The Intersection of Juvenile Psychopathy,Protective Factors,Treatment Change,and Diversity in Justice-Involved Youth

Participants

The study sample was developed initially by identifying 451 consecutive admissions to a community mental health facility in Saskatoon, Saskatchewan, Canada within a 5-year catchment period (2007–2012). Youth were referred either by the courts or a community youth worker for assessment and/or treatment services. Of these 451 cases, 257 files were of sufficient quality to code the study measures and comprised this study sample described as follows. Owing to incomplete or missing details within files, the n fluctuated for some variables within the sample and exact variable counts are provided in parentheses. Overall, about three quarters of the sample was male (76.7%, 197/257) and about one quarter was female (23.3%, 60/257). On average, the youth were 16.5 (SD = 1.6, range age: 12–20) years of age at the time of services. Most youth had a history of school suspension or expulsion (83.7%, 195/233), approximately half were attending school at the time of arrest (54.5%, 132/242), and the last grade completed was grade 8 (M = 8.3, SD = 1.4). Information concerning ethnoracial heritage was available for most of the sample (89.9%, 231/257), most of whom were identified as having Indigenous ancestry (75.3%, 174/231) from self-report and/or legal processes (e.g., inclusion of Gladue considerations in presentence reports), and the remainder were predominantly White or other non-Indigenous heritage (24.7%, 57/231). More than half (62.6%, 161/257) of youth had an index conviction for a nonsexual violent offense (e.g., assault, robbery), while a smaller proportion had convictions for sexual offenses (13.2%, 34/257), or exclusively nonviolent offenses (24.1%, 62/257). About two thirds (68.8%, 150/218) had any previous conviction and 42.2% (92/218) had a previous conviction for a violent offense.

Almost half the sample (43.2%, 111/257) had participated in some form of forensic treatment service such as substance management (43.2%, 48/111), violence reduction treatment or anger management (39.6%, 44/111), general forensic therapeutic services (23.4%, 26/111), sexual offense treatment (17.1%, 19/111), or other unspecified services (5.4%, 6/111); fairly frequently, youth attended multiple programs (27.9%, 31/111). For individual services (n = 93), youth attended an average number of 15.1 individual sessions (SD = 14.4) and cumulative individual sessions spanned an average of 251.8 days (SD = 233.3). For group services (n = 25), youth attended an average of 18.8 total sessions (SD = 20.0) over a cumulative duration of 187.6 days (SD = 219.4). Table 1 presents descriptive statistics and frequencies of the sample for all demographic, criminal history, clinical, and treatment variables and study measures.

OPEN IN VIEWER

Table 1:

Descriptive Statistics and Frequencies of Study Sample and Measures

Measure	N	% (n)	M (SD)
Gender	257
Male		76.7 (197)
Female		23.3 (60)
Age	257		16.5 (1.6)
Ethnoracial heritage	231
White/non-Indigenous		24.7 (57)
Indigenous		75.3 (174)
School history
Attending school	242	54.5 (132)
Suspended/expelled	233	83.7 (195)
Current offenses	257
Nonsexual violent index		62.6 (161)
Sexual offense index		13.2 (34)
Nonviolent index only		24.1 (62)
Offense history	218
Prior conviction		68.8 (150)
Prior violent offense		42.2 (92)
Treatment history	111
Substance management		43.2 (48)
Violence reduction		39.6 (44)
General forensic		23.4 (26)
Sexual offense treatment		17.1 (19)
Unspecified		5.4 (6)
Treatment attendance
Individual length days	93		251.8 (233.3)
Individual sessions	93		15.1 (14.4)
Group length days	25		187.6 (219.4)
Group sessions	25		18.8 (20.0)
Psychopathy Checklist: Youth Version	257
Interpersonal			1.6 (1.8)
Affective			4.8 (2.3)
Lifestyle			5.9 (2.3)
Antisocial			5.4 (2.6)
Total			18.6 (7.0)
Structured Assessment of Protective Factors–Youth Version	257
Resilience			1.3 (1.2)
Motivational			2.2 (2.4)
Relational			1.8 (1.4)
External			2.2 (1.0)
Total			7.4 (4.7)
Violence Risk Scale-Youth Version
Stable	257		5.3 (2.6)
Dynamic (pre)	257		30.7 (10.1)
Dynamic (post)	88		30.0 (10.1)
Total (pre)	257		36.8 (11.9)
Total (post)	88		36.0 (12.0)
Change	88		1.5 (3.2)

Measures

Procedure

Ethical and/or operational approval to conduct the present research was obtained from the University of Saskatchewan Behavioral Research Ethics Board (Beh ID # 18–30), the Provincial Ministry of Justice (Corrections and Policing Division), and the Provincial Law Courts. This study is a retrospective archival investigation that involved rating the study measures from Court and clinical files. A well-established literature supports the validity and reliability of file-based ratings for measures of psychopathy, violence risk, and protective factors across youth and adult populations (e.g., de Vries Robbé et al., 2015a; Stockdale et al., 2014; Wong et al., 2012), so long as the files are sufficiently detailed. All files captured between 2007 and 2012 were stored at the agency and data collection occurred on site. The files were accessed, reviewed for completeness, and if sufficient information was available (which usually required at a minimum a psychological assessment report and a presentence report), the structured forensic rating scales, criminal history, sentencing, treatment, and demographic variables were coded.

All files and study measures were coded by the first author, who was a graduate clinical psychology student at the time of the research, and a senior undergraduate honors student. The raters were trained on rating the study measures and supervised by two registered psychologists (the second and third authors), each of whom had more than 15 years of experience in research and clinical practice with youth justice populations and had been trained on, or were involved in the development of, the study measures. The coder training for this study consisted of a full day of youth violence risk assessment training on the VRS-YV, PCL:YV, and SAPROF-YV and completing two redacted cases, one female and one male. The research supervisors then co-coded four study cases on all measures from file with the two raters to calibrate their ratings, with rating discrepancies resolved through consensus. When a satisfactory level of consensus had been obtained (i.e., through reviewing and discussing ratings), the raters began independent coding. Approximately 10% of files were periodically selected at random and independently double-coded throughout the study to establish interrater reliability and to prevent rater drift.

