Validating Responsivity Assessments for Correctional Populations: Evaluating the Association With Program Participation,Dosage,and Completion

Abstract

While specific responsivity factors are believed to either inhibit or facilitate successful involvement in programming, relatively little attention has been given to the validation of responsivity assessments used for correctional populations. Based on a sample of nearly 2,100 individuals confined in Minnesota prisons, this study examines the relationship between a needs and responsivity assessment system and multiple measures of program participation, dosage, and completion. The responsivity domains—childhood trauma, mental health, religiosity, motivation and learning style—were significantly associated with at least one measure of programming. All but two of the needs domains also had a significant association with programming involvement. We discuss the implications of the findings for correctional research and practice.

Keywords

responsivity assessment validation risk-needs-responsivity model programming

Over the last several decades, the risk-needs-responsivity (RNR) model has gained traction as the guiding framework for the delivery of programming to correctional populations. The risk principle holds that programming should be calibrated to a person’s risk of reoffending, with greater dosages reserved for higher-risk individuals. The needs principle indicates that programming should target a person’s criminogenic needs, which are dynamic recidivism risk factors that are susceptible to change. The responsivity principle maintains that programming should be tailored to an individual’s learning style, strengths, and abilities (Andrews et al., 2006; Bonta & Andrews, 2017).

To help determine who, what, and how programming should be delivered, the RNR model relies on assessments. Some tools have been designed to focus only on predicting a risk-related outcome (e.g., recidivism, prison misconduct, failure to appear, etc.) or assessing specific domains for criminogenic needs (e.g., substance use, anti-social thinking, anti-social peers, etc.) or responsivity (e.g., childhood trauma). Other instruments, however, have been created to simultaneously assess for risk, needs, and, less frequently, responsivity. Even when there has been an attempt to account for responsivity within an RNR assessment instrument, it has seldom been more than a checklist of items such as language, gender, and culture.

Irrespective of whether an instrument was created to assess a specific domain or all three areas of the RNR model, it is imperative to evaluate its validity. Perhaps because responsivity has attracted the least amount of attention among the three principles within the RNR model, validations of responsivity assessments have, to our knowledge, been virtually nonexistent. Instead, given that existing validations of assessment tools for correctional populations have focused mostly on their ability to predict recidivism, prior validation research has been confined to risk and, to a lesser extent, need assessments.

When evaluating the predictive validity of instruments that have been designed to assess recidivism risk, observed recidivism is the measure that should be used. Moreover, a criminogenic need is, by definition, expected to not only have a significant, direct impact on reoffending, but interventions that successfully target this need will lower recidivism. For instance, substance use is a criminogenic need with a significant, direct impact on recidivism (Gendreau et al., 1996), and substance use disorder (SUD) treatment has been shown to reduce reoffending (Mitchell et al., 2007). On the contrary, when needs are not effectively addressed, the likelihood of recidivism increases (Krushas et al., 2024). Thus, as with risk assessments, recidivism is the outcome measure that should be used to determine the predictive validity of instruments designed to assess criminogenic needs.

Specific responsivity factors, on the contrary, are not expected to have a significant, direct impact on criminal behavior. While general responsivity refers to types of programming that are most effective in reducing recidivism, such as cognitive-behavioral interventions, specific responsivity includes individual barriers that may limit the likelihood of program participation and successful completion (Bonta & Andrews, 2017). Examples of specific responsivity factors include motivation, anxiety, learning style, mental health, religiosity, language, gender, and culture (Cullen, 2002; Duwe & Johnson, 2023; McCormick et al., 2017; Mowen et al., 2018; Pinals et al., 2021; Sachs & Miller, 2018).

The association between specific responsivity factors and recidivism, if it exists, likely reflects the influence of criminogenic needs. For example, in their discussion of mental illness, Andrews and colleagues (2006) point out that it has only a modest, indirect impact on reoffending. Whatever effect mental illness has on recidivism, Andrews et al. (2006) continued, is likely due to the impact of criminogenic needs such as substance use, criminal thinking, and antisocial personality pattern.

Rather than having a significant, direct association with recidivism, specific responsivity factors are believed to affect whether individuals can successfully complete programming that targets criminogenic needs (McCormick et al., 2017; Pinals et al., 2021). For example, the presence of mental illness can interfere with the ability to complete SUD treatment. Because specific responsivity factors are expected to predict involvement in, and completion of, programming, validating a responsivity assessment should involve examining whether an empirical relationship exists with program involvement and completion. To date, there has been very little, if any, prior research that has validated responsivity assessments by evaluating their association with measures of program participation and completion.

The Present Study

In 2021, the Minnesota Department of Corrections (MnDOC) piloted a needs and responsivity assessment that was designed to be used in tandem with the Minnesota Screening Tool Assessing Recidivism Risk (MnSTARR) 2.0, a recidivism risk assessment instrument that has been validated on Minnesota’s prison population (Duwe, 2021). The needs and responsivity components consist of assessments for eight criminogenic needs domains (anti-social thinking, anti-social peers, SUD, education, employment, housing stability, family/domestic relationships, and self-identity) and five responsivity domains (religious faith and spirituality, mental health, childhood trauma, self-identity, motivation, and learning style). Of the 13 needs and responsivity domains, the MnDOC has existing assessment processes for SUD, education, and mental health. For the remaining 10 domains, we developed a self-reported assessment that was administered in March 2021 to individuals confined in Minnesota prisons.

