The Minnesota Sex Offender Screening Tool-Revised (MnSOST-R) was one of the earliest validated, actuarial risk assessment tool for sexual offenders. This article describes the development and validation of the MnSOST-R in the context of evolving risk assessment strategies and methods, which are essential for effective decisions in the criminal justice system.
Risk Assessment With Sexual Offenders
Determinations of risk occur across the criminal justice system. From decisions for setting bail, to sentencing, to establishing custody level, to determining conditions of release—implicit or explicit judgments about risk must be made. This has been true throughout modern correctional history.
Sexual offenders present even more challenges to the legal system because of additional risk assessment to inform additional high-stakes decisions unique to them. These include registration, community notification, and civil commitment.
In 1994, the U.S. Congress passed the Jacob Wetterling Crimes Against Children’s Act, which required states to set up sexual offender registries in order to better track sexual offenders. Megan’s Law, which took just 2 days to pass through the U.S. Congress in 1996 as an amendment to the Jacob Wetterling Act, authorized law enforcement agencies to notify the public about convicted sex offenders living, working, or visiting in their communities. It also required all states to develop a procedure for notifying the public when a convicted sex offender is released into their community.
To address these statutes, many states incorporated a three-tier system of risk (low, medium, or high) rather than identifying the specific types of offenders that are to be subject to notification. In Minnesota, for example, the Community Notification Act of 1996 called for a risk assessment scale to be implemented by January 1, 1997, and identified a series of factors to be used in assessing risk and assigning risk levels. As in many other states, those sex offenders determined to present the greatest risk received the highest level of community notification as well as supervision.
Also in the 1990s, states began to implement existing, or passed new, civil commitment statutes, often referred to as sexually violent predator laws. These statutes enabled states to commit the most dangerous sexual offenders to forensic hospitals rather than release them back into the community.
The policy justification for registration, community notification, and civil commitment statutes was protection of the public from the most dangerous sexual offenders. Clearly, then, the assessment and prediction of dangerousness is essential for the constitutional and effective implementation of these statutes.
While accurate assessment has always been important, the high-stakes decisions regarding sexual offenders helped focus attention on the objectivity, consistency, and accuracy of existing methods of risk assessment. The practice of sexual offender assessments in the 1980s and early 1990s was largely clinical and subjective in nature because of the standards of practice at that time and the absence of research on empirically validated risk indicators.
However, a compendium of studies, particularly in the 1980s, demonstrated that the clinical prediction of any type of longer term violence, including sexual violence, was essentially no better than chance. Consequently, reliance on clinical judgment alone resulted in numerous inaccurate predictions of sexual recidivism. On one hand, false-positive offenders were unnecessarily deprived of liberty, and the incorrect placement of offenders in prolonged and expensive treatment facilities put a great financial burden on the state. On the other hand, false-negative predictions risked placing dangerous sexual offenders in the community without appropriate supervision. There was a clear and compelling need for more accurate forms of risk assessment.
Since the 1950s, a growing body of evidence indicated that fully empirical approaches to diagnosis and prediction of future behavior outperformed clinical judgment. Such instruments included only variables that were linked to the target behavior by sound research, and each of these variables was assigned a relative weight empirically based on the strength of its relationship with the target behavior. This approach to assessment is now widely referred to as actuarial assessment.
By the early 1990s, several actuarial risk assessment tools had already established moderate predictive accuracy with general criminal recidivism (future commission of any criminal offense). However, those instruments were generally less successful at predicting violent recidivism, including sexually violent recidivism. Clearly, a different set of variables were needed for the actuarial assessment of risk of sexual recidivism.
Minnesota Sex Offender Screening Tool
Doug Epperson, Jim Kaul, Steve Huot, and Robin Goldman began development of MnSOST in response to a 1991 Minnesota Department of Corrections special report calling for a more formal and uniform process to identify predatory and violent sex offenders. Few prediction models existed at that time, and many of those were dependent on clinical and/or phallometric data. A major limitation of such models is the fact that many sex offenders are not offered or refuse treatment and/ or phallometric assessment, making clinical and phallometric data unavailable. From the outset, then, the goal was to create a reliable and valid predictive instrument that could be easily scored by correctional case managers using only information routinely available in correctional records.
Based on an extensive review of the relevant literature, an initial inventory of 14 items was developed. Although the items were empirically selected, the weights assigned to each item on this initial tool, the MnSOST, were based on clinical judgment. The initial MnSOST, thus, represented a structured clinical judgment approach to risk assessment in that it included only items with an empirical relationship to sexual recidivism but still involved a degree of clinical judgment. Results of two preliminary studies indicated that the MnSOST had acceptable reliability and modest predictive accuracy, demonstrating value to taking a more empirically structured approach to clinical judgment.
A major revision of the MnSOST was initiated in 1996. This was based on updated review of the literature to identify additional potential predictor variables and a commitment to making the revised tool a fully actuarial one in which both item selection and item scoring/weighting was empirically determined.
Minnesota Sex Offender Screening Tool–Revised
Development
Analyses to select and weight items for the MnSOST-R were performed on case files from 256 incarcerated sexual offenders in Minnesota from three cohorts: a random subsample of sex offenders released in 1988 (n = 107), a random subsample of sex offenders released in 1990 (n = 107), and a subsample of offenders readmitted to the Minnesota Department of Corrections for any reason during the time the sample was being put together, regardless of release year (n = 42). Most of the people in the last subsample were sexual recidivists (n = 32). This group was deliberately oversampled to provide more stability to any observed relationships between sexual recidivism and potential predictor variables.
