Examining the Reliability and Validity of the Criminal Sentiments Scale–Modified and the Pride in Delinquency Scale in Black and White Canadian Justice System-Involved Youth

Abstract

Procriminal attitudes, key recidivism predictors in the Risk–Need–Responsivity framework, differ among Black and White individuals. However, the procriminal attitudes domain and associated measures have not been critically examined based on race. We examined psychometric properties of the Criminal Sentiments Scale–Modified (CSS-M) and the Pride in Delinquency Scale (PIDS) in 473 justice system-involved Black and White youth. Both scales showed acceptable internal reliability, construct validity, and convergent validity across groups. Procriminal attitudes aligned with antisocial behavior and substance use for White youth, but this was less pronounced for Black youth. The five-factor CSS-M predicted recidivism for White youth, and the PIDS improved predictions. For Black youth, the CSS-M Total and Police subscale, but not the PIDS, predicted recidivism. Results suggest caution in the use of these tools with Black youth. Future research should investigate the role of procriminal attitudes in offending for marginalized groups.

Keywords

attitudes race forensic assessment Risk–Need–Responsivity risk assessment youth validity criminal thinking

The Risk–Need–Responsivity framework (RNR; Bonta & Andrews, 2023) holds that eight criminogenic needs strongly and directly predict reoffending, and these should therefore be assessed to understand recidivism risk, inform sentencing, and guide rehabilitation plans for youth in the criminal justice system. Professionals use multiple information sources (e.g., interviews, questionnaires, standardized measures) to evaluate the relevance of each criminogenic need for a given youth—youth who present with higher needs (in number and severity) are rated as higher risk to reoffend.

Antisocial/procriminal attitudes, as captured in RNR-based risk assessment tools, have been upheld as robust predictors of reoffending (Olver et al., 2014). Bonta and Andrews (2023) describe antisocial/procriminal attitudes as “thoughts, feelings, and beliefs that are supportive of criminal conduct.” They group attitudes in three categories: (1) techniques of neutralization (justifications for crime); (2) identification with criminal others (adoption of a criminal identity); and (3) rejection of convention (devaluation of education, work, and legal institutions). The breadth of constructs captured in this domain has been critiqued (Costaris et al., 2022), as “procriminal attitudes” encompass attitudes toward help seeking/acceptance, authority, concern for others, law, crime, the criminal justice system, and people who commit crime (Bonta and Andrews, 2023). Each of the abovementioned attitudes may have different distal (historical and societal) or proximal (direct life experience, personality) origins rooted in broader perspectives and experiences that vary among ethnoracial groups (e.g., Harris & Jones, 2020; Samuels-Wortley, 2022). For example, Black people are less likely than White people to report confidence in the criminal justice system and are more likely to report justice system-based discrimination (e.g., Department of Justice Canada, 2022). Racial discrimination has been linked to lower trust in the justice system both quantitatively and qualitatively (Fine & van Rooij, 2021; Harris & Jones, 2020; Jackson et al., 2023; Samuels-Wortley, 2021). The racial gap in perceptions of the justice system has persisted over decades, and it is growing more profound (Fine et al., 2020).

Study results indicate good psychometric properties for two procriminal attitudes measures used in forensic services and in research (Department of Justice Canada, 2021): the Criminal Sentiments Scale–Modified (CSS-M; Simourd, 1997) and the Pride in Delinquency Scale (PIDS; Shields & Whitehall, 1994). However, there is a dearth of research examining their utility in different ethnoracial groups. Using tools that are not studied at the subgroup level poses a risk for inaccurate theory development and invalid recidivism risk assessment (e.g., Olver et al., 2024; Skeem & Lowenkamp, 2016). Thus, establishing that the construct of procriminal attitudes and the tools used to measure it have acceptable reliability and validity for specific groups is an urgent priority. This is especially critical for groups who are overrepresented in the criminal justice system and have higher rates of procriminal attitudes tied to systemic discrimination. To begin to address this need, the purpose of the present study was to examine the reliability and validity of the CSS-M and PIDS in Black and White justice system-involved youth in Canada.

Psychometric Properties of Two Commonly Used Measures of Procriminal Attitudes

The CSS-M

The 41-item CSS-M questionnaire measures procriminal attitudes on five subscales: Attitudes toward the Law, Courts, and Police (sometimes aggregated into one “LCP” subscale), Tolerance for Law Violations (TLV), and Identification with Criminal Others (ICO) (Simourd, 1997). Items are rated on a 3-point scale, with higher scores indicating more procriminal attitudes. A one-factor CSS-M Total score and a 3-factor model are commonly used in research (Walters, 2016), but a 4-factor model has been used as well (Simourd & Olver, 2002).

In research with majority White male samples, the CSS-M has shown acceptable internal reliability, construct, convergent, and concurrent validity with measures of personality, substance use, and risk for reoffending (Andrews & Wormith, 1984; Shields & Simourd, 1991; Simourd, 1997). The CSS-M’s predictive validity for recidivism varies based on factors like offense type, operationalization of recidivism, and the CSS-M subscale in question (Simourd, 1997; Simourd & van de Ven, 1999; van der Put et al., 2020). In a meta-analysis that pooled 13 samples using the original CSS and the CSS-M (Walters, 2016), all subscales predicted reoffending and reincarceration with small effect sizes, with the CSS-M Total and the LCP subscale being the strongest predictors (pooled effect sizes of .14 and .17, respectively).

Measurement invariance analyses and structural equation modeling suggested acceptable, but differing, construct validity and item functioning of the CSS-M in a mixed-race sample of male and female young adults (Vaske et al., 2017). In a mixed-race sample of Canadian justice system-involved male youth, Skilling and Sorge (2014) reported good internal reliability for the CSS-M Total and the LCP and TLV subscales, but the Courts and ICO subscales had weaker internal reliability (α < .70). The scales showed good convergent and concurrent validity, and the total score significantly predicted new convictions (area under the ROC curve [AUC] = .69), although when regressed together with another procriminal attitudes measure (the Pride in Delinquency Scale; PIDS), only the PIDS significantly predicted reconvictions. Some researchers have begun to examine the impact of race on CSS-M scores, finding that non-White justice-involved adults had higher CSS-M scores than their White counterparts (Pauselli et al., 2024; Soyer et al., 2017). Good reliability and validity of the CSS-M were evidenced in an Urdu version of the CSS-M, which had acceptable internal and test–retest reliability and good construct validity for Pakistani male and female university students (Mazher et al., 2022). In a study of youth in Hong Kong with Chinese ethnicity, Chui and Cheng (2017) found that all CSS-M scales were positively related to self-reported previous criminal behavior, while the TLV scale was negatively correlated with criminal behavior; internal reliability was acceptable. Despite initial findings from these studies, their samples preclude generalization to justice-involved Black youth, who face some of the highest rates of discrimination and distrust of legal systems in North America (e.g., Department of Justice Canada, 2024). In addition, none of these studies examined the CSS-M’s ability to predict future reoffending based on official documentation.

The PIDS

Developed to complement the CSS-M, the 10-item PIDS assesses, on a −10 to +10 scale, how proud a respondent would be to commit various crimes (Shields & Whitehall, 1994). Item scores are summed and added to a constant of 100, with higher scores indicating more procriminal attitudes. While the one-factor model is often used clinically, a two-factor model has also been supported (Simourd, 1997). The PIDS was developed on White male adolescents and subsequently validated alongside the CSS-M in mixed-race and mixed-gender adult samples, showing good reliability, convergent validity, and concurrent validity with respect to aggression, peers, personality, externalizing problems, and substance use challenges (Shields & Whitehall, 1994; Simourd, 1997; Simourd & van de Ven, 1999; Skilling & Sorge, 2014). The PIDS also showed good predictive validity for new convictions (AUC = .59–.70; O’Hagan et al., 2019; Skilling & Sorge, 2014), rearrest (r = .23), and reincarceration (r = .27) (Simourd & van de Ven, 1999).

