Unfortunately, assessments that consist of only, or mostly, static items have been shown to increase racial/ethnic and gender bias (Butler et al., 2022). Research has shown that an over-reliance on criminal history items inflate risk scores for racial/ethnic minorities, due to disproportionate minority contact reflected through local policing practices and prosecutorial decisions (Eckhouse et al., 2019). Similarly, females offend at lower rates than men, and tools that use predominantly criminal history indicators inflate the risk posed by women, especially for violent crimes (Salisbury et al., 2009). Subsequently, contemporary assessments tend to mistakenly ‘overclassify’ risk of future offending, assigning a defendant a higher than appropriate risk rating (Hamilton et al., 2023). Further, current PRAs often fail to gather information on defendants’ pretrial circumstances (e.g., residential stability, substance use, mental health needs, & employment), which are often used by judges when considering release and/or applying conditions that allow for successful pretrial supervision.

Balancing utility at the expense of efficiency is a formidable challenge faced by assessment developers and the agencies their tools service. Fortunately, the development of a new PRA, the Personal Recognizance Interview and Needs Screen (PRINS), provides a different perspective, using improved development methods aimed at resolving these limitations.

Overcoming Limitations

PRAs applied off-the-shelf are adopted ‘as is’ and many agencies may feel like a ‘square peg in a round hole’, needing to gather potentially irrelevant information from pretrial defendants to score a tool. Tools that include items that the court system does not routinely collect can create complications for the agency and extend the workloads of staff. Further, items that are consequential to judicial decisions, representing the defendant’s release circumstances, are commonly ignored (DeMichele et al., 2018; Scott-Hayward et al., 2022). Conversely, the PRINS was developed via a localized approach. Localization means that a tool is created with the context of a local jurisdiction in mind. Localized tools overcome limitations of off-the-shelf assessments by using jurisdiction-specific data to inform their development and are crafted by carefully considering the agency’s needs and the population they serve (Hamilton et al., 2022). For example, short assessments are sometimes viewed as not especially useful because they are more likely to leave out information that judges consider important when making pretrial release decisions (DeMichele et al., 2018). When judges and the courtroom work group are skeptical of what the tool captures, this often results in the assessment recommendations being overridden, which reduces accuracy (Lowder et al., 2023). Conversely, the creators of the PRINS collaborated with local stakeholders to ensure that all the important information is commonly gathered by the court system, such as details about defendants’ involvement in substance abuse treatment, employment, and living situation. This approach also helped to tailor the outcomes the PRA predicts. In particular, the King County stakeholders sought more specific outcomes than the three commonly predicted by PRAs (failure to appear [FTA], new criminal arrest [NCA], and new violent criminal arrest [NVCA]), resulting in the offense types being further broken down into ‘any’, felony, violent, property, drug, and domestic violence sub-categories. Overall, the efficiency and utility of PRAs can be greatly increased via localization.

Tailoring assessment items to an agency’s policies, jurisdiction, resources, and population can also increase accuracy. For one, tailoring involves strategically selecting the most useful items for that agency (Hamilton et al., 2016). As a result, buy-in for judges and stakeholders is improved, reducing the likelihood of overrides (Hamilton et al., 2016). Further, by using local data, the strongest predictors for that population can be given more weight (i.e., larger scores), and thus, more importance (Stevenson, 2018).

Similar methods can be used to reduce bias in the assessment outcomes (Duwe, 2014; Hamilton et al., 2022). To reduce female overclassification, the PRINS was developed using gender-responsive modeling. This method recognizes that there are different pathways to crime for males and females, and as a result, not all assessment items predict FTA and arrest outcomes equally across genders (Van Voorhis et al., 2010). For example, among individuals arrested for drug-related crimes, significantly more females attribute the onset of their substance abuse disorder to symptoms of post-traumatic stress and/or prior victimization (Reichert & Bostwick, 2010). PRINS developers created male and female-specific samples, then selected and weighted items that were the most important for the respective gender-specific samples. Gender-responsive models have proven to be capable of reducing gender bias (Duwe & Rocque, 2019; Hamilton et al., 2022). This is due, in part, to their ability to identify and incorporate more dynamic itemsthat are important to that agency’s population (Andrews & Bonta, 2010; Hamilton et al., 2022). Unlike static items, dynamic items are changeable measures of a defendant’s personal circumstances and can be used to establish pretrial supervision services that reduce their risk of FTA or arrest (Scott-Hayward et al., 2022). Further, including dynamic items reduces the need for criminal history indicators, which can exacerbate racial/ethnicity bias (Butler et al., 2022). Specifically, bias can be reduced by removing predictors that are correlated with being non-white and supplementing their removal by including a greater number of dynamic-need items (Butler et al., 2022).

Creating the PRINS

In 2017, King County, Washington stakeholders sought to adopt a PRA as a response to growing concerns of bias and overuse of cash bail in pretrial decision-making. When doing so, they wanted a tool that could demonstrate reductions in bias, while maintaining efficiency and accuracy. Subsequently, researchers from the Nebraska Center for Justice Research (NCJR) were asked to develop a PRA that was specific to the King County population. Creating a local tool offset concerns of predictive shrinkage and allowed for more sophisticated development methods, increased efficiency, and reduced bias.