Specialized Mental Health Supervision: Revocations and Risk Composition

Abstract

Although many studies have investigated the disproportionate representation and negative experiences of justice-involved persons with mental illness (MI), we know less about probation/parole revocations among this population. Using statewide data and propensity score matching, we compare rates of rearrests and revocations between individuals with and without MI and assess the effectiveness of Specialized Mental Health Supervision (SMHS) in reducing the likelihood of revocation. In addition, we examine whether the risk score composition differs among participants of the SMHS program from clients with MI not placed on SMHS. Findings reveal that persons with MI are more likely to have a revocation, specifically for technical violations. In addition, SMHS does not significantly lower revocations after controlling for other caseload characteristics. Finally, those with higher risk scores were significantly less likely to be placed on SMHS caseloads. Considerations for SMHS implementation are discussed.

Keywords

community supervision mental illness probation parole revocations risk assessment

There is a significant overrepresentation of people with mental illnesses in the U.S. legal system. Compared with the general population, rates of mental illness are at least twice as high among justice-involved persons (Al-Rousan et al., 2017; Prins, 2014). More specifically, between 700,000 and 1.2 million people under community supervision (i.e., probation or parole) have mental health concerns (Crilly et al., 2009; Ditton, 1999; Van Deinse et al., 2018). Facing complex challenges (e.g., homelessness, addiction, and victimization), individuals with mental illnesses experience increased odds of supervision failure, such as arrests and technical violations (Cloyes et al., 2010; Ostermann & Matejkowski, 2014). Specialized Mental Health Supervision (SMHS) is a strategy aimed at augmenting Community Supervision Officers’ (CSOs) capacity to assist persons with mental health needs, enhance mental health treatment engagement among supervisees, and improve outcomes for justice-involved people with mental illnesses (Council of State Governments, 2002; Skeem et al., 2006).

Although mental health supervision shows promise, there are mixed findings about the impact of SMHS programs on supervisees’ outcomes. Overall, SMHS models show improvements in mental health–related outcomes (e.g., treatment engagement); however, there has been inconsistent impact of mental health supervision on criminal justice outcomes, such as probation violations and rearrests (Manchak et al., 2014; Skeem et al., 2017; Van Deinse et al., 2021; Wolff et al., 2014). There are a few factors that may contribute to SMHS’s inconsistent impact on criminal justice outcomes, including differences in implementation of the model, different degrees of officer discretion in determining violations and sanctions, questions about whether treatment engagement as a core feature of the model can reduce recidivism, and the risk profiles of those assigned to mental health caseloads. Although exploring all of these potential factors is essential, this study focuses on the risk profiles of those on mental health supervision.

There has been extensive empirical research on risk factors associated with criminal behavior, called criminogenic risk factors (Andrews & Bonta, 2010; Andrews et al., 2006; Bonta et al., 1998). There are eight risk factors (referred to as the “central eight”) for criminal recidivism, including (a) a history of antisocial behaviors, (b) antisocial cognitions, (c) antisocial personality patterns, (d) antisocial peers and associates, (e) family or marital strain, (f) poor performance or satisfaction in school or work, (g) low levels of leisure or recreational involvement, and (h) substance use/misuse (Andrews & Bonta, 2010; Andrews et al., 2006). Each of the eight risk factors is either dynamic or static (i.e., factors that change and those that are immutable), such that the dynamic risk factors (e.g., antisocial associates, substance misuse) are targets for interventions to reduce criminal recidivism. Notably, empirical evidence suggests that the eight central risk factors predicting criminal recidivism are the same for people with mental illnesses as for the general population (Bonta et al., 2014; Bonta et al., 1998). That is, mental illness itself is not a predictor of criminal recidivism. Rather, the risk factors are the same regardless of mental health status; however, those with mental illnesses may experience these risk factors with higher frequency or intensity (e.g., higher rates of substance use/misuse among people with serious mental illnesses; Morgan et al., 2010; Wilson et al., 2014; Wolff et al., 2011).

For several reasons, administrators of SMHS programs must understand the roles that criminogenic risk and need play in SMHS implementation and outcomes. First, from a program implementation standpoint, given that mental illness is not a predictor of recidivism, mental illness alone should not be the sole determinant in SMHS caseload assignment (Eno Louden et al., 2020). That is, someone can not only have a mental illness but also have a low risk of reoffending, and therefore, the impact on criminal justice outcomes may be marginal for those with a mental illness and low scores on risk assessments. Second, the risk-needs-responsivity framework instructs agencies to match risk level to supervision terms and programming, prioritize criminogenic needs that drive recidivism (e.g., substance use/misuse), and address the individual characteristics and needs of supervisees that inhibit engagement in and adherence to supervision (Andrews & Bonta, 2010; Andrews et al., 2006; Bonta & Andrews, 2017; Taxman, 2014). This latter principle, which refers to responsivity, is a core feature of the SMHS model in its focus on connecting supervisees to the resources and treatment they need to promote stability. Third, given that the central eight risk factors are the same for people with mental illnesses as they are for the general population, it follows that focusing only on treatment engagement will not reduce criminal recidivism. Instead, the dynamic risk factors (i.e., criminogenic needs) should be viable targets for SMHS intervention (e.g., addressing substance use/misuse among supervisees, promoting prosocial relationships with peers and family members, engaging supervisees in seeking employment and vocational activities).