The Current Study

While psychometric studies of the CSS-M and the PIDS have progressed from using primarily White male samples to include mixed gender and race samples, there is no research that examines the psychometric properties of these scales separately in ethnoracial subgroups of justice system-involved youth. Black youth are among the most disproportionately impacted by systemic racism and resulting distrust in North America’s criminal justice systems (Department of Justice Canada, 2024). Procriminal attitudes scales must be examined to ensure that their use with Black youth is evidence-based. In the current study, we examined the internal reliability and construct, convergent, concurrent, and predictive validity of the CSS-M and PIDS separately in Black and White youth serving probation terms in the Greater Toronto Area, Canada.

Given initial findings suggesting that the CSS-M and PIDS have acceptable psychometric properties in samples including diverse racial groups (Chui & Cheng, 2017 ; Mazher et al., 2022; Skilling & Sorge, 2014), as well as previous research suggesting their efficacy with White youth (e.g., Shields & Whitehall, 1994; Simourd, 1997), we hypothesized that the CSS-M and PIDS would show good internal reliability, construct validity, and convergent validity for Black and White youth. We predicted that both measures would be related to constructs associated with procriminal attitudes in previous research, including externalizing behaviors (Helmond et al., 2014), aggression (Tangney et al., 2012), and substance use (Timko et al., 2017). We did not predict whether reliability and validity would be stronger for either group. We hypothesized that both scales would predict reoffending (Skilling & Sorge, 2014; Walters, 2016), but that the PIDS would predict better than the CSS-M (Simourd, 1997; Skilling & Sorge, 2014).

Method

Participants

The clinic where the data were collected had a database of 2,065 cases at the time of the study. Of these cases, 1,065 were either Black (n = 612) or White (n = 453). Research consent data were missing for 4.1% (n = 25) of Black youth and 4.0% (n = 18) of White youth; those youth were not included in the sample. Consent to use assessment data for research was obtained from 803 Black and White youth. White youth consented at a slightly higher rate (81.8%; n = 356) than Black youth (76.2%; n = 447), χ²(1, N = 1,022) = 4.80, p = .03. There were no differences in consent rate based on gender χ²(1, N = 1,021) = .05, p = .83, recidivism χ²(1, N = 520) = 1.11, p = .29, age (r = −.02, p = .50), or procriminal attitudes scores, PIDS: r (767) = .00, p = .92; CSS-M: r (375) = −.02, p = .64. Of the 803 consenting Black and White youth, 473 had PIDS and/or CSS-M data available.

The final sample included 473 justice system-involved youths (n = 410 young men and n = 63 young women; n = 247 Black, n = 226 White) from Ontario, Canada, who underwent court-ordered psychological assessments during their court proceedings between 2001 and 2014 at a mental health center in Toronto, Canada. Mean age at time of assessment was 16.39 years (SD = 1.56, range = 12–24; see Table 1 for detailed demographic information). There were transitional age youth in the sample; 87 (18.4%) were 18 years old, 17 (3.6%) were 19 years old, and 5 (1.1%) were aged 20–24 when assessed. Youths aged 18 and older were prosecuted as youth because they incurred charges when they were under 18. Given this and evidence supporting the use of youth risk assessment measures in emerging adults (e.g., Vincent et al., 2019), these youth were kept in the sample to better reflect clinic demographics.

Table 1:

Sample Demographic and Background Characteristics by Gender and Race

Characteristic	Overall	Black	White	Statistic	p	Effect size	Young men	Young women	Statistic	p	Effect size
% Gender (n)
Young men	86.7 (410)	90.7 (224)	82.3 (186)	χ² = 7.19	.007	φ = .12
Young women	13.3 (63)	9.3 (23)	17.7 (40)	χ² = 7.19	.007	φ = .12
% Index offense (n)
Nonviolent	26.8 (125)	21.6 (53)	32.4 (72)	χ² = 6.57	.01	φ = .12	27 (107)	29 (18)	χ² = .17	.68	φ = .02
Violent	58.2 (272)	63.7 (156)	52.3 (116)	χ² = 6.76	.009	φ = -.12	57 (231)	65 (41)	χ² = 1.71	.19	φ = .06
Sexual	15.0 (70)	14.7 (36)	15.3 (34)	χ² = .02	.89	φ = .007	16 (66)	6 (4)	χ² = 4.12	.04	φ = -.09
% Reoffended (n)	54.5 (258)	61.5 (152)	46.9 (87)	χ² = 10.20	.001	φ = -.15	56 (231)	43 (27)	χ² = 4.01	.04	φ = -.09
Mean age in years (SD)	16.39 (1.56)	15.61 (1.51)	15.51 (1.45)	t (470) = 1.89	.06	d = .07	16.39 (1.60)	16.33 (1.36)	t (470) = .29	.75	d = .04

We excluded cases for individual analyses if they were missing more than 20% of items on a relevant questionnaire, or if they exhibited responses that showed straight lining or inconsistency (e.g., responding “no” to all items, including reverse-coded items) (see Supplemental Table 1 for the number of valid cases for each measure). There was a 30.2% sample overlap (143 cases) in the present study and Skilling and Sorge (2014). Research ethics board approval was obtained in the context of a large-scale project under which this project took place.

Procedure

Under Canada’s youth justice legislation, youth court judges may order assessments for youth who are suspected to have a mental illness, a learning or intellectual disability, have repeated findings of guilt, or who have allegedly committed a serious violent offense (Youth Criminal Justice Act, 2002, s. 34); a very small proportion of the overall youth justice population receive such assessments (e.g., Jack & Ogloff, 1997). Most youth in our sample resided in the community at the time of their assessment, and most were assessed just prior to sentencing. Trained clinicians conducted assessments, which comprised interviews with youth and caregivers, validated questionnaires, standardized testing, and reports from third parties. Using these data, clinicians interpreted domain-level and overall recidivism risk scores to provide the court with a report describing risk level, criminogenic and mental health needs, and rehabilitation recommendations.

Variables and Measures

The Adolescent Alcohol Involvement Scale

The 14-item Adolescent Alcohol Involvement Scale (AAIS; Mayer & Filstead, 1979) measures adolescent alcohol use; higher scores indicate more risky alcohol use. It has good reliability and validity (Mayer & Filstead, 1979; McKay & Dempster, 2016). Its internal reliability in the current study was good (α = .80).

The Drug Abuse Screening Test for Adolescents

We measured drug use with the Drug Abuse Screening Test for Adolescents (DAST-A; Martino et al., 2000), a 27-item self-report questionnaire. Items are summed into a total score, where higher scores indicate more drug use challenges. The DAST-A has shown good reliability and validity (Martino et al., 2000; Mogadam et al., 2024). It had good internal reliability in the current study (α = .91).

The Aggression Questionnaire

We measured aggression with the Aggression Questionnaire (AQ; Buss & Warren, 2000), a 34-item self-report questionnaire about youth’s aggression, anger, and hostility. Youth rate, on a scale from 1 (“not at all like me”) to 5 (“completely like me”), whether a characteristic describes them. Total scores were used in the current study. The AQ has shown good reliability and validity (Buss & Warren, 2000). The internal reliability of the AQ in the current study was good (α = .95).

The Youth Self-Report—Externalizing Problems

The Youth Self-Report (YSR; Achenbach, 2013) is a 112-item self-report measure of youth emotion and behavior. Items are rated on a 3-point scale from “not true” to “very true”; higher scores indicate more challenges. The Externalizing Problems scale, comprised of rule-breaking and aggression, was used. The YSR’s reliability and validity are well established (e.g., Achenbach, 2013; Achenbach & Rescorla, 2001), and its internal reliability in the current study was good (α = .92).

