The Relationship Between Psychopathy and Violence: A Comparison Across Birth Cohorts

Abstract

The current study responds to calls to consider whether individual characteristics (e.g., psychopathy) interact with broader structural conditions (e.g., economic, social, and legal contexts) in ways that influence offending. We examined whether the psychopathy-violence relationship varies across birth cohorts. Data came from 535 male and female participants from the Incarcerated Serious and Violent Young Offender Study who were interviewed in adolescence and received a rating on the Psychopathy Checklist: Youth Version (PCL:YV). Participants were born between 1979 and 1997 and were followed into adulthood to examine offending outcomes. Generalized estimating equations examined the relationship between psychopathy, birth cohort membership, and annual violent convictions between ages 18 and 40. PCL:YV scores were significantly associated with violent convictions across adulthood. Individuals from recent birth cohorts had higher probabilities of violence. Interaction analyses indicated that youth scoring high on the PCL:YV had similar probabilities of violent convictions regardless of birth cohort, whereas youth from recent cohorts who scored low on the PCL:YV had higher probabilities of violent convictions than low-scoring youth from earlier cohorts. These findings contrast with U.S. research showing that low-risk youth born in the 1980s had similar probabilities of arrest to high-risk youth born in the 1990s. This difference may reflect cross-national variation in the intensity of macro-level changes. People with elevated psychopathy traits may be more resistant to their social conditions. We discuss how changes in the composition of study populations may attenuate observed effect sizes through selection processes and recommend improving how effect sizes are contextualized.

Introduction

Psychopathy is a maladaptive, multidimensional syndrome encompassing interpersonal, affective, and behavioural traits (Cleckley, 1976). Forensic psychology and related fields have given substantial attention to psychopathy due to its association with violence (Hart, 1998) and its implications for risk assessment and risk communication (Gatner et al., 2022). Research has examined whether a person's demographic background and immediate social environment influence the psychopathy–offending relationship (Klein Haneveld et al., 2022; McCuish et al., 2018; McCuish & Lussier, 2023). Less attention has been given to broader macro-level circumstances, including structural conditions (e.g., unemployment rates), that might affect this relationship (Crocker & Leclair, 2025; McCuish, 2026; Tonry, 2019; Verona & Fox, 2025). Bronfenbrenner's (1979) ecological model and Elder's (1975) life-course perspective provide useful frameworks for considering the role of macro-level structural conditions in influencing the relationship between psychopathy and offending. For example, both perspectives emphasize that macro-level systems change over time (e.g., chronosystems; Bronfenbrenner, 1979). These changing macro-level systems suggest that the relationship between psychopathy and offending may vary across different birth cohorts that come of age under different structural conditions. Addressing this possibility has practical implications, as legal systems would need to rethink generalizing findings about the psychopathy–violence relationship to more contemporary birth cohorts if such findings were based on studies of earlier birth cohorts (Montana et al., 2023).

The current study examined whether the psychopathy–violence relationship varied across birth cohorts exposed to different structural conditions. We used data from 535 male and female participants from the Incarcerated Serious and Violent Young Offender Study (ISVYOS) who were assessed using the Psychopathy Checklist: Youth Version (PCL:YV; Forth et al., 2003). Participants were followed into adulthood, allowing us to examine the relationship between psychopathy and longer-term patterns of violence. Birth years ranged between 1979 and 1997, a span that is consistent with prior birth cohort comparisons (Neil & Sampson, 2021; Payne & Piquero, 2020) and that enabled us to evaluate whether the psychopathy–violence relationship differs across birth cohorts that came of age under distinct legal, social, and economic conditions. We focused on psychopathy for two main reasons. First, it is a construct that features prominently in violence risk assessment instruments and justice system decision-making (e.g., Viljoen et al., 2010). Second, this focus helps address debate about whether people with psychopathy traits are especially over-reactive or under-reactive to changes in their social environment (McCuish, 2026). These findings should not be generalized to different constructs (e.g., birth cohort differences in the relationship between gender and violence), different birth cohorts (e.g., Gen Z), different samples (e.g., non-incarcerated populations), or different jurisdictions (e.g., beyond British Columbia, Canada).

Overview of Birth Cohort Research

The macro-level structural conditions in which a person is embedded have implications for individual development (Elder, 1975). For example, a person's likelihood of employment is not only shaped by individual-level factors (e.g., work experience and level of education) but also by structural conditions (e.g., the Great Depression, unemployment and post-secondary education rates). Birth cohorts are groups of individuals born in the same or similar years and help approximate groups who are exposed to similar structural conditions at the same developmental stage ¹ (Ryder, 1965). Comparing birth cohorts allows researchers to examine whether changes in structural conditions affect individual development (Bronfenbrenner, 1979). For example, simply being a member of a larger birth cohort (i.e., one born during a period of higher fertility) implies fewer economic opportunities (Easterlin, 1978), which in turn has implications for offending. Birth cohort research thus helps account for historical context, including criminal justice policies, crime rates, and social and economic conditions (e.g., unemployment) that may influence offending (Neil & Sampson, 2021). Birth cohort comparisons are common in psychiatry and epidemiology (Arseneault et al., 2000; Filatova et al., 2019; Haukka et al., 2001), but less attention has been given to birth cohort differences in offending (cf. Neil & Sampson, 2021; Wolfgang et al., 1972). ² This gap has left unaddressed a question with implications for forensic psychology and the criminal justice system: Do risk factors that predicted offending in the 1990s also predict offending in the 2010s and 2020s? Three studies using data from the Project on Human Development in Chicago Neighbourhoods (PHDCN+) addressed this question (see Montana et al., 2023; Neil & Sampson, 2021; Neil et al., 2021).