Our sample for this study consists of nearly 2,100 people who completed the self-reported assessment. We evaluate the validity of the needs and responsivity domains by examining their empirical relationship with multiple measures of program participation, dosage, and completion following the administration of the self-reported assessment (SRA). Because the SRA was a pilot, the results were not used to influence any decisions relating to program assignment. In addition to controlling for gender and race/ethnicity, we account for length of stay in prison and prior program participation in the statistical analyses.

The Development of the MnDOC Risk-Needs-Responsivity Assessment System

The MnDOC developed a needs and responsivity assessment to supplement existing RNR assessments used by the agency. Prior to 2021, the MnDOC used the Level of Service/Case Management Inventory (LS/CMI) to assess for criminogenic needs, and the MnSTARR 2.0 to assess for recidivism risk. Separately, individuals admitted to the MnDOC were also assessed for SUDs, level of education, and physical and mental health issues.

Since 2016, the MnSTARR 2.0 has been fully automated and can be administered to everyone who enters Minnesota’s prison system (Duwe & Rocque, 2017). This gender-specific tool draws data from multiple databases to assign individuals to one of four risk levels: (a) low, (b) medium, (c) high, and (d) very high. Twenty percent of individuals fall in the highest risk category (very high), followed by the next highest 20 percent (high), the middle 20% (medium), and the bottom 40% (low). A recent revalidation study of the MnSTARR 2.0 found that it accurately predicts recidivism for Minnesota’s prison population, with an overall area under the curve (AUC) of 0.74 for females and 0.73 for males (Duwe, 2021).

In addition to the MnSTARR 2.0, nearly all individuals who enter Minnesota prisons are screened for mental or physical health issues as well as level of education. Behavioral health staff screen individuals for intellectual disabilities, mental illness, mood disorders, traumatic brain injuries, suicide and self-harm potential, the potential for causing harm to others, and whether the person is likely to be disruptive, among other issues. Education staff verify earned secondary and post-secondary credentials, and most individuals are tested for adult basic education skills.

Due to limited resources, the SUD and LS/CMI assessments have been administered more selectively. The SUD assessments are limited to individuals who will be confined for at least 150 days. Because the LS/CMI requires lengthy interviews and file reviews to complete, only those individuals who were confined for 6 months or more and fell in the high or very high risk categories on the MnSTARR 2.0 were assessed. Even with these limitations, a review of releases over 2016 and 2017 found that only a little more than half of the individuals who should have been assessed with the LS/CMI were assessed (Duwe, 2021).

To ensure that most individuals admitted to Minnesota prisons receive a comprehensive RNR assessment, the MnDOC developed an assessment to be used in combination with the MnSTARR 2.0 and existing SUD, education, and mental and physical health assessments. To create this new assessment, MnDOC staff assembled a series of smaller existing validated assessments or created new assessments based on relevant literature. This new supplemental assessment can be self-administered by incarcerated persons in less than 1 hr.

The new self-reported assessment includes six criminogenic needs domains: Anti-Social Cognition, Anti-Social Associates, Family/Domestic, Self-Identity, Employment, and Housing/Homelessness. The second iteration of the Texas Christian University-Criminal Thinking Scales (TCU-CTS) were used to measure Anti-Social Thinking (Knight et al., 2006), and the Attitudes toward Associates scale within the Measures of Criminal Attitudes and Associates was used to measure Anti-Social Peers (Mills et al., 2002).

To measure relationships with family, significant others, and friends, the self-reported assessment includes the Multidimensional Scale of Perceived Social Support (MSPSS; Zimet et al., 1988). To supplement measures of support received from family and significant others included in the MSPSS, a four-item scale that measures relations with children was also included. This scale includes questions about whether children lived with the respondent prior to incarceration, the quality of relationships with children, whether the respondent has active child support orders, and whether the respondent’s parental rights have been terminated.

Self-identity is measured using two scales derived from the literature that measure both how an individual sees himself or herself and the individual’s openness to change. The first scale, adapted from Paternoster and Bushway’s (2009) “feared self” concept, includes six items. These items include, for example, “I like the person I have become” and “I have hit rock bottom in my life.” The second scale is based on Giordano and colleagues’ (2002) concept of cognitive transformation and includes six items related to how open individuals are to change and the desire to pursue a new way of life. This second measure includes items such as “I am open to a new way of life” and “I want to avoid criminal behaviors.” Higher scores on the Self-Identity domain reflect more entrenched antisocial identities.

Because there were no existing validated assessments for employment or housing issues, MnDOC staff created measures for each domain. The Employment domain is measured using questions about work history, career planning, and job search skills. A higher score for this domain indicates poor work history, a lack of job search or interview skills, and/or uncertainty about employment prospects. The Housing/Homelessness domain is measured with a series of questions about where individuals lived prior to incarceration, where they expect to live after incarceration, and past experiences with housing instability. A higher score in this domain indicates past experiences with housing instability and/or a higher likelihood of homelessness upon release from incarceration.

The self-reported assessment also measures four responsivity domains, including Childhood Trauma, Religiosity, Motivation, and Learning Style. The Childhood Trauma domain is measured using the Adverse Childhood Experiences (ACEs) index adapted from the Family Health History questionnaire developed by Felitti and colleagues (1998). Using 18 questions, the ACEs questionnaire measures 10 forms of trauma. Individuals receive a score of 0 to 10 based on the number of forms of trauma they reported.

