Latent Profile Analysis of Personality Dimensions Among Juvenile Offenders: Relevance for Predicting Offending Seriousness

Abstract

There is a burgeoning base of research identifying personality as a predictor of offending. However, research has focused on personality dimensions, rather than full personality profiles as predictors. The present study utilized the Pathways to Desistance data to examine the relationship between personality profiles and offending. This sample comprised 1,354 juvenile offenders followed during the study period of 2000 to 2010. Latent profile analysis was used to identify patterns across dimensions to elucidate personality profiles. Negative binomial regression was used to examine profiles as predictors of offending. Results indicated that a two-profile model fit the data. Participants assigned to the Undercontrolled profile (high in neuroticism and low in all other dimensions) engaged in more serious offending than the Resilient profile.

Keywords

personality latent profile analysis juvenile offenders

Introduction

There is a burgeoning literature focused on the importance of personality in the field of criminology. In the past decade, it has reached the point where meta-analyses are able to identify key themes within the literature to better understand which facets of personality are important for understanding offending (Jones et al., 2011; Vize et al., 2018). While this progress has demonstrated the great importance of personality for things like psychoticism and antisocial behavior (Decuyper et al., 2013; Miller & Lynam, 2001; Vaughn et al., 2008), there remain some gaps in understanding the manifestation of personality and how differential clustering of personality dimensions may help to understand offending. While past research has indicated the relevance of individual personality dimensions predicts delinquency and offending (Lynam et al., 2003; Miller & Lynam, 2001), the relevance of full personality profiles has, thus far, been understudied. While individual dimensions of personality may be relevant, a lack of understanding of how they cluster with other personality dimensions to form full profiles may obscure their true effects on offending. It may be that the personality-offending relationship only exists for these individual dimensions when observed with other distinct personality dimensions as part of a profile. Despite the potential for this, there is a dearth of research focused in this area. The present study sought to clarify this relationship further in a novel manner.

The Five-Factor Model of Personality and Clustering of Personality Dimensions

While numerous personality scales exist (Cloninger et al., 1994; Hogan, 1995; Lee & Ashton, 2004), perhaps the most prominent is the NEO five-factor personality inventory. This scale was developed in response to a focus on traits as cognitive categories and a lack of focus on the actual systematic description of said traits, that is, personality. The NEO five-factor typology of personality grew out of the need to systematically describe the actual components of personality. The scale rose to prominence as it was validated across both observers and instruments as a strong framework for describing underlying cognitive traits as actual personality dimensions (McCrae & Costa, 1987). This scale captures the “big five” personality dimensions of agreeableness, conscientiousness, openness to experience, extraversion, and neuroticism (Costa & McCrae, 1992). Each of these dimensions represents a distinct component of personality which individuals demonstrate to a greater or lesser extent. Agreeableness refers to the degree to which someone is likable, pleasant, and harmonious as it pertains to social relations with others. Conscientiousness is the degree to which individuals are able to control impulses, delay gratification, and be future goal oriented. Openness to experience is related to an individual’s capacity for permeability of cognitive and affective states when exposed to novel experiences and ideas. This dimension is also related to capacity for self-reflection. Extraversion refers to the degree to which individuals have a tendency to demonstrate positive affect, decisiveness in decision-making, assertiveness, and a desire for social attention. Finally, neuroticism refers to a consistent disposition associated with negative affect, such as anxiety, depression, anger, and guilt. Neurotic individuals may also have a general inability to manage stress and relieve negative emotions when aroused. While past research has indicated that these dimensions of personality may be important for understanding antisocial behavior and aggression individually (Meier et al., 2006; Robinson, 2007; Vize et al., 2018), it may also be highly relevant to attain a broader understanding of how the differential clustering of variance in each of these dimensions may come together to form a complete personality profile. In doing so, this would allow for a more holistic understanding of how certain personality dimensions influence crime together, rather than individually. While this type of profiling has been the subject of past research, there also remain additional areas that warrant further attention.