The PHDCN+ is a study of over 8,000 children and youth born in the 1980s and 1990s (Sampson et al., 2022) and includes individual-level risk factors that are commonly included in risk assessment instruments (RAIs; Montana et al., 2023). A subsample of 1,057 PHDCN+ participants was followed into adulthood to measure arrest trajectories. The PHDCN+ is unique because few datasets include multiple birth cohorts assessed using identical interview protocols and measures of offending across the same ages. ³ Neil and Sampson (2021) found that PHDCN+ participants who were born in the 1990s came of age during periods with lower crime rates and themselves had lower levels of arrest than cohorts born in the 1980s. Critically, the era in which someone was born was more informative for predicting arrest in adulthood than individual-level factors commonly included in RAIs (e.g., low self-control). The difference between cohorts in arrest rates was so dramatic that more contemporary cohorts who were high-risk had arrest rates similar to older cohorts who were low-risk (Neil & Sampson, 2021). Montana et al. (2023) further reported that re-arrest was systematically over-predicted for PHDCN+ participants deemed high-risk who were born in the 1990s. Thus, criminal justice decision-making may systematically disadvantage contemporary birth cohorts.

Actuarial assessment uses statistical methods to categorize people with similar characteristics who have similar probabilities of an outcome such as recidivism. Actuarial RAIs typically sum item scores to produce a total score, which is then translated into a probabilistic estimate of recidivism risk. A purely actuarial approach considers only the score produced from the RAI and gives no consideration to other information (Quinsey et al., 1998). Actuarial RAIs operate on the assumption that low- and high-scoring individuals differ in their probability of re-offending (Imrey & Dawid, 2015). Montana et al.'s (2023) research challenged this assumption by demonstrating that high-risk youth born in the 1990s had a probability of offending similar to that of lower-risk youth born in the 1980s. Psychopathy is commonly assessed as part of violence risk assessment protocols (Viljoen et al., 2010) and affects criminal justice system decision-making (DeMatteo et al., 2014). Given Montana et al.'s (2023) findings, it is important to examine whether the psychopathy–violence relationship varies across birth cohorts exposed to different macro-level structural conditions. For example, two people of identical age and identical RAI scores may nevertheless vary in their probability of re-offending because they were from different birth cohorts. Such information may be important to communicate in legal settings and as part of clinical case formulation.

Current Study

The current study responds to calls for research in forensic psychology that gives greater attention to the structural conditions within which people are embedded (Crocker & Leclair, 2025; Verona & Fox, 2025). Examining whether the psychopathy–violence relationship varied across birth cohorts contributes to debate about whether people with psychopathy traits are less responsive to changes in their social environment (Hoppenbrouwers et al., 2015; McCuish, 2026). Addressing this theoretical question has practical implications for the criminal justice system. If people with psychopathy traits differ in their likelihood of violence based on when they were born, this would suggest that risk assessments, risk communication, and case formulation should more precisely contextualize violence risk by considering macro-level social conditions (e.g., crime rates, unemployment rates, and legal systems).

We used data from 535 participants from the ISVYOS (British Columbia, Canada) who received a PCL:YV assessment. Participants were recruited from youth custody centres between 1998 and 2011 and were born between 1979 and 1997. Although we do not focus on total birth cohorts (i.e., all individuals born in a single year in a single jurisdiction), this range exposed participants to distinct macro-level structural conditions and is comparable to prior research (e.g., Haukka et al., 2001; Neil & Sampson, 2021; Payne & Piquero, 2020). Longer-term follow-up periods are essential given that cohort effects may vary across developmental stages (Steffensmeier et al., 1992). Participants were followed into adulthood, allowing comparisons of offending at equivalent ages across birth cohorts. For example, violent convictions measured at age 18 occurred in 1997 for individuals born in 1979 but in 2015 for those born in 1997. Thus, we could compare the adulthood outcomes of participants who entered adulthood during different historical periods. This facilitated the examination of whether the psychopathy–violence relationship was stable across cohorts or varied with historical context. Analyses are not meant to be generalized to the relationship between other risk factors and outcomes, other birth cohorts, or other jurisdictions.

Method

Open Science and Transparency

The study was not pre-registered. Benning and Smith's (2023) post-registration guidelines were used to improve transparency. STROBE reporting guidelines were used to draft this manuscript (von Elm et al., 2007) and the STROBE reporting checklist (see Supplemental Material) was used to revise the manuscript (von Elm et al., 2025).

The dataset is not publicly available because of the confidential and sensitive nature of the self-report information provided by incarcerated youth. Access requests should be directed to Evan McCuish, the Principal Investigator of the ISVYOS. A codebook is available at https://doi.org/10.13140/RG.2.2.24702.75847.