The Religiosity domain was measured using three questions about religious involvement previously administered to correctional populations (Jang et al., 2017). The questions used for this index include: (a) “In general, how important is religion to you?,” (b) “How often do you prefer to attend religious services?,” and (c) “About how often do you spend time alone praying or reading the Bible, Koran, Torah, or other sacred book?.” Higher scores on this domain indicate that religion is a salient factor in the respondent’s life.

The Motivation domain is based on a series of questions that ask incarcerated persons to first identify a change they want to make to avoid future involvement in the criminal justice system, and then how confident they are in making that change. Individuals could choose from eight possible changes (e.g., quit using drugs or alcohol, find stable housing, complete more education) or insert their own answer.¹ Confidence levels ranged from 0 to 10, with values closer to 10 indicating greater levels of confidence. Individuals could choose two changes and corresponding confidence levels.

Finally, the Visual, Auditory, Reading, Kinesthetic (VARK) instrument was used to assess Learning Style (Fleming & Mills, 1992). The VARK presents respondents with real-life scenarios and four ways to respond to each scenario that align with the four learning styles. Higher scores with one learning preference over all others reveal the individual’s preferred learning style.

Because needs and responsivity domains are latent constructs, it was necessary to analyze the self-reported assessment for validity. Two separate studies have tested the convergent, discriminate, internal, concurrent, and predictive validity of the self-reported assessment using a sample of approximately 2,100 incarcerated persons who took the assessment in the spring of 2021. Clark and Duwe (2023) conducted exploratory and confirmatory factor analyses (EFA and CFA, respectively) on eight of the ten domains included in the self-reported assessment in addition to the Mental Health domain. Results from the EFA and CFA of the TCU-CTS were consistent with prior research (Taxman et al., 2011), revealing weak discriminate validity for three of the six sub-scales. However, the CFA found overall acceptable psychometrics with a six-factor solution, consistent with the TCU-CTS’s six subscales.

The 10 items included in the Attitudes toward Associates scale loaded strongly on to one factor with acceptable fit statistics, and the 12 items included in the MSPSS loaded strongly on three factors, which in turn loaded on a higher-level construct representing social support. The four items that measure any discord between respondents and their children loaded on to one factor with acceptable fit statistics.

To measure the Employment domain, the self-reported assessment originally included 16 employment-related items. However, the EFA found that only half of these items loaded onto two factors. The first factor includes questions pertaining to the respondent’s career history, including whether they were employed in the year prior to incarceration, they have a lot of work experience, they liked the jobs they had in the past, and their employers would give them a favorable job reference. The second factor is geared toward the respondent’s career future, including the respondent’s ability to search and apply for jobs and whether they know what they want to do for work when they are released from custody. The two-factor solution had acceptable fit statistics; however, there were some significant cross-loadings and the correlation between the two factors exceeds conventional standards.

The Self-Identity domain originally included four scales derived from the literature on identity, crime, and desistance. The EFA found that only the feared self scale (Paternoster & Bushway, 2009) had acceptable psychometric properties with a two-factor solution. Three items on the scale relating to realization of the consequences of deviant behavior loaded on to one factor. These items include (a) “I worry that I will eventually burn out if I stay on the same path.,” (b) “I have hit rock bottom in my life.,” and (c) “I need to make a change for the better to become more satisfied in my life.” The second factor includes items that measure whether the respondent is satisfied with the direction of his or her life, including: (a) “I am headed in the right direction in life.,” (b) “I like the person I have become.,” and (c) “I am becoming a better person.” In addition, only three of the original six items included in the cognitive transformation scale (Giordano et al., 2002) loaded on to one factor: (a) “It is important to me that family and friends think of me as a good person.,” (b) “I am open to a new way of life.,” and (c) “I can envision a new life for myself.” Ultimately, the CFA found that the first factor of the feared self scale had optimal fit statistics as a stand-alone construct, and the second factor of the feared self scale along with three-item cognitive change scale fit well on a two-factor solution.

The Mental Health domain originally included eight indicators derived from standard mental and physical health screenings conducted on all new prison admissions. The EFA and CFA found that only four of these items loaded well together on a single factor. These items include whether the respondent has general mental illness, has been diagnosed with a mood disorder, is at heightened risk of suicide, and may engage in self-injurious behaviors. The remaining two domains that were analyzed in (Clark and Duwe, 2023)’s study—Religiosity, and Housing/Homelessness—demonstrated optimal psychometric properties, with each domain loading on a single factor.

Four needs and responsivity domains included in the prison admission process—SUD, Education, Motivation, and Learning Style—were not included in [Clark and Duwe, 2023]’s study. The SUD domain was not included because the authors did not have access to the original assessments used to make SUD treatment decisions. The limited number of indicators used to determine need level for the Education and Motivation domains made these domains ineligible for EFA and CFA. Finally, the Learning Style domain (the VARK assessment) was not included because it is not used to inform programming decisions and it has been validated in previous research (Leite et al., 2010).

In the second validity study, Duwe and Clark (2023) analyzed the concurrent and predictive validity of all 13 need and responsivity domains included in the MnDOC’s new intake process. Concurrent validity was assessed by measuring the extent to which these thirteen domains were correlated with MnSTARR 2.0 risk levels. Predictive validity was based on the association between the domains and the observed recidivism patterns for a sub-sample of incarcerated persons who were released from custody by the end of 2022.

