Risk assessments under fire, but critics don't differentiate uses

Grits has been a supporter of risk assessments as part of pretrial detention determinations in lieu of bail, which available evidence suggests leads to far better outcomes. But lately there has been much weeping and gnashing of teeth among liberals over risk-assessment instruments, charging that they result in racial bias depending on which variables are used.

See a critique of risk assessments from ProPublica, a defense of risk assessments from the Marshall Project, and related critiques.

In all these debates, Grits has seldom if ever seen commentators focus on a key distinction between how risk assessments are used at different stages in the process.

Grits is not a fan of their use in any but an advisory capacity for predictive policing or in sentencing. In policing, probable cause must always be individualized to a specific person. And punishments should be decided by judges and juries, not algorithms (although if we're honest, that process is really controlled by prosecutors and certainly isn't immune from racial bias).

But using risk assessments to decide pretrial detention is a horse of a different color, because the alternative is that money bail is required of all. That means people who can't afford bail must rot in jail until their case is resolved, putting immense pressure on even-innocent people to plea. In that circumstance, the benefits to liberty from applying a risk-assessment tool that lets the majority of poor people out pretrial outweighs any detriment from bias in the instrument.

Risk assessments also have well-recognized uses for probation systems which are key to reducing recidivism and encouraging probationers' success. Low-risk probationers subjected to intensive programming (or for that matter, who are revoked to prison) tend to escalate in criminality in ways which are counterproductive, a growing body of research indicates. So assessing risk/needs is key to deciding what programming should be assigned.

In those circumstances, Grits remains unconvinced that bias from these instruments is greater or more harmful than the bias being exhibited by judges. For example, the videos of bail hearings before magistrate judges which arose out of the Harris County bail litigation demonstrated near-universal disrespect for misdemeanor defendants' pretrial liberty interests, whereas the risk assessment would have freed most of them. So which is more biased? And which bias is more concerning?

Predicting "future dangerousness" (as the courts have dubbed risk assessment by juries in the sentencing phase of capital cases) is always a dubious prospect, but also a fundamental part of what the justice system does. In most phases of the process, those decisions should probably be left to the decision-makers traditionally in charge of them.

In the case of pretrial detention, though, where the defendants are still presumed innocent by the system, IMO opposing the use of risk-assessment tools amounts to allowing the perfect to become the enemy of the good. Ditto for probation where there's a pragmatic need to assess probationers' risk in order to make decisions about programming, and risk levels can change significantly over time.

The desire to purge the system of racial bias is understandable, but please let's not throw the baby out with the bath water.