In Connecticut, a formula helps make bail decisions

Print More

The state’s bail system aims to protect public safety and make sure people show up for court using the “least restrictive” manner possible, so understanding the risks posed by each defendant is a key part of that system. It turns out, there’s a formula to help with that.

Bail reform is again a legislative priority of the Malloy administration. Whether or not the governor’s proposals – or even more radical changes – are adopted, the need to quantify risk won’t go away.

When police arrest and detain someone who is unable to post the bond set by the police department, bail staff from the Judicial Branch Court Support Services Division review the police department’s decision and can override it. Bail staff base their decisions on a scorecard called a pretrial risk assessment, which determines their likelihood of skipping a court date or having to be rearrested.

While there is a formulaic approach to predicting risk, Gary Roberge, director of the CSSD’s Adult Probation and Bail Services, said the decision-making is never automatic. “That’s not how we do it. Each case is considered individually here in Connecticut.”

Here’s what that risk assessment looks like:

The scorecard assigns different point values to 14 factors, such as marital status, employment, and previous failures to appear in court — and the sum of those point assignments is the defendant’s risk score. Studies in 2003 and 2015 validated this as a method of predicting failures to appear in court and rearrest. When a risk score is zero or higher (the higher a score, the lower the risk), the recommendation is usually for non-financial release. When a score is below zero a financial bond might be recommended. However, in many cases, even a negative score will result in a non-financial release recommendation, according to a lengthy report to the governor on bail published by the Connecticut Sentencing Commission this month.

When financial bond is recommended, bail staff determine a dollar amount starting with a table of bond amounts, where each row is a category and class of offense, such as a class A misdemeanor or a class C felony, and each column represents a risk score from -6 or lower to +6 or higher. Unlike the risk score, which has been validated through study, the bond amounts in the table haven’t been shown to have any relationship to court appearance or rearrest rates, according to the Sentencing Commission report.

Roberge pointed out that the table of bail amounts differs from a more rigid “bail schedule” that some other states have. He said the table is a “starting point” that in large part is meant to make more uniform decisions.

When courts are not operating, on nights and weekends, the bail staff makes the decision. When courts are operating, or if a defendant is held unable to post the bail staff’s bond, a judge receives the bail staff’s recommendation and considers it along with other factors — namely arguments by the prosecution and defense.

How is the risk assessment score ‘validated?’

The pretrial risk assessment is referred to as an “empirically validated” tool. What that means is, researchers have shown there is a statistically significant link between a defendant’s risk score and his or her likelihood of skipping a court appearance.

In 2003 and 2015, Central Connecticut State University researchers contracted by the state Judicial Branch showed that the score generated by the risk assessment correlates to failure-to-appear rates. Researchers did this by looking at data on defendants from past years, grouping them by risk score, and seeing how many in each group failed to appear in any of the first five court appearances, when most failures to appear occur.

They found that the lower the risk score – indicating higher risk – the larger the percentage of defendants who failed to appear. In both validation studies, some modifications were recommended to the risk assessment to strengthen that correlation.

A work in progress

Just because the assessment has been validated doesn’t mean it’s perfect, or that officials are finished altering it. The system was analyzed by researchers in 2003 and then again with more data in 2015, and some changes have been recommended each time. The 2015 recommendations are expected to be rolled out in April or May, Roberge said.

The 2003 CCSU report found that race and gender — called “extralegal” variables — didn’t have a “significant impact on bail decisions.” In 2015, researchers found that the portion of defendants whose risk scores were above or below zero differed by up to 5.8 percent among black, Hispanic and white defendants. Blacks had the highest portion of below-zero scores, followed by whites and then Hispanics.

Researchers said that their suggested changes would reduce those differences, while also improving the risk score’s predictive power.

“One of the goals of developing and implementing standardized tools is to reduce the potential for extra-legal factors to influence bail decisions,” the report states. “More specifically, we sought to avoid contributing to the potential disparate treatment of clients based on their race/ethnicity. […] With the current points, some group differences were as much as 5.8% but with the [recommended] points, the largest group difference was 1.9%. These findings suggest that the new points have the potential to further reduce disparities in bail decisions.”

The following chart shows first how the risk scores varied among racial groups based on the scoring criteria in 2015, and the subsequent chart shows how those differences would be reduced by the CCSU researchers’ recommended changes to the point system.

Recommended changes in the 2015 report were fairly subtle — among them, altering some of the point values associated with each factor and allowing three rather than two possible levels of education by separating the current “high school or less” response into “less than high school” and “high school.”

The validation study also found that while most factors did correlate with failure to appear and rearrest likelihood, severity of charge didn’t.

“It may be worth exploring in the future whether Charge Severity is a suitable criterion to include in the point scale since it failed to exhibit a consistent relationship with either FTA or NA. In fact, those with felony charges are less likely to fail to appear or acquire a new arrest which seems counterintuitive. At this time, no changes are recommended to these factors because state statute requires that they be included as part of the weighted release criteria.”

What do you think?