Despite the importance of understanding risk within the context of mental health supervision, the focus on risk in SMHS research has been insufficient. Our study aims to fill this gap by analyzing statewide administrative data to examine criminogenic risk, rearrests, and revocations of those on SMHS caseloads, as well as a matched sample of people with mental illnesses who were on standard caseloads. Specifically, we employ a quasi-experimental design using propensity score matching and logistic regression to examine (a) whether persons with mental illness are more likely to be rearrested or have their probation/parole term revoked than supervisees without mental illness, (b) whether placement on an SMHS caseload reduces the likelihood of rearrest or revocation among persons with mental illnesses, and (c) whether people placed on SMHS caseloads have higher criminal risk scores according to a validated risk instrument than individuals with mental illnesses not placed on SMHS caseloads.

Method

Data and Sample

Our initial cohort drew from administrative data spanning nearly 3 years of supervision comprising 54 million records, which included 27,396 unique individuals placed on probation or parole from July 1, 2017, to February 28, 2018. Individuals were followed for 2 years (up to February 28, 2020) to allow reasonable judicial processing time for supervision violation hearings because the average length of stay on supervision before an arrest event is 9 months, and the average length of stay is 14 months before a revocation event. In addition, this study chose dates to avoid the start of the COVID-19 pandemic and its impact on supervision practices. A subset of 7,947 individuals were identified based on whether a mental health need existed, whether a mental health assessment from prison existed, or whether the individual participated in the SMHS program. Of these, 1,383 were under the supervision of the SMHS program, and the remaining 6,564 formed the basis for a comparison cohort.

Eligibility for the SMHS Program

There are four eligibility criteria for placement on the specialized mental health caseload at the Georgia Department of Community Supervision (DCS): (a) individuals assessed in state prison by the Georgia Department of Corrections (GDC) as having a moderate or severe mental illness requiring placement in supportive housing or the crisis stabilization units; (b) individuals discharging from a residential substance use treatment facility for co-occurring disorders; (c) individuals who have been diagnosed with schizophrenia, bipolar disorder, major depressive disorder, severe post-traumatic stress disorder (PTSD), or “any diagnosis accompanied by serious functional impairment”; or (d) individuals currently receiving mental health services who are unstable or show signs of decompensation.

Mental Health Screening

The DCS utilizes an 11-question mental health screening instrument to assess new placements for mental health needs (see the appendix for screening instrument). Not all the individuals with a mental health indicator on record would have had a severe and persistent mental illness that qualified them for supervision under the specialized program. However, we hypothesized that some individuals sentenced directly to probation might have comparable mental illnesses without having a more formal diagnostic assessment. There was insufficient diagnostic information in the administrative data to determine the nature or severity of mental illness; only the presence of a mental health need, as identified in the screening instrument (see Supplemental Materials, available in the online version of this article for details), or having a mental health profile, which was a mental health level identified during an incarceration event with the GDC, was available to flag individuals with a history of mental illness.

Risk Assessment

DCS utilizes the Unified Risk Assessment, a validated risk instrument custom-built by a third-party research partner, to calculate a risk score. This score represents a person’s likelihood of rearrest and is used to determine the intensity of services they receive. Comprising a 1 to 10 scale, those who score between eight and 10 receive the most involved supervision services. Integrated into the DCS case management system, the tool uses logistic regression to assess the odds of reoffending based on an array of static and dynamic variables, such as age, employment status, and criminal history. The complete list of variables used in the algorithm is available upon request.

Treatment and Control Groups

For our quasi-experimental design, the SMHS program is the intervention or treatment condition, whereas nonparticipation is the control condition. More specifically, the treatment group was based on identifying individuals supervised on an SMHS caseload at any point in their current supervision term. The control group comprised individuals identified as having either a mental health need or a mental health profile but not placed on SMHS. In addition, individuals were only included in the control group if they had a history of high-risk scores and did not have a history of sex offenses, as individuals with sex offenses are generally not eligible for SMHS. These refinements ensured that the subsequent propensity score matching would be more effective. The final sample sizes for the regression analysis consisted of 959 individuals in the treatment group and 4,212 in the control group.

Measures

Specialized Mental Health Supervision

The primary independent variable in our study is placement in the SMHS program. The variable is dichotomous, where 1 represents placement on SMHS at any point in their current supervision term, and 0 represents people not placed on SMHS.