Recidivism

Recidivism data were obtained from a national police criminal records database. Youth were considered to have reoffended if they had any youth or adult reconviction within 3 years from the assessment date. As few youth (<10%) in the Canadian youth justice system receive custodial sentences or are detained during the court process (Department of Justice Canada, 2024), the opportunity to reoffend extends from the time of assessment.

Analytic Plan

To examine the construct validity of the CSS-M and PIDS, we conducted a Confirmatory Factor Analysis (CFA). As the most commonly examined model, we examined the one-factor model of the PIDS. For the CSS-M, we examined both the one-factor (i.e., total score) model (which is concise and comprehensive) and the five-factor model (which allows a more nuanced examination of attitude types). We assessed goodness of fit using the Comparative Fit Index (CFI; Bentler, 1990), the Tucker–Lewis coefficient (TLI; Tucker & Lewis, 1973), and the root mean square error of approximation (RMSEA; Steiger & Lind, 1980). All structural models were tested using the weighted least squares mean and variance adjusted (WLSMV) estimator, as the CSS-M items are rated on a 3-point scale and are thus categorical (Flora & Curran, 2004). CFAs for the PIDS were computed using the Satorra–Bentler chi-square (MLM Estimator) as item responses were not normally distributed (Satorra & Bentler, 2010). CFA assumptions were evaluated for all items.

Internal consistency was assessed with Cronbach’s alpha and McDonald’s omega. The latter was included as it performs more accurately than Cronbach’s alpha when dealing with scales comprised of few items (Hayes & Coutts, 2020), and it relies on fewer assumptions (Dunn et al., 2013; Kalkbrenner, 2024). Values > 0.7 were acceptable (Nunnally & Bernstein, 1994). Mean inter-item correlations (MICs) indicated whether items measured the same constructs without being redundant, with .2 < r < .4 being acceptable (Piedmont, 2014). Mean corrected item-total correlations (MCITCs) allowed assessment of whether items represented the construct (indicated by the total score), with r > .3 being acceptable (Nunnally & Bernstein, 1994).

We examined correlations between the CSS-M and PIDS to assess convergent validity. To assess which CSS-M factor model had better convergent validity, we compared correlation coefficients using Fisher’s r to z transformations with Steiger’s equations to make the transformations suitable for a dependent sample (Fisher, 1921; Steiger, 1980). We assessed concurrent validity by correlating the CSS-M and PIDS with YSR Externalizing, AQ, AAIS, and DAST-A scores. To examine whether the strength of relationships differed between Black and White youth, we conducted Fisher’s r to z transformations (Fisher, 1915).

We assessed predictive validity through AUCs for the PIDS, CSS-M Total, and each CSS-M subscale score (> .556 = small effect, > .639 = medium effect, and > .714 = large effect; Rice & Harris, 2005). To examine the developers’ intention that the PIDS provides additional predictive power to the CSS-M, we conducted hierarchical logistic regressions, with the CSS-M (total scores and subscales) at Step 1 and the PIDS entered at Step 2. We assessed goodness of fit using the Hosmer and Lemeshow test (2000). The assumption of linearity of the logit was met if the interaction term between each variable and the natural log of itself was non-significant (Field, 2017). The assumption of no multicollinearity was met with Tolerance values ≥ 1 and variance inflation factor (VIF) values <10 for the combined sample and Black and White groups.

We used MPlus Version 8.10 (Muthén & Muthén, 2017) to compute CFAs, the Quantpsy Calculator (Lee & Preacher, 2013) for convergent validity comparisons, and the Significance of the Difference between Two Correlations Calculator (Soper, 2025) for concurrent validity comparisons. We used SPSS Version 27 (IBM Corp, 2021) for all other analyses. We bootstrapped analyses to produce robust estimates in score differences and used Spearman’s rho as a correlation coefficient due to non-normal distributions and outliers for many variables. Outliers were examined to ensure they were legitimate data points.

Results

Preliminary Analyses

In this sample of predominantly young men, there was a higher proportion of young women in the White subgroup than the Black subgroup (Table 1). Young men were more likely to have sexual index offenses—and reoffended at higher rates—than young women. Black youth reoffended at higher rates than White youth, were more likely to have violent index offenses, and less likely to have nonviolent index offenses (Table 1). There were few race or gender differences in CSS-M and PIDS scores (see Table 2 in Supplemental Online Material), but Black youth scored higher than White youth, and young women scored higher than young men, on the CSS-M Police subscale. The low number of young women meant that analyses could not be conducted separately by gender. As there were no gender differences in the measures’ AUC scores (.20 ≤ p ≤ .95; Supplemental Table 3), gender was not included as a covariate in the regression analyses.

CFAs

The one- and five-factor models of the CSS-M showed acceptable fit for all subgroups (CFIs and TLIs between .84 and .94, RMSEAs of .06 and .05; Supplemental Table 4). The PIDS Item 2 (“committing sexual assault”) was removed as 85% of participants rated it “−10.” The one-factor model fit the combined sample and the Black subgroup (CFIs and TLIs above .9, RMSEAs of .08 and .07) but did not fit the White subgroup (CFI = .94, TLI = .92, RMSEA = .099). The largest modification index (MI = 21.84) suggested correlating the error variance between Item 7 (“selling drugs”) and Item 8 (“carrying a concealed weapon without a license”), which resulted in a better fit (CFI = .96, TLI = .94, and RMSEA = .09; Supplemental Table 4).

Internal Reliability

As Table 2 shows, the CSS-M Law, Police, and TLV subscales showed acceptable alpha and omega reliability for the combined sample and both subgroups; coefficients for the Courts and ICO subscales were below acceptable for Black youth and the combined sample. All MICs fell within the acceptable range, except for the CSS-M Courts and ICO scales for Black youth, where MICs were low. All MCITCs were acceptable. The PIDS showed good internal reliability (acceptable MCITCs and MICs) for all groups.

Table 2:

Internal Reliability Statistics for the CSS-M and PIDS.

Measure	MIC (r)			MCITC (r)			Cronbach’s alpha (α)			McDonald’s omega (ω)
Measure	Black	White	Total	Black	White	Total	Black	White	Total	Black	White	Total
CSS-M Law	.26	.29	.28	.45	.48	.47	.77	.80	.79	.79	.81	.80
CSS-M Courts	.18	.25	.21	.33	.42	.37	.63	.73	.68	.57	.72	.65
CSS-M Police	.37	.44	.41	.54	.60	.58	.80	.85	.83	.81	.85	.83
CSS-M TLV	.24	.30	.27	.43	.49	.46	.76	.81	.79	.77	.81	.79
CSS-M ICO	.18	.26	.22	.31	.41	.36	.55	.68	.62	.58	.70	.65
CSS-M Total	.20	.25	.22	.42	.48	.46	.91	.93	.92	.91	.93	.92
PIDS	.35	.43	.39	.55	.62	.58	.84	.87	.85	.86	.89	.88

Note. Bolded items represent those that do not meet the acceptable internal consistency threshold. CSS-M = Criminal Sentiments Scale–Modified; TLV = Tolerance for Law Violations subscale; ICO = Identification with Criminal Others subscale; PIDS = Pride in Delinquency Scale.

Convergent Validity

For both Black and White youth, CSS-M subscales and the Total score showed mostly small to moderate significant correlations with the PIDS (Table 3). In both groups, the CSS-M Total showed a significantly smaller relationship with the PIDS compared to that of the Law subscale: Black, z (158) = 4.04, p < .001; White, z (157) = 5.43 p < .001, Police subscale: Black, z (158) = 2.60, p = .009; White, z (157) = −3.75, p < .001, TLV subscale: Black: z (158) = 3.52, p < .001; White: z (157) = 4.05, p < .001, and ICO subscale: Black: z (158) = 2.38, p = .02; White: z (155) = 2.22, p = .03. The CSS-M Total was also less strongly correlated to the PIDS compared to the Courts subscale for White youth, z (157) = 3.02, p = .003.