The sample included sex offenders with a history of extrafamilial child molestation or rape of unrelated victims. For the purposes of this research, offenses involving vaginal or anal penetration of a victim aged 13 or younger were classified as rape. Forcible penetration was required for offenses against older victims to be classified as rape. Thus, the only group of sexual offenders excluded from the sample was those with an exclusive history of intrafamilial offenses that would not be classified as rape using the aforementioned criteria. This group was excluded because preliminary analyses indicated that this group was substantially different from the other groups on multiple dimensions.
Sexual offenders in the sample ranged in age from 17 to 70 years (mean = 32.42), and the ethnic mix of the sample was 66% European American, 23% African American, 5% Hispanic, 4% Native American, 2% other minority groups. On average, offenders in the sample had 0.82 prior sex convictions, with a range from 0 to 8.
To ensure that the tool could be used with a wide range of male sexual offenders, only variables that would be routinely documented in correctional case file were considered as potential items on the tool (predictors of sexual recidivism). The criterion variable, sexual recidivism, was defined as a formal charge for a new sexual offense within 6 years of release from prison.
Complex analyses identified 16 items that each made a unique contribution to the prediction of sexual recidivism (maximum identification of recidivists and nonrecidivists based on MnSOST-R total scores). With these 16 items in the predictive model, no other items significantly added to the prediction of sexual recidivism. Twelve of the items were historical/static in nature, meaning that a score could only increase with time and not decrease. For example, once a person had one conviction for a sexual offense, the person could never have fewer in the future. However, future convictions for sexual offenses could add to that score. The historical/static items were number of convictions for sexual offenses, length of sexual offending history, commission of a sexual offense while under court supervision, commission of a sexual offense in a public place, use or threat of force in any sexual offense, perpetration of multiple sex acts in a single event contact, offending against victims from multiple age groups, offending against a 13- to 15-year-old victim with more than a 5-year age difference between the offender and the victim, victimization of a stranger in a sexual offense, persistent pattern of adolescent antisocial behavior, recent pattern of substantial substance abuse, and recent employment history. The remaining four items were institutional/ dynamic items, meaning that they were reset to 0 at the beginning of any new incarceration for a sexual offense. These variables included discipline history, chemical dependency treatment recommendations and outcomes, sex offender treatment recommendations and outcomes, and age of the offender at the time of release from prison.
The MnSOST-R demonstrated good overall accuracy within the development sample. The most common statistic used to reflect overall accuracy of a risk assessment tool is the area under the receiver operating characteristics curve (AUC). An AUC value of .50 reflects chance levels of overall predictive accuracy, so a risk assessment tool must demonstrate an AUC value that is significantly greater than .50 using conventional statistical standards. In the development sample, the AUC was .77, significantly greater than chance. Furthermore, the tool performed equally well for various subgroups of offenders (rapists vs. child molesters and minorities vs. nonminorities).
Reliability
At least seven studies have assessed the reliability of MnSOST-R scores (agreement between raters on the scores of sexual offenders). Reliability coefficients can range from 0 (no reliability) to 1.00 (perfect reliability). The overall average reliability coefficient was .81. Five of the seven studies yielded reliability coefficients of .80 or higher, and the other studies produced coefficients of .68 and .76. There was minimal training of scorers in the latter two studies. Collectively, these studies demonstrated that, with modest training (a 1-day workshop), the MnSOST-R can be scored reliably.
Predictive Accuracy
Cross-validation studies are essential to demonstrate the predictive accuracy of the MnSOST-R with samples other than the development sample. In nine of 11 such studies, the predictive accuracy of the MnSOST-R was significantly greater than chance. The median AUC value across all 11 studies was .70, which reflects a moderate to high level of predictive accuracy.
The Current Context
The MnSOST-R always compared very favorably in reliability and predictive accuracy to other contemporary actuarial risk assessment tools for sexual offenders. In fact, the MnSOSTR and the Static-99 (and later the Static-99R), developed by Robert Karl Hanson and David Thornton, were the two most widely used tools of their type for about a decade in the United States. The latter tool was largely developed in Canada, which has a national tracking system for sexual offenders, making outcomes easier to determine. Because of this, the Static-99R established the most robust research base of all similar tools and is now the most widely used risk assessment tool for sexual offenders. Although the MnSOST-R was a pivotal tool in the transition to actuarial risk assessment, it is no longer supported by its authors. Anyone still using the MnSOST-R is encouraged to switch to the Static-99R.
References:
- Epperson, D. L., & Ralston, C. A. (2015). Conceptual model of risk versus threat and risk management versus risk reduction: Clarification of terms and implications for risk assessment. In A. Phenix & H. Hoberman (Eds.), Sexual offenders: Diagnosis, risk assessment, and management (pp. 397–415). New York, NY: Springer.
- Phenix, A., & Epperson, D. L. (2015). Overview of the development, reliability, validity, scoring, and uses of the Static-99, Static-99R, Static-2002, and Static2002R. In A. Phenix & H. Hoberman (Eds.), Sexual offenders: Diagnosis, risk assessment, and management (pp. 397–415). New York, NY: Springer.