Given that files often had multiple opening dates and youth could have multiple admissions to the agency, the raters pulled each other’s files, reviewed each one, and then removed any information that could be potential source of criterion contamination before the file was subsequently reviewed and coded by the other rater. When files had multiple opening dates, the last admission was used and all information contained from that point forward was utilized in coding the study measures. In this manner, the potential for criterion contamination was considerably decreased. For clinical files that contained both assessment and treatment progress information, the researcher conducting the screening separated the two sets of information for the other coder so that ratings on the measures completed at pretreatment (i.e., using baseline assessment and pretreatment information) and posttreatment (i.e., using treatment progress information) could be done as independently as possible. All files were coded prior to obtaining recidivism data, and thus blind to outcome. The recidivism data, in turn, were coded from CPIC by the three study authors.

Planned Analyses

All analyses were conducted using SPSS for Widows v. 28 and were conducted to examine the intersection of PCL:YV measured juvenile psychopathy, protective factors, violence risk, and change to recidivism as a function of gender and ethnoracial heritage. Whenever power permitted, we conducted intersectional analyses; however, the small number of non-Indigenous female youth in the sample precluded most analyses with this subgroup.

First, we reported PCL:YV facet and total scores and conducted gender (male–female) and ethnoracial (Indigenous–non-Indigenous) group comparisons on the measures employing standardized mean difference (Cohen’s d), in which d values of 0.20 correspond to small, 0.50 medium, and 0.80 large (Cohen, 1992). Second, to examine the relation of protective factors to juvenile psychopathy, we subsequently compared high and low PCL:YV scoring youth (based on a cut score of 20) within gender and ethnoracial groups on the SAPROF-YV total score and its four subscales, again employing Cohen’s d to capture the magnitude of difference. Third, we subsequently examined the convergent properties of PCL:YV facet and total scores by correlating these with the VRS-YV and SAPROF-YV total and subscale scores, with correlation magnitudes (between two continuous variables) corresponding to small r = .10, medium r = .30, and large r = .50 (Cohen, 1992); analyses were conducted on the aggregate sample for parsimony.

Fourth, we examined the predictive properties of PCL:YV scores for the three recidivism outcomes employing fixed 3- and 5-year follow-ups via ROC (receiver operating characteristic) analyses. These fixed follow-ups are commonly employed in the correctional risk assessment literature (Olver et al., 2024) to control for differences in follow-up time and to capture prediction at different timepoints (e.g., shorter vs. intermediate or longer term), respectively. The follow-ups also reflect the developmental context and time limited duration of adolescence. These analyses were conducted on the aggregate sample, among broad gender and ethnoracial subgroups, and among specific gender × ethnoracial subgroups. ROC analyses generate an area under the curve (AUC) statistic representing the probability that a randomly selected recidivist scores higher on a given measure than a randomly selected nonrecidivist. With possible values ranging from 0 to 1.0, and .50 representing chance-level prediction, AUC magnitudes can be interpreted using effect size language of small .56, medium .64, and large .71 (Rice & Harris, 2005). Of note, for SAPROF-YV analyses, the direction of the criterion variable was reversed (i.e., increasing protection associated with decreased recidivism) so that all AUC values would be in the positive direction (i.e., above .50).

The remaining sets of analyses examined the incremental predictive validity of measures of protective factors and violence risk and change, controlling for PCL:YV total score, through a series of hierarchical Cox regression survival analyses. These analyses featured a pretreatment model, a posttreatment model, and a change model and were conducted on the aggregate sample using continuous predictor scores given the power requirements of multipredictor regression models, particularly given that analyses using change information were limited to one-third of the sample. The pretreatment Cox regression model examined the relative contributions of baseline risk and protection for predicting recidivism over and above psychopathy, with the order of entry switched in alternating blocks to directly examine the relative contributions of static and dynamic risk (VRS-YV) and protection (SAPROF-YV), prior to entry into the final predictor model (block 2). The analyses were intended to examine: (a) the risk-mitigating effects of protective factors over and above juvenile psychopathy, (b) how a purposively developed violence risk scale with dynamic factors compares to the juvenile psychopathy construct in the prediction of outcome, and (3) whether protective factors and violence risk measures uniquely predict outcome over and above juvenile psychopathy. The baseline pretreatment model was supplemented by a set of Kaplan–Meier survival analyses examining recidivism trajectories employing mean splits of high and low psychopathy (i.e., PCL:YV ≥20 vs. ≤19, respectively) and high and low protection (SAPROF-YV total ≥8 vs. ≤7, respectively). Mean splits were employed to evenly distribute cell n to maximize power; given that the sample had a high proportion of youth with substantial psychopathy features, the high scoring PCL:YV group reflected an abundance of such traits. A risk-mitigating effect would be illustrated by steeper recidivism trajectories for youth who are high on psychopathy and low on protective factors, compared to high-psychopathy youth who are high(er) on protective factors.

The posttreatment Cox regression model substituted in posttreatment dynamic item ratings to examine to what extent dynamic variables incorporating change information and protective factors, were uniquely associated with outcome controlling for psychopathy and static risk. The change Cox regression model incorporated VRS-YV change scores as a final covariate in the analyses, to directly examine to what extent treatment-related changes in risk versus a baseline measure of protective factors, are associated with decreased recidivism after accounting for individual differences in levels of psychopathy and on static and dynamic risk factors.

The final set of analyses were exploratory and featured a set of targeted Cox regressions for the Indigenous, non-Indigenous, male, and female groups. Given that any gender or ethnoracial subgroups employing posttreatment or change information would be prohibitively small to support the Cox regression models, variations on the baseline models were examined; to partly account for changes that occurred in risk, scores from the most recent available dynamic assessment were employed.

Profiles of Juvenile Psychopathy Among Gender and Ethnoracial Groups

Table 2 reports PCL:YV descriptives and gender × ethnoracial group comparisons on PCL:YV total and facet scores. Indigenous youth scored significantly higher than non-Indigenous youth, overall and in the male subgroup, on PCL:YV total score (small effect) and its Lifestyle (large effect) and antisocial (medium effect) facets; by contrast, non-Indigenous youth scored higher overall (small effect) than Indigenous youth on the Interpersonal facet. Indigenous youth scored nonsignificantly higher on the Affective facet. The same pattern of ethnoracial differences was observed for female youth but none were significant owing to the extremely small cell size of White female youth.