Supervision Success and Failure

Supervision success was our primary dependent variable. Events were categorized as “supervision failures” when individuals experienced either (a) a rearrest or (b) had their supervision sentence revoked to a probation detention center, county jail, or state prison. Conversely, “supervision success” refers to the absence of these specific events. An individual can be sent to a probation detention center for up to 6 months to address noncompliance issues. County jails can be used for sanctions and incarcerating someone for a misdemeanor conviction while on felony supervision. While a rearrest or incarceration while under supervision does not necessarily require a formal revocation, incarceration is disruptive to whatever treatment or other progress occurs in the community—this broad definition of failure measures further criminal justice involvement for individuals with mental illness.

Covariates

Our variables included race, sex, age, highest education level achieved, type of supervision location (rural or urban), probation status (either direct probation or post-incarceration), history of substance use, presence of a violent offense in their criminal background, the number of previous felonies, and the number of prior arrests. Race was dummy-coded to represent individuals who reported being White, Black, or from another racial group, with White being the reference group. Age was treated as a continuous variable, representing the age of the individuals in years. Education was segmented into two categories: less than a high school diploma and a high school diploma or higher. Supervision location was coded as rural or urban, depending on the client’s residence. History of substance use was coded dichotomously, with 1 indicating the presence of substance use history and 0 indicating no such history. The presence of a violent offense in the participant’s criminal history was also dichotomously coded, with 1 indicating the presence of a violent crime and 0 otherwise. The number of previous felonies and prior arrests were treated as continuous variables, noting the respective counts for each client.

Procedure

The project received IRB approval through the University of Southern Maine (protocol 20-10-1577, approved December 2, 2020). This section details our approach, starting with an initial analysis where baseline recidivism rates and critical differences between our groups were studied. Then, we employed our quasi-experimental design, utilizing propensity score matching and logistic regression to estimate the impact of the SMHS program on supervision success.

Initial Analysis

Using Stata 16, we calculated baseline recidivism rates with univariate descriptive statistics for the entire population and among the mental illness treatment and control cohorts. In addition, we assessed differences between treated and control groups through several bivariate analyses, including chi-square tests of correlation and t tests. The analysis focused on the supervision failure rate for people with mental illness participating in the SMHS program compared with people with mental illness supervised on other caseload types.

Quasi-Experimental Design

After determining initial failure rates for individuals with mental illness for the SMHS and non-SMHS populations, we conducted a more robust, quasi-experimental analysis featuring both propensity score matching and logistic regression using the R statistical language. The purpose of the propensity score match was to balance the observed covariates between the treatment and control groups, thereby reducing selection bias and allowing for a more reliable estimation of the program’s effect on supervision success (Harris & Horst, 2019). Propensity scores were estimated by a logistic regression model with placement on SMHS as the dichotomous variable and our observed covariates as predictors.

Before matching, we observed an overlap of propensity scores for the treatment and control groups, signifying an overlapping range of values for the measured covariates between these two groups, as seen in Figure 1. This overlap suggests that each group has enough similar individuals to make meaningful comparisons (Murnane & Willett, 2011). This condition of common support not only enhances the study’s validity but also improves the efficacy of the propensity score matching process (Murnane & Willett, 2011).

Figure 1:

Overlap of Case Characteristics in Treatment and Control Groups

As shown in Table 2, we experimented with various propensity score matching techniques available in the R package MatchIt to identify the most balanced approach across the covariates, which included nearest neighbor without replacement, nearest neighbor with caliper matching of 0.1 σp, optimal matching, and both probit and logistic full matching (Ho et al., 2011). We used the weighted propensity scores in a logistic regression model to estimate the average treatment effects for the treated (ATT) and the entire cohort (ATE), taking supervision success as our outcome of interest. We also incorporated additional covariates and matching weights into the model. We used the glm function to estimate the treatment effect and the vcovCL function from the sandwich package to calculate cluster-robust standard errors (Zeileis et al., 2020). We used the twang R package to compute the ATT for further validation, and the results across techniques were compared (Cefalu et al., 2021).

Most Balanced Approach

For evaluating the ATT, the nearest neighbor with a caliper matching of 0.1 σp was the most balanced method (Ho et al., 2011). This technique matched 810 SMHS clients to 810 individuals in the control group. All standardized mean differences for the covariates were near or below 0.1 postmatching, indicating a balanced comparison (Ho et al., 2011). Full matching with logit regression was the most balanced method to estimate the ATE. This approach matched approximately 219.82 SMHS clients to 3,106.19 individuals in the control group.

Results

Differences in Outcomes for People with and without a Mental Health Need

As shown in Figure 2, just more than half of the cohort of 27,396 individuals were rearrested within 2 years (51.9%, n = 14,210), and less than a quarter were revoked (23.5%, n = 6,424). Once the sample was subset into people without mental illness and people with mental illness, the overall failure outcomes were almost 50% higher for individuals with mental illness. In addition, 68.5% (n = 5,440) of people with an identified mental health need were rearrested or revoked within 2 years of being placed on supervision compared with 46.1% (n = 8,962) of people without any identified mental health need. There was a statistically significant difference between the failure rates for the cohorts across all three measures: rearrests, revocations, and the combined failure rate (p < .001). Moreover, individuals with identified mental health needs comprised 42.4% (n = 2,727) of all revocation events despite representing only 29.0% (n = 7,947) of the total cohort population.