Table 3:

Correlations Between Measures for Black and White Youth

Note. CSS-M = Criminal Sentiments Scale–Modified; TLV = Tolerance for Law Violations; ICO = Identification with Criminal Others; PIDS = Pride in Delinquency Scale; AQ = Aggression Questionnaire; YSR Ext. = Youth Self Report Externalizing Scale; AAIS = Adolescent Alcohol Involvement Scale; DAST-A = Drug Abuse Screening Test for Adolescents. Shaded correlations = Black youth, unshaded = White youth. Concurrent validity correlations that were stronger for White youth are bolded: Courts-DAST-A (z = -3.6, p < .001); Police-AAIS (z = -2.43, p = .02); Police-DAST-A (z = -2.21, p = .03); Police-AQ (z = -2.54, p = .01); Police-YSR Ext. (z = -2.50, p = .01); TLV-AQ (z = -2.03, p = .04); PIDS-DAST-A (z = -2.28, p = .02); and PIDS-YSR Ext. (z = -3.57, p < .001).

Concurrent Validity

As Table 3 shows, for Black youth, the CSS-M Law, Police, TLV, and ICO subscales showed small to moderate correlations with at least one aspect of substance use (drug use, alcohol use, or both) and at least one aspect of antisocial behavior (aggression, externalizing problems, or both). The Courts scale and CSS-M Total showed small correlations with aggression. A more consistent, and at times stronger, pattern of relationships emerged for White youth (Table 3); all CSS-M subscales and the CSS-M Total were moderately to strongly related to at least one aspect of both substance use and antisocial behavior, with all subscales except Courts significantly related to all four behaviors. The Courts subscale was more strongly related to drug use, TLV was more strongly related to aggression, and the Police subscale was more strongly related to all four behaviors for White versus Black youth. In Black youth, the PIDS showed small to moderate correlations with alcohol and drug use, aggression, and externalizing, with the PIDS-externalizing and PIDS-drug use correlations being stronger for White youth.

Predictive Validity

Regression and AUC results are presented in Tables 4 and 5, respectively. For all models, the Hosmer and Lemeshow (HL) test was not significant, indicating good model fit. Interaction terms between variables and their natural logs were non-significant, indicating the assumption of linearity of the logit was met. Multicollinearity was shown not to be an issue for any group (Tolerance ≥ .32, VIF ≤ 3.09). Prior to disaggregating the data by race, we examined relationships for the combined sample; the overall model of the CSS-M split up into its five subscales was predictive of recidivism, but only the Police subscale emerged as a significant predictor (p = .02). Adding the PIDS improved the model fit, χ²(1, N = 309) = 5.33, p = .02, and overall classification accuracy increased from 57.6% to 60.8%. When the CSS-M subscales and the PIDS were used together, the Police subscale and the PIDS positively predicted recidivism. Similarly, the CSS-M Total positively predicted recidivism alone, but adding the PIDS improved the model, χ²(1, N = 313) = 8.39, p = .004, with overall classification accuracy increasing from 57.2% to 62.0%. In fact, when the PIDS was added to the model, the CSS-M no longer predicted recidivism. The AUCs, which measure the predictive ability of each subscale independently, showed slightly different results; the PIDS, CSS-M Total, and all CSS-M subscales except ICO predicted recidivism independently in the combined sample.

Table 4:

Hierarchical Logistic Regression Results for the PIDS and CSS-M

Model	Variable	B	SE	p	95% CI
Combined
Model 1a	χ²(5, n = 309) = 14.53, p = .013, R² = .061.
	CSS-M Law	–.001	.04	.99	[–.08, .08]
	CSS-M Courts	.01	.05	.80	[–.09, 0.1]
	CSS-M Police	.10	.05	.02	[.02, .21]
	CSS-M TLV	.02	.04	.53	[–.06, .11]
	CSS-M ICO	–.04	.07	.57	[–.17, .09]
Model 1b	χ²(6, n = 309) = 19.86, p = .003, R² = .08.
	CSS-M Law	–.02	.05	.62	[–.11, .06]
	CSS-M Courts	.00	.05	.93	[–.10, .10]
	CSS-M Police	.10	.05	.02	[–.02, .20]
	CSS-M TLV	–.00	.05	.92	[–.09, .08]
	CSS-M ICO	–.07	.07	.29	[–.22, .05]
	PIDS	.01	.01	.02	[.00, .02]
Model 2a	χ²(1, n = 313) = 7.93, p = .005, R² = .03.
	CSS-M Total	0.05	.02	.002	[.02, .08]
Model 2b	χ²(2, n = 313) = 16.33, p < .001, R² = .07.
	CSS-M Total	.03	.02	.06	[–.00, .07]
	PIDS	.01	.00	.007	[.00, .02]
Black
Model 3a	χ²(5, n = 156) = 9.60, p =.09, R² = .08.
	CSS-M Law	.02	.07	.81	[–.11, .14]
	CSS-M Courts	.04	.07	.49	[–.10, .17]
	CSS-M Police	.15	.07	.03	[.02, .31]
	CSS-M TLV	–.08	.06	.18	[–.22, .04]
	CSS-M ICO	–.05	.11	.63	[–.27, .15]
Model 3b	χ²(6, n = 156) = 10.75, p = .10, R² = .09.
	CSS-M Law	.00	.07	.96	[–.13, .13]
	CSS-M Courts	.04	.07	.51	[–.11, .17]
	CSS-M Police	.15	.07	.03	[.02, .31]
	CSS-M TLV	–.10	.07	.13	[–.25, .03]
	CSS-M ICO	–.07	.11	.52	[–.29, .14]
	PIDS	.01	.01	.30	[–.01, .02]
Model 4a	χ²(1, n = 157) = 8.17, p = .004, R² = .07.
	CSS-M Total	.07	.03	.005	[.03, .13]
Model 4b	χ²(2, n = 157) = 8.74, p = .01, R² = .07.
	CSS-M Total	.07	.03	.01	[.02, .12]
	PIDS	.00	.01	.41	[–.01, .01]
White
Model 5a	χ²(5, n = 153) = 11.73, p = .04, R² = .10.
	CSS-M Law	–.03	.07	.59	[–.17, .10]
	CSS-M Courts	–.00	.07	.91	[–.15, .13]
	CSS-M Police	.04	.07	.55	[–.11, .19]
	CSS-M TLV	.13	.07	.03	[–.02, .29]
	CSS-M ICO	–.02	.10	.86	[–.21, .17]
Model 5b	χ²(6, n = 153) = 16.77, p = .01, R² = .14.
	CSS-M Law	–.07	.07	.30	[–.22, .07]
	CSS-M Courts	–.03	.07	.73	[–.18, .12]
	CSS-M Police	.04	.07	.54	[–.10, .19]
	CSS-M TLV	.09	.07	.17	[–.03, .25]
	CSS-M ICO	–.09	.11	.39	[–.32, .11]
	PIDS	.02	.01	.03	[.00, .04]
Model 6a	χ²(1, n =156) = .70, p = .40, R² = .01.
	CSS-M Total	.02	.02	.40	[–.03, .07]
Model 6b	χ²(2, n = 156) = 12.24, p = .002, R² = .10.
	CSS-M Total	–.01	.03	.75	[–.07, .04]
	PIDS	.02	.01	.003	[.01, .03]

Note. CSS-M = Criminal Sentiments Scale–Modified; TLV = Tolerance for Law Violations subscale; ICO = Identification with Criminal Others subscale; PIDS = Pride in Delinquency Scale.