As expected, eight of the domains previously identified as criminogenic needs had at least one significant, positive association with the MnSTARR 2.0 risk level and actual recidivism. Among these criminogenic needs, Anti-Social Peers had the strongest association with both outcomes. Childhood Trauma, which was identified as a responsivity issue and thus would not be expected to have a direct effect on recidivism risk or observed recidivism, had a small significant relationship with the MnSTARR 2.0. However, the remaining four domains initially identified as specific responsivity factors—Mental Health, Religiosity, Motivation, and Learning Style—did not have significant, positive associations with assessed and observed recidivism, which is consistent with what was hypothesized. With few exceptions, the results from both studies (Clark and Duwe, 2023; Duwe and Clark, 2023) were consistent for both males and females and across racial and ethnic groups.

The abovementioned studies have generally supported the construct validity of the needs and responsivity domains included in the MnDOC’s assessment process, particularly the needs domains. Given that specific responsivity factors are not expected to directly impact assessed or actual recidivism, additional tests are required to assess the influence of these domains on program participation. In the ensuing sections of this paper, we first describe the data used for this research and how program participation was operationalized. Next, we present analyses that test the impact of all 13 domains on program participation, followed by a discussion of the implications of these results.

Data and Method

The needs and responsivity assessment was self-administered on desktop computers using Snap computer-assisted survey software. Incarcerated individuals selected to participate in the pilot were notified in writing by their case manager about 1 week prior to taking the assessment. Individuals were advised that their participation in the pilot was completely voluntary, and they could refuse to participate or skip any questions that they did not want to answer. Incarcerated individuals signed a consent form prior to beginning the pilot, and respondents were offered a small incentive in exchange for their participation.

The SRA was administered to incarcerated individuals at all 11 adult prisons in Minnesota in the spring of 2021 (please see the appendix for the SRA). These facilities include a range of custody levels from minimum to maximum throughout the state. All but one of the facilities house men, while the remaining facility houses women. In an effort to achieve the largest sample possible without unduly burdening staff at the men’s facilities, half of the approximately 6,700 men who were incarcerated at the time of the survey were randomly selected. Given the relatively small number of incarcerated women (400), all individuals housed in the lone women’s facility were invited to participate. Of the 3,335 men and 400 women who were invited to participate, 1,758 men and 337 women completed the assessment, resulting in a total participation rate of 56% (53% for men, and 84% for women).

To further validate the needs and responsivity assessment system, we created multiple measures of program participation, dosage, and completion. The program outcome variables were measured from the time the SRA was administered in March 2021 to the point at which individuals were released from prison or December 31, 2022, for those yet to be released from prison. As shown later in Table 1, the average post-SRA follow-up period for the 2,095 individuals in our sample was 423 days.

Table 1.

Sample Descriptive Statistics.

Variables	Variable description	M/%	SD	N
Substance Use Disorder	Substance Use Disorder (SUD) Need Level
High	High Need	0.696		1,458
Medium	Medium Need	0.044		93
Low	Low Need	0.260		544
Education	Education Need Level
High	Less than a secondary degree	0.161		337
Medium	Secondary degree or diploma	0.644		1,350
Low	Post-secondary degree or certificate	0.195		408
Anti-Social Thinking	Average overall TCU-CTS score	22.265	5.237
Self-Identity	Self-identity scales	9.248	1.896
Employment Scale	Employment scale	33.590	8.773
Housing/Homelessness	Housing/homelessness scale	1.978	1.973
Family/Domestic	Family/domestic score	5.897	1.885
Anti-Social Peers	Attitudes toward Associates scale	35.737	10.469
Childhood Trauma	Adverse Childhood Experiences (ACE) score	3.796	2.943
Mental Health	Mental health score	0.543	0.858
Religiosity	Religious Faith/Spirituality Scale	9.39	5.73
Motivation	Motivation score	8.445	1.971
Learning Style	Preferred learning style
Kinesthetic	Kinesthetic learning style score	2.517	1.412
Audio	Audio learning style score	1.997	1.394
Read	Read learning style score	2.471	1.342
Visual	Visual learning style score	0.947	0.947
Gender	Men = 1; Women = 0	0.839	0.367
Race/Ethnicity	White is the reference category
White	White	0.473	0.499
Black	Black	0.291	0.455
Hispanic or Latino/a	Hispanic or Latino/a	0.089	0.284
American Indian	American Indian	0.126	0.332
Asian/Native Hawaiian	Asian or Native Hawaiian	0.030	0.171
Post-Assessment Days	Post-Assessment Days in Prison	422.70	243.99
Pre-Assessment W/P	Pre-Assessment Work/Programming	0.645	0.298
Work/Program Percent	Percentage of time in work/programming	0.804	0.285
Work/Program Time	Days in work/programming	342.67	232.34
Education Classes	Number of education classes	1.870	3.866
Education Awards	Number of education awards	0.710	1.343
Number of Programs	Number of programs	0.472	0.741
N		2,095

Note. TCU = Texas Christian University; CTS = Criminal Thinking Scales.

Given that Learning Style is one of the responsivity domains, we included several variables specific to education programming. The number of education classes measures participation in education programming from the time the SRA was administered to the end of the follow-up period. The number of education awards, on the contrary, quantifies completing classes and earning degrees, certificates, or credentials.

For a broader measure of program participation, we created two additional outcome variables. The first, Work/Program Days, counts the number of post-SRA days in which individuals were either working a job in prison and/or participating in programming. The second, Work/Program Percent, measures the percentage of post-SRA days in which people were working a prison job and/or participating in programming. We also measured program completion more broadly by creating a variable that counts the number of interventions completed from the time the SRA was administered to the end of the follow-up period. The interventions include SUD treatment, a correctional boot camp, and employment, cognitive–behavioral, and faith-based programming.