Figure 2:

Failure Rates for Full Cohort, for People With or Without an Identified Mental Health Need, and for People on SMHS Caseload or Not on SMHS Caseload

To examine trends among people with mental health needs, we conducted a more in-depth look at the nature of arrests and revocations. People with mental illness had higher rearrest rates for probation or parole violations (38.6%, n = 2,066) than those without mental illness (34.2%, n = 3,032; p < .001). Similarly, people with mental illness had higher revocation rates for technical violations of supervision (51.0%, n = 1,432) than those without mental illness (46.2%, n = 1,654; p < .001).

Characteristics of SMHS Participants and Comparison Group

As shown in Table 1, the treatment and control groups had statistically significant differences for all covariates except education and location (urban vs. rural). The comparison group had significantly higher risk scores, more prior felony convictions, and more prior charges on record than individuals in the treatment group. In addition, almost two thirds of the comparison group had been previously incarcerated in the GDC and supervised on high-risk caseloads.

Table 1:

Characteristics of SMHS Population and Comparison Group

	Caseload type, total no. (%)
	SMHS(n = 1,383)			Comparison group(n = 6,564)
Demographics	n	%	SD	n	%	SD	p Value for group differences
Age—M	40.25		12.02	38.19		11.00	<.001
Sex—Male	975	72		5,173	80		<.001
Sex—Female	377	28		1,283	20		<.001
Race—Black	575	43.5		3,243	51		<.001
Race—White	747	56.5		3,122	49		<.001
Education—Some HS or less	423	40		2,310	41		.282
Education—HS diploma or GED	614	58		3,287	58
Education—Some college or more	21	2		77	1
Location—Urban	1,050	76		4,892	75		.265
Location—Rural	329	24		1,656	25		.265
Overall score—M	6.5		3.01	8.0		2.44	<.001
Low risk (Levels 1–3)	296	22		565	9
Medium risk (Levels 4–7)	406	30		1,235	19
High risk (Levels 8–10)	636	48		4,651	72
Direct probation	681	49		2,356	36		<.001
Post-incarceration	702	51		4,208	64		<.001
Substance use flag (yes)	644	47		4,898	75		<.001
Violent offense flag (yes)	482	35		1,923	29		<.001
Num prior felonies—M	6.23		6.32	9.56		7.90	<.001
Num prior charges—M	16.36		16.82	24.43		20.17	<.001

Note. SMHS = Specialized Mental Health Supervision; SD = standard deviation; HS = high school; GED = general educational development.

Table 2:

Outcomes of Propensity Score Model

Outcomes	ATT using MatchIt	ATT using twang	ATE using MatchIt
Outcome coefficient	.14	.18	.05
SE	.124	.115	.165
p value	.242	.126	.745

Note. ATT = average treatment effect on the treated; ATE = average treatment effect.

Outcome Differences for SMHS Participants and Comparison Cohort

The next step in the analysis focused mainly on the 7,947 people with identified mental health needs to assess whether supervision on the SMHS caseload resulted in fewer rearrests and revocations than individuals not on an SMHS caseload. As shown in Figure 2, a bivariate analysis indicated that placement on SMHS resulted in lower revocations and slightly lower rearrests than the comparison group of individuals with mental health needs but supervised on other caseload types (χ² = 22.63, p < .001).

Outcomes to Predict Placement on SMHS Caseload

We found that individuals supervised on SMHS caseloads had a lower risk profile on average than individuals with mental illness not supervised on specialized caseloads. As demonstrated in Figure 3, a disproportionate percentage of individuals in the comparison cohort have a high-risk score.

Figure 3:

Assessed Risk Level Across Study Cohorts

To examine how risk score affects SMHS placement, we developed a logistic regression model with SMHS as the dependent variable and risk score, sex, race, age, education status, and a flag for whether a mental health profile exists as the independent variables. We limited the sample population to individuals with an identified mental health need. Table 3 shows the regression output for the model. An individual’s assessed risk score was a statistically significant predictor for placement on the SMHS caseload (p < .001). With each one-unit increase in the risk score, individuals have 17% lower odds of being placed on the SMHS caseload. Other statistically significant covariates in predicting placement on an SMHS caseload include sex, education level, and whether GDC identified the individual as having a mental illness.