Table 5:

ROC AUC Scores for Relationship between PID, CSS-M, and Recidivism

	Black	White	Combined	Difference (p)
Measure	AUC
PIDS Total	.51, p =.89	.66, p < .001	.59, p =.001	–3.04 (.002)
CSS-M Total	.62, p =.01	.55, p =.34	.59, p =.007	1.11 (.27)
CSS-M Law	.57, p =.14	.60, p =.02	.59, p =.004	.59 (.56)
CSS-M Courts	.55, p =.31	.60, p =.03	.58, p =.02	.87 (.38)
CSS-M Police	.61, p =.01	.61, p =.01	.62, p < .001	–.04 (.97)
CSS-M TLV	.49, p =.84	.66, p < .001	.58, p =.01	–2.7 (.007)
CSS-M ICO	.50, p =.93	.60, p=.03	.55, p =.10	–1.5 (.13)

Note. CSS-M = Criminal Sentiments Scale–Modified; TLV = Tolerance for Law Violations subscale; ICO = Identification with Criminal Others subscale; PIDS = Pride in Delinquency Scale.

Disaggregating the data by race revealed different relationships between the measures and recidivism. For Black youth, the CSS-M Total score, but not the five-factor model, predicted recidivism. Adding the PIDS did not improve the CSS-M five-factor model, χ²(1, N = 156) = 1.15, p = .28, or the one-factor model, χ²(1, N = 157) = .57, p = .45. Apart from the AUC finding that attitudes toward Police (but no other CSS-M subscales) predicted recidivism, regression and AUC results aligned; the CSS-M Total predicted recidivism, but the five-factor CSS-M and the PIDS did not.

For White youth, the five-factor regression model did predict recidivism, but only the TLV subscale was a significant predictor. When the PIDS was added to the model, predictive ability improved significantly, χ²(1, N = 153) = 5.04, p = .03, increasing classification accuracy from 62.7% to 64.7%, with the PIDS (p = .03), but not the CSS-M subscales (ps ≥ .17), emerging as a significant predictor of recidivism. The CSS-M Total was not a significant predictor of recidivism, in line with the AUC results (AUC = .55, p =.34), but adding the PIDS to the model improved its predictive ability, χ²(1, N = 156) = 11.55, p < .001, leaving the PIDS as the only significant predictor in the model and increasing classification accuracy from 54.5% to 60.3% (Table 4). The AUC results differed from the regression results in that all CSS-M subscales and the PIDS were found to predict recidivism when used independently (Table 5).

Discussion

In the present study, we examined the reliability and validity of two measures of procriminal attitudes—the CSS-M and the PIDS—in Black and White Canadian justice system-involved youth. Overall, the factor structure and internal reliability of the CSS-M and the PIDS were acceptable in both groups, with some concerns around the reliability of the CSS-M Courts and ICO subscales for Black youth, which is consistent with Pauselli et al.’s (2024) findings in a majority Black sample of adults. While low internal reliability of the ICO subscale has previously been explained as resulting from the small number of items (Forero, 2014; Mazher et al., 2022), McDonald’s omega accounts for this (Hayes & Coutts, 2020) and results were still poor for Black youth, suggesting that items in this scale may not measure a unified construct.

Scores on the CSS-M and PIDS Hold Different Meaning for Black and White Youth

As hypothesized, the CSS-M and PIDS were correlated, showing good convergent validity. However, when subscales were considered together in the form of the Total score, the procriminal attitudes measured in the CSS-M aligned less with the procriminal attitudes measured in the PIDS than when the subscales were examined individually. Particular items/subscales may have differentially impacted the CSS-M Total, resulting in an overall weaker correlation with the PIDS. Although examining differential item functioning was beyond the scope of this study, an item like “you’re crazy to work for a living if there’s an easier way, even if it means breaking the law” might be expected to more strongly associate with pride in committing crime than an item like “the police are honest,” given the content of these attitudes (Fine et al., 2020). Although future research should analyze differential item functioning in detail, the findings suggest differences in the meaning of various types of procriminal attitudes. This interpretation lends support to the argument for examining distinct attitude types rather than a single “antisocial/procriminal attitudes” factor. Distinguishing between attitudes can provide a more accurate understanding of the attitudes-offending relationship (e.g., Costaris et al., 2022).

The importance of examining the content of procriminal attitudes aligns with the concurrent validity findings, which revealed racial differences in the way procriminal attitudes relate to antisocial behaviors. As predicted based on previous work (Skilling & Sorge, 2014; Tangney et al., 2012; Timko et al., 2017), there were significant relationships between the CSS-M, PIDS, and antisocial behaviors (substance use and aggression) for both groups. However, for White youth, there was a clearer picture of procriminal attitudes pairing with antisocial behaviors, whereas Black youth’s procriminal attitudes (and particularly their attitudes toward the police) were less consistently—and less strongly—related to antisocial behaviors than White youth’s. This finding suggests that the procriminal attitudes measured in this study may be less reflective of Black youth’s individual behavior and more reflective of widespread sentiments stemming from long-held injustices and racism (Samuels-Wortley, 2021).

The Relationship Between Attitudes Toward Police and Recidivism

As hypothesized, in the combined sample, the CSS-M one- and five-factor models predicted recidivism. The PIDS improved the predictive ability of the CSS-M to the point of rendering the CSS-M redundant when used alongside the PIDS. Similarly, in the White sample, the CSS-M five-factor model predicted recidivism, and its predictive ability was improved by the addition of the PIDS. These results are consistent with the findings in Skilling and Sorge’s (2014) mixed-race sample and suggest that the PIDS may be a better recidivism predictor than the CSS-M for White youth. Effect sizes were small, aligning with previous criminal justice research in which relationships to recidivism tend to be small to moderate (e.g., Bonta & Andrews, 2023; Walters, 2016). Together, these results reflect how procriminal attitudes are situated within a constellation of needs considered in risk assessment, and they underscore the importance of studying factors that strengthen and weaken the attitudes-offending relationship.

The only CSS-M subscale that predicted recidivism for Black youth was the most racialized: attitudes toward police. Consistent with previous research in youth justice samples and the general population (Fine et al., 2017, 2020; Skilling & Sorge, 2014), Black youth had more negative attitudes toward police than White youth. Although attitudes toward police predicted recidivism for Black youth, the finding that attitudes toward police showed significantly smaller, and sometimes nonexistent, relationships with substance use, aggression, and externalizing behavior for Black youth compared to White youth suggests that these attitudes may not reflect the same antisociality across races. Rather than these attitudes reflecting deviant intentions for Black youth, they may represent understandable responses to unjust social contexts (e.g., Jackson et al., 2023; Samuels-Wortley, 2021). Previous studies have also suggested that the CSS-M reflects constructs other than antisociality. Soyer et al. (2017) posited that the CSS-M reflected inmates’ cognitive ability to interpret the implications of the questionnaire and respond desirably, while Pauselli et al. (2024) suggested that the CSS-M-offending relationship may reflect how negative attitudes toward the justice system develop through discriminatory experiences. Critically, negative attitudes toward police among Black youth have been attributed to their experience of frequent and hostile police encounters (Harris & Jones, 2020). Youth with negative attitudes toward police may be at higher risk for rearrest and reconviction, not necessarily because they reoffend more, but because they are under closer surveillance.