Sample and Procedures

The Ministry of Children and Family Development oversees youth incarceration facilities in the province of British Columbia, Canada, and granted the research team access to custody centres to recruit participants. The ISVYOS received ethics approval from the Simon Fraser University Research Ethics Board. Among the 1,719 youth recruited, 1,640 participated in an intake interview conducted by ISVYOS research assistants (RAs). On average, participants were 16.08 years old (SD = 1.33). Interviews were conducted in English and occurred in a private room away from other youth and custody staff. To obtain assent, an information sheet was read to participants explaining the purpose of the study, how data would be collected, their right to withdraw at any time, and that the information they provided would be kept confidential unless they made a direct threat against themselves or others. RAs explained that participation would not affect sentencing or any other aspect of their stay in custody. Among youth who completed an intake interview, 535 subsequently participated in a semi-structured interview used to rate the PCL:YV. The PCL:YV interview lasted up to three hours and was typically conducted across multiple sessions to maintain participant focus. ⁴

Follow-up data were collected through the Corrections Network (CORNET), an electronic software program for client management and supervision. CORNET includes data on all individuals involved in the provincial justice system in British Columbia. Each person has a searchable electronic file containing multiple sources of information, including criminal history. Because CORNET is centralized and integral to BC Corrections’ operations, the current study was able to collect comparable follow-up data for participants born over a decade apart without the need to merge or link data sources. Criminal history follow-up data were missing for 18 participants, either due to a sealed correctional file (n = 17) or mortality before age 18 (n = 1). Additionally, two participants lacked data on ethnicity. Imputation was not conducted for these types of missing data (Derose et al., 2012). Thus, the analytic sample was reduced from 535 to 515 participants (96% retention).

Measures

For the analytic sample (n = 515), approximately 90% of participants had complete data across all measures. Missing data were addressed through multiple imputation by chained equations (MICE) via the mi impute chained command in Stata (StataCorp, 2019). Item-level values were imputed prior to constructing scales. The mi impute chained command does not require the assumption that data are missing completely at random (Shin et al., 2017). Descriptive statistics in Table 1 reflect post-imputation values.

Table 1.

Descriptive Statistics for the Analytic Sample (n = 515).

Variable	Cohort 1 (1979–1983) n = 228 (M/SD; n/%)	Cohort 2 (1984–1988) n = 122 (M/SD; n/%)	Cohort 3 (1989–1993) n = 141 (M/SD; n/%)	Cohort 4 (1994–1997) n = 24 (M/SD; n/%)	Full Sample (1979–1997) n = 515 (M/SD; n/%)
Justice system involvement a
Violent convictions per year of age	0.09 (0.11)	0.11 (0.15)	0.15 (0.17)	0.17 (0.25)	0.11 (0.15)
Months incarcerated per year of age	1.88 (2.42)	2.24 (2.54)	2.73 (2.69)	3.85 (3.80)	2.29 (2.64)
Social environment
Family adversity scale	2.82 (1.59)	2.95 (1.71)	2.90 (1.61)	2.79 (1.29)	2.87 (1.61)
Psychopathy
PCL:YV total scores	20.68 (6.42)	20.40 (5.90)	21.67 (6.75)	26.75 (6.22)	21.17 (6.50)
Control variables
Youth justice legislation
Young Offenders Act	228 (100%)	105 (86.1%)	0 (0%)	0 (0%)	333 (64.7%)
Youth Criminal Justice Act	0 (0%)	17 (13.9%)	141 (100%)	24 (100%)	182 (35.3%)
Prior violence
Violent convictions (ages 12–17)	1.47 (1.58)	1.64 (1.72)	2.02 (1.94)	2.04 (2.34)	1.68 (1.77)
Gender
Female	40 (17.5%)	27 (22.1%)	19 (13.5%)	3 (12.5%)	89 (17.3%)
Male	188 (82.5%)	95 (77.9%)	122 (86.5%)	21 (87.5%)	426 (82.7%)
Ethnicity
White	131 (57.5%)	77 (63.1%)	84 (59.6%)	15 (62.5%)	307 (59.6%)
Indigenous	54 (23.7%)	31 (25.4%)	40 (28.4%)	6 (25.0%)	131 (25.4%)
Non-Indigenous minority	43 (18.9%)	14 (11.5%)	17 (12.1%)	3 (12.5%)	77 (15.0%)

Note. ^aOlder cohorts had a significantly longer follow-up period. To account for differences in the length of follow-up, violent convictions/months incarcerated was divided by the number of years in the follow-up period. Differences in exposure time were accounted for in the multivariate analyses. PCL:YV= Psychopathy Checklist: Youth Version.

Violent Convictions

CORNET records the date of each criminal conviction in British Columbia. ⁵ The outcome of interest in the current study was violent convictions, defined as any instance of hands-on violence (e.g., assault, robbery, and homicide). For each participant, violent convictions were coded for each year of age, beginning at age 18, allowing comparisons of convictions at the same ages across all birth cohorts. However, the calendar year of each conviction varied. For example, for participants born in 1979, convictions at age 18 occurred in 1997–1998. For participants born in 1997, convictions at age 18 occurred in 2015–2016. On average, participants were 35.69 years old (SD = 4.31) at the end of the follow-up period and had accrued an average of 1.87 violent convictions (SD = 2.41). About 60% of the sample incurred at least one violent conviction. The probability of a violent conviction at each age is shown in Figure 1. During the follow-up period, 9.1% of participants died (n = 47), and 8.2% emigrated outside of British Columbia (n = 42). For these participants, the date of emigration or death represented the end of the follow-up period. For all others, the follow-up period extended to Winter 2023. Reoffending could lead to incarceration, which in turn limited opportunities to commit new violent offences. This was accounted for by creating an exposure variable that represented the proportion of each year of age spent in the community rather than in custody (see Figure 1). Participants averaged 38.47 (SD = 45.15) months in custody during the follow-up period.

Figure 1.