Table 3:

Outcomes to Predict Placement on SMHS Caseload

Variables	(1)Simple model	(2)seEform	(3)Demographics model	(4)seEform	(5)Full model	(6) seEform
Highest risk score	0.82***	(0.01)	0.83***	(0.01)	0.83***	(0.01)
Sex			0.90	(0.07)	1.72***	(0.19)
Race			0.96	(0.06)	1.04	(0.09)
Age			0.99	(0.02)	1.01	(0.02)
Age squared			1.00	(0.00)	1.00	(0.00)
Education					0.72*	(0.10)
Education squared					1.01*	(0.01)
Has mental health profile					28.08*	(2.65)
Constant	0.87	(0.07)	1.15	(0.40)	0.95*	(0.83)
Observations	7,789		7,651		6,583
Pseudo-R²	0.0478		0.0490		0.305

Note. SMHS = Specialized Mental Health Supervision; seEform in parenthesis.

p < .05. **p < .01. ***p < .001.

Postmatching Regression Analysis

We found consistent results using the MatchIt and twang R packages. The estimated ATT, represented by the coefficient for SMHS, was 0.14 (SE = 0.124, p = .242), indicating that after controlling for differences associated with our covariates, SMHS has a negligible positive effect on supervision success. This effect also lacks statistical significance, implying that SMHS does not significantly impact supervision success when controlling for the observed covariates. Similarly, the ATE was estimated to be 0.05 (SE =0.165, p = .745), indicating no statistically significant average treatment effect across the entire population. We found similar results using the twang R package, yielding an ATT of 0.18 (SE =0.115, p = .126). This coefficient estimate is also negligible and statistically not significant.

Discussion

The high rates of individuals with mental illness in the legal system (Al-Rousan et al., 2017; Prins, 2014; Van Deinse et al., 2018), coupled with increased odds of rearrests and technical violations (Cloyes et al., 2010; Ostermann & Matejkowski, 2014), make the intersection of mental health and community supervision worthy of attention for practitioners, scholars, and administrators. This study notably fills a gap in current research as one of the few to focus on revocations as a criminal justice outcome for people with mental illness, particularly in the context of SMHS programs. Our findings demonstrate that individuals with mental health needs not only fare worse in rearrests, which is consistent with previous findings in the research literature and a vital replication, but they are also more likely to have their probation/parole term revoked than those without mental health concerns. Moreover, people with mental illnesses experienced revocations for technical violations significantly more often than those without, underscoring the importance of developing specialized strategies for assisting clients with mental health needs.

Thus, SMHS programs aim to enhance service delivery among clients with mental illness (Council of State Governments, 2002; Skeem et al., 2006). However, it remains unclear whether SMHS is an effective intervention strategy for improving criminal justice outcomes (Manchak et al., 2014; Skeem et al., 2017; Van Deinse et al., 2021; Wolff et al., 2014). Therefore, this study’s second question asked whether participating in the SMHS program decreased the likelihood of rearrest and revocation among clients with mental illness. Although initial findings showed a significant effect of SMHS, the association disappeared once we controlled for caseload characteristics, such as criminal risk factors. These results highlight the importance of understanding who receives the SMHS intervention.

For this reason, comparing criminal risk scores of SMHS participants to clients with mental health needs not placed on specialized caseloads (i.e., supervision services as usual) deserves observation. As such, the third research question of this project examines the relationship between criminal risk scores and placement on SMHS caseloads. Our results indicate that the higher the risk score is among clients with mental illness, the less likely they are to receive SMHS services, contrary to the risk principle (Bonta & Andrews, 2017).

Implications

Our study advances knowledge of SMHS as an intervention strategy in a few meaningful ways. Following the risk need responsivity framework, effective supervision with any client entails matching intervention intensity with their likelihood of committing new crimes (i.e., the risk principle) according to a validated risk assessment tool (Bonta & Andrews, 2017). However, the relationship between supervisees with mental illness and risk assessment is complicated. For example, on one hand, the tendency to overemphasize the risk of reoffending among persons with mental illness (Eno Louden et al., 2018) may lead to a lack of fidelity when officers are implementing risk assessment results among this group (see Miller & Maloney, 2013). On the other hand, if officers view SMHS as less punitive or more lenient supervision because it focuses on referrals to treatment and less on surveillance, they may underestimate the mental health needs of clients at an elevated risk of reoffending because they are paying more attention to risk than mental health symptoms. Our findings demonstrate a target area for improving SMHS implementation by reinforcing the importance of explicit agency policies, clarifying that mental illness should not automatically prompt placement into an intensive intervention (Eno Louden et al., 2020). Instead, practitioners should adhere to all three principles of the Risk-Needs-Responsivity (RNR) model when considering eligibility. Thus, in addition to targeting criminogenic needs that drive recidivism (i.e., the need principle) and tailoring interventions to the abilities of individuals (i.e., the responsivity principle), agencies should include criminal risk factors as a critical part of the SMHS selection process (i.e., the risk principle).

Next, our findings that people with mental health needs fare worse on supervision and are more often assigned to standard caseloads than SMHS caseloads highlight this population’s challenges. With the high volume of persons with mental health needs and the limited mental health resources in the criminal justice system, most people with mental health needs are still supervised in nonspecialized settings (Wolff & Pogorzelski, 2005). Therefore, it is safe to conclude that organizations should train all officers, not only specialized officers, to assist people with mental illnesses.