In addition, the finding that no other attitude types captured in the two measures predicted recidivism for Black youth calls into question the validity of using these tools with Black youth in general. Studies suggest that the relationships between attitudes, criminogenic needs, and offending may depend on factors including race, sex, gender, psychosocial maturity/self-control, intelligence, adverse childhood experiences, offense type, and criminal justice system entrenchment (e.g., Soyer et al., 2017; van der Put et al., 2020), but research regarding the role of procriminal attitudes in offending for Black youth is scant. Inadequately recognizing the role of discrimination in attitude development, particularly if higher-risk labels can contribute to more intrusive sentencing, could perpetuate a cycle of discrimination, criminal justice system entrenchment, and increasing negative attitudes (Fine et al., 2017, 2021; Prins, 2019). In the procedural justice framework, justice system fairness as a driver of perceived legitimacy and desistance is emphasized (Ginsburg-Kempany & Kaiser, 2016; Yasrebi-De Kom et al., 2021).

Conceptualizing Procriminal Attitudes

Our findings give rise to questions about how to best conceptualize procriminal attitudes for a more nuanced understanding of their role in offending. Bonta and Andrews (2023) highlight that although some “rejection of convention” attitudes (e.g., devaluation of education/work) are not directly supportive of crime, they leave one with “less to lose” if they do commit crime, and they predict offending. Here, the authors begin to distinguish attitudes that support crime in terms of their content (on their face) from those that predict crime (whether directly or through mediating factors) but do not overtly endorse crime. We suggest that exploring these nuances in attitudes toward law and legal systems is a fruitful direction for research and practice. Attitudes that explicitly promote crime would include examples like Item 2 on the PIDS: “how proud/ashamed would you be to commit sexual assault,” whereas those that have predictive ability but do not explicitly endorse crime would include examples like devaluation of education/work and distrust of law enforcement. Recognizing this distinction is more theoretically precise, allowing context behind procriminal attitudes to be acknowledged (e.g., trauma, desire for justice), while maintaining their place in risk assessment and rehabilitation if they are found to be predictive of recidivism. Distinguishing attitudes in this way also opens inquiry into contexts, mediators, and moderators of the attitudes-offending relationship, leaving potential to identify more proximal predictors of offending (i.e., identifying what connects distrust of police to offending behavior). In addition, although research is mixed, there is not a reliable causal link between reduction in procriminal attitudes and reduction in offending (Banse et al., 2013). Improving understanding of the attitudes-offending relationship, especially for attitudes that do not explicitly support crime, may improve rehabilitation efforts.

Limitations and Future Directions

Study results and limitations provide directions for future research. A key implication of our findings is that there is a need to better understand how procriminal attitudes relate to youth profiles based on their identities and their history with the criminal justice system. In addition, any salient findings in this study were initially masked when analyses were completed with a combined sample, underscoring the importance of disaggregating data based on race.

Research examining what attitudes—and in which contexts—are more likely to increase risk would support practical recommendations for forensic clinicians. Differential item functioning and measurement invariance analyses would support further understanding of the CSS-M’s functioning based on race. Qualitative and mixed-methods research would allow youth to describe how their attitudes relate to their identity and behavior and how systemic trust can be regained. Studies highlighting youth perspectives about their procriminal attitudes have been conducted in the criminology and social justice literature (Fine et al., 2017; Samuels-Wortley, 2021); they would be a welcome addition to the RNR and criminal justice psychology field (see Pauselli et al., 2024 for similar recommendations).

Some studies indicate differences in criminal attitudes and measurement validity for girls (e.g., Mazher et al., 2022; O’Hagan et al., 2019), so examining how these measures work for different genders, and intersectionally, would be a useful endeavor. Similarly, examining how these tools work for other groups that have been overrepresented in the criminal justice system (e.g., Indigenous youth) is an important step in assessing their validity. Small sample sizes for girls and non-Black racialized groups precluded analyses for these groups. Deepening investigations to consider other factors that may impact youth’s experiences of discrimination and reconviction (e.g., neighborhood, family attitudes, immigration status, physical features, sexual orientation) could shed more light on how relationships clouded by discrimination are impacted by factors related to identity beyond race alone. Although findings are mixed, research suggests that procriminal attitudes vary based on offense type (Simourd & van de Ven, 1999; van der Put et al., 2020). In the present study, Black youth had more violent offenses, and this may have impacted results, but given that previous research is inconclusive, it cannot be said whether that would lead to over- or underpredictions in procriminal attitudes.

The use of real-world forensic assessment and recidivism data contributes to the study’s ecological validity. However, there are valid reasons to conduct future research using data outside the context of risk assessments. First, most justice-involved youth do not undergo court-ordered assessments. Although youth in our sample may be of particular interest due to their complex profiles of history, risk, and offending, findings cannot be generalized to all youth who offend. Second, using self-report data in the context of an assessment meant to guide sentencing could introduce desirability bias in responding despite validity checks (e.g., Soyer et al., 2017). Finally, the data set includes cases from 2001 to 2014, after which administration of the CSS-M was discontinued in the clinic out of an abundance of caution against using an instrument that did not add benefit over and above the PIDS and had potential negative implications for Black youth. Attitudes about the justice system have changed during that time (Fine et al., 2020). To maintain an up-to-date risk assessment and case management policy, research should explore whether the role of procriminal attitudes in offending changes with shifts in perceptions of criminal justice.

Supplemental Material

sj-docx-1-cjb-10.1177_00938548251407576 – Supplemental material for Examining the Reliability and Validity of the Criminal Sentiments Scale–Modified and the Pride in Delinquency Scale in Black and White Canadian Justice System-Involved Youth

Supplemental material, sj-docx-1-cjb-10.1177_00938548251407576 for Examining the Reliability and Validity of the Criminal Sentiments Scale–Modified and the Pride in Delinquency Scale in Black and White Canadian Justice System-Involved Youth by Cassandra R. Stevenson, Michele Peterson-Badali and Tracey A. Skilling in Criminal Justice and Behavior

Footnotes

Authors’ Note:

We have no known conflict of interest to disclose. This research was supported by the Canada Graduate Scholarship–Master’s (Canadian Institute of Health Research), the Canada Graduate Scholarship–Doctoral (Social Sciences and Humanities Research Council), and the Ontario Graduate Scholarship, held by the first author.

ORCID iDs

Cassandra R. Stevenson

Michele Peterson-Badali

Tracey A. Skilling

Supplemental Material

Supplemental Tables 1–4 are available in the online version of this article at .

Cassandra R. Stevenson, MA, is a doctoral student in the School and Clinical Child Psychology Program at the Ontario Institute for Studies in Education, University of Toronto. Her research, conducted under the supervision of Dr. Michele Peterson-Badali and Dr. Tracey Skilling, focuses on the role of attitudes, education, and mental health in risk assessment, recidivism, and desistance for criminal justice-involved youth.

Michele Peterson-Badali, PhD, is a professor in the Department of Applied Psychology and Human Development at the Ontario Institute for Studies in Education, University of Toronto. Her research focuses on children’s and adolescent’s knowledge, reasoning, perceptions, and experiences of the youth justice system; their understanding of rights; and their evolving legal capacities. Current projects focus on the implementation of the Risk–Need–Responsivity framework in community-sentenced youth and the role of mental health in youth criminal justice.

Tracey A. Skilling, PhD, is a psychologist and associate scientist at the Center for Addiction and Mental Health and an assistant professor in the Departments of Psychiatry and Applied Psychology and Human Development at the University of Toronto. Her research interests include mental health issues in justice-involved youth and risk assessments in both male and female justice-involved youth from diverse backgrounds. Her current projects include studies evaluating risk assessment practices and understanding the desistance process with adolescents involved in the justice system.

References

Achenbach

T. M.

(2013). DSM guide for the ASEBA. University of Vermont, Research Center for Children, Youth, & Families. https://tinyurl.com/2x52xyyv

Achenbach

T. M.

Rescorla

L. A.

(2001). Manual for the ASEBA school-age forms & profiles. University of Vermont, Research Center for Children, Youth, & Families.