Distribution of the probability of incurring a violent conviction (top panel) and of the number of months incarcerated (bottom panel) at each year of age.

Psychopathy

Psychopathy was assessed using the PCL:YV, a 20-item expert-rating scale (Forth et al., 2003). Items are rated on a three-point scale (0 = item does not apply; 1 = item applies somewhat; 2 = item definitely applies) based on evidence that the trait is characteristic of the individual across time and social contexts. PCL:YV items reflect interpersonal (e.g., impression management and manipulation), affective (e.g., shallow affect and lack of remorse), lifestyle (e.g., stimulation seeking and parasitic orientation), and anti-social behaviour (e.g., early behaviour problems and poor anger control) domains. RAs rated the PCL:YV following a semi-structured interview and a review of participants’ correctional files. Participants were, on average, 16.48 years old (SD = 1.39) at assessment. To evaluate interrater reliability, six RAs were assigned to three pairs to score 30 randomly selected cases. Each pair interviewed and rated 10 cases. For each pair, both RAs participated in the interview, with one RA taking the lead. Ratings were completed independently. Based on intraclass correlation coefficients from a two-way random effects model for absolute agreement, reliability was excellent (intraclass correlation coefficient = 0.92). Main analyses focused on PCL:YV total scores (M = 21.17; SD = 6.50; polychoric ordinal alpha = 0.84). Given concerns about the PCL:YV being oversaturated with items indicating prior antisocial and criminal behaviour, we also report findings using the three-factor model, which includes only the Interpersonal, Affective, and Lifestyle facets. Differences in findings were not expected because of the strong correlation between total scores and the three-factor model (r = .929; p < .001).

Birth Cohorts

Birth cohort membership was operationalized in two ways to assess the sensitivity of the findings. First, birth cohort status was treated as a continuous variable ranging from 1979 to 1997 (M = 1985.48; SD = 4.39). However, macro-level events and historical contexts do not necessarily change consistently over time. For example, unemployment (bottom-left of Figure 2) fluctuated across cohorts. When participants born in 1979 turned 18, unemployment was approximately 9%. When participants born in 1989 turned 18, unemployment had declined to about 4%. However, when participants born in 1997 turned 18, unemployment had increased to about 6%. Since the relationship between birth cohort membership and violence might not be linear, we also created four birth cohort groups: 1979–1983 (n = 228), 1984–1988 (n = 122), 1989–1993 (n = 141), and 1994–1997 (n = 24). These categories resembled the cohort groups compared in analyses of PHDCN+ data (Neil & Sampson, 2021). Although ISVYOS participants and PHDCN+ participants were born in similar eras, PHDCN+ participants were drawn from a single city, whereas ISVYOS participants were drawn from a single province. Unlike the PHDCN+, the ISVYOS lacked neighbourhood-level data that could provide a more nuanced understanding of the structural contexts in which participants were embedded.

Figure 2.

Macro-historical environmental characteristics in the Province of British Columbia, Canada when participants were age 18.

Notes. All participants were age 18 for purposes of these historical comparisons; however, participants turned age 18 at different years. The top left panel helps visualize when data were recorded for participants at age 18. Data were derived from the Statistics Canada website. Police incidence rate and rate of police officers is per 100,000 persons. Rate of post-secondary education is per 10,000 persons.

Figure 2 displays macro-level structural conditions in the province of British Columbia when ISVYOS participants were age 18. Figure 2 is not intended to suggest that these conditions persisted throughout adulthood, but rather to show that social, economic, and legal environments differed across historical periods. Overall, recent cohorts came of age under conditions associated with lower offending. For example, among participants born in 1979, who turned 18 in 1997, the post-secondary education rate was about 4,400 per 10,000. For those born in 1997, who turned 18 in 2015, the rate was about 6,000 per 10,000. Although older cohorts also experienced rising post-secondary enrolment rates, they were older (e.g., age 36 in 2015 for the 1979 cohort) and therefore less likely to be affected by changes in enrolment than those entering early adulthood. As another example, participants born in the 1980s turned age 18 between 1998 and 2007, when police incidents were much higher than during the period when participants born in the 1990s were age 18 (beginning in 2008). Although older cohorts also experienced declining police incidents, these changes occurred in their late 20s, when offending risk typically declines according to the age-crime curve.

Legislative changes further distinguished cohorts: older cohorts were adjudicated under the Young Offenders Act (YOA; 1984–2003), whereas contemporary cohorts were adjudicated under the Youth Criminal Justice Act (YCJA; 2003–present). Youth incarceration rates in Canada were substantially lower following the implementation of YCJA than under the YOA. The YCJA permitted the use of custody sentences only when the index offence met certain severity criteria and the judge was able to justify why all non-custody sentencing options were unsuitable for protecting the public (Bala et al., 2009). Unlike the YOA, the YCJA placed greater emphasis on rehabilitation and service provision as the principal goals of incarceration, and deterrence was purposefully excluded from its sentencing principles (Bala et al., 2009).

Family Adversity

Although birth cohort status indicates exposure to broader social conditions, family circumstances reflect a more immediate social environment. We included a measure of family adversity because the impact of macro-level structural conditions on individual development may depend on family circumstances (Elder, 1998). During the intake interview, participants were asked whether a family member had a: (1) problem with alcohol, (2) problem with drugs, (3) history of physical abuse, (4) history of sexual abuse, (5) criminal record, or (6) mental illness. These indicators were summed to create a cumulative index of family adversity (M = 2.87; SD = 1.61; tetrachoric ordinal alpha = 0.72).

