Risk assessment tools are now a standard part of how criminal justice professionals make decisions, occurring in both the adult and juvenile system. Currently, several federal criminal justice bills propose to use risk assessment, including the FIRST STEP Act, which passed the House by an overwhelming bipartisan vote in May and is due to be taken up for consideration in the Senate. The FIRST STEP Act proposes to use risk assessment in the federal prison system to assign programming and to determine who may be able to safely finish out their sentence in home confinement or a halfway house. Prison Fellowship’s Vice President of Government Affairs Heather Rice-Minus recently asked Zachary Hamilton, Ph.D., Associate Professor of the Department of Criminal Justice & Criminology at Washington State University, to discuss how risk assessment works, the accuracy of such tools, and the potential benefits and drawbacks as it relates to racial bias. Dr. Hamilton has published over two dozen peer-reviewed articles on a variety of criminal justice issues and served as the principal investigator for the Washington State Offender Risk Assessment Project to develop a consolidated system of prediction instruments using both dynamic and static factors.
Prison Fellowship: Why and when were risk assessment tools developed? How has their use expanded over time?
Dr. Hamilton: Risk assessments have mostly been used to determine supervision levels in the community (that is, how often a person is to have contact with their probation or parole officer). Since their creation in the early 1990's, there has been substantial advancement to the point where it would be difficult for a person involved with the criminal justice system to avoid assessment. While they are now used at pretrial, classification, and release decisions, the original intent was not to use risk assessment tools for sentencing.
What is a needs assessment?
A needs assessment is a collection of dynamic or changeable items, (e.g. the offender can improve or get worse overtime). Typically needs assessments are broken into several 'domains' (i.e. substance use, employment, family, attitudes/behaviors, aggression) that contain a subset of related items. Each set of items compiles a domain score, which is divided into low, moderate and high. Typically moderate and high need individuals are identified for programming. However, a tool's ability to do that is based on the quality and number of items in each subset. Tools that have few or low quality needs items tend to have greater problems identifying people that need programming or the best type of programming for each person.
Some states using risk assessment during incarceration have incorporated needs assessments into a comprehensive system to serve multiple purposes, including enhancing effective programming assignments and predicting recidivism in order to inform release decisions.
What is the difference between a static and dynamic risk factor?
Static factors are items that only change in one direction or stay the same. Commonly, these include gender, age, and criminal history. Dynamic items are those that can move up and down a scale, such as level of conflict with one's family, number of months in a stable residence, using substances with the last six months, or currently participating in educational programming.
How accurate are risk assessment tools?
It depends on the tool and what population it is applied to. Risk assessments predictive accuracy is measured by the Area Under the Curve (AUC) statistic. It ranges from 0.5-1.0 and values can be viewed as a percentage. Most risk assessment tools, when developed for a specific population, provide moderate predictive accuracy (i.e. 64%-70%). When tools developed for one population are then applied to a new population, accuracy often shrinks (i.e. 56%-63%). However, when a researcher can collect data for a specific population and then select and weigh items, a tool is optimized for that jurisdiction and accuracy often improves to strong levels (71%+).
When using a risk assessment tool for release decisions, overrides (allowing a human actor such as a warden to overrule the risk assessment prediction) must be restricted. There are several studies that suggest that overrides should be kept to 10 percent or less to ensure accuracy.
What factors in risk assessment are arguably linked to race, and how can we control for further unwarranted racial disparities when using risk assessment tools?
Due to what are viewed as unintended biases of law enforcement practices, static criminal history measures often possess the most racial biases. Unfortunately, past behavior is the best predictor of future behavior and thus, criminal history measures are typically the most predictive. These are often the measures that judges and other correctional officials are most familiar with and comfortable using. This creates a bit of a paradox. To date there have been few solutions for reducing bias. While it is unlikely that one could create an accurate risk assessment tool that does not include prior criminal history, recent findings have indicated that the greater use of dynamic items increases tool accuracy while reducing the impact of racial bias within a tool. Although not devoid of bias, dynamic items, such as substance abuse, employment, and aggression, have less bias. Their inclusion reduces bias within a tool and gives credit to prisoners that complete programming and effectively demonstrate behavior change at the time of reassessment. Therefore, tools with a greater number of dynamic items are likely to have less bias and improve case management and program prioritization.
Would a traditional parole system be more likely to avoid racial disparity than using risk assessment for release decisions?
No. Risk assessments are designed to remove idiosyncrasies of decision making and to make decisions more consistently, whereas parole board decisions often get entangled with inadvertent racial biases of individual decision makers. There are decades of studies that bear out these findings. Traditional systems that are built to not have decision makers, such as giving good time based on sentence and infraction behavior, will still suffer from racial biases as these models perpetuate the use of a static-only, criminal history metric of assessing risk. A comprehensive risk and needs assessment instrument helps to improve prediction over and beyond these two methods by incorporating more race-neutral factors into release decision consideration. When an assessment tool is not used, the alternative is to depend on human actors dispersed across the nation to make consistent and unbiased release decisions; the totality of evidence indicates that humans decision makers create greater racial bias.
For more information about the application of risk assessment in the FIRST STEP Act, please visit Prison Fellowship’s legislative primer page here. Click here learn more about the risk assessment tools currently being used in various jurisdictions.