Andrews

D. A.

Wormith

J. S.

(1984). Criminal sentiments and criminal behaviour. Ministry of the Solicitor General of Canada. https://tinyurl.com/424jmzv8

Banse

Koppehele-Gossel

Kistemaker

L. M.

Werner

V. A.

Schmidt

A. F.

(2013). Pro-criminal attitudes, intervention, and recidivism. Aggression and Violent Behavior, 18(6), 673–685. https://doi-org/10.1016/j.avb.2013.07.024

Bentler

P. M.

(1990). Comparative fit indexes in structural models. Psychological Bulletin, 107(2), 238–246. https://doi.org/10.1037/0033-2909.107.2.238

Bonta

Andrews

D. A.

(2023). The psychology of criminal conduct (7th ed.). https://doi.org/10.4324/9781003292128

Buss

A. H.

Warren

W. L.

(2000). Aggression Questionnaire. Western Psychological Services. https://www.wpspublish.com/aq-aggression-questionnaire

Chui

W. H.

Cheng

K.-Y.

(2017). Criminal sentiments and behaviors among young people in Hong Kong. International Journal of Adolescence and Youth, 22(1), 57–67. https://doi.org/10.1080/02673843.2013.875479

Costaris

Peterson-Badali

Skilling

T. A.

(2022). A latent class analysis of the Antisocial Attitudes domain of the Youth Level of Service/Case Management Inventory. Criminal Justice and Behavior, 49(8), 1192–1209. https://doi.org/n9j9

10.

Department of Justice Canada. (2021). Youth risk-need assessment: An overview of issues and practices. https://tinyurl.com/bdze7ee8

11.

Department of Justice Canada. (2022). The national justice survey 2022 methodological report. https://bit.ly/3F9ElTl

12.

Department of Justice Canada. (2024). State of the criminal justice system report: A focus on youth. https://www.justice.gc.ca/eng/cj-jp/state-etat/2024rpt-rap2024/index.html

13.

Dunn

T. J.

Baguley

Brunsden

(2013). From alpha to omega: A practical solution to the pervasive problem of internal consistency estimation. British Journal of Psychology, 105(3), 399–412. https://doi.org/10.1111/bjop.12046

14.

Field

A. P.

(2017). Discovering statistics using IBM SPSS statistics (5th ed., North American ed.). Sage.

15.

Fine

Cavanagh

Donley

Frick

P. J.

Steinberg

Cauffman

(2017). Is the effect of justice system attitudes on recidivism stable after youths’ first arrest? Race and legal socialization among first-time youth offenders. Law and Human Behavior, 41(2), 146–158. https://doi.org/10.1037/lhb0000229

16.

Fine

Simmons

Cavanagh

Rowan

Cauffman

(2020). Implications of youths’ perceptions of police bias and the code of the street for violent offending. Psychology of Violence, 10(5), 473–482. https://doi.org/10.1037/vio0000286

17.

Fine

A. D.

van Rooij

(2021). Legal socialization: Understanding the obligation to obey the law. Journal of Social Issues, 77(2), 367–391. https://doi.org/10.1111/josi.12440

18.

Fisher

R. A.

(1915). Frequency distribution of the values of the correlation coefficient in samples from an indefinitely large population. Biometrika, 10(4), 507–521. https://doi.org/cx2wt3

19.

Fisher

R. A.

(1921). On the probable error of a coefficient of correlation deduced from a small sample. Metron, 1, 3–32.

20.

Flora

D. B.

Curran

P. J.

(2004). An empirical evaluation of alternative methods of estimation for confirmatory factor analysis with ordinal data. Psychological Methods, 9(4), 466–491. https://doi.org/10.1037/1082-989X.9.4.466

21.

Forero

C. G.

(2014). Cronbach’s alpha. In Michalos

A. C.

(Ed.), Encyclopedia of quality of life and well-being research (pp. 1357–1359). Springer. https://doi.org/n9j7

22.

Ginsburg-Kempany

K. G.

Kaiser

K. A.

(2016). Incorporating procedural justice and legitimacy into the RNR model to improve risk-need assessment. In Taxman

, (Ed.), Handbook on risk and need assessment (pp. 285–307). Routledge.

23.

Harris

J. W.

Jones

M. S.

(2020). Shaping youths’ perceptions and attitudes toward the police: Differences in direct and vicarious encounters with police. Journal of Criminal Justice, 67, Article 101674. https://doi.org/10.1016/j.jcrimjus.2020.101674

24.

Hayes

A. F.

Coutts

J. J.

(2020). Use omega rather than Cronbach’s alpha for estimating reliability. But. . .. Communication Methods and Measures, 14(1), 1–24. https://doi.org/ggwd6m

25.

Helmond

Overbeek

Brugman

Gibbs

J. C.

(2014). A meta-analysis on cognitive distortions and externalizing problem behavior: Associations, moderators, and treatment effectiveness. Criminal Justice and Behavior, 42(3), 245–262. https://doi.org/gf4frz

26.

Hosmer

D. W.

Lemeshow

(2000). Applied logistic regression (2nd ed.). John Wiley & Sons. https://doi.org/c96pnx

27.

IBM Corp. (2021). IBM SPSS statistics for Mac (version 28.0). https://tinyurl.com/48spxu8e

28.

Jack

L. A.

Ogloff

J. R. P.

(1997). Factors affecting the referral of young offenders for medical and psychological assessment under the Young Offenders Act. Canadian Journal of Criminology, 39(3), 247–273. https://tinyurl.com/5seznrzm

29.

Jackson

McKay

Cheliotis

Bradford

Fine

Trinkner

(2023). Centering race in procedural justice theory: Structural racism and the under- and overpolicing of Black communities. Law and Human Behavior, 47(1), 68–82. https://doi.org/10.1037/lhb0000524

30.

Kalkbrenner

M. T.

(2024). Choosing between Cronbach’s coefficient alpha, McDonald’s coefficient omega, and coefficient h: Confidence intervals and the advantages and drawbacks of interpretive guidelines. Measurement and Evaluation in Counseling and Development, 57(2), 93–105. https://doi.org/10.1080/07481756.2023.2283637

31.

Lee

I. A.

Preacher

K. J.

(2013). Calculation for the test of the difference between two dependent correlations with one variable in common. Quantpsy. http://quantpsy.org

32.

Martino

Grilo

C. M.

Fehon

D. C.

(2000). Development of the drug abuse screening test for adolescents (DAST-A). Addictive Behaviors, 25(1), 57–70. https://doi.org/b2rmht

33.

Mayer

Filstead

W. J.

(1979). The Adolescent Alcohol Involvement Scale: An instrument for measuring adolescents’ use and misuse of alcohol. Journal of Studies on Alcohol, 40(3), 291–300. https://doi.org/10.15288/jsa.1979.40.291

34.

Mazher

Masood

Simourd

D. J.

(2022). A cross-cultural examination of the Criminal Sentiments Scale–Modified. Criminal Justice and Behavior, 49(9), 1295–1310. https://doi.org/10.1177/00938548221088812

35.

McKay

M. T.

Dempster

(2016). The reliability and factorial validity of the Adolescent Alcohol Involvement Scale in a large sample in the United Kingdom. Journal of Substance Use, 21(2), 177–178. https://doi.org/10.3109/14659891.2014.995725

36.

Mogadam

Skilling

T. A.

Peterson-Badali

Hannah

(2024). Examining the measurement invariance and psychometrics of the Drug Abuse Screening Test for Adolescents (DAST-A) in justice-involved youth. Criminal Justice and Behavior, 51, 1413–1432. https://doi.org/10.1177/00938548241246437

37.

Muthén

L. K.

Muthén

B. O.

(2017). MPlus user’s guide (Version 8). https://bit.ly/4i47PAv

38.