OPEN IN VIEWER

Table 2:

Psychopathy Checklist: Youth Version (PCL:YV) Score Comparisons for Gender and Ethnoracial Subgroups

PCL:YV measure	Indigenous	Non-Indigenous	Overall	d,I v. NI
	M (SD)	M (SD)	M (SD)
Male
Interpersonal	1.6 (1.6)	2.1 (2.2)	1.7 (1.8)	−0.28
Affective	4.9 (2.2)	4.7 (2.4)	4.8 (2.3)	0.11
Lifestyle	6.4 (2.1)	5.1 (2.8)	5.9 (2.5)	0.53
Antisocial	6.3 (2.6)	4.1 (2.4)	5.4 (2.8)	0.86
Total	19.7 (6.6)	17.0 (8.0)	18.6 (7.2)	0.39
Female
Interpersonal	1.3 (1.3)	2.4 (2.6)	1.4 (1.6)	−0.77
Affective	5.0 (2.1)	4.3 (2.9)	4.9 (2.3)	0.33
Lifestyle	6.3 (1.9)	4.6 (2.0)	6.0 (2.0)	0.93
Antisocial	5.5 (2.0)	5.1 (2.1)	5.4 (2.1)	0.20
Total	19.0 (5.6)	17.1 (7.8)	18.6 (6.1)	0.31
Overall
Interpersonal	1.5 (1.6)	2.1 (2.2)	1.6 (1.8)	−0.36
Affective	4.9 (2.2)	4.6 (2.5)	4.8 (2.3)	0.15
Lifestyle	6.4 (2.1)	5.1 (2.7)	5.9 (2.3)	0.58
Antisocial	6.1 (2.4)	4.2 (2.4)	5.4 (2.6)	0.76
Total	19.5 (6.3)	17.0 (7.9)	18.6 (7.0)	0.37

Note. Significant d values in bold font. I = Indigenous, NI = non-Indigenous; PCL:YV = Psychopathy Checklist: Youth Version.

The Interrelations of Psychopathy, Protective Factors, Violence Risk, and Change

As seen in Table 3, youth scoring with a higher preponderance of psychopathic traits (PCL:YV total score 20+) scored significantly lower on SAPROF-YV measures of protective factors across gender and ethnoracial groups. In particular, high-psychopathy youth scored lower with uniformly large effects on the Motivational domain (d = −0.84 to −1.11), medium to large effects on the Resilience domain and total score (d = −0.65 to −1.11), and medium effects for scores on the Relational domain (d = −0.43 to −0.72). The only exception were frequently weak and variable nonsignificant psychopathy group differences on the External domain for female and Indigenous youth and small to medium effects for male and non-Indigenous youth.

OPEN IN VIEWER

Table 3:

High and Low Juvenile Psychopathy Group Comparisons on Measures of Protective Factors Among Gender and Ethnoracial Subgroups

SAPROF measure	Gender			Ethnoracial ancestry
	PCL:YV < 20	PCL:YV 20+	d	PCL:YV < 20	PCL:YV 20+	d
	M (SD)	M (SD)		M (SD)	M (SD)
	Male			Indigenous
Resilience	1.9 (1.4)	0.71 (0.83)	−1.01	1.6 (1.3)	0.69 (0.83)	−0.91
Motivational	3.6 (2.7)	1.1 (1.6)	−1.11	2.5 (2.2)	0.99 (1.3)	−0.88
Relational	2.4 (1.5)	1.4 (1.1)	−0.73	2.0 (0.96)	1.3 (1.1)	−0.55
External	2.3 (1.0)	2.1 (0.88)	−0.30	2.1 (0.85)	2.0 (0.96)	−0.05
Total	10.3 (5.3)	5.3 (3.1)	−1.14	8.2 (4.1)	5.0 (2.7)	−0.94
	Female			Non-Indigenous
Resilience	1.3 (0.92)	0.78 (0.75)	−0.65	2.3 (1.6)	0.82 (0.73)	−1.11
Motivational	2.4 (2.2)	0.96 (1.0)	−0.84	4.9 (2.7)	1.6 (2.1)	−1.35
Relational	1.4 (1.2)	0.93 (0.96)	−0.43	2.7 (1.6)	1.6 (1.3)	−0.72
External	2.1 (0.90)	2.1 (1.1)	0.05	2.8 (1.1)	2.3 (0.94)	−0.54
Total	7.3 (3.8)	4.8 (2.3)	−0.78	12.8 (5.7)	6.2 (3.7)	−1.32

Note. Significant d values in bold font. PCL:YV = Psychopathy Checklist: Youth Version.

Table 4 is a convergent validity correlation matrix of PCL:YV factor and total scores with violence risk and protection measure scores. As anticipated, higher levels of PCL:YV measured juvenile psychopathy were associated with greater levels of violence risk as measured by the VRS-YV and fewer protective factors as measured by the SAPROF-YV. Specifically, large in magnitude associations were found for PCL:YV total score and Affective, Lifestyle, and Antisocial factor scores with VRS-YV dynamic and total scores (pre and post), and SAPROF-YV total scores and Resilience, Motivational, and Relational scores. VRS-YV static/stable scores had particularly strong associations with PCL:YV Lifestyle and Antisocial factor and total scores. PCL:YV Interpersonal factor scores, by contrast, had small in magnitude (or lower) associations with the VRS-YV and SAPROF-YV measures.

OPEN IN VIEWER

Table 4:

Correlation Matrix of PCL:YV Associations With Violence Risk and Protection Measures

Measure	PCL:YV
	Interpersonal	Affective	Lifestyle	Antisocial	Total
PCL:YV
Interpersonal	(.64)
Affective	.43***	(.77)
Lifestyle	.24***	.54***	(.71)
Antisocial	.18**	.47***	.59***	(.72)
Total	.58***	.81***	.80***	.76***	(.83)
VRS-YV
Stable	.10	.38***	.49***	.75***	.59***
Dynamic (pre)	.24***	.60***	.73***	.80***	.79***
Dynamic (post)	.20	.54***	.74***	.78***	.75***
Change	.08	.17	.04	.06	.12
Total (pre)	.22***	.57***	.71***	.82***	.78***
Total (post)	.18	.52***	.71***	.80***	.73***
SAPROF-YV
Resilience	−.14*	−.44***	−.56***	−.50***	−.56***
Motivational	−.15*	−.42***	−.64***	−.62***	−.60***
Relational	−.04	−.30***	−.45***	−.48***	−.44***
External	−.05	−.10	−.27***	−.25***	−.18**
Total	−.11	−.43***	−.66***	−.64***	−.62***

Note. N = 257, except VRS-YV post and change measures (n = 88).