Demographic Characteristics

Gender and ethnicity were self-reported by participants to RAs during the intake interview. Among this sample, 82.7% self-identified as male (n = 426) and 17.3% self-identified as female (n = 89). Regarding ethnicity, 59.6% of participants self-identified as White (n = 307), 25.4% as Indigenous (n = 131), and 15.0% as members of a non-Indigenous minority group (n = 77). The latter group includes participants who identified as Black, East Asian, Southeast Asian, Hispanic, Middle Eastern, or reported that a single ethnicity inadequately represented their background. These groups were combined because each individual group comprised less than 5% of the sample. This aggregation does not imply homogeneity across these diverse groups.

Prior Violence

To account for the relationship between past and future violence, we measured the number of violent convictions incurred between age 12 (the age of criminal responsibility in Canada) and age 17 (the year prior to the start of the follow-up period; M = 1.68; SD = 1.77; R = 0–10).

Analytic Strategy

Binomial logit generalized estimating equation (GEE) models were used to examine between-group differences in the probability of a violent conviction at each year of age (ages 18–40). Like logistic regression, the logit link function models the log-odds of the outcome as a linear combination of the predictor variables. Unlike logistic regression, GEE models take fuller advantage of longitudinal data by examining at each year of age whether the participant incurred a violent conviction. GEEs also account for the clustering of convictions within individuals (Hardin & Hilbe, 2007). ⁶ Analyses estimated the average association between PCL:YV scores and violent convictions across each year of age during the follow-up period, increasing confidence that the findings are robust across developmental stages.

Interaction effects were examined to determine whether the relationship between PCL:YV scores and violence varied by birth cohort. This analysis allowed us to examine whether two participants with identical PCL:YV scores nevertheless had different probabilities of violence because they were born during periods with distinct social, economic, and legal conditions. Conducting interaction analyses was necessary rather than simply controlling for birth cohort, as the relationship between birth cohort and violence may vary depending on whether a person has higher or lower levels of psychopathy traits.

Results

Model 1 of Table 2 presents the main effect GEE model with birth cohort measured as an interval-level variable. A one-unit increase in the year a participant was born was associated with a 5% higher odds of incurring a violent conviction in a given year of the follow-up period. Thus, participants born recently were more likely to incur a violent conviction even though they came of age during periods with lower crime rates, lower unemployment, higher post-secondary enrolment, and more rehabilitative approaches to youth justice.

Table 2.

GEE Models With Violent Conviction per Year of Age (Ages 18–40) as the Outcome of Interest (n = 515).

Variable	Model 1	Model 2
	Main Effects OR (Robust SE)	Interaction Term OR (Robust SE)
Psychopathy
PCL:YV total scores	1.03** (0.01)	1.03 (0.01)**
Birth cohort measure
Birth cohort (1979–1997)	1.05*** (0.02)	1.10 (0.05)
Interaction term
PCL:YV total score*birth Cohort (1979–1997)	-	1.00 (0.002)
Immediate social environment
Family adversity scale	1.12** (0.05)	1.12** (0.05)
Control variables
Violent convictions (Ages 12–17)	1.12*** (0.03)	1.12*** (0.03)
Male participanta	-	-
Female participant	0.26*** (0.06)	0.27*** (0.06)
Whitea	-	-
Indigenous	1.72*** (0.24)	1.73*** (0.24)
Non-Indigenous minority	0.91 (0.18)	0.92 (0.18)
Model fit	χ2 (7) = 105.49, p < .001	χ2 (8) = 108.16, p < .001

Note.^aDenotes reference category.

* p < .05; **p < .01; ***p < .001.

For the interaction term, birth cohort was mean centered to avoid the main effect of PCL:YV scores reflecting participants born in year “0000.” GEE=generalized estimating equation; OR = odds ratio; PCL:YV= Psychopathy Checklist: Youth Version; SE = standard error.

With respect to psychopathy, model 1 of Table 2 shows that a one-unit increase in PCL:YV total score was associated with significantly higher odds of a violent conviction within a given year of age (odds ratio [OR] = 1.03). Some researchers have suggested that the strength of the relationship between psychopathy and offending outcomes is weak. This assertion was based on comparing ORs for PCL total scores to ORs for dichotomous risk factors. This comparison is potentially misleading because ORs for interval and dichotomous predictors are interpreted differently. Psychopathy was conceptualized as a dimensional construct. Accordingly, analyses focused on PCL:YV total scores, with effect sizes interpreted per one-unit increases on the 40-point PCL:YV scale. Although clinically meaningful comparisons are based on potentially arbitrary cut-points, they illustrate that differences across the distribution are substantially larger than the one-unit OR suggests. For example, participants who scored one standard deviation above the mean on the PCL:YV (i.e., ≈28) had 1.98 times the odds of incurring a violent conviction during a given year of age compared to those who scored one standard deviation below the mean on the PCL:YV (95% confidence interval [CI] = 1.31–2.99). In a bivariate GEE analysis, the OR was smaller (OR = 1.54; 95% CI = 1.24–1.91) than in the analysis that included covariates. Accounting for gender was especially important given differences in offending among the highest-risk girls and highest-risk boys (Gushue & McCuish, 2021).