Nunnally

J. C.

Bernstein

I. H.

(1994). Psychometric theory (3rd ed.). McGraw-Hill. http://catdir.loc.gov/catdir/toc/mh022/93022756.html

39.

O’Hagan

H. R.

Brown

S. L.

Jones

N. J.

Skilling

T. A.

(2019). The reliability and validity of the Measure of Criminal Attitudes and Associates and the Pride in Delinquency Scale in a mixed sex sample of justice-involved youth. Criminal Justice and Behavior, 46(5), 751–769. https://doi.org/10.1177/0093854818810459

40.

Olver

M. E.

Stockdale

K. C.

Helmus

L. M.

Woods

Termeer

Prince

(2024). Too risky to use, or too risky not to? Lessons learned from over 30 years of research on forensic risk assessment with Indigenous persons. Psychological Bulletin, 150(5), 487–553. https://doi.org/10.1037/bul0000414

41.

Olver

M. E.

Stockdale

K. C.

Wormith

J. S.

(2014). Thirty years of research on the Level of Service Scales: A meta-analytic examination of predictive accuracy and sources of variability. Psychological Assessment, 26(1), 156–176. https://doi.org/10.1037/a0035080

42.

Pauselli

Zern

Ashekun

Ellis

Jackson

Pope

L. G.

Watson

A. C.

Wood

J. D.

Compton

M. T.

(2024). Predictors of Criminal Sentiments Scale-Modified scores in outpatients with legal system involvement. The Journal of the American Academy of Psychiatry and the Law, 52(2), 176–185. https://doi.org/10.29158/JAAPL.230120-23

43.

Piedmont

R. L.

(2014). Inter-item correlations. In Michalos

A. C.

(Ed.), Encyclopedia of quality of life and well-being research (pp. 3303–3304). Springer. https://doi.org/kc23

44.

Prins

S. J.

(2019). Criminogenic or criminalized? Testing an assumption for expanding criminogenic risk assessment. Law and Human Behavior, 43(5), 477–490. https://doi.org/n9j4

45.

Rice

M. E.

Harris

G. T.

(2005). Comparing effect sizes in follow-up studies: ROC Area, Cohen’s d, and r. Law and Human Behavior, 29(5), 615–620. https://doi.org/d2xh3f

46.

Samuels-Wortley

(2021). To serve and protect whom? Using composite counter-storytelling to explore Black and Indigenous youth experiences and perceptions of the police in Canada. Crime & Delinquency, 67(8), 1137–1164. https://doi.org/hhfv

47.

Samuels-Wortley

(2022). Youthful discretion: Police selection bias in access to pre-charge diversion programs in Canada. Race and Justice, 12(2), 387–410. https://doi.org/gr9jtv

48.

Satorra

Bentler

P. M.

(2010). Ensuring positiveness of the scaled difference chi-square test statistic. Psychometrika, 75(2), 243–248. https://doi.org/bzjmjk

49.

Shields

I. W.

Simourd

D. J.

(1991). Predicting predatory behavior in a population of incarcerated young offenders. Criminal Justice and Behavior, 18(2), 180–194. https://doi.org/10.1177/0093854891018002006

50.

Shields

I. W.

Whitehall

G. C.

(1994). Neutralization and delinquency among teenagers. Criminal Justice and Behavior, 21(2), 223–235. https://doi.org/10.1177/0093854894021002003

51.

Simourd

D. J.

(1997). The Criminal Sentiments Scale-Modified and Pride in Delinquency Scale: Psychometric properties and construct validity of two measures of criminal attitudes. Criminal Justice and Behavior, 24(1), 52–70. https://doi.org/b2v37s

52.

Simourd

D. J.

Olver

M. E.

(2002). The future of criminal attitudes research and practice. Criminal Justice and Behavior, 29(4), 427–446. https://doi.org/dqztd6

53.

Simourd

D. J.

van de Ven

(1999). Assessment of criminal attitudes: Criterion-related validity of the Criminal Sentiments Scale-Modified and Pride in Delinquency Scale. Criminal Justice and Behavior, 26(1), 90–106. https://doi.org/crx4ht

54.

Skeem

J. L.

Lowenkamp

C. T.

(2016). Risk, race, and recidivism: Predictive bias and disparate impact. Criminology, 54(4), 680–712. https://doi.org/10.1111/1745-9125.12123

55.

Skilling

T. A.

Sorge

G. B.

(2014). Measuring antisocial values and attitudes in justice-involved male youth: Evaluating the psychometric properties of the Pride in Delinquency Scale and the Criminal Sentiments Scale–Modified. Criminal Justice and Behavior, 41(8), 992–1007. https://doi.org/10.1177/0093854814521415

56.

Soper

D. S.

(2025). Significance of the difference between two correlations calculator. Free Statistics Calculators Index. https://tinyurl.com/5dhxjvue

57.

Soyer

McNeeley

Zajac

Bucklen

K. B.

(2017). Measuring the criminal mind: The relationship between intelligence and CSS-M results among a sample of Pennsylvania prison inmates. Criminal Justice and Behavior, 44(11), 1444–1461. https://doi.org/gcqj78

58.

Steiger

J. H.

(1980). Tests for comparing elements of a correlation matrix. Psychological Bulletin, 87, 245–251. https://doi.org/10.1037/0033-2909.87.2.245

59.

Steiger

J. H.

Lind

J. C.

(1980). Statistically based tests for the number of common factors [Conference session]. Psychometric Society Annual Meeting, Iowa City, IA, United States.

60.

Tangney

J. P.

Stuewig

Furukawa

Kopelovich

Meyer

Cosby

(2012). Reliability, validity, and predictive utility of the 25-item Criminogenic Cognitions Scale (CCS). Criminal Justice and Behavior, 39, 101340–101360. https://doi.org/ggq2d9

61.

Timko

Booth

B. M.

Han

Schultz

N. R.

Blonigen

D. M.

Wong

J. J.

Cucciare

M. A.

(2017). Criminogenic needs, substance use, and offending among rural stimulant users. Rural Mental Health, 41(2), 110–122. https://doi.org/10.1037/rmh0000065

62.

Tucker

L. R.

Lewis

(1973). A reliability coefficient for maximum likelihood factor analysis. Psychometrika, 38(1), 1–10. https://doi.org/10.1007/BF02291170

63.

van der Put

C. E.

Assink

Gubbels

. (2020). Differences in risk factors for violent, nonviolent, and sexual offending. Journal of Forensic Psychology Research and Practice, 20(4), 341–361. https://doi.org/10.1080/24732850.2020.1735248

64.

Vaske

J. C.

Gehring

K. S.

Lovins

(2017). Gender differences in the measurement of criminal thinking. Criminal Justice and Behavior, 44(3), 395–415. https://doi.org/f9x798

65.

Vincent

G. M.

Drawbridge

Davis

(2019). The validity of risk assessment instruments for transition-age youth. Journal of Consulting and Clinical Psychology, 87(2), 171–183. https://doi.org/10.1037/ccp0000366

66.

Walters

G. D.

(2016). Predicting recidivism with the Criminal Sentiments Scale: A Meta-analysis of a putative measure of criminal thought content. Criminal Justice and Behavior, 43(9), 1159–1172. https://doi.org/10.1177/009385481664900

67.

Yasrebi-De Kom

F. M.

Dirkzwager

A. J.

Van Der Laan

P. H.

Nieuwbeerta

. (2021). The effect of sanction severity and its interaction with procedural justice. Criminal Justice and Behavior, 49(2), 200–219. https://doi.org/10.1177/0093854821103835

68.

Youth Criminal Justice Act, S.C. (2002). https://laws-lois.justice.gc.ca/eng/acts/Y-1.5/

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

0.04 MB