Participation in the education classes, treatment, programming, and work measured in this study was, for the most part, voluntary. Nevertheless, there are incentives for participation in many of these structured activities. For example, securing a prison labor assignment is usually contingent on having at least a secondary degree, which may encourage some individuals to take adult basic education classes in pursuit of this educational requirement. Similarly, individuals who complete the first phase of the correctional boot camp are eligible for early release from prison, which can be a strong motivation for program participation. In general, however, individuals can refuse to participate in programming if that is their prerogative.

To control for participation in work/programming prior to the SRA, we created a pre-SRA Work/Program Percent variable that measured the percentage of pre-SRA days in prison in which individuals were working and/or participating in programming. We further controlled for the amount of time in prison from the administration of the SRA to the end of the follow-up period. We also included control variables for gender and race/ethnicity. While gender was measured as a binary variable (1 = male; 0 = female), race/ethnicity had five categories: (a) White, (b) Black, (c) Native American, (d) Hispanic, and (e) Asian. Individuals who identified as Hispanic were placed in that category, while those identifying as non-Hispanic were assigned to one of the four remaining categories.

Because both of the education programming measures were count variables, we used a negative binomial regression model for these analyses. We did not use Poisson regression considering that the mean and variance for both outcomes were not equal. Given that the work/programming variables were ratio-level measures, we used Ordinary Least Squares (OLS) regression for these analyses. The errors for both outcome measures were normally distributed, and the OLS model minimized all four measures of fit—deviance, Pearson, Akaike information criterion (AIC), and Bayesian information criterion (BIC)—better than generalized linear models with non-normal distributions. Due to the ordinal structure of the interventions variable, we used ordinal regression to evaluate the association between the needs and responsivity domains and this outcome measure, net of the effects from the control variables.

Results

As shown in Table 1, which provides the descriptive statistics for our sample, 84% were male and the remaining 16% were female. Similar to the racial/ethnic breakdown for the MnDOC prison population in general, nearly half of the respondents were non-Hispanic White individuals, while the other half were Black, Indigenous, or People of Color. The median number of days in prison prior to taking the SRA was 663, with 65% of the time spent in work/programming. After taking the SRA, the amount of time in work/programming was 80%. For education programming, 43 percent of the sample participated in classes after the SRA while 37% earned at least one award. A little more than one-third (35%) of the sample completed an intervention after the SRA, with 10% completing multiple programs.

In Table 2, we present results that analyzed the association between the 13 needs and responsivity domains and education programming. Not surprisingly, the Education domain was positively associated with the number of education classes and awards; that is, following the completion of the SRA, people with a higher need for this domain were enrolled in more education classes and earned more education awards. The Housing/Homelessness domain was positively associated with education classes, while Religious Faith/Spirituality was negatively associated; in other words, as religiosity increased so did subsequent involvement in education programming.

Table 2.

Negative Binomial Regression Models for Education Classes and Awards.

Measures	Education classes		Education awards
Measures	B	SE	B	SE
Substance Use Disorder	0.057	0.053	0.182**	0.052
Education	0.529**	0.070	0.528**	0.067
Anti-Social Thinking	0.002	0.010	−0.004	0.010
Self-Identity	0.029	0.027	0.037	0.026
Employment	0.012	0.009	−0.005	0.008
Housing/Homelessness	0.078**	0.024	−0.003	0.022
Family/Domestic	0.017	0.027	−0.073**	0.025
Anti-Social Peers	−0.005	0.005	0.009	0.005
Childhood Trauma	0.003	0.017	−0.011	0.015
Mental Health	−0.040	0.056	−0.145**	0.052
Religiosity	−0.028**	0.008	0.001	0.007
Motivation	−0.026	0.027	0.011	0.024
Learning Style
Kinesthetic	0.100	0.121	0.222*	0.107
Audio	0.135	0.138	−0.103	0.128
Read	0.172	0.125	−0.035	0.116
Visual	0.055	0.288	0.207	0.256
Pre-Assessment W/P Percent	−0.562**	0.161	−0.110	0.144
Post-Assessment Days	0.003**	0.000	0.000	0.000
Male	−0.188	0.128	−0.266*	0.112
Race/Ethnicity
Black	0.245*	0.108	−0.281**	0.101
Hispanic	0.281	0.159	0.181	0.140
Native American	0.080	0.139	−0.033	0.124
Asian	0.009	0.254	−0.402	0.245
Constant	−1.851**	0.515	−1.512**	0.474
N	2,095		2,095

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

The SUD, Family/Domestic, and Childhood Trauma domains were negatively associated with earning education awards. People rated higher for these domains had significantly fewer education awards. The results also show that individuals with a Kinesthetic preferred Learning Style were, compared with multi-style learners, more likely to obtain education awards. Furthermore, greater involvement in work and/or programming prior to the SRA was negatively associated with education classes, while longer post-SRA follow-up periods were positively associated. Compared with women, men were significantly less likely to earn education awards. While Black individuals were significantly more likely to participate in education classes compared with non-Hispanic White people, they were significantly less likely to earn education awards.

In Table 3, we present the findings from the work/programming participation analyses. The results show that Motivation was positively associated with the number of post-SRA days that individuals were involved in work/programming. On the contrary, the ratings for Anti-Social Thinking, Childhood Trauma, and SUD were negatively associated; that is, people scoring higher for Anti-Social Thinking, Childhood Trauma, and SUD had significantly fewer days of involvement in work/programming. Anti-Social Thinking and SUD were also the only domains that were significantly and negatively associated with the percentage of prison time involved in work/programming.

