Predicting the risk of being placed in solitary confinement

Predicting the risk of being placed in solitary confinement

A robust, easy-to-use, simple scale can be developed to assess the risk for being placed in administrative segregation among inmates. This is the bottom line of a recently published article in Psychology, Public Policy, and Law. Below is a summary of the research and findings as well as a translation of this research into practice.


Featured Article | Psychology, Public Policy, and Law | 2019, Vol. 25, No. 4, 284–302


Developing and validating a tool to predict placements in administrative segregation: Predictive accuracy with inmates, including indigenous and female inmates



Maaike Helmus, Simon Fraser University
Sara Johnson, Correctional Service of Canada, Ottawa, Ontario, Canada
Andrew J. R. Harris, Forensic Assessment Group, Ottawa, Ontario, Canada


The last 15 years have witnessed considerable concern regarding the use of segregation. Attempts to reduce segregation would benefit from being able to identify which inmates are at greatest risk for being placed in segregation. The goal of the current research project was to develop an actuarial scale to assess the risk of being placed in administrative segregation for inmates in federal prison in Canada. A total sample of inmates (N = 16,701) was randomly divided into a development sample (N = 11,110) and a validation sample (N = 5,591). Of the 413 potential predictor variables examined, 86% significantly predicted segregation placements. The item pool was reduced using both conceptual and empirical methods. Although several scales were developed and tested, the most efficient option was a simple scale with six static items. Predictive accuracy of this scale was high in the validation sample, as well as for subgroups (e.g., Indigenous and female inmates), and it also significantly outperformed other assessments already used by the prison service. We found that it is possible to develop a simple and easy-to-use scale that would be effective in identifying inmates at risk for placements to administrative segregation, which would be an important first step in efforts to intervene to reduce risk of segregation placement. Implications for developing risk assessment tools and applying to subgroups are discussed.


administrative segregation, female inmates, Indigenous inmates, prediction, risk scales

Summary of the Research

“Segregation (also referred to as solitary confinement or restrictive housing) typically refers to any condition where an inmate is confined to a cell for at least 22 hr a day with limited contact with others. Segregation is often classified as administrative (i.e., intended as a proactive measure to ensure the safety and security of the institution and its inmates) or disciplinary (a punitive response to a specific incident). This paper focuses solely on administrative segregation.” (p. 284)

“Generally, placement in segregation is to be avoided or used as a last resort as it is presumed to have negative effects on the inmates, although two recent systematic reviews have challenged the presumption that segregation produces severe adverse consequences for the individual’s mental and emotional health, at least any more so than the effects of routine incarceration. […] Regardless of whether segregation is harmful for inmates, however, there are additional reasons to reduce its use. There are a limited number of segregation cells; consequently, they should be reserved for those who need them the most. Additionally, maintaining an inmate in a segregation placement can cost two to three times more than traditional incarceration.” (p. 285)

“Efforts to reduce segregation can rely on two routes: changes in policy/legislation, or interventions targeted at individuals to prevent the behaviors that tend to lead to segregation. Some combination of both routes is ideal, but they are likely to occur independently and the authority for each rests at different levels of prison management. […] Both routes are worthwhile, but this study is designed to inform efforts at the individual level (although this could involve prison policy changes as well).” (p. 285)

“Although there is little research on segregation specifically, there is a wide body of research to broadly inform effectiveness of correctional interventions. Specifically, the Risk/Need/Responsivity model of effective correctional practice provides numerous principles to guide rehabilitative and case management efforts, with the ultimate goal of reducing reoffending. […] Consequently, efforts to reduce the use of segregation should start by identifying and targeting those individuals most likely to end up there (i.e., the risk principle) and should target behaviors that are related to segregation (i.e., the need principle). This study was undertaken to develop an actuarial tool to flag, at intake, federal inmates in Canada who may be of elevated risk for a segregation placement.” (p. 285)

“Though small, the existing research suggests that there may be enough predictors to develop a risk scale for predicting segregation placements. What is unknown, however, is whether the strongest predictors of segregation would overlap with general risk factors or whether unique risk factors would emerge. Additionally, it is not known whether a scale predicting placements in administrative segregation would provide value-added over traditional risk scales (i.e., developed to predict reoffending).” (p. 286)

“The goal of this research project was to determine the feasibility of creating a practical and accurate tool gauging the risk an individual inmate faces of being placed in administrative segregation. This would also involve identifying primary risk factors for segregation placements. A secondary goal was also to demonstrate one example of how to develop an empirical-actuarial risk scale, including producing normative data. […] We additionally sought to develop a scale that was sensitive to expense and workforce demands in correctional facilities. […] In addition, to ensure the applicability of the tool in a Canadian federal corrections context, it was also important to demonstrate that the proposed risk assessment protocol would be valid for male, female, Indigenous, and non-Indigenous inmates. […] Lastly, this study sought to examine whether a risk scale developed specifically for segregation would outperform assessment scales already used by CSC for other purposes (e.g., to assess risk for recidivism).” (p. 286)

“The total sample size of men and women was 16,701. From the overall sample, two thirds of the cases were randomly selected to form the development sample (N = 11,110). The remaining cases (N = 5,591) were set aside as a validation sample. In the overall sample, 16% of inmates were female (n = 2,694) and 20% self-reported being of Indigenous ancestry (n = 3,385; information unavailable for 135 cases). For the overall sample as well as the development and validation samples, the sample consisted of 68% non-Indigenous men, 16% Indigenous men, 12% non-Indigenous women, and 4% Indigenous women.” (p. 287)

“The outcome examined was whether the inmate was placed into administrative segregation for at least six consecutive days for reasons of “inmate’s own safety” or “jeopardizing security of the institution” within two years of admission. Separate outcomes were also created for the two different reasons for segregation. Other reasons for administrative segregation (e.g., interfering with an investigation) were excluded; notably, they represented a trivial proportion of segregation incidents. Disciplinary segregation was also excluded.” (p. 288)

“The initial goal was to determine whether it was feasible to develop a scale to identify an inmate’s risk of being placed in administrative segregation. The large and diverse number of variables that significantly predicted this outcome (>350) demonstrated that this goal was feasible; moreover, it was practical (i.e., high predictive accuracy can be achieved using information that is already available for each inmate, which is sensitive to expense and workforce demands). […] Potential scale items were refined by considering overall predictive accuracy, stability of predictive accuracy among subgroups, face validity, practical considerations (i.e., availability of information and ease of scoring), and incremental accuracy (i.e., unique contribution over other predictors).” (p. 296)

“A simple six-item static risk scale (called the RAST [Risk of Administrative Segregation Tool]) was able to predict admission to administrative segregation for at least six consecutive days for reasons of jeopardizing security or the inmate’s own safety, within two years of admission, and evidenced a very large effect size. This straightforward scale showed similarly large predictive accuracy for non-Indigenous men, Indigenous men, non-Indigenous women, and Indigenous women. Furthermore, each item was also a significant predictor for each subgroup. More comprehensive assessment scales were tested but did not provide meaningfully better accuracy than the RAST.” (p. 296)

“Additionally, the scale also showed large effect sizes in predicting both subtypes of segregation (jeopardize security or inmate in danger) and in predicting segregation placements within one and two years of admission. There was no evidence of statistical shrinkage in a large validation sample. Together, these findings suggest this static risk scale is robust across different groups of inmates, segregation types, and observation periods.”( p. 296)

“It is possible to achieve similar or slightly (though not significantly) better predictive accuracy by adding an item to the static risk scale that considers the inmate’s overall level of dynamic need (this can be done using measures of dynamic need that are specific to each subgroup). Incorporating an assessment of dynamic risk factors allows for identification of potential treatment targets that could improve institutional adjustment, which would be consistent with the Need principle of effective correctional practices.” (pp. 296–297)

“Attempts to develop risk scales specifically for subgroups (i.e., Indigenous men, non-Indigenous women, and Indigenous women) were not particularly successful. In other words, specialized scales did not have consistent or meaningful advantages (in terms of predictive accuracy) beyond the overall scales that were developed. There are likely two reasons for this. The first reason is that, compared with how other risk assessment scales have been developed, the overall scales developed to predict segregation were more responsive to differences across gender and ethnicity. […] Another reason for the lack of success of the specialized scales pertains to overfitting. […] Future research on unique risk scales should consider not just whether it is possible to create scales for a specific subgroup, but whether these scales show better performance than standard practice in new validation samples.” (p. 297)

“The results indicated that the largest and most robust relationships with segregation tended to fall under the general domains of the Central Eight risk factors (specifically, criminal history, procriminal personality pattern, procriminal attitudes, procriminal associates, family/marital difficulties, school/work difficulties, leisure/recreation problems, and substance abuse.” (p. 297)

“Although this scale overlapped with the broader risk domains examined in other risk scales, analyses found the RAST significantly outperformed the assessment scales currently being used by CSC [Correctional Service of Canada], suggesting that there is value in the creation of a scale specifically designed to predict segregation placements. This likely demonstrates both the universality and specificity of the risk factors. Although they tap into similar domains, they are chosen to maximize prediction of segregation.” (p. 298)

“A straightforward six-item static risk scale (called the RAST) that can be scored immediately upon admission can predict administrative segregation placements in CSC within two years of admission with large effect sizes. Upon validation, the RAST was found to be robust across different subgroups, types of segregation, and length of observation periods. This scale could be scored upon admission to an institution with little time investment from correctional staff to identify a small group of inmates who have a high probability of being placed in segregation.” (p. 299)

“Using a risk scale for early identification of inmates who are at risk of being placed in segregation creates important opportunities to provide these individuals with additional support and intervention in their adjustment to incarceration, reducing segregation placements (although it also provides opportunities for this information to be misused in ways that could hamper adjustment and rehabilitation). Identifying and diverting inmates from administrative segregation has benefits not just for the inmates (e.g., avoiding negative effects of segregation), but also for the institution (e.g., fewer setbacks to the correctional plan, cost savings), and for public safety (e.g., better adjustment to incarceration can help inmates focus on their correctional plan, potentially leading to quicker releases and safer transitions to the community).” (p. 299)


Translating Research into Practice

“In addition to the caveats above about research and modifications needed before applying this scale to new settings, some additional comments about potential uses and misuses of the scale are warranted. First, there is little point in assessing risk for segregation unless you plan to do something to mitigate that risk. What should be done, however, is a complicated question that this research study cannot address.” (p. 298)

“Equally important in considering how this risk information can be used, however, is to consider how it should not be used, as good intentions can inadvertently lead to practices that are ultimately harmful or unethical. The scale was designed to identify inmates who could perhaps benefit most from efforts to improve correctional adjustment and reduce likelihood of being placed in segregation. The scale should not be used to preemptively segregate inmates, and efforts should be to avoid labeling effects or practices, which can be more harmful than helpful.” (p. 298)

“Even phrasing in policy can mitigate (though likely not eliminate) some of these potential labeling effects. For example, a high RAST score should not be interpreted as “this person will end up in segregation”; rather, it should signify that this person is an ideal candidate to benefit from additional resources in acclimating to the prison environment and in managing potential behavioral issues.” (p. 298)

“The hope is that thoughtful and ethical implementation of risk management strategies will result in improved institutional adjustment and reduced segregation placements, which can free up resources and opportunities to better focus on rehabilitation, benefitting the inmates (during and postincarceration), as well as institutional and public safety, in addition to cost-savings. Achieving these potential benefits, however, requires further empirical work and evidence-based interventions.” (pp. 298–299)

“A tool such as the RAST could be used to inform alternate housing placements for high risk inmates, although such a practice might become an analogue for segregation. Any such placements should be short-term placements with focus on evidence-based coping and adjustment training, rather than prolonged security/supervision or isolation. Alternately, it is possible that similar goals could be accomplished with short, motivational-based interventions delivered at intake (ideally before security classification is finalized), such as skills-based prison orientation workshops.” (p. 299)

Other Interesting Tidbits for Researchers and Clinicians

“There were also a few limitations to the development of the scale. To oversample women, data were obtained dating back to 1999. This means the sample of women is not as recent as the sample of men. Additionally, the scope of this scale development was restricted in several ways. First, only administrative segregation incidents for reasons of jeopardizing security or inmate in danger were examined. The outcome was restricted to placements in segregation of at least six days duration. It is unknown whether the scale would be applicable to other types of segregation or to shorter stays in segregation.” (p. 298)

“This scale was also designed specifically to be used at intake. As such, no attempts were made to explore changes in risk for segregation, or to develop a scale that would reflect such changes. With future research, it may be possible to create revisions to this scale that incorporate dynamic risk factors and would be capable of assessing changes in risk.” (p. 298)

“Although the scale had large effect sizes in the current validation sample, it is not clear how well the scale would generalize to applied settings or to correctional settings outside of CSC. Applications to other jurisdictions would likely require additional research to make adjustments to at least two items. First, the sentence length item is specific to CSC in that all inmates had been sentenced to at least two years of custody. Given jurisdictional differences in sentencing policies (including jurisdictions that include inmates with less than 2-year custodial sentences), it would likely be necessary to recalibrate this item. Additionally, the prior segregation placements item was specific to previous federal sentences. Other jurisdictions would have to explore how to define and measure previous segregation placements most optimal for their jurisdiction.” (p. 298)

“There has been one validation of the RAST in a forensic psychiatric sample. Although some items showed promise (more after some modification), the study largely demonstrated that many of the items were not applicable to a hospital setting, particularly as many patients were either not criminally responsible or unfit to stand trial, so all the items requiring convictions demonstrated little endorsement and low variability. Applying the scale in a hospital setting would benefit from some of the modifications recommended by Hilton and colleagues (2019). It is likely that the current version of the RAST would be more applicable to convicted inmates in other correctional settings, but some modification might still be required (e.g., sentence length), and replication studies in new settings are recommended.” (p. 298)

Join the Discussion

As always, please join the discussion below if you have thoughts or comments to add!

Authored by Kseniya Katsman

Kseniya Katsman is a Master’s student in Forensic Psychology program at John Jay College of Criminal Justice. Her interests include forensic application of dialectical behavior therapy, cultural competence in forensic assessment, and risk assessment, specifically suicide risk. She plans to continue her education and pursue a doctoral degree in clinical psychology.

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.