*

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

Predictive Accuracy of PCL:YV Scores as a Function of Gender and Ethnoracial Heritage

Table 5 reports predictive accuracy results (ROC analyses, AUC) for PCL:YV total scores with 3- and 5-year violent, nonviolent, and general recidivism overall and among gender × ethnoracial subgroups. In the aggregate sample overall, PCL:YV scores had broadly medium effects in the prediction of each of the recidivism outcomes over both follow-ups. AUC magnitudes were also largely equivalent between male and female groups (medium effects), with the exception of a small effect in the prediction of 3-year violence for male youth. Across broad ethnoracial groups, PCL:YV scores showed the greatest parity in the prediction of violence and some disparity in the prediction of nonviolent and general recidivism. Specifically, PCL:YV scores had small effects in the prediction of 3-year violence and medium effects in the prediction of 5-year violence across Indigenous and non-Indigenous groups. It had small effects in the prediction of nonviolent recidivism for Indigenous youth (compared to medium to large effects for non-Indigenous youth), and small to medium effects in the prediction of general recidivism for Indigenous youth (compared to medium to large effects for non-Indigenous youth); none of these AUC differences reached statistical significance.

OPEN IN VIEWER

Table 5:

ROC Analyses: Prediction of Fixed 3- and 5-Year Recidivism Outcomes by PCL:YV Total Score Among Ethnoracial and Gender Subgroups

PCL:YV group and recidivism criterion	3-year						5-year
	Indigenous		Non-Indigenous		Overall		Indigenous		Non-Indigenous		Overall
	AUC	95%CI	AUC	95%CI	AUC	95%CI	AUC	95%CI	AUC	95%CI	AUC	95%CI
Violent recidivism
Male	.57	[.47, .67]	.63	[.45, .80]	.62**	[.54, .70]	.61*	[.51, .72]	.74**	[.59, .88]	.66***	[.58, .74]
Female	.79***	[.63, .95]	−	−	.66	[.49, .83]	.72*	[.55, .89]	−	−	.66*	[.50, .82]
Overall	.61**	[.53, .70]	.59	[.43, .75]	.62***	[.55, .69]	.64***	[.56, .73]	.69**	[.55, .84]	.66***	[.59, .72]
Nonviolent recidivism
Male	.58	[.47, .70]	.74***	[.61, .88]	.66***	[.59, .74]	.58	[.45, .71]	.71**	[.57, .85]	.65***	[.56, .73]
Female	.69*	[.52, .86]	−	−	.66*	[.51, .81]	.65	[.49, .81]	−	−	.65*	[.52, .80]
Overall	.61**	[.53, .70]	.73***	[.60, .86]	.66***	[.60, .73]	.60*	[.51, .70]	.70**	[.56, .83]	.65***	[.58, .72]
General recidivism
Male	.61*	[.50, .72]	.73***	[.60, .87]	.68***	[.60, .76]	.61	[.49, .74]	.70**	[.56, .85]	.66***	[.58, .75]
Female	.71*	[.55, .87]	−	−	.68*	[.53, .83]	.65	[.49, .81]	−	−	.66*	[.52, .80]
Overall	.64***	[.55, .72]	.72***	[.59, .85]	.67***	[.61, .74]	.63**	[.54, .72]	.69**	[.55, .83]	.66***	[.59, .73]

Note. PCL:YV = Psychopathy Checklist: Youth Version; AUC = area under the curve; CI = confidence interval.

*

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

When predictive properties were examined within gender-ethnoracial subgroups, for female Indigenous youth, large prediction effects were observed for violent recidivism and 3-year general recidivism, and medium effects were found for the remaining outcomes (AUCs are not reported for female non-Indigenous youth owing to the extremely small cell n). A similar pattern emerged for male non-Indigenous youth, for whom PCL:YV scores had large effects in the prediction of nonviolent recidivism, 3-year general, and 5-year violent recidivism, and a small effect for 3-year violence and medium effect for 5-year general recidivism. By contrast, for male Indigenous youth, PCL:YV total scores had small prediction effect sizes for all three outcomes across both follow-ups. For the PCL:YV facets (not reported for space considerations), the Lifestyle and Antisocial factor scores tended to have the most consistent associations with violent, nonviolent, and general recidivism (see Supplementary Table S1 for full facet level AUC prediction findings).

Incremental Associations of Psychopathy, Protective Factors, Violence Risk, and Change

A series of hierarchical Cox regression survival analyses were conducted examining the incremental associations of PCL:YV, SAPROF-YV, and VRS-YV scores to violent, nonviolent, and general recidivism across three sets of models: a pretreatment (i.e., baseline risk and protection) model, posttreatment model, and change model. A set of Kaplan–Meier survival analyses followed to unpack the psychopathy, protective factors, and recidivism associations.

Psychopathy, Protective Factors, and Recidivism

The first block (1a) of the Cox regression survival analyses found both PCL:YV scores and SAPROF-YV total scores incremented the prediction of the three recidivism outcomes in their expected direction. That is, higher levels of psychopathy predictably were associated with higher and faster rates of reoffending across all outcomes, while this was offset by higher levels of protective factors, which were associated with lower and slower rates of reoffending. A set of Kaplan–Meier survival analyses further illustrated how the risk aggravating properties of psychopathy are offset by the risk-mitigating properties of protective factors, by way of comparing recidivism trajectories in high- versus low-psychopathy (PCL:YV total 20+ vs <20) and high- versus low-protection groups (SAPROF-YV total 8+ vs <8). As seen in Figure 1, youth with high psychopathy and low protection had significantly higher rates of each recidivism outcome over time compared to similarly high-psychopathy youth who also scored above the mean on protective factors (see Supplementary Table S2 for full survival analysis results). Moreover, youth who were either high on psychopathy or low on protection, had virtually indistinguishable trajectories of recidivism, while youth who were both low psychopathy and high on protection had the shallowest trajectories of each recidivism outcome.