Table 3.

OLS Regression Models for Work/Programming Participation.

Measures	Days		Percent
Measures	B	SE	B	SE
Substance Use Disorder	−8.148**	2.935	−0.013*	0.007
Education	−0.562	4.319	0.006	0.010
Anti-Social Thinking	−2.841**	0.569	−0.006**	0.001
Self-Identity	0.714	1.466	0.006	0.003
Employment	−0.733	0.481	−0.002	0.001
Housing/Homelessness	0.756	1.320	0.000	0.003
Family/Domestic	−0.097	1.486	−0.004	0.003
Anti-Social Peers	−0.331	0.299	0.000	0.001
Childhood Trauma	−7.740*	3.055	−0.013	0.007
Mental Health	−0.835	0.911	−0.002	0.002
Religiosity	0.008	0.441	0.000	0.001
Motivation	2.909*	1.394	0.005	0.003
Learning Style
Kinesthetic	0.684	6.605	−0.003	0.015
Audio	−1.977	7.577	−0.002	0.017
Read	−3.847	6.862	−0.007	0.016
Visual	−6.059	15.528	0.001	0.035
Pre-Assessment W/P Percent	137.635**	8.735	0.416**	0.020
Post-Assessment Days	0.791**	0.010	0.000	0.000
Male	−1.952	7.050	−0.029	0.016
Race/Ethnicity
Black	−30.189**	5.925	−0.059**	0.013
Hispanic	0.922	8.925	−0.010	0.020
Native American	−9.509	7.624	−0.024	0.017
Asian	2.173	14.254	0.017	0.032
Constant	12.674	27.868	0.715	0.063
N		2,095		2,095

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

The extent to which people were involved in work/programming prior to the SRA was positively associated with both measures of post-SRA work/programming participation. Not surprisingly, longer post-SRA follow-up periods were positively associated with the number of days in which individuals were involved in work/programming. Compared with non-Hispanic White individuals, participation in work/programming was significantly lower for people identifying as Black for both outcome measures.

As shown in Table 4, a number of the domains were negatively associated with program completions. Among the responsivity domains, completing programs was less likely for people rated higher for Mental Health and Childhood Trauma. In addition, compared with multi-style learners, individuals with a Read preferred Learning Style were less likely to complete programs. For the needs domains, individuals rated higher for SUD, Education, Anti-Social Thinking, and Family/Domestic completed significantly fewer programs. Conversely, people scoring higher for Anti-Social Peers were significantly more likely to complete programs.

Table 4.

Ordinal Regression Model for Programs Completed.

Measures	Estimate	SE
Programs Completed
Programs Completed = 0	0.230	0.543
Programs Completed = 1	2.134**	0.544
Programs Completed = 2	5.283**	0.584
Substance Use Disorder	0.844**	0.066
Education	−0.253**	0.083
Anti-Social Thinking	−0.032**	0.011
Self-Identity	−0.022	0.028
Employment	−0.017	0.009
Housing/Homelessness	−0.010	0.025
Family/Domestic	−0.065*	0.029
Anti-Social Peers	0.026**	0.006
Childhood Trauma	−0.037*	0.017
Mental Health	−0.203**	0.062
Religiosity	−0.006	0.008
Motivation	0.025	0.028
Learning Style
Kinesthetic	0.031	0.125
Audio	−0.068	0.145
Read	−0.326*	0.134
Visual	0.219	0.293
Pre-Assessment W/P Percent	1.435**	0.177
Post-Assessment Days	−0.003**	0.000
Male	−0.534**	0.132
Race/Ethnicity
Black	−0.249*	0.116
Hispanic	−0.378*	0.177
Native American	−0.096	0.143
Asian	−0.263	0.282
N	2,095

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

While pre-SRA involvement in work/programming was positively associated with program completion, the length of the post-SRA follow-up period was negatively associated, which likely reflects the MnDOC’s preference for backloading programming closer to an individual’s release date; that is, the people in our sample with shorter post-SRA follow-up periods had been released from prison, which also means they were more likely to have been prioritized for programming in comparison to individuals released after 2022. Men were less likely to complete programs compared with women, whereas the same was true for Black and Hispanic individuals relative to non-Hispanic White people.

Discussion

Given that the assessment results did not influence program placement decisions, the findings reported above provide a rigorous validation of the needs and responsivity assessment system. Except for Self-Identity and Employment, the needs and responsivity domains were significantly associated with at least one measure of program participation, dosage, or completion. Among the responsivity domains, the associations observed between Religiosity and Motivation and at least one measure of programming were in the expected positive direction. Likewise, as we hypothesized, Childhood Trauma and Mental Health were negatively associated with at least one programming measure. Aside from the finding that people with a hands-on learning style preference were more likely to obtain education awards, Learning Style did not have a strong association with program participation, dosage and completion.

These results generally dovetail with the recent academic literature on learning style. For example, research on students from the United States (Dobson, 2010), Iran (Mozaffari et al., 2020), Saudi Arabia (Almigdal, 2015), and Jamaica (Mlambo, 2011) have found that learning style does not have a significant relationship with academic achievement. Although learning style has often been assumed to be a specific responsivity factor since the introduction of the RNR model more than 30 years ago, the findings from this study and the recent literature suggest that it may be worth reconsidering this assumption.

Among the needs domains, Education, Anti-Social Thinking, and SUD were significantly associated with at least three program measures. While Education was, as expected, positively associated with education classes and awards, it was negatively associated with the number of programs completed. People with a higher need level for Anti-Social Thinking had less participation in work/programming and completed fewer programs. Individuals with a greater need for SUD treatment were more likely to participate in education classes and earn education awards but less likely to participate in work/programming.