Finally, given the mixed recidivism outcomes for SMHS, it is essential to continue academic and organizational research into the factors that influence not only its efficacy on supervision success but also fidelity outcomes for agencies and mental health outcomes for supervisees. For instance, prior research links the fidelity of SMHS programs, even partially, to increased treatment engagement among supervisees with mental health needs (Van Deinse et al., 2021). Variability in these outcomes may be due to inconsistent program implementation, an overreliance on treatment to reduce recidivism, or differences in risk attention. Gaining a deeper understanding of elements that promote or hinder SMHS implementation will inform policy decisions that may improve supervision services for persons with mental illnesses.

Limits and Future Directions

To guide further research on this topic, we must identify and address the limitations of this investigation. First, consistent diagnostic data were not available. Therefore, it was not possible to determine the severity of individuals’ mental illnesses, which would have made the matching between the treatment and control groups stronger. Extending research in this area should include linking data from probation/parole agencies to scores on a standardized mental health scale, which will better illustrate clients’ overall experience with mental illness.

Furthermore, people change supervision types, risk levels, and locations throughout a supervision period. In this study, we assigned the most frequently occurring supervision type, risk level, and judicial circuit location to each case in the cohort. Few individuals assigned to the SMHS caseload remained on specialized supervision for their entire supervision period, whereas others spent as little as 3 months. Thus, we could not incorporate differences in treatment dosage into the model because it was impossible to discern whether a shortened length of stay was related to failure (revocation or rearrest) or transferring to a different type of supervision. As an essential next step, future studies should consider using time series analyses to assess how participants enter and exit SMHS caseloads over time.

Finally, some supervision circuits had lower recidivism rates than the statewide average. It was outside the scope of this study to incorporate a nested model by supervision circuit or officer, and some of the rural circuits would have needed more cases to support such a model. An analysis by the supervision circuit or assigned officer would increase our understanding of the relationship between adherence to the prototypical SMHS model and reduced failure rates.

Conclusion

As many studies have illustrated the negative supervision experience for individuals with mental illness (Babchuk et al., 2012; Lurigio et al., 2012), this is not a discovery but a critical replication. With few exceptions, however, no studies have examined revocations among supervisees with mental illness (see Baillargeon et al., 2009) or the role of criminal risk factors in SMHS caseload placement. Therefore, this investigation contributes to this topic by comparing failure rates (i.e., rearrest or revocation) between clients with mental illness and those without and whether SMHS reduces the risk of failure among participants. Consistent with prior studies, our findings show that clients with mental illness fare worse on supervision than others, and there is some support for SMHS as a potential intervention for breaking the cycle. However, the extent to which SMHS effectively improves criminal justice outcomes may depend on caseload characteristics, such as criminal risk factors. To this end, this project revealed lower criminal risk score composition among SMHS caseloads than among clients with mental illness not placed on SMHS caseloads. Since prior research supports applying the “risk principle” to persons with mental illnesses (Skeem et al., 2015), these findings underscore the potential for probation and parole agencies to improve SMHS services by ensuring their programs align with the RNR model.

Supplemental Material

sj-docx-1-cjb-10.1177_00938548241232562 – Supplemental material for Specialized Mental Health Supervision: Revocations and Risk Composition

Supplemental material, sj-docx-1-cjb-10.1177_00938548241232562 for Specialized Mental Health Supervision: Revocations and Risk Composition by Nicholas K. Powell, Angela Gunter, Mari Roberts and Tonya Van Deinse in Criminal Justice and Behavior

Footnotes

Appendix

Table A1:

Chi-Square and Cramer’s V Tests for Bivariate Analyses

Variables	Sample size	p value	Chi-square (χ²)	Cramer’s V (φ_c)
Arrests and mental health flag	27,396	<.001	1.1e+03	−0.1970
Revocations and mental health flag	27,396	<.001	736.33	−0.1639
All failures and mental health flag	27,396	<.001	1.1e+03	−0.2033
Arrests and placement on SMHS	7,947	<.001	23.79	0.0547
Revocations and placement on SMHS	7,947	<.001	22.18	0.0528
All failures and placement on SMHS	7,947	<.001	22.63	0.0534

Note. SMHS = Specialized Mental Health Supervision.

Authors’ Note:

We have no conflicts of interest to disclose.

ORCID iD

Nicholas K. Powell

Supplemental Material

Supplemental Material is available in the online version of this article at .

Nicholas K. Powell, PhD, is a professor in the Criminal Justice and Criminology Department at Georgia Southern University and director of Strategic Planning & Analysis at the Georgia Department of Community Supervision. He studies the social context of crime. Specifically, his research focuses on patterns surrounding the intersections of legal system involvement (i.e., probation, parole, corrections) and contextual factors, such as addiction, mental health, and homelessness.