Model 2 of Table 2 includes an interaction term between PCL:YV total score and birth cohort, which was not significant. Figure 3 illustrates the lack of variation in the slope of the relationship between psychopathy and violence across birth years. To more directly address risk assessment questions about whether the relationship between high levels of psychopathy and violence is similar across birth cohorts, we examined marginal effects for participants who scored either at least 1 SD above (≈28) or at least 1 SD below (≈14) the mean PCL:YV total score (≈21). For participants scoring at least 1 SD above the mean, predicted probabilities were similar across cohorts (i.e., the CIs overlapped; see Figure 4, top panel). However, for participants scoring at least 1 SD below the mean, recent cohorts had a higher predicted probability of a violent conviction than participants in the 1980 birth cohort (Figure 4, bottom panel).

Figure 3.

The average probability of incurring a violent conviction at a given year of age (ages 18–40).

Notes. The relationship between increases in the Psychopathy Checklist: Youth Version (PCL:YV) scores and the probability of incurring a violent conviction at a given year of age did not significantly vary depending on the year in which a participant was born. The 95% confidence intervals are represented by the lower and upper bounds of each point on the graph.

Figure 4.

The average probability of incurring a violent conviction at a given year of age (ages 18–40) among participants with a Psychopathy Checklist: Youth Version (PCL:YV) score one standard deviation above the mean (top panel) or below the mean (bottom panel).

Notes. The 95% confidence intervals are represented by the lower and upper bounds of each point on the graph. A score that is one standard deviation above the mean reflects a score of approximately 28 or higher. A score that is one standard deviation below the mean reflects a score of approximately 14 or lower.

Macro-level structural conditions do not necessarily change linearly over time. As shown in Figure 2, unemployment rates in British Columbia declined from 1998 (age 18 for the 1979 cohort) to 2008 (age 18 for the 1990 cohort) but increased by 2009 (age 18 for the 1991 cohort). To avoid assuming linearity, another GEE model was used to examine birth cohort groupings (see Table 3). Participants in the oldest cohort (1979–1983) had significantly lower probabilities of violent convictions than participants in the 1984–1988 and 1989–1993 birth cohorts. Higher PCL:YV total scores were associated with significantly greater odds of a violent conviction within a given year of age.

Table 3.

GEE models With Violent Conviction per Year of Age (Ages 18–40) as the Outcome of Interest (n = 515).

	Model 1	Model 2
Variable	Main Effects OR (Robust SE)	Interaction Term OR (Robust SE)
Psychopathy
PCL:YV total scores	1.03** (0.01)	1.04** (0.01)
Birth cohort measure
1979–1983 birth cohorta	-	-
1984–1988 birth cohort	1.36* (0.21)	2.05 (1.31)
1989–1993 birth cohort	1.65*** (0.25)	2.30 (1.22)
1994–1997 birth cohort	1.48 (0.47)	1.31 (2.41)
Interaction term
PCL:YV total score*1979–1983 birth cohorta	-	-
PCL:YV total score*1984–1988 birth cohort	-	0.98 (0.03)
PCL:YV total score*1989–1993 birth cohort	-	0.99 (0.02)
PCL:YV total score*1994–1997 birth cohort	-	1.00 (0.07)
Immediate social environment
Family adversity scale	1.12** (0.05)	1.15** (0.05)
Control variables
Violent convictions (Ages 12–17)	1.12*** (0.03)	1.12*** (0.03)
Male participanta	-	-
Female participant	0.27*** (0.06)	0.27*** (0.06)
Whitea	-	-
Indigenous	1.74*** (0.24)	1.74*** (0.24)
Non-Indigenous minority	0.90 (0.18)	0.91 (0.18)
Model fit	χ2 (9) = 109.72, p < .001	χ2 (12) = 112.25, p < .001

Note. ^aDenotes reference category.

*p < .05; **p < .01; ***p < .001.

GEE=generalized estimating equation; OR = odds ratio; PCL:YV= Psychopathy Checklist: Youth Version; SE = standard error.

Model 2 of Table 3 added interaction terms between PCL:YV scores and birth cohort groups. The interaction terms were not significant, indicating that the psychopathy–violence relationship did not differ by birth cohort. Marginal effects analyses showed that among participants scoring one standard deviation below the mean on the PCL:YV, the probability of a violent conviction was significantly lower in older cohorts. In contrast, among participants scoring one standard deviation above the mean, the probability of a violent conviction did not differ across birth cohorts. Because these findings overlapped with those presented in Figures 3 and 4, the output is not shown but is available from the first author upon request.

Supplemental Analyses

Analyses were repeated using the three-factor model, which avoids tautological concerns when predicting violence by excluding items from the Antisocial factor. Substantive interpretations remained unchanged: (a) higher scores on the three-factor model significantly increased the odds of a violent conviction during a given year of age and (b) this relationship did not vary by birth year or birth cohort grouping. We did not examine the relative contribution of individual factors based on the view that they do not adequately represent the broader psychopathy construct (Salekin, 2016).

The primary analyses examined the probability of a violent conviction at each year of age up to age 40. However, participants from recent birth cohorts had not reached their late 30s. As a result, older cohorts contributed a greater number of person-period observations to the models, which could potentially have influenced the estimated ORs. Additional analyses restricted the follow-up period to ages 18 to 25 to ensure that participants born in different years contributed the same number of person-period observations. This narrower age range was also important because aggregating observations across multiple developmental stages might mask birth cohort effects that are specific to a single developmental stage (Steffensmeier et al., 1992). When restricting the follow-up period to ages 18 to 25 (i.e., emerging adulthood), the substantive conclusions remained the same.