Conclusion

The results from this study illustrate the potential value of using program measures to validate responsivity assessments, although the extent to which they are generalizable to other correctional populations is unclear given that our sample consisted of individuals from one state’s prison system. Nevertheless, the findings hold several implications for correctional research and practice. First, because mental disorders and childhood trauma impede program participation and completion, these specific responsivity factors may need to be addressed before individuals enter programming that targets their criminogenic needs. Thus, in terms of program sequencing, individuals with a high rating in these domains would be prioritized for mental health treatment and trauma-informed programming. Although effective treatment would help mitigate the severity of mental disorders, childhood trauma is a static factor for an adult correctional population. Although trauma-informed interventions would not be able to alter the presence of adverse childhood experiences, they may be able to help individuals successfully participate in programming that targets criminogenic needs.

Second, while people scoring higher in the Mental Health and Childhood Trauma domains may not be ready for criminogenic need programming, the opposite may be true for those rated higher for Religiosity and Motivation. As expected, people are more likely to be involved in programming when they are more motivated and confident in their ability to change their lives. Moreover, religious involvement has been shown to be positively associated with optimism and hope (Ciarrochi & Heaven, 2012; Hood et al., 2009; King et al., 2020), and it has been linked with greater educational attainment (Jeynes, 2003; Regnerus, 2003). Thus, people scoring higher for Religiosity may have a greater willingness to participate in programming, especially education classes.

Third, Learning Style is believed to be important but there is not strong empirical support that it influences educational achievement. We did not find that it has a strong association with program participation and completion. As noted above, it could be that Learning Style may not be as important as originally thought. Or it could be the assessment we used does not adequately tap into preferred learning styles that matter for correctional populations.

Fourth, when using program participation and completion as the validation measures, the results can also be used to help determine whether programming aligns with the RNR model. In theory, program participation and dosage should be greater for individuals scoring high in needs domains. People with a high need for Education were more likely to be enrolled in classes and earn awards. In addition, positive associations were observed for Substance Use and Housing/Homelessness for education programming. Otherwise, the results suggest that higher-need individuals did not have increased involvement in programming. Granted, the MnDOC piloted the self-reported assessment, and the assessment results were not used to help influence program assignment decisions. Still, the findings shown here suggest the agency may need to focus on better aligning its programming with assessed criminogenic needs.

Finally, this study may offer a template for how responsivity assessments can be validated. To be sure, some specific responsivity factors, such as gender or language, do not require the use of a validated assessment. As shown here, however, assessments designed to measure constructs such as motivation, religiosity, learning style, mental health, and childhood trauma should be subjected to validation research. We believe that program participation, dosage, and completion provide the outcome measures that should be used to validate responsivity assessments for correctional populations. Responsivity has been the overlooked principle within the RNR model, but more validation research could eventually lead to a better understanding of this principle and, ultimately, improved program outcomes.

Footnotes

Appendix

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Grant Duwe

Notes

Author Biographies

Grant Duwe is the Director of Research and Evaluation for the Minnesota Department of Corrections, where he evaluates correctional programs, develops assessment instruments, and forecasts the state’s prison population. He is the author of two books and has more than 90 peer-reviewed publications on a wide variety of topics within corrections.

Valerie Clark is the Research and Evaluation Supervisor at the Minnesota Department of Corrections. She completed her PhD at the Pennsylvania State University. Dr. Clark’s research has focused on prisoner reentry, prison programing, and victimization.

References

Almigdal

T. H.

(2015). Relationship between the learning style preferences of medical students and academic achievement. Saudi Medical Journal, 36(3), 349–355. https://doi.org/10.15537/smj.2015.3.10320

Andrews

D. A.

Bonta

Wormith

S. J.

(2006). The recent past and near future of risk and/or need assessment. Crime & Delinquency, 52(1), 7–27. https://doi.org/10.1177%2F0011128705281756

Bonta

Andrews

D. A.

(2017). The psychology of criminal conduct. Routledge.

Ciarrochi

Heaven

P. C. L.

(2012). Religious values and the development of trait hope and self-esteem in adolescents. Journal for the Scientific Study of Religion, 51, 676–688.

Clark

V. A.

Duwe

(2023). Evaluating the convergent and discriminate validity of a needs and responsivity assessment system [Manuscript under review]. Minnesota Department of Corrections.

Cullen

F. T.

(2002). Rehabilitation and treatment programs. In Wilson

J. Q.

Petersilia

(Eds.), Crime: Public policies for crime control (2nd ed., pp. 235–289). ICS Press.

Dobson

J. L.

(2010). A comparison between learning style preferences and sex, status, and course performance. Advances in Physiology Education, 34, 197–204.

Duwe

(2021). Evaluating bias, shrinkage, and the home-field advantage: Results from a revalidation of the MnSTARR 2.0. Corrections: Policy, Practice and Research, 9, 20–42. https://doi.org/10.1080/23774657.2021.2011802

Duwe

Clark

V. A.

(2023). The concurrent and predictive validity of a needs and responsivity assessment system. Corrections: Policy, Practice and Research. Advance online publication. https://doi.org/10.1080/23774657.2023.2298807

10.

Duwe

Johnson

B. R.

(2023). New insights for “what works”: Religiosity and the Risk-needs-responsivity model. Crime & Delinquency. Advance online publication. https://doi.org/10.1177/00111287231160

11.