Angela Gunter, MPA, is a policy researcher for the Council of State Governments. Her research interests include recidivism, reentry, pretrial, reform policy, and community supervision. Before working with the CSG Justice Center, she assisted the Texas Criminal Justice Policy Council with forecasting the impact of juvenile justice policy options for the legislature.

Mari Roberts, MS, is a data scientist for the Council of State Governments (CSG). Her research interests include behavioral health, sentencing forms, and corrections. Before joining the CSG Justice Center, she tracked COVID-19 in New York State prisons and reported findings to the legislature and the broader public.

Tonya Van Deinse, PhD, MSW, is a research associate professor at the University of North Carolina at Chapel Hill School of Social Work. She teaches courses in the community, management, and policy practice concentration, including social work with groups and communities, interorganizational and community-level theory and practice, and research and evaluation. Her research focuses on nationwide approaches to probation for adults with mental illnesses, using implementation science to enhance the uptake of mental health interventions within criminal justice settings.

References

Al-Rousan

Rubenstein

Sieleni

Deol

Wallace

R. B.

(2017). Inside the nation’s largest mental health institution: A prevalence study in a state prison system. BMC Public Health, 17(342), 1–9. https://doi.org/10.1186/s12889-017-4257-0

Andrews

D. A.

Bonta

(2010). Rehabilitating criminal justice policy and practice. Psychology, Public Policy, and Law, 16(1), 39–55.

Andrews

D. A.

Bonta

Wormith

J. S.

(2006). The recent past and near future of risk and/or need assessment. Crime and Delinquency, 52(10), 7–27. https://doi.org/10.1177/0011128705281756

Babchuk

L. C.

Lurigio

A. J.

Canada

K. E.

Epperson

M. W.

(2012). Responding to probationers with mental illness. Federal Probation, 76(2), 41–48. https://www.uscourts.gov/sites/default/files/76_2_8_0.pdf

Baillargeon

Williams

B. A.

Mellow

Harzke

A. J.

Hoge

S. K.

Baillargeon

Greifinger

R. B.

(2009). Parole revocation among prison inmates with psychiatric and substance use disorders. Psychiatric Services, 60(11), 1516–1521. https://doi.org/10.1176/appi.ps.60.11.1516

Bonta

Andrews

D. A.

(2017). The psychology of criminal conduct (6th ed.). Routledge.

Bonta

Blais

Wilson

H. A.

(2014). A theoretically informed meta-analysis of the risk for general and violent recidivism for mentally disordered offenders. Aggression and Violent Behavior, 19, 278–287. https://doi.org/10.1016/j.avb.2014.04.014

Bonta

Law

Hanson

(1998). The prediction of criminal and violent recidivism among mentally disordered offenders: A meta-analysis. Psychological Bulletin, 123(2), 123–142. https://doi.org/10.1037/0033-2909.123.2.123

Cefalu

Ridgeway

McCaffrey

Morral

Griffin

B. A.

Burgette

(2021). Twang: Toolkit for weighting and analysis of nonequivalent groups (R package version 2.5). https://CRAN.R-project.org/package=twang

10.

Cloyes

K. G.

Wong

Latimer

Abarca

(2010). Time to prison return for offenders with serious mental illness released from prison a survival analysis. Criminal Justice and Behavior, 37(2), 175–187. https://doi.org/10.1177/0093854809354370

11.

Council of State Governments. (2002). Criminal justice/mental health consensus project. https://csgjusticecenter.org/wp-content/uploads/2020/01/consensus-project-full-report.pdf

12.

Crilly

J. F.

Caine

E. D.

Lamberti

J. S.

Brown

Friedman

(2009). Mental health services use and symptom prevalence in a cohort of adults on probation. Psychiatric Services, 60(4), 542–544. https://doi.org/10.1176/ps.2009.60.4.542

13.

Ditton

(1999). Mental health and the treatment of inmates and probationers. U.S. Department of Justice, Bureau of Justice Statistics. https://bjs.ojp.gov/content/pub/pdf/mhtip.pdf

14.

Eno Louden

Adair

Skeem

J. L

. (2020). Moving past the myths: Research-informed practices for supervising clients with serious mental illness. Perspectives, 44(2), 18–31. https://www.appa-net.org/eWeb/docs/APPA/pubs/Perspectives/Perspectives_V44_N2/index_20.html#page=10

15.

Eno Louden

Manchak

S. M.

Ricks

E. P.

Kennealy

P. J

. (2018). The role of stigma toward mental illness in probation officers’ perceptions of risk and case management decisions. Criminal Justice and Behavior, 45(5), 573–588. https://doi.org/10.1177/0093854818756148

16.

Harris

Horst

J. S.

(2019). A brief guide to decisions at each step of the propsensity score matching process. Practical Assessment, Research, adn Evaluation, 21(4). https://doi.org/10.7275/yq7r-4820

17.

