited to): sex, age, family background, and prior criminal history.

Not Everyone is Pleased

The risk assessment trend is controversial. For example, does using historical crime statistics to predict future crimes have the potential to equate past patterns of policing with the predisposition of people in certain groups to commit crimes?

Critics raise numerous questions, such as:

  • Is it fair to decide an offender’s fate based on the actions of similar offenders in the past?
  • How do you eliminate factors that might be associated with race?
  • Should characteristics that might be associated with an offender’s socioeconomic status be used?
  • What do you do about unreliable data?
  • Which of the many available tools — some of them licensed by for-profit companies — should policymakers choose?

Some argue risk assessment could become a real life version of the movie “Minority Report” imprisoning people for crimes they might commit in the future.

Supporters counter that judges and other decision makers are already making their own risk assessments. And people aren’t as good as computer generated statistics at predicting who is likely to commit crimes in the future.

Final Thoughts

There is little question that well-designed risk assessment tools are better at predicting behavior than unaided expert opinions. Dozens of scientific studies have been published comparing professional predictions of risk to predictions made by statistics.

The results are remarkably similar according to a study by psychologists at the University of Minnesota. Statistical actuarial tools are about 10% more accurate than experts assessing without the assistance of such a tools. Additionally, state released statistics show there are benefits to letting low-risk offenders avoid jail time alongside high-risk offenders.

But assessing the real world impact of risk assessment has proved difficult. States tend to release limited data that’s too new to draw conclusions about long-term effects. Additionally, risk assessment tools are often part of larger criminal justice reforms, making it difficult to identify isolate their effect.

The core questions around risk assessment tools don’t involve data. The discussion is more about about the goals of criminal justice reforms. Some see the primary goal as:

  • reducing incarceration
  • reducing recidivism
  • eliminating racial disparities

But there is no question that predictive modeling generated risk assessments have drawn widespread interest because in theory they can accomplish all three goals.