Clustering and Personality Profiles

Past research has examined personality profiles extensively. This type of research has generally focused on examining correlation between each personality dimension to identify latent personality profiles and exhibit average levels of each dimension within each profile. Results from studies generally vary in the number of distinct personality profiles that are identified, but there are some similarities in some of the commonly identified profiles. One common profile is characterized by high levels of neuroticism and low levels of all other personality dimensions, indicating a negative correlation between neuroticism and other personality dimensions (Chen et al., 2013, 2017; Morgan et al., 2017; Zhang et al., 2015). Similarly, a group demonstrating the opposite is often observed also, with low levels of neuroticism and relatively high levels of all other personality dimensions (Chen et al., 2017; Claes et al., 2006; Kinnunen et al., 2012; Morgan et al., 2017; Specht et al., 2014). This provides further indication of the negative correlation between these personality dimensions. While identification of commonalities across studies is important, it is also highly relevant to note the ways that these studies differ in their findings and the potential reasons why they may differ across groups.

Research has indicated that the identification of the number and types of profiles may vary cross-culturally (Kinnunen et al., 2012; Specht et al., 2014). Specht et al. (2014) found that the number of distinct behavior profiles differed between an Australian and German sample, as did the general composition of the behavior dimensions of the identified profiles between nationality groups. This type of variance has also been observed between distinct clinical populations as well (Claes et al., 2006, 2013; Thomas et al., 2014). All of this provides further indication of the importance of investigating differences within distinct subgroups. These types of analyses have yet to be conducted among juvenile offenders. This is a major omission for the field of criminology, as these individuals have already demonstrated propensity for offending and little is known about whether or not distinct personality profiles may help to understand etiology. The present study sought to fill this gap in the literature by examining the clustering of personality dimensions among this population and examine the relevance of profiling personalities in this way for predicting offending.

Importance of Personality for Predicting Offending

Numerous studies have implicated distinct personality dimensions as being relevant predictors of offending (Lynam et al., 2003; Miller & Lynam, 2001; Vize et al., 2018). For example, low levels of agreeableness and conscientiousness have been demonstrated to increase risk for aggressive and antisocial behavior (Vize et al., 2018). Low agreeableness may increase offending propensity due to a lack of harmony in relations with others potentially leading to tumultuous social situations and interpersonal relationships. By definition, conscientiousness is related to self-control and low self-control is a robust risk factor for offending (Hay & Meldrum, 2015; Turanovic et al., 2015). Furthermore, the lack of future orientation that is characteristic of low levels of this personality dimension may restrict an individual’s ability to see the potential consequences of criminal behavior. Neuroticism may also play a large role, as the lack of emotional self-regulation and increased negative emotions that are characteristic of this personality dimension are consistent with Agnew’s (1992) postulations regarding the roles that strain exposure and the arousal of negative affect have for increasing risk for offending and the empirical evidence supporting these relationships (Cudmore et al., 2017; Moon & Morash, 2017; Snyder et al., 2016). Individuals who demonstrate high degrees of neuroticism would be expected to react to strain with high levels of anger, thus precipitating a need to cope and increasing risk for offending. Similarly, the consistent positive affect characteristic of highly extraverted individuals would seem to clash with this last point, indicating that individuals who demonstrate low levels of extraversion may be at-risk for demonstrating negative affect and elevated offending propensity. Finally, openness to experience may also be related to offending. Being closed to new ideas and cognitive change may result in interpersonal conflict when exposed to such experiences, potentially increasing risk for aggressive offending in response. It is perhaps with this last point that the importance of full personality profiling for understanding offending is clearest. While being closed off from new experiences and changes may not be a risk factor for criminality in and of itself, combining that with high neuroticism and/or low agreeableness may result in increased negative affect and conflict does indeed increase offending risk.

It is here that it is clear that the convergence of different dimensions, not just the individual dimensions explored by past research, becomes important for understanding the role that personality plays for predicting offending. As mentioned above, there is wide variance in the number and types of personality profiles identified based on the distinct subgroup of interest. Profiles high and low in neuroticism may be particularly important here given that, by definition, juvenile offenders have already demonstrated risk for offending. Profiles that are high in neuroticism may be prevalent among this subpopulation and may act to predict increased recidivism risk. Alternatively, there may also be other profiles that are present that may indicate diminished recidivism risk. If profiles like this are identified, then this would indicate that juvenile offenders with certain personality clusters may not warrant as much attention for rehabilitation. Despite this potential that personality has demonstrated, there has yet to be an examination of the relevance of holistic personality profiles for predicting offending behavior. There also is a dearth of research focused on how these latent profiles may help to predict offending behavior. This study sought to bridge these gaps in the extant literature.