Imai

King

Stuart

(2011). MatchIt: Nonparametric preprocessing for parametric causal inference. Journal of Statistical Software, 42(8), 1–28. https://doi.org/10.18637/jss.v042.i08

18.

Lurigio

A. J.

Epperson

M. W.

Canada

K. E.

Babchuk

L. C.

(2012). Specialized probation programs for people with mental illness: A review of practices and research. Journal of Crime and Justice, 35(2), 317–326. https://doi.org/10.1080/0735648X.2012.682845

19.

Manchak

S. M.

Skeem

J. L.

Kennealy

P. J.

Eno Louden

(2014). High-fidelity specialty mental health probation improves officer practices, treatment access, and rule compliance. Law and Human Behavior, 38(5), 450–461. https://doi.org/10.1037/lhb0000076

20.

Miller

Maloney

(2013). Practitioner compliance with risk/needs assessment tools: A theoretical and empirical assessment. Criminal Justice and Behavior, 40(7), 716–736. https://doi.org/10.1177/0093854812468883

21.

Morgan

R. D.

Fisher

W. H.

Duan

Mandracchia

J. T.

Murray

(2010). Prevalence of criminal thinking among state prison inmates with serious mental illness. Law and Human Behavior, 34(4), 324–336. https://doi.org/10.1007/s10979-009-9182-z

22.

Murnane

R. J.

Willett

J. B.

(2011). Methods matter: Improving causal inference in educational and social science research. Oxford University Press.

23.

Ostermann

Matejkowski

(2014). Exploring the intersection of mental health and release status with recidivism. Justice Quarterly, 31(4), 746–766. https://doi.org/10.1080/07418825.2012.677465

24.

Prins

S. J.

(2014). Prevalence of mental illnesses in US State prisons: A systematic review. Psychiatric Services, 65(7), 862–872. https://doi.org/10.1176/appi.ps.201300166

25.

Skeem

J. L.

Emke-Francis

Eno Louden

(2006). Probation, mental health, and mandated treatment: A national survey. Criminal Justice and Behavior, 33(2), 158–184. https://doi.org/10.1177/009385480528440

26.

Skeem

J. L.

Manchak

Montoya

(2017). Comparing public safety outcomes for traditional probation vs specialty mental health probation. JAMA Psychiatry, 74(9), 942–948. https://doi.org/10.1001/jamapsychiatry.2017.1384

27.

Skeem

J. L.

Steadman

H. J.

Manchak

S. M.

(2015). Applicability of the Risk-Need-Responsivity model to persons with mental illness involved in the criminal justice system. Psychiatric Service, 66(9), 916–922. https://doi.org/10.1176/appi.ps.201400448

28.

Taxman

. (2014). Second generation of RNR: The importance of systemic responsivity in expanding core responsivity principles. Federal Probation, 78(2), 32–40. https://www.uscourts.gov/sites/default/files/78_2_5_0.pdf

29.

Van Deinse

T. B.

Cuddeback

G. S.

Wilson

A. B.

Lambert

Edwards

. (2018). Using statewide administrative data and brief mental health screening to estimate the prevalence of mental illness among probationers. Probation Journal, 66(2), 236–247. https://doi.org/10.1177/0264550518808369

30.

Van Deinse

T. B.

Givens

Cowell

Ghezzi

Murray-Lichtman

Cuddeback

G. S

. (2021). A randomized trial of specialty mental health probation: Measuring implementation and effectiveness outcomes. Administration and Policy in Mental Health and Mental Health Services Research, 49, 415–428. https://doi.org/10.1007/s10488-021-01172-0

31.

Wilson

A. B.

Farkas

Ishler

K. J.

Gearhart

Morgan

Ashe

(2014). Criminal thinking styles among people with serious mental illness in jail. Law and Human Behavior, 38(6), 592–601. https://doi.org/10.1037/lhb0000084

32.

Wolff

Epperson

Shi

Huening

Schumann

B. E.

Sullivan

I. R.

(2014). Mental health specialized probation caseloads: Are they effective? International Journal of Law and Psychiatry, 37(5), 464–472. https://doi.org/10.1016/j.ijlp.2014.02.019

33.

Wolff

Morgan

R. D.

Shi

Fisher

Huening

(2011). Comparative analysis of thinking styles and emotional states of male and female inmates with and without mental disorders. Psychiatric Services, 62(12), 1485–1493. https://doi.org/10.1176/appi.ps.000432011

34.

Wolff

Pogorzelski

(2005). Measuring the effectiveness of mental health courts: Challenges and recommendations. Psychology, Public Policy, and Law, 11(4), 539–569. https://doi.org/10.1037/1076-8971.11.4.539

35.

Zeileis

Köll

Graham

(2020). Various versatile variances: An object-oriented implementation of clustered covariances in R. Journal of Statistical Software, 95(1), 1–36. https://doi.org/10.18637/jss.v095.i01

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.49 MB