Birth cohort membership is a proxy for exposure to macro-level structural conditions. The structural conditions shown in Figure 2 were included in a series of GEE models. These models tested whether a specific indicator (e.g., the unemployment rate when a participant was age 18) moderated the association between PCL:YV scores and violent conviction. None of the macro-level indicators moderated the association between PCL:YV scores and violence. Birth cohorts also differed with respect to the youth justice legislation under which participants were adjudicated (i.e., YOA [1984–2003] vs. YCJA [2003–present]). Nevertheless, the relationship between PCL:YV total scores and violent conviction did not differ according to exposure to youth justice legislation.

Discussion

Broader social conditions shape individual development (Bronfenbrenner, 1979; Elder, 1975; Ryder, 1965). Yet, the implications of structural influences for criminal behaviour have been understudied in forensic psychology and psychiatry (Crocker & Leclair, 2025; Verona & Fox, 2025). Addressing such questions is relevant for forensic research and practice given the frequent use of psychopathy assessments to inform risk classification and justice system decision-making (Viljoen et al., 2010). The current study examined whether the relationship between psychopathy and violent convictions varied across birth cohorts exposed to different structural conditions. We used data from 535 incarcerated youth assessed with the PCL:YV and followed from adolescence into adulthood. Participants were recruited between 1998 and 2011 and were born between 1979 and 1997. Those who were born recently came of age under more favourable social conditions in British Columbia and under less punitive youth justice policies (Bala et al., 2009).

We found that annual probabilities of violent convictions varied across birth cohorts, but the association between psychopathy and violence was consistent across birth cohorts. If the predictive validity of RAIs varies across birth cohorts, the stability of risk classifications becomes uncertain, complicating the use of “high-risk” designations in justice system decision-making (Montana et al., 2023). However, our findings did not support this possibility. Participants born recently were significantly more likely to be convicted of violent offences during the follow-up period. Yet, the relationship between higher PCL:YV scores and violence did not vary as a function of birth cohort. We consider the implications of our findings for birth cohort research and for psychopathy research.

Implications for Birth Cohort Research

Research using data from the PHDCN+ reported that participants born recently were significantly less likely to incur arrests during the follow-up period (Neil & Sampson, 2021). Similar patterns were reported in Finland (Sivertsson et al., 2025). In contrast, we found that participants born recently were more likely to incur violent convictions. Despite these contrasting findings, we do not view them as contradictory. Rather, the differences highlight the importance of “situated knowledge” (Simandan, 2019) and the difficulty of generalizing findings across Canada and the United States, across incarcerated youth and general population samples, and across different risk factors and offending outcomes. Although ISVYOS and PHDCN+ participants were born in the 1980s and 1990s, cohort differences may be larger in the United States, where policies and social conditions have historically shifted more dramatically (Tonry, 2013). Canada may be closer to what Swidler (1986) described as a “settled period,” in which changes to macro-level social conditions unfold gradually, making it difficult to detect independent effects of cultural or structural factors on individual behaviour. Canadian policy changes have generally occurred more slowly than those in the United States (Doob & Webster, 2016; Tonry, 2013). Detecting birth cohort differences in the association between individual-level risk factors and violence may require examining cohorts born farther apart than those examined in the current study.

Participants born in the 1990s came of age during social, economic, and legal conditions that were conducive to lower levels of offending. Yet, compared to older cohorts, those born in the 1990s were more likely to be convicted of violent offences in adulthood. This finding could be interpreted as evidence that youth are becoming more violent despite more favourable social conditions. This would be a misinterpretation. Instead, our results highlight the importance of considering period effects alongside birth cohort effects. Specifically, in the current study, Canada's transition from the YOA to the YCJA represented a unique period effect. Although the ISVYOS consistently sampled incarcerated youth, the transition from the YOA to the YCJA reduced the likelihood that lower-risk youth would receive custody sentences. Consequently, lower-risk members of more recent birth cohorts were less likely to be included in the study, even though recruitment procedures remained unchanged.

Our study focused more narrowly on the relationship between psychopathy and violence. Nevertheless, these findings are also useful as a case study illustration of why research comparing birth cohorts should consider period effects. The ISVYOS used the same data collection procedures and sampled from the same locations throughout the study period. Yet, due to legislative changes, the population under study changed in ways that were independent of birth cohort effects. Supplemental analyses supported this interpretation. There was no longer a significant positive association between birth year and violent convictions after accounting for whether participants were adjudicated under the YOA or YCJA. These findings suggest that, among participants who scored low on the PCL:YV, the higher levels of violence observed for more contemporary cohorts were no longer evident after accounting for youth justice legislation. Instead, among participants with low PCL:YV scores, those who were members of older birth cohorts were less likely than participants from more contemporary birth cohorts to report other risk factors (e.g., substance use, gang involvement, and family adversity; see McCuish et al., 2021). This difference in risk profile could reflect legislative changes and changes in who enters the study population as opposed to differences between birth cohorts. The potential for legislative changes to impact the composition of custody populations has implications for psychopathy research and risk communication.