OPEN IN VIEWER

Figure 1:

Survival Analysis: Trajectories of Recidivism as a Function of Psychopathy (PCL:YV) and Protection

Psychopathy, Protective Factors, Violence Risk, and Change

The remaining Cox regression survival analyses in Table 6 examined the intersection of psychopathy, protective factors, violence risk, and change to recidivism over time. The first block (1b) of the baseline pretreatment model found that only dynamic factor scores uniquely predicted any of the recidivism outcomes, notably violent and nonviolent recidivism, controlling for PCL:YV and stable factor scores. When SAPROF-YV scores were added to the baseline pretreatment model (block 2), although most predictor-criterion associations were in the expected direction, none attained significance. In the posttreatment model, despite featuring only one third of the sample (n = 88), posttreatment VRS-YV dynamic scores incrementally predicted all three recidivism outcomes, including violence over time, while none of the other model predictors had significant unique associations with any outcomes. In the final change model, which disaggregates the VRS-YV posttreatment dynamic score into separate pretreatment dynamic and change components, dynamic factor scores continued to uniquely predict increases in each recidivism outcome. VRS-YV change scores were also significantly associated with decreased nonviolent and general recidivism after comprehensive controls for baseline risk; however, none of PCL:YV scores, stable factors, or protective factor scores incremented predictions of any recidivism outcomes.

OPEN IN VIEWER

Table 6:

Hierarchical Cox Regression Survival Analyses: Incremental Associations of Psychopathy, Protective Factors, Violence Risk, and Change Measures With Recidivism

Regression model	Violent recidivism				Nonviolent recidivism				Any recidivism
	B	SE	p	eB [95% CI]	B	SE	p	eB [95% CI]	B	SE	p	eB [95% CI]
Pretreatment Model
Block 1a
PCL:YV total	.031	.015	.045	1.031 [1.001, 1.063]	.030	.014	.033	1.030 [1.002, 1.058]	.036	.014	.010	1.036 [1.009, 1.065]
SAPROF-YV total	−.079	.027	.003	0.924 [0.877, 0.974]	−.063	.022	.005	0.939 [0.899, 0.981]	−.054	.022	.013	0.947 [0.908, 0.989]
Block 1b
PCL:YV total	.013	.019	.496	1.013 [0.976, 1.052]	.020	.018	.252	1.021 [0.986, 1.057]	.030	.018	.090	1.030 [0.995, 1.066]
VRS-YV stable	.008	.049	.876	1.008 [0.915, 1.109]	.005	.045	.911	1.005 [0.921, 1.097]	−.008	.044	.858	0.992 [0.911, 1.081]
VRS-YV dynamic (pre)	.042	.017	.011	1.043 [1.010, 1.078]	.029	.014	.042	1.029 [1.001, 1.059]	.026	.014	.067	1.026 [0.998, 1.055]
Block 2
PCL:YV total	.011	.019	.572	1.011 [0.973, 1.050]	.020	.018	.267	1.020 [0.985, 1.056]	.029	.018	.095	1.030 [0.995, 1.066]
SAPROF-YV total	−.052	.031	.094	0.949 [0.893, 1.009]	−.050	.027	.061	0.951 [0.902, 1.002]	−.046	.026	.078	0.955 [0.907, 1.005]
VRS-YV stable	−.007	.050	.890	0.993 [0.900, 1.096]	−.012	.046	.797	0.988 [0.903, 1.082]	−.024	.045	.596	0.976 [0.893, 1.067]
VRS-YV dynamic (pre)	.030	.018	.096	1.031 [0.995, 1.068]	.016	.016	.303	1.016 [0.985, 1.048]	.014	.016	.368	1.014 [0.984, 1.045]
Posttreatment Model
PCL:YV total	−.015	.033	.641	0.985 [0.922, 1.051]	−.044	.032	.169	0.957 [0.899, 1.019]	−.026	.031	.388	0.974 [0.917, 1.034]
SAPROF-YV total	−.011	.055	.846	0.989 [0.887, 1.103]	.026	.048	.583	1.027 [0.935, 1.128]	.032	.047	.496	1.032 [0.942, 1.131]
VRS-YV stable	−.085	.080	.290	0.919 [0.785, 1.075]	.033	.072	.652	1.033 [0.897, 1.191]	−.001	.071	.991	0.999 [0.870, 1.148]
VRS-YV dynamic (post)	.074	.031	.019	1.076 [1.012, 1.144]	.079	.029	.006	1.082 [1.023, 1.146]	.076	.029	.008	1.079 [1.020, 1.141]
Change Model
PCL:YV total	−.016	.035	.641	0.984 [0.918, 1.054]	−.040	.033	.219	0.961 [0.901, 1.024]	−.021	.032	.503	0.979 [0.920, 1.042]
SAPROF-YV total	−.004	.057	.945	0.996 [0.892, 1.113]	.031	.048	.527	1.031 [0.938, 1.133]	.031	.047	.512	1.031 [0.941, 1.130]
VRS-YV stable	−.086	.083	.301	0.918 [0.781, 1.080]	.042	.076	.581	1.043 [0.898, 1.211]	.013	.074	.863	1.013 [0.876, 1.171]
VRS-YV dynamic (pre)	.078	.035	.026	1.081 [1.010, 1.158]	.077	.032	.016	1.080 [1.014, 1.151]	.069	.031	.026	1.071 [1.008, 1.138]
VRS-YV change	−.085	.054	.111	0.918 [0.826, 1.020]	−.107	.047	.025	0.899 [0.819, 0.987]	−.092	.045	.040	0.912 [0.836, 0.996]

Note. Significant p values for model predictors in bold font. Pretreatment model N = 257; posttreatment and change models n = 88. Blocks 1a and 1b of the pretreatment model illustrate the predictive associations of either VRS-YV (stable and dynamic) or SAPROF-YV scores (respectively) after controlling for PCL:YV score alone. Block 2 is the final step of the pretreatment model incorporating all model predictors that illustrates the unique predictive associations of VRS-YV (stable and dynamic), SAPROF-YV, controlling for PCL:YV score within a common model. For the posttreatment and change models, only the final step of the model incorporating all predictors (akin to Block 2) is presented for parsimony. All models employ continuous predictor scores. PCL:YV = Psychopathy Checklist: Youth Version; SAPROF-YV = Structured Assessment of Protective Factors–Youth Version; VRS: YV = Violence Risk Scale–Youth Version; SE = standard error; CI = confidence interval.