Method

Specific Aims

This study sought to accomplish several important goals. First, this study focused on identifying distinct clusters of variance in personality dimensions. By examining correlation between five distinct personality dimensions (neuroticism, agreeableness, conscientiousness, extraversion, and openness to experience), differential clustering of average levels of each dimension could be identified. These identified personality profiles were then examined as predictors of offending. This would allow for greater understanding of entire personality profiles of individuals predict offending, rather than simply individual personality dimensions; as this has been the major focus of past research.

Procedures

The present study utilizes data from the first and fifth waves of the Pathways to Desistance study. This dataset comprised responses from 1,354 juvenile offenders who had recently been adjudicated for a felony or serious misdemeanor offense prior to baseline measurements.¹ Serious misdemeanors which qualified participants for inclusion were sexual assault and weapons-related charges. Offenses which qualified participants for the study had to have been committed when participants were between the ages of 14 and 17 and all participants were between the ages of 14 and 19 at baseline measurement. Data collection resulted in a total of 11 data points for each participant across 84 months, with the study period spanning from 2000 to 2010. Recruitment sites were located in Maricopa County, Arizona and Philadelphia, Pennsylvania. Of all potential participants, 20% declined the opportunity to be a part of the study. Retention of participants at the final data point was 83.2% of the original sample. The proportion of the sample that consisted of male drug offenders was capped at 15% to maintain heterogeneity in the sample based on these characteristics.

Data utilized in the present study were collected via participant self-report using computer-assisted interview technology. Members of the research team met with participants in convenient locations (participants’ homes, libraries, criminal justice facilities, etc.) to conduct interviews. Participants were provided with a laptop computer which they used to manually input responses to verbal prompts from the interviewer. Past research has indicated that computer-assisted interviewing like this does indeed increase completeness of reporting as it pertains to deviant behavioral outcomes (O’Reilly et al., 1994).

Variables

Offending

The main dependent variable in this study was offending seriousness. This was measured using a variable capturing variety of offending at the fifth wave of data collection. Past research has indicated that variety measures like this are good proxy measures of offending seriousness and do not suffer from similar measurement bias issues often observed with frequency measures (Brame et al., 2004; Steinberg et al., 2015). Beyond this, past research has indicated a high correlation between offending variety and arrest records in the Pathways to Desistance data (Brame et al., 2004). This measure was calculated as the quotient of the total number of types of offenses participants reported engagement in during the observation period divided by the total of number of offenses that they possibly could have engaged in. While the original coding of the variable resulted in proportional scores potentially ranging from “0” to “1,” this necessitated some transformation to model correctly. All scores were multiplied by 100 to allow for modeling of the scores using the modeling strategy that best suited distributional concerns of the variable. Higher scores indicated more serious levels of offending.

Personality

Personality variables were used as independent variables predicting offending seriousness. The NEO Five-Factor Personality Inventory Short Form was used to measure each of the following dimensions of personality: neuroticism, extraversion, openness to experience, agreeableness, and conscientiousness. This is a validated tool that is widely used to tap the “big five” dimensions of personality (Carvalho & Nobre, 2019; Claes et al., 2014; Van Dam et al., 2005). This instrument consisted of 120 individual ordinal items which asked participants to rate the degree to which a given statement is true of themselves on a 5-point Likert-type scales, with higher scores indicating greater agreement with the statement. A mean of the items corresponding to each personality dimension was calculated for each of the personality dimensions. This resulted in a single score for each participant on each dimension. This variable was measured at the fifth wave of data collection.

Control variables

A number of control variables were also selected to be included in model estimation to mitigate the risk of bias. The first of these variables was participants’ gender. This was measured at baseline, delineating male and female participants using a binary measure (0 = Male; 1 = Female). This variable was included as a control variable because past research has indicated that males generally demonstrate higher risk for offending (Loeber et al., 2017).

Another control variable included in analyses was racial identity. This is because past research indicates that prevalence of offending may differ by race (Piquero, 2015). Race was measured nominally at baseline, delineating four racial identity categories: Black, Hispanic, White, and Other Race. Dummy variables delineating each race category from all other participants were coded from this variable (e.g., 1 = Black; 0 = all other participants). The dummy variable corresponding to White