The issue

In the United States, approximately 1.6 million individuals are being held in federal and state prisons.

In 2015, of the more than 700,000 individuals detained in local jails, 63% were unconvicted and awaiting trial – also known as pre-trial detention. [1]

The cost of detaining so many individuals is over $80 billion – just to operate the jails and prisons. [2]

Moreover, persons of color make up nearly 70% of individuals in pre-trial detention, despite representing only a 25% of the United States population.

Next

Pre-trial risk assessment

There are various systemic factors contributing to this issue. Here we will be exploring one in particular – the use of algorithms and data via pre-trial risk assessment instruments, or PRAIs.

PRAIs are intended to standardize pre-trial decision-making by providing information about or “predicting” a defendant’s likelihood of re-offending or failing to appear in court.

PRAIs can be qualitative, quantitative, or a combination of the two. Jurisdictions using qualitative or combination instruments experience more jail overcrowding than those that do not, as pre-trial detention is disproportionately recommended. [3]

Next

The problem

PRAIs are not only used to predict a defendant's likelihood of re-offending. These algorithms are increasingly being used to set the terms of the defendant's detention, such as setting the amount of their bond.

However, there is a concern that the data that these algorithms are trained on may be biased, due to the association between many risk factors and systemic social and economic inequities.

But how can we know, when we are unable to inspect the algorithms themselves?

Next

The goal

Our goal with this webpage is to study the impacts of pre-trial risk assessment algorithms indirectly, by investigating the racial disparities in the pre-trial inmate population.

The purpose of this webpage is educational: We aim to raise general awareness of these issues.

We do not claim that the results of our statistical analysis are conclusive. But they should make us ask more questions, and demand more answers about the use of pre-trial risk assessment algorithms throughout the country.

Next

The case study

The state of Connecticut has been using a pre-trial risk assessment algorithms since 2003. The algorithm was revalidated in 2015.

The state of Connecticut has also, since July 2016, made available data on all its pre-trial inmates – all individuals being held overnight in its correctional facilities while awaiting trial.

We do not know everything about these inmates that their risk assessment algorithm did. But we do know the amount of their bond, the numbers of days they were detained awaiting trial, and the offense for which they were arrested.

We also know their race.

Next

Overall racial demographics

Since 2016, Black people have always made up the largest number and proportion of pre-trial inmates In Connecticut state correctional facilities. In 2016, White people made up the next largest number, but by 2020 White and Latinx people were incarcerated in roughly equal numbers.

Next

But is this any different from Connecticut’s racial demographics?

Yes. C’mon. This is Connecticut. The overwhelming majority of the state's population is White. Yet the overwhelming majority of its pre-trial inmate population are people of colour.

Race Connecticut population Pre-trial inmate population
White 71.2 % 31.6 %
Black 10.1 % 40.9 %
Latinx 6.4 % 26.6 %

*State population numbers come from the 2010 Census.

Next

Racial disparities in bond amounts

It’s not only that inmates of colour are disproportionately represented in the pre-trial inmate population. Inmates of colour also tend to be assigned higher bond amounts than White inmates.

*The above chart shows the median bond amount for White inmates and inmates of colour. The vertical bars indicate the 95% confidence interval of our estimate of the median. In other words, we can be 95% confident that the true median lies somewhere in the range indicated by this bar.

Next

But is this because inmates of colour tend to be arrested for different, more serious offenses?

No. Even if we keep the specific offense fixed, the median bond amount for inmates of colour is often significantly higher than for White inmates.

(This chart highlights the offenses that discriminate the most severely overall. This pattern is not, however, limited to these offenses. Of the 40 most frequently occurring offenses, the median bond amount was higher for inmates of colour in 31 of them, or 78%.)

Next

But is this because inmates of colour tend to be repeat offenders?

No. The relative proportions of first-time and repeat offenders are nearly identical between White inmates and inmates of colour, and the difference in their median bond amounts has nothing to do with whether they have been arrested for a repeat offense.

As this plot illustrates, while White inmates tend to be assigned higher bond amounts for a repeat offense, inmates of colour tend to be assigned the same, much higher bond amount, regardless if it was a first-time or a repeat offense.

Next

Is gender an issue?

Yes and no. For male inmates of colour the situation looks the same: Their median bond amounts are higher in 60% of offense types, and their maximum bond amounts are higher in 56%. Here are some of the offenses for which male inmates of colour are discriminated against:

The situation for women of colour looks a little better. The minimum bond amounts are usually lower for female inmates of colour, and the median and maximum bond amounts are usually comparable to what White female inmates are assigned. Still, there are some offenses where female inmates of colour are discriminated against:

Next

What might explain these differences?

It's hard to say, but one notable difference between the male and female inmate populations is that, while in the male population Black inmates make up the majority, in the female population it is White inmates that make up the majority.

Next

So what does Connecticut know that we don’t?

What explains these clear racial disparities? Connecticut’s PRAI explicitly ignores the defendant’s race when making its risk assessments. It does, however, consider other data about them.

According to the state’s latest report, its PRAI makes its assessments based on the defendant’s current charge, criminal history, marital status, employment, education, substance abuse, mental health, and prior failure to appear in court. [4]

The question is: Should the terms of a defendant’s pre-trial detention be determined by factors such as their employment, education, health, and living situation – given the association between these factors and systemic social and economic inequities?

Is this justice? Or is it perpetuating society’s existing inequities?

Next

The national landscape

Connecticut is not unique in its use of PRAIs. According to research done by Mapping Pretrial Injustice, PRAIs "are in use all over the country, in at least one county in almost every state."

(Visualization courtesy of Mapping Pretrial Injustice)

Next

Is there anything that can be done?

This all may seem alarming. But successful adoption and ethical use of PRAIs is possible.

The District of Columbia’s Pre-trial Services Agency (PSA) has used PRAIs since it began operating in 1967. The PSA not only administers the risk assessments but provides the supervision of pre-trial defendants. PSA demonstrates a commitment to the use of well designed, tested, and validated risk assessment tools to reduce bias related to race and socio-economic status – improving outcomes overall for persons involved in the justice system.

In particular, PSA has found that the proper use and vetting of PRAIs, combined with pro-social interventions (recommendations for mental health and substance abuse screening appropriately) and minimally restrictive non-financial release options (conditional release with low, moderate, or high level supervision) fosters positive results for defendants.

As a result of this approach, an average of 88% of pre-trial defendants are released. Nearly 90% of those individuals are not re-arrested and appear in court. Only 12% of those detained in DC jails are pre-trial defendants, preventing overcrowding, and the cost to taxpayers is reduced, as the cost of supervision of a released defendant is less than that of detention.

Next

Interested to do something about these issues? Get involved!

Pre-trial risk and the resulting recommendations can be handled differently at the state and local levels. To find out if PRAIs are being used in your local or state judicial systems, visit Mapping Pretrial Injustice.

And to explore other issues related to AI and racial disparities, check out the Algorithmic Justice League and Algorithmic Justice League and Data 4 Black Lives.

Next