Gender and Ethnoracial Moderators: Psychopathy, Protection, and Violence Risk

The final set of hierarchical Cox regressions examined variations on the baseline pretreatment models employing the most recent dynamic factors assessment across gender and ethnoracial subgroups (Table 7). The results are interpreted with caution given that none of these regression models could directly examine changes in risk and owing to fluctuating and some small cell sizes, some of regression models would be underpowered. Furthermore, combining gender groups within an ethnoracial category and vice versa due to sample size limits, invariably muddies interpretation and the results may not generalize to the individual gender or ethnoracial groups subsumed within the larger aggregate category. We believed, however, that the nature and strength of some of the observed predictor-criterion associations and the importance of the conclusions emanating from these analyses both justified and necessitated their execution, and offset the limitations inherent within these analyses.

OPEN IN VIEWER

Table 7:

Hierarchical Cox Regression Survival Analyses: Incremental Associations of Psychopathy, Protective Factors, and Violence Risk Measures With Recidivism among Gender and Ethnoracial Groups

Regression model	Indigenous			Non-Indigenous			Male			Female
	B	p	eB [95% CI]	B	p	eB [95% CI]	B	p	eB [95% CI]	B	p	eB [95% CI]
Block 1a	Violent recidivism
PCL:YV total	.020	.283	1.02 [0.98, 1.06]	.027	.494	1.03 [0.95, 1.11]	.014	.389	1.01 [0.98, 1.05]	.084	.045	1.09 [1.002, 1.18]
SAPROF-YV total	−.110	.002	0.90 [0.83, 0.96]	−.055	.340	0.95 [0.85, 1.06]	−.106	<.001	0.90 [0.85, 0.95]	−.002	.975	1.00 [0.86, 1.16]
Block 1b
PCL:YV total	.011	.622	1.01 [0.97, 1.06]	.011	.813	1.01 [0.93, 1.11]	.001	.952	1.00 [0.96, 1.04]	.025	.636	1.03 [0.93, 1.14]
VRS-YV stable	−.002	.969	1.00 [0.90, 1.11]	−.101	.463	0.90 [0.69, 1.18]	−.043	.408	0.96 [0.86, 1.06]	.243	.056	1.28 [0.99, 1.64]
VRS-YV dynamic (MR)	.045	.025	1.05 [1.01, 1.09]	.058	.102	1.06 [0.99, 1.14]	.058	<.001	1.06 [1.03, 1.10]	.003	.949	1.00 [0.92, 1.09]
Block 2
PCL:YV total	.006	.807	1.01 [0.96, 1.05]	.009	.853	1.01 [0.92, 1.11]	−.004	.847	1.00 [0.96, 1.04]	.029	.575	1.03 [0.93, 1.14]
SAPROF-YV total	−.094	.023	0.91 [0.84, 0.99]	−.015	.834	0.99 [0.86, 1.13]	−.084	.016	0.92 [0.86, 0.99]	.071	.418	1.07 [0.91, 1.27]
VRS-YV stable	−.031	.590	0.97 [0.87, 1.09]	−.104	.455	0.90 [0.69, 1.18]	−.067	.217	0.94 [0.84, 1.04]	.254	.045	1.29 [1.01, 1.65]
VRS-YV dynamic (MR)	.028	.197	1.03 [0.99, 1.07]	.054	.197	1.06 [0.97, 1.14]	.039	.043	1.04 [1.001, 1.08]	.017	.721	1.02 [0.93, 1.11]
Block 1a	General recidivism
PCL:YV total	.021	.244	1.03 [0.99, 1.06]	.021	.451	1.02 [0.97, 1.08]	.023	.122	1.02 [0.99, 1.05]	.059	.114	1.06 [0.99, 1.14]
SAPROF-YV total	−.073	.019	0.93 [0.87, 0.99]	−.078	.064	0.93 [0.85, 1.01]	−.073	.002	0.93 [0.89, 0.97]	−.068	.325	0.93 [0.82, 1.07]
Block 1b
PCL:YV total	.034	.117	1.03 [0.99, 1.08]	−.016	.663	0.98 [0.91, 1.06]	.014	.436	1.02 [0.98, 1.05]	.029	.550	1.03 [0.94, 1.13]
VRS-YV stable	.021	.667	1.02 [0.93, 1.13]	−.194	.066	0.82 [0.67, 1.01]	−.035	.450	0.97 [0.88, 1.06]	.119	.289	1.13 [0.90, 1.40]
VRS-YV dynamic (MR)	.007	.693	1.01 [0.97, 1.04]	.106	<.001	1.11 [1.05, 1.18]	.041	.005	1.04 [1.01, 1.07]	.022	.583	1.02 [0.95, 1.11]
Block 2
PCL:YV total	.029	.194	1.03 [0.99, 1.07]	−.017	.656	0.98 [0.91, 1.06]	.013	.475	1.01 [0.98, 1.05]	.027	.577	1.03 [0.94, 1.13]
SAPROF-YV total	−.084	.020	0.92 [0.86, 0.99]	−.009	.871	0.99 [0.89, 1.10]	−.058	.041	0.94 [0.89, 0.998]	−.030	.701	0.97 [0.83, 1.13]
VRS-YV stable	−.011	.839	0.99 [0.89, 1.10]	−.195	.066	0.82 [0.67, 1.01]	−.053	.272	0.95 [0.86, 1.04]	.111	.332	1.12 [0.89, 1.40]
VRS-YV dynamic (MR)	−.009	.635	0.99 [0.96, 1.03]	.103	.004	1.11 [1.03, 1.19]	.025	.117	1.03 [0.99, 1.06]	.016	.695	1.02 [0.94, 1.10]

Note. Significant p values in bold font. MR = most recent assessment. Group n: Indigenous = 174, non-Indigenous = 57, male = 197, female = 60. Blocks 1a and 1b are provided to illustrate the predictive associations of either VRS-YV (stable and dynamic) or SAPROF-YV scores (respectively) after controlling for PCL:YV score alone. Block 2 is the final step incorporating all model predictors that illustrates the unique predictive associations of VRS-YV (stable and dynamic), SAPROF-YV, controlling for baseline psychopathy score within a common model. All models employ continuous predictor scores. PCL:YV = Psychopathy Checklist: Youth Version; SAPROF-YV = Structured Assessment of Protective Factors-Youth Version; VRS: YV = Violence Risk Scale-Youth Version; SE = standard error; CI = confidence interval.