Practice and Policy Impact: Implications for Psychopathy Research

Treatment effect heterogeneity, sometimes referred to as life-course interdependence, describes how people faced with similar situations may respond differently due to differences in personality traits. The amplification perspective on treatment effect heterogeneity (Caspi & Moffitt, 1995; Hirtenlehner & Schulz, 2021) suggests that people with psychopathy traits are more likely than others to experience changes in their risk of offending following changes in their social environment. The social resistance perspective on treatment effect heterogeneity (McCuish & Lussier, 2023) suggests that people with psychopathy traits are less likely than others to experience changes in their risk of offending following changes in their social environment. Results aligned more closely with the social resistance perspective. Overall, more contemporary birth cohorts had higher probabilities of incurring violent convictions. However, among youth with high PCL:YV scores, the probability of a violent conviction in a given year of age did not differ across birth cohorts. In other words, unlike participants with lower PCL:YV scores, participants with higher PCL:YV scores appeared to be less affected by differences in macro-level structural conditions and changes to youth justice legislation. Consequently, their probability of a violent conviction in a given year of age was similar across birth cohorts. Response modulation refers to the tendency for people with psychopathy traits to exhibit reduced affective reactivity (Hoppenbrouwers et al., 2015). Although not directly tested in the current study, reduced affective responsiveness to the social environment could help explain why the probability of violence did not fluctuate across different macro-level contexts.

In Canada, largely due to legislative reform, youth born in the 1980s were more likely to be incarcerated than youth born in the 1990s (Bala et al., 2009). Psychopathy and RAI research should account for policy change, as legislative context, and the birth cohorts exposed to it, can have implications for interpreting the psychopathy–violence relationship. In the current study, the probability of violent conviction among incarcerated youth with high PCL:YV scores remained consistent across birth cohorts and legislative contexts. By contrast, following the transition from the YOA to the YCJA, the incarcerated population of youth with low PCL:YV scores appeared to have a more severe overall risk profile. Accordingly, in the context of declining youth incarceration rates, participants from recent cohorts (incarcerated under the YCJA) with low PCL:YV scores had higher predicted probabilities of violent convictions than participants from older cohorts with similarly low PCL:YV scores (incarcerated under the YOA).

The higher probabilities of violent convictions for incarcerated youth from recent birth cohorts who scored low on the PCL:YV may reflect shifts in the composition of custody populations following legislative reform. Consider two hypothetical subgroups of youth scoring low on the PCL:YV: (1) those with few or no other risk factors (“low psychopathy, low-risk”) and (2) those with additional risk factors such as gang involvement (“low psychopathy, high-risk”). When jurisdictions reduce their reliance on custody, “low psychopathy, low-risk” youth are disproportionately diverted to community-based sentences, leaving a larger proportion of “low psychopathy, high-risk” youth within custody populations. This selection process may explain why recent birth cohorts, who were adjudicated under the YCJA, showed higher probabilities of a violent conviction than their counterparts adjudicated under the YOA. In effect, it does not seem that more contemporary birth cohorts are more likely to be violent. Instead, legislative reforms may have reshaped the composition of youth custody populations such that lower-risk youth from more contemporary cohorts were less likely to experience incarceration.

Greater consideration should be given to how changes in the composition of the population under study, whether through changes to legislation or the inclusion of different birth cohorts, may lead to selection-induced attenuation bias. In this context, we refer to selection-induced attenuation as a weakening of the magnitude of an effect size between a risk factor and an outcome that arises because the composition of the sampled population changed as opposed to a change in the underlying relationship. To illustrate, even though the probability of a violent conviction among participants with high PCL:YV scores remained unchanged, the gap in the probability of violent conviction between incarcerated youth with high PCL:YV scores and those with low PCL:YV scores narrowed. In the current study, this narrowing may be explained by the fact that youth incarcerated under the YCJA presented with a wider range of risk factors than youth incarcerated under the YOA (McCuish et al., 2021). In jurisdictions that have reduced their reliance on custody sentences, one implication may be that, over time, the base rate of violent convictions increased for incarcerated youth scoring low on the PCL:YV. Accordingly, although the probability of violence among people with high PCL:YV scores may remain unchanged, the association between psychopathy and violence in incarcerated samples may appear weaker because of changes in the composition of custody populations. ⁷ Rising base rates of violence among youth scoring low on the PCL:YV can attenuate effect sizes even if the base rates of violence among youth scoring high on the PCL:YV remain unchanged. This possibility has been overlooked in some recent critical commentaries on psychopathy.

Limitations and Future Research

Our sample included over 500 youth, which is typically sufficient to avoid sign errors (e.g., a positive coefficient in the sample when the coefficient in the population is negative; Castillo et al., 2026). Moreover, interaction analyses were theory-driven rather than post hoc attempts to explain non-significant main effects. However, tests of interaction effects ideally use larger samples to increase confidence in coefficient stability (e.g., reducing the likelihood that a positive coefficient changes from significant to non-significant; Castillo et al., 2026). We found that birth cohort did not moderate the relationship between high scores on the PCL:YV and violent convictions. However, our findings should not be interpreted as evidence that birth cohort membership is unimportant more broadly. The current study's generalizability is limited due to its focus on the psychopathy–violence relationship among youth born in the 1980s and 1990s who were incarcerated in British Columbia, Canada. Future research can address these generalizability concerns by examining constructs other than psychopathy (e.g., gang involvement, RAIs, and gender), different jurisdictions, and birth cohorts from other generations. Finally, future research on psychopathy and birth cohorts should consider other measures of offending (e.g., dangerousness; Gillespie et al., 2023) and other consequences associated with psychopathy (e.g., negative health outcomes; Rose et al., 2025).

Data Availability Statement

Data used in this study may be accessed by completing a confidentiality form available from the first author. An in-progress codebook for measures from the ISVYOS is available at: https://www.researchgate.net/publication/347516754_Working_Document_ISVYOS_Codebook