Duwe

Rocque

(2017). The effects of automating recidivism risk assessment on reliability, predictive validity, and return on investment (ROI). Criminology & Public Policy, 16, 235–269.

12.

Felitti

V. J.

Anda

R. F.

Nordenberg

Williamson

D. F.

Spitz

A. M.

Edwards

Marks

J. S.

(1998). Relationship of childhood abuse and household dysfunction to many of the leading causes of death in adults: The Adverse Childhood Experiences (ACE) Study. American Journal of Preventive Medicine, 14(4), 245–258.

13.

Fleming

Mills

(1992). Not another inventory, rather a catalyst for reflection. To Improve the Academy, 11, 137–155.

14.

Gendreau

Little

Goggin

(1996). A meta-analysis of the predictors of adult offender recidivism: What works!. Criminology, 34, 575–607.

15.

Giordano

P. C.

Cernkovich

S. A.

Rudolph

J. L.

(2002). Gender, crime, and desistance: Toward a theory of cognitive transformation. American Journal of Sociology, 107(4), 990–1064.

16.

Hood

R. W.

Hill

P. C.

Spilka

(2009). The psychology of religion: An empirical approach (4th ed.). Guilford Press.

17.

Jang

S. J.

Johnson

B. R.

Hays

Hallett

Duwe

(2017). Religion and misconduct in “Angola” prison: Conversion, congregational participation, religiosity, and self-identities. Justice Quarterly, 35, 412–442. https://doi.org/10.1080/07418825.2017.1309057

18.

Jeynes

W. H.

(2003). The effects of religious commitment on the academic achievement of urban and other children. Education and Urban Society, 36(1), 44–62. https://doi.org/10.1177/0013124503257206

19.

King

P. E.

Vaughn

J. M.

Yoo

Tirrell

J. M.

Dowling

E. M.

Lerner

R. M.

Geldof

G. J.

Lerner

J. V.

Iraheta

Williams

Sim

A. T. R.

(2020). Exploring religiousness and hope: Examining the roles of spirituality and social connections among Salvadoran Youth. Religions, 11, 1–19. https://doi.org/10.3390/rel11020075

20.

Knight

Garner

B. R.

Simpson

D. D.

Morey

J. T.

Flynn

P. M.

(2006). An assessment for criminal thinking. Crime & Delinquency, 52(1), 159–177. https://doi.org/10.1177%2F0011128705281749

21.

Krushas

Hamilton

Kigerl

Mei

(2024). Finding an answer in time: Assessing change in needs scores on time to recidivism among justice-involved youth. Journal of Criminal Justice, 90, 102146. https://doi.org/10.1016/j.jcrimjus.2023.102146

22.

Leite

W. L.

Svinicki

Shi

(2010). Attempted validation of the scores of the VARK: Learning styles inventory with multitrait–multimethod confirmatory factor analysis models. Educational and Psychological Measurement, 70(2), 323–339.

23.

McCormick

Peterson-Badali

Skilling

T. A.

(2017). The role of mental health and specific responsivity in juvenile justice rehabilitation. Law and Human Behavior, 41, 55–67.

24.

Mills

J. F.

Kroner

D. G.

Forth

A. E.

(2002). Measures of Criminal Attitudes and Associates (MCAA): Development, factor structure, reliability and validity. Assessment, 9, 240–253.

25.

Mitchell

Wilson

D. B.

MacKenzie

D. L.

(2007). Does incarceration-based drug treatment reduce recidivism? A meta-analytic synthesis of the literature. The Journal of Experimental Criminology, 3, 353–375.

26.

Mlambo

(2011). An analysis of some factors affecting student academic performance in an introductory biochemistry course at the University of the West Indies. Caribbean Teaching Scholar, 1(2), 75–88.

27.

Mowen

T. J.

Stansfield

Boman

J. H.

(2018). During, after or both? Isolating the effect of religious support on recidivism during reentry. Journal of Quantitative Criminology, 34, 1079–1101.

28.

Mozaffari

H. R.

Janatolmakan

Sharifi

Ghandinejad

Andayeshgar

Khatony

(2020). The relationship between the VARK learning styles and academic achievement in dental students. Advances in Medical Education and Practice, 11, 15–19.

29.

Paternoster

Bushway

(2009). Desistance and the “feared self”: Toward an identity theory of criminal desistance. The Journal of Criminal Law and Criminology, 99, 1103–1156.

30.

Pinals

D. A.

Gaba

Shaffer

P. M.

Andre

M. A.

Smelson

D. A.

(2021). Risk-need-responsivity and its application in behavioral health settings: A feasibility study of a treatment planning support tool. Behavioral Science and the Law, 39, 44–64.

31.

Regnerus

M. D.

(2003). The influence of religion upon the academic performance of youth in disadvantaged communities. Center for Research on Religion and Urban Civil Society, University of Pennsylvania.

32.

Sachs

N. M.

Miller

(2018). Beyond responsivity: Client service engagement in a reentry demonstration program. International Journal of Offender Therapy and Comparative Criminology, 62(13), 4295–4313.

33.

Taxman

F. S.

Giuranna

R. A.

Dumenci

(2011). Construct and predictive validity of criminal thinking scales. Criminal Justice and Behavior, 38(2), 174–187. https://doi.org/10.1177%2F0093854810389550

34.

Zimet

G. D.

Dahlem

N. W.

Zimet

S. G.

Farley

G. K.

(1988). The multidimensional scale of perceived social support. Journal of Personality Assessment, 52(1), 30–41.