The male subgroup, and the largest of the subgroups, found that VRS-YV dynamic scores and SAPROF-VY scores each incremented predictions of violence controlling for the VRS-YV and that in the final model (block 2), both dynamic risk factors and protective factors each uniquely predicted future violence in their expected direction. A similar theme was observed for general recidivism, only in the second block of analyses, dynamic scores did not reach significance. Paradoxically, in the female only regression models, psychopathy ratings trumped protective factor ratings in the prediction of violence (block 1a) and only VRS-YV stable scores uniquely predicted future violence in the final regression model (block 2); none of the predictors demonstrated incremental prediction of general recidivism. Among Indigenous youth in a combined male–female subgroup, dynamic factor and protective factor ratings each incremented the prediction of violent recidivism, while PCL:YV score did not, and in the final model (block 2), only SAPROF-YV scores reached significance; dynamic factor ratings, although in the expected direction, did not retain significance. Furthermore, only SAPROF-YV scores incremented predictions of general recidivism for Indigenous youth. By contrast, for non-Indigenous youth, VRS-YV dynamic item ratings consistently incrementally predicted general recidivism, while none of the other model predictors had significant associations with outcome.

Juvenile Psychopathy and Protective Factors: Unpacking the Association

This study examined the intersection of psychopathy, violence risk, protective factors, and treatment change with recidivism in an ethnoracially diverse sample of male and female court-adjudicated youth. Protective factors were fewer in number for high-psychopathy youth; the greater the number of juvenile psychopathy features, the fewer protective factors a given youth tended to have. This inverse association between juvenile psychopathy and protection was observed across ethnoracial and gender subgroups; all of which have implications for the adaptation of services suited to the unique characteristics of clientele to cultivate and harness strengths to increase treatment buy-in, minimize attrition, and maximize gain, per the responsivity principle. The disparity was most notable for the resilience, motivational, and relational domains (and by default, the total score) but less so for the external domain. The external domain refers to the availability of external supports and resources to the youth, whether they are fully utilized or not, and although these should be prioritized for higher risk, higher psychopathy youth (per the risk principle), they may not be. The other protective domains reflect the use of internal, psychological, and behavioral resources such as self-control and healthy coping (resilience), positive, and effortful engagement in prosocial systems such as engaging in school/work, meaningful leisure, or having a positive attitude toward authority (motivational), or close connectedness with prosocial others (relational). To the extent that youth struggle with motivation, task completion, prosocial orientation, emotional bonding, or have largely dysfunctional familial and social networks, will they score lower on these domains; and high-psychopathy youth are unfortunately likely to have many such concerns that undermine the development and expression of protective factors. This may explain for instance, why inverse associations were most notable with the affective, lifestyle, and antisocial features, but not so with the interpersonal domain. Arguably, some of the features of the interpersonal domain, if prosocially channeled, may have adaptive qualities (e.g., charm and volubility, persuasive qualities, and intact self esteem).

Protective Factors and Risk Mitigation in High-Psychopathy Youth

The above being said, youth with higher levels of psychopathy did have protective factors, and per Olver and Riemer’s (2021) findings with a large treated adult sexual offending sample, higher levels of protection appeared to have risk-mitigating properties, although the baseline was low. On average, youth in this sample, who tended to be broadly medium to high risk and had committed serious crimes, scored only 7 out of what would be a maximum possible 32 points, when summing the items on the SAPROF-YV; and for youth lower in psychopathy versus those who were high, the ratio of protective factors was about 2 to 1. For high-psychopathy youth scoring above average on protective factors, however, the results of survival analysis demonstrated that they had significantly lower rates of future violence and other outcomes over time than similarly high-psychopathy youth without the same number and quality of protective factors. Overall, both PCL:YV scores and SAPROF-YV ratings uniquely predicted recidivism, that is, SAPROF-YV scores mitigated risk, but youth with higher PCL:YV scores still posed a greater recidivism risk than youth with lower scores; put another way, with enough protective factors, it is possible that these can offset the risk aggravating qualities of juvenile psychopathy.

Direct comparisons between high-psychopathy youth who were high versus low on protective factors demonstrated significant differences on each of the SAPROF-YV domains (around or exceeding a full standard deviation), and these differences were cumulative; however, high-psychopathy high-protection youth also had fewer static and dynamic risk factors, and hence were also lower risk. The key seems to be risk relevance—in order for protective factors to have risk-mitigating qualities, these factors must be antithetical to risk—prosocial, realistic, attainable, and time limited, such as engaging meaningfully with school or work, having an influential prosocial caregiver or support person, or ready access to prosocial cultural, spiritual, or leisure pursuits. A further candidate is access to risk-reduction treatment that can promote the development of prosocial competencies, skills, and strategies that can offset risk for crime and violence, and increase the potential for positive outcomes, per the risk and need principles.

Dynamic Violence Risk and Treatment Change in Juvenile Psychopathy

A growing literature, primarily with high-psychopathy adult populations, has demonstrated meaningful engagement in treatment to be associated with decreased recidivism outcomes (e.g., Langton et al., 2006; Olver & Riemer, 2021; Skeem et al., 2002; Wong et al., 2012). This too was found in this study, whereby changes in violence risk, as measured by the VRS-YV, in a subsample of youth who participated in treatment services and had sufficient documentation available to evaluate change, were significantly associated with decreased nonviolent and general recidivism, after controlling for individual differences in static and dynamic risk factors, as well as PCL:YV score. Moreover, in these regression models, PCL:YV score and static/stable factors did not increment predictions of recidivism, while dynamic factors and change did. In other words, youth who engaged meaningfully and were rated as having reduced their violence risk had lower rates of adverse outcomes, irrespective of psychopathy or static/historical factors. The results extend findings elsewhere that show serious youth justice populations change on dynamic risk factors with appropriate intervention (e.g., Viljoen et al., 2017). Furthermore, those youth who were higher risk and thus stood to benefit the most from services (i.e., had greater criminogenic need and hence, the most room for change), would have derived greater benefit than lower risk youth, per the risk principle (Bonta & Andrews, 2024).

Ethnoracial and Gender Implications for Protective Factors and the Juvenile Psychopathy Construct in Violence Risk Assessment and Management

Consistent with the Olver et al. (2024) meta-analysis (which included unpublished data from this study among the studies included in the final aggregation), PCL:YV scores were significantly higher overall, by about one-third of a standard deviation (small effect), for Indigenous male and female youth, with the difference explained primarily by higher scores on the lifestyle and affective facets. Furthermore, PCL:YV scores predicted their targeted outcomes with small to medium effects for violence in Indigenous and non-Indigenous subgroups, small effects for nonviolent recidivism (vs medium to large for non-Indigenous youth), and small to medium effects for general recidivism (vs medium to large for non-Indigenous youth). When these bivariate associations were examined intersectionally, there was a notable disparity in the predictive accuracy of PCL:YV scores for male versus female Indigenous youth; specifically, the PCL:YV registered consistently small effect sizes for male Indigenous youth, but medium- to large-effect sizes for female Indigenous youth.

A set of targeted Cox regressions examining variations on a baseline pretreatment model generally supported the value of dynamic risk and protective factors even from a pure prediction standpoint. Among male youth, close to the prototypical or ideal scenario statistically was observed, with dynamic risk factors uniquely predicting increased violent recidivism, and protective factors, decreased violent recidivism. A similar trend for the aggregate Indigenous sample was observed with respect to violence, although for any future recidivism, only protective factors uniquely predicted this outcome. The opposite seemed to be the case for non-Indigenous youth and female youth; protective factor ratings did not incrementally predict any outcomes. For non-Indigenous youth, dynamic factor ratings had unique prominence (but the associations achieved significance only in the general recidivism model), and for female youth, stable factor and psychopathy ratings had the most predictive strength, and only for future violence. Given that dynamic risk and change could not be examined within the subgroups owing to sample size limits, which also precluded specific gender-ethnoracial subgroup analyses, and the fact that some of these regressions were underpowered, caution is warranted in interpreting these associations, but we believe some conclusions are clear.

One conclusion is that protective factors can have value added to measures of static and dynamic risk and vice versa, and this was most evident in sufficiently powered regression models; however, when the dynamic measures are assessed in a dynamic manner (i.e., at two or more timepoints) and incorporated measures of risk change, baseline assessments of protective factors added little to predictive validity. Elsewhere, in another Canadian sample, Finseth et al. (2023) found SAPROF-YV scores incrementally predicted decreased general recidivism controlling for scores on a general risk-need measure among Black youth, but not for White youth. Of note, both sets of measures were administered at only one time point, so conclusions could not be drawn about situating protective factors within dynamic assessments of risk and need. As such, further research is warranted to compare changes in protection to changes in risk to see if each add unique information toward predicting recidivism outcome (e.g., de Vries Robbé et al., 2015a; Olver & Riemer, 2021), including in diverse samples.

A second conclusion is to underscore the value of dynamic risk factors, both from a recidivism prediction as well as risk reduction and recidivism prevention standpoint. This was also readily apparent from the capacity for dynamic factor ratings to incrementally predict most outcomes, particularly violence, above and beyond static/stable factors, but not vice versa. Especially for Indigenous correctional populations, the incorporation of dynamic and protective factors can offset biases inherent in unduly weighting historic factors, and provide a positive and rehabilitative angle toward prosocial community reintegration and the prevention of recidivism.

A third conclusion is casting the juvenile psychopathy construct within a realistic light; when valid measures of risk and protection are available, the unique predictive properties of the construct seemed to take a backseat. Longitudinal investigations of juvenile psychopathy in youth populations also show that measures of the construct can change over time and follow different developmental trajectories, one of which includes decreases in psychopathy features and concordant crime and aggression (Hawes et al., 2018). This does not diminish the relevance of juvenile psychopathy in forensic risk assessment, but rather, that high juvenile psychopathy ratings are not all defining; they do not place limits on the capacity for prosocial change or determine outcome for youth. This is important to bear in mind, given that juvenile psychopathy evidence has increased in youth court cases in North America in recent years and has been found to be influential for some decisions (Viljoen et al., 2010).

With all this said, a note is warranted about the hierarchical Cox regression findings for female youth; the initial report from this data set demonstrated that VRS-YV and SAPROF-YV ratings each predicted recidivism in bivariate analyses with female youth (Lovatt et al., 2022), but in this study, when entered into a Cox regression model along with PCL:YV ratings, they did not uniquely predict outcome. We do not believe this finding diminishes the value of dynamic or protective factor ratings for female youth, and given the small female subsample and its diversity, the results of a multipredictor regression need to be interpreted with caution, and the analyses should be replicated on a larger sample. From our clinical experience working with justice-involved youth, we believe the findings reflect the nature and lived experiences of this court-adjudicated female sample, in which elevated levels of juvenile psychopathy and severity of history, notably violence background and an unstable chaotic upbringing, are prominent, and have important clinical and forensic relevance.

Strengths, Limitations, and Future Directions

This study has notable strengths and limitations with implications for further research and practice. This work featured a diverse sample of male and female youth, comprehensive information sources to obtain quality data, and sufficient intervention information to evaluate changes in risk for about one third of the sample. These conditions facilitated gathering data regarding the predictive properties of risk, treatment change, and protective measures situated within the clinical construct of juvenile psychopathy. The findings we hope, offer cautious optimism and a rejoinder about the immutability of high-psychopathy youth and the presence of strengths and resiliencies to mitigate risk. Furthermore, the diverse nature of sample enabled examination of key clinical research questions among diverse subgroups. Most analyses were sufficiently powered to make inferences pertaining to broad gender and ethnoracial groups, and in some cases (e.g., PCL-R bivariate recidivism associations), specific intersectional gender-ethnoracial subgroups (e.g., Indigenous male and female persons). That said, although the sample was sufficiently large to power most analyses, perhaps the most significant limitation was insufficient n to conduct a greater number of specific gender-ethnoracial analyses, particularly for multipredictor regression models. Moreover, only one-third of the sample had participated in some form of structured forensic treatment intervention with sufficient information available to evaluate changes in violence risk, which reduced statistical power to some degree and also made subgroup analyses of treatment-related risk change untenable.

Thus, these findings need to be replicated and extended to additional samples and settings to permit more nuanced examination of risk, treatment change, and protection, in an intersectional manner. To start, the dynamic and responsive properties of measures of protection in youth samples require further investigation. Research on criminogenic risk factors for adults and youth has begun exploring the relevance of other culturally salient factors (e.g., gender, Indigeneity), but these lines of inquiry have yet to be pursued with measures of protective factors to better understand if there are culturally specific areas of protection that may not be tapped by extant tools. Further research is also required on high risk-need subsamples of justice-involved youth; these are the very youth that stand to benefit the most from risk-reduction services, and these youth often present with specific responsivity considerations to be effectively addressed in treatment. As such, a better understanding of the relationship between risk-need and protection, in treated samples is critical to inform clinical practice and juvenile justice decision making.