Ethics and limitations of risk assessment instruments in sentencing

By: Marija Buzanin

Introduction

In 2022 and 2023, around 40 in 100,000 people in Canada were incarcerated in the federal system. In the provincial and territorial systems, that number jumps to 71.59. In the Canadian provincial and territorial carceral systems, 30% of a total of 163,387 custodial admissions were identified as Indigenous. A 2021 census found that Indigenous people make up roughly 5% of the Canadian population (Statistics Canada, 2024). 46.6% of the total prison population in Canada (35,485) are pre-trial detainees—all while the occupancy level of the prison system sits at about 102.2% (World Prison Brief, 2021). In 2014, the Canadian Civil Liberties Association reported that 54.5% of all people in Canada’s provincial and territorial jails were not there to serve a sentence after a conviction—they were awaiting bail, the determination of bail, or the resolution of their charges—legally innocent. By 2022, that proportion was 70.5% (Canadian Civil Liberties Association, 2024).   
These statistics hint at a grim reality: There are too many people in prison. Thus, the problem becomes tailoring sentences more precisely to each offender, partly according to the level of risk they present to reoffend, especially violently. This is why some jurisdictions have implemented the use of risk assessment tools in sentencing, aiming to determine the likelihood of recidivism, with the goal of reserving custodial sentences for high-risk offenders, and alternative sentences for others (Mugford et al., 2017). 

While the aim of risk assessment in sentencing is a noble one, the idea of algorithmically predicting future violence to modify jail sentences raises ethical concerns and practical implementation problems. This paper will examine some of these tools, examining their roots in psychiatric knowledge and mental health to determine their limitations and what, if any, future they may have for decision-makers in law. ​​

​

Section 1: Risk Assessment Tools; Origins and Development

​​

Offender risk assessment tools are used to assist in sentencing, as well as aid correctional officers in deciding the type and intensity of correctional interventions. They are used to assess offenders on items relating to the risk of reoffending, providing a measure of that individual’s “risk level” and informing decisions on sentencing, treatment during incarceration, and even release or parole. The goal, considering that carceral punishment is not always necessary or desired when aiming to reduce the risk of reoffence, is to match the supervision strategies to the level of risk of each offender (Mugford et al., 2017). 

The type of risk assessment tool used varies on a case-by-case basis, and is usually specific to the type of crime committed by an offender. For example, the Violence Risk Scale (VRS) is used to assess and target the risks of violence (Wong and Gordon, 2019) while the Spousal Assault Risk Assessment (SARA) is administered by external evaluators on males to specifically assess the risk of spousal violence perpetration (Kropp et al., 2000). This section will examine the history of risk assessment tools and their foundations in the law, evaluating their roots to understand their implementation in the sentencing stage of the legal process.

​

1.1 Early Models: Professional Discretion

The history of risk assessment tools starts in the mid-20th century, in which these first generation tools were facilitated solely by correctional staff and clinical professionals (Bonta & Andrews, 2007). Probation officers, prison staff, psychologists, psychiatrists, and more were tasked with making professional judgments regarding offenders’ specific risk level or likelihood to reoffend. This is what the Department of Justice Canada (DOJ), in collaboration with the British Columbia Institute Against Family Violence (BCIFV), has termed “unstructured clinical decision-making,” citing it as “probably still the most widely used approach to spousal violence risk assessment” (Department of Justice Canada, 2022). With no constraints or guidelines for the evaluators, decisions about risk level are often based entirely on professional discretion. This approach has long been criticized for concerns regarding reliability, validity, and accountability due to its informal and unstructured nature. 

​

1.2 Early Actuarial Tools

​

In the mid-1970s, it became apparent to mental health professionals and correctional staff alike that assessment of risk needed to be based on actuarial evidence and science; it needed to be standardized (Bonta & Andrews, 2007). This marks what researchers Bonta and Andrews (2007) term the second generation of risk assessment tools, in which individual items demonstrated to increase the risk of reoffending, such as criminal history, are measured and quantified. Eventually, actuarial risk assessment tools, such as the Salient Factor Score in the United States and the Statistical Information on Recidivism scale in Canada, were developed, assigning quantitative scores according to the presence of risk factors. These went on to extend to “diverse” offender groups, such as mentally disordered or sex offenders. Though these risk assessment tools were found to reliably differentiate lower from higher risk offenders with greater accuracy than professional judgments had, they were criticized as being atheoretical and relying on static, immutable factors. Offenders could not expect to see a decrease in their risk level as the categories did not allow it: even if they had in fact been rehabilitated. This led to third generation risk assessment tools: evidence-based and dynamic ones. 

1.3 Dynamic and Needs-Based Models

​

Research developing in the late 1970s and 1980s began to include dynamic risk factors, such as situational or circumstantial ones (Bonta & Andrews, 2007). While criminal history remained important in an assessment of an offender’s likelihood to reoffend, they based additional risk-need instruments on things like employment status, criminal friendships, and supportive family relationships. Bonta and Andrews (2007) developed a Level of Service Inventory-Revised tool, utilizing these dynamic factors, sensitive to changes in an offender’s circumstances. Additionally, the Brief Spousal Assault Form for the Evaluation of Risk (B-SAFER) worksheet, created by the DOJ and the BCIFV, would fall under this category. Aiming to prevent violence by systematically identifying risk factors, it asks users to assign each risk factor a number: 0 for “No, Omit”; 1 for “Possible”; and 2 for “Yes” (Department of Justice Canada, 2022). Higher scores were shown to equate with higher risk. These factors fit into one of two categories, Spousal Assault and Psychosocial Adjustment, and include threats of or acts of serious physical or sexual violence, negative attitudes about spousal assault, substance abuse, and mental disorder. Not only did the development of these kinds of tools allow for a more nuanced rather than two-dimensional or static assessment of risk, but it also informed correctional staff as to what needs should be targeted in interventions (Bonta & Andrews, 2007). These changes prioritized rehabilitation and decreasing recidivism, even allowing staff to monitor the effectiveness of the supervision strategies being employed. 

1.4 Current Generation: RNR and Comprehensive Integration 

​

Finally, the fourth and current generation risk assessment tools are systematic and comprehensive. They integrate systematic intervention with a broader range of offender risk factors, not just criminal history, as well as other, more personal factors that may be important to treatment. This includes the Risk-Need-Responsitivity (RNR) model, first formalized in 1990 and undergoing recontextualization and improvement even today (Bonta & Andrews, 2007). The following section will deal in detail with this model and others like it to gain an understanding of their implementation in Canadian correctional jurisdictions and their use in sentencing.

​

Section 2: Risk Tools in Sentencing; Practice and Purpose

​

During a sentencing process, after the defendant has been found guilty of the crime in question, judges are tasked with determining their punishment. They weigh mitigating and aggravating factors, according to submissions by both crown and defence counsels, to determine a just punishment according to several specific guidelines. In Canada, judges are bound by the Criminal Code and the Charter when sentencing offenders. Section 718 specifically lays out the purpose of sentencing “to contribute, along with crime prevention initiatives, to respect for the law and the maintenance of a just, peaceful and safe society by imposing just sanctions,” placing the focus squarely on the foundational principles of justice and proportionality (Criminal Code, 1985). In determining what this means for each individual, judges may use risk assessment tools to determine an individual’s likelihood of reoffending, and the danger they may pose to the community. 

Risk assessment tools evaluate risk factors, or items, to predict outcomes. These outcomes can range from technical violations of probation or parole, to violent re-offense. In the case of predicting the likelihood of recidivism, risk assessment instruments use items that statistically correlate with the outcome (such as criminal history or mental health issues), which occur beforehand. These items don’t need to cause specific outcomes, rather, they are determined by various psychological tools and statistical analysis to be correlated to the predicted outcome (Garrett and Monahan, 2019). This section will examine the various uses of risk assessment tools, such as in presentence reports or during sentencing, in custodial and non-custodial decisions, and how sentence lengths can be adjusted based on projected risk to determine potential equitability issues and concerns. ​​

​

2.1 Types of Tools Used in Sentencing Contexts 

The most commonly used risk assessment tools in sentencing vary depending on the type of offence committed and the jurisdiction. The Level of Service/Case Management Inventory (LS/CMI) is a general risk-need tool designed to assess both dynamic and static factors across a wide range of possible offences. Most commonly seen in presentence contexts, it evaluates items such as criminal history, level of education, employment status, family or marital situation, and substance abuse or mental health issues, in order to assess risk (Bonta & Andrews, 2007).

In domestic violence contexts, the Ontario Domestic Assault Risk Assessment (ODARA) or the SARA are often used. The former is a static, actuarial tool, predicting the likelihood of future intimate partner violence by scoring offenders on 13 empirically validated items, such as prior domestic incidents, confinement, threats, and the victim’s concern level (Ontario Domestic Assault Risk Assessment, 2019). The latter combines both static and dynamic elements, and is administered by trained evaluators who consider past assault behaviour along with current psychosocial factors, such as substance abuse or mental disorder (Department of Justice Canada, 2021). Both tools assess the likelihood of intimate partner violence recidivism.  

For sexual offences, the Static-99R is the most widely used tool across the globe. Based on entirely static factors such as age, gender of offender and victim, prior offenses, and victim relationship, it produces a numerical score corresponding to categories of recidivism risk (Fazel et al., 2022). As it is based on static factors, individual offenders’ risk levels cannot change over time, meaning this tool does not consider potential rehabilitation or escalating risk level. However, it remains a significant aid in sentencing where the protection of the public is at the forefront. 
 

Section 3: Limitations of Risk-Based Sentencing

3.1 Lack of Standardization

The purported goal of risk assessment tools across the board is to align supervision intensity or sentencing type and length to the level of risk to reoffend. However, the existence of multiple risk tools across Canada raises concerns about the equivalence of the risk classifications each tool produces. Different tools may define the same individual to be either low or high risk, some tools have different standards altogether; varying from “very low” to “medium” and more (Mugford et al., 2017). This lack of standardization across the board leads to a large number of hypotheticals, each positing different levels of concern: what if two individuals in similar circumstances, having committed similar crimes, are classified in two widely different risk categories, and treated substantially differently? What if two individuals pose vastly differing levels of risk, but are classified in the same manner by two different tools? What are the implications for offender rehabilitation? How do institutions administer correctional responses when faced with such scenarios? How do judges account for this equitability issue in sentencing? Should they? 

A study by Mugford et al. (2017) asked 20 professionals from correctional agencies across Canada about the risk assessment policies in their jurisdictions, including which instruments they used, how they understood the respective risk categories provided, how these risk categories were used to inform decisions about release and supervision, and more. The findings not only included that risk assessment has become commonplace in the Canadian correctional systems, but that at least 20 different risk tools are in use. They differ in everything from the number of categories used to the meaning of each category, leading Mugford et al. (2017) to conclude that, even if risk categories were labelled similarly, they may have differed in how they communicated risk level. 

3.2 Ethical Concerns

According to Mugford et al. (2017), the lack of uniform procedures for assigning risk levels can result in scenarios where two offenders with similar profiles may receive entirely different risk level scores, and therefore, sentencing outcomes, depending solely on the tool used. Furthermore, the myriad tools used mean that the labels of “high” or “low” risk may not carry the same meaning in different jurisdictions. These labels vary in number (e.g. including moderate and medium risk on some tools, but not all) or definition, (e.g. the threshold for “high” risk) two individuals classified as “high risk” might have completely different profiles, risks of reoffending, and treatment needs. Such inconsistencies raise fairness concerns as well as ideals of equitability and justice, challenging the legitimacy of an inconsistent algorithm-based sentencing process. 

3.3 Transparency and Due Process

Risk assessment tools are proprietary, in some cases. Companies can manufacture algorithms to be sold to correctional institutions and used alongside psychiatric assessments and in sentencing. This opens the door to the issue of transparency, a recent focal point of judicial criticism. In State v. Loomis (2016), the Supreme Court of Wisconsin was tasked with deciding a due process challenge to the use of the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS), a commercial risk assessment tool developed by a company called Equivant (Garrett & Monahan, 2019). The defendant argued that because he could not examine the basis for the risk score due to the algorithm's proprietary nature, he could not determine whether factors such as race or gender—impermissible in consideration of sentencing—had been relied upon. Thus, he argued that his right to due process, his right to an individualized sentence, and his right to be sentenced on accurate information, had all been violated (Harvard Law Review, 2017). Though the court rejected the claim, the issue of lack of transparency raised by Loomis highlights the danger of commercialized risk assessment in sentencing. 

Without full disclosure of the algorithms, psychiatric tools, weightings, and validation studies, parties cannot accurately determine whether assessments provided to them are free from bias. Courts and the broader justice system are at risk of making sentencing decisions informed by hidden variables that may disproportionately disadvantage certain groups, contravening both the right to fair and open trials, as well as the right to freedom from discrimination. 

3.4 Built-in Bias
​​

3.4(1) Racial Bias

​

Studies have demonstrated that racialized offenders can often score higher on certain risk assessment tools, labelling them at a higher risk of reoffending—even when these tools control for actual criminal behaviour (Sreenivasan et al., 2022). This may be because certain items measured, such as prior convictions, mental disorders, substance abuse, or unemployment, are disproportionately influenced by the systemic discrimination, socioeconomic inequality, and over-policing plaguing these communities. When these tools are used without any adjustment for structural factors, they risk perpetuating the cycle of overrepresentation in prisons. 

​

3.4(2) Socioeconomic Bias

​

Research has also shown that, where risk assessment tools measure items like unstable housing, unemployment, or education level, individuals from lower socioeconomic backgrounds are more likely to receive a higher risk classification (Sreenivasan et al., 2022). These traits reflect broader structural disadvantages rather than direct criminality, highlighting yet another potential issue of basing risk level classification on items correlated with rather than causally related to criminal behaviour. Relying on these items in risk scoring can, as previously mentioned, further compound the marginalization of disadvantaged populations rather than addressing the root causes of both inequality and criminal behaviour. 

​Section 4: Recommendations for Ethical Risk-Based Sentencing Reform
 

4.1 Standardization 

Standardizing the meaning of risk categories across the country could help mitigate disparities found when using myriad risk assessment tools, or tools with differing categorizations of certain items. A “common language of risk” could increase consistency in how risk is both assessed and communicated. This could help remind decision-makers that “we are talking about the same people, regardless of the risk scale that is being used or the jurisdiction where offenders are serving their sentence” (Mugford et al., 2017). If fewer tools or more consistent ones are used, both evaluators and offenders will be better able to understand the scope of these tools, allowing for more fairness and equality in the sentencing process. Without such reforms, the potential exists for scenarios in which offenders are punished according to geography or institutional preference, rather than on the actual threat posed. 

4.2 Culturally Specific Alternatives

In one study conducted by Garrett and Monahan (2019), judges in Virginia were surveyed about their thoughts on risk assessment methods in sentencing. In general, they believed that the risk assessment was “useful,” but cited a “lack of useful alternatives.” The conclusion was that a lack of local treatment programs as an alternative to carceral punishment was “a major factor constraining alternative sentencing for low-risk and property offenders”. Thus, the study presented a “treatment resource hypothesis:” the probability of a judge assigning an alternative sentence to a low-risk offender increases with the availability of community programs. Indeed, this positive relationship was found. Judges’ likelihood of imposing a sentence consisting of alternative or nonjail sentences increased “from 44 percent in the most resource-poor jurisdictions to 71 percent in the most resource-rich jurisdictions.” This signalled to the researchers that the wide variation in sentences for drug and property offenders assessed as low risk could be due to the distribution of alternative resources among various jurisdictions. Thus, where ethical concerns arise as to the proportionality of lengthy prison sentences for minor or non-violent crimes, alternative or community programs for offenders present as a potential solution. The socioeconomic problem presented by this research signals another potential recommendation: a “justice reinvestment model of corrections”. 

This model sees states enacting statutes with two particular branches of reform: reducing mass incarceration and allocating the resulting fiscal benefits to “fund expansion of community alternatives to incarceration”. According to judges and researchers, non-jail alternatives such as mental health programs, substance abuse clinics, and other rehabilitative efforts for nonviolent offenders, must be accessible to judges in sentencing (Garrett & Monahan, 2019). 
 

Conclusion

Psychiatric risk assessment tools present both an opportunity and a challenge to the criminal justice system and sentencing law. Their integration into judicial decision-making is a technical advancement that may promise future consistency, but that may also risk undermining transparency, due process, and equitability if not accompanied by safeguards and standardization. As courts seek to increasingly rely on such tools in an attempt to protect potential victims and survivors, the law must ensure that predictive psychiatry serves justice to the offenders as well, to avoid undermining the very rights it attempts to protect. ​

​

​

​

References

​

1. Berger, L., Myers N.M, and Deshman, A. (February 2024). Still failing: The deepening crisis of bail and pre-trial detention in Canada [final report]. Canadian Civil Liberties Association and Education Trust. https://ccla.org/wp-content/uploads/2024/02/CCLA_Bail-Reform-Report-2024.pdf 

2. Bonta, J., & Andrews, D. A. (2007). Risk-need-responsivity model for offender assessment and rehabilitation 2007-06. Rehabilitation, 56–86. https://doi.org/10.4324/9780203962176-9

3. Fazel, S., Burghart, M., Fanshawe, T., Gil, S. D., Monahan, J., & Yu, R. (2022). The predictive performance of criminal risk assessment tools used at sentencing: Systematic review of validation studies. Journal of Criminal Justice, 81, 101902. https://doi.org/10.1016/j.jcrimjus.2022.101902 

4. Garrett, B., & Monahan, J. (2019). Assessing risk: The use of risk assessment in sentencing. Judicature. https://judicature.duke.edu/articles/assessing-risk-the-use-of-risk-assessment-in-sentencing/ 

5. Government of Canada, Statistics Canada. (2025, March 19). Average counts of adults in provincial and territorial correctional programs. https://www150.statcan.gc.ca/t1/tbl1/en/tv.action?pid=3510015401  

6. Government of Canada. (2021, December 8). Intimate partner violence risk assessment tools: A review. Department of Justice Canada. https://www.justice.gc.ca/eng/rp-pr/cj-jp/fv-vf/rr12_8/a.html  

7. Government of Canada. (2022, December 28). The development of the brief spousal assault form for the evaluation of risk (B-SAFER): A tool for Criminal Justice Professionals. Department of Justice Canada. https://www.justice.gc.ca/eng/rp-pr/fl-lf/famil/rr05_fv1-rr05_vf1/p4.html  

8. Government of Canada. (2024, June 13). Indigenous peoples and communities. Crown-Indigenous Relations and Northern Affairs Canada. https://www.rcaanc-cirnac.gc.ca/eng/1100100013785/1529102490303  

9. Institute for Crime & Justice Policy Research . (2021, January 1). Canada. World Prison Brief. https://www.prisonstudies.org/country/canada 

10. Kropp, P. R., & Hart, S. D. (2000). The Spousal Assault Risk Assessment (SARA) guide: Reliability and validity in adult male offenders. Law and human behavior, 24(1), 101-118. https://doi.org/10.1023/A:1005430904495

11. Lee, S. C., Hanson, R. K., & Blais, J. (2023, February 24). Evaluation of risk assessment tools for indigenous individuals with a history of sexual crimes: A response to Ewert Case (2018). Public Safety Canada. https://www.publicsafety.gc.ca/cnt/rsrcs/pblctns/2023-r002/index-en.aspx 

12. Mugford, R., Bourgon, G., Hanson, R. K., & Coligado, M. (2017). Offender risk assessment practices vary across Canada (Research Report 2017-R015). Public Safety Canada: Ottawa, ON.

13. Ontario Domestic Assault Risk Assessment; Maine guidelines and general scoring criteria. (2019). https://dirigosafety.com/wp-content/uploads/2020/09/ODARA-guidelines-booklet-FINAL-x3.pdf  

    Sreenivasan, S., DiCiro, M., Rokop, J., & Weinberger, L. E. (2022). Addressing systemic bias in violence risk assessment. PubMed, 50(4), 626–635. https://doi.org/10.29158/jaapl.220031-2

14. State v. Loomis. Harvard Law Review. (2023, March 24). https://harvardlawreview.org/print/vol-130/state-v-loomis/#footnote-ref-2  

15. Wong, S., & Gordon, A. (2019). Violence risk scale. The SAGE encyclopedia of criminal psychology, 4, 1641-1644. https://doi.org/10.4135/9781483392240.n532