When a judge decides whether a defendant accused of a crime should receive bail, and what the amount of bail should be, they will evaluate the risk of the defendant failing to appear in court. They also will consider the risk of the defendant continuing their criminal activity or committing an additional crime before the trial date. In making this decision, the judge often will go beyond reviewing the charges and criminal record of the defendant. They can review reports by bail investigators, which will describe the defendant’s reputation. They also may hear the opinions of people who know the defendant. Ultimately, however, the decision remains at the discretion of the judge. A bail algorithm is a way to simplify the process and make it more transparent.
Bail algorithms are meant to guide judges by providing them with a statistical analysis based on several factors. These programs conduct an objective assessment of the defendant’s flight risk. A bail algorithm often will result in a specific score, but sometimes it will present only a recommendation for whether the defendant should be released. In some cases, an algorithm will evaluate the risk of the defendant breaking the law and the risk of the defendant failing to appear in court separately. It might evaluate the risk of perpetrating a violent crime separately from the risk of perpetrating a non-violent crime. Factors that algorithms typically consider include the charges against the defendant, their criminal record, their age, and any history of previously failing to appear at a court proceeding.
Many observers have welcomed the use of bail algorithms as a way to eliminate bias and improve consistency. Some research suggests that people who live in poverty and people who belong to minority groups receive less favorable treatment from judges during the bail process than people from privileged backgrounds. In addition to these unfair results, a high rate of denying bail can lead to unnecessary overcrowding in jails.
However, bail algorithms are not a perfect solution. The factors that they consider may be more limited than those that a judge would consider. For example, an algorithm may not integrate information about substance abuse in a defendant’s past, or it might not account for their employment status. (Whether these factors correlate strongly to flight risk remains disputed.)
Sometimes an algorithm will place too much weight on the formal charge, failing to account for the distinctive facts of the situation. Not every person who is charged with a certain crime poses the same level of risk. For example, a shoplifter who panicked and punched a security guard when they were caught may be facing a robbery charge, and so may a group of gang members who held bank tellers at gunpoint. Most people would realize that these two situations are different, even though the label of the charge may be the same.
While bail algorithms are meant to defeat racial prejudice and other biases, some people have questioned whether they achieve this goal. People in minority groups are disproportionately likely to be arrested and charged. Since bail algorithms take arrest records and criminal history into account, this could mean that they inherently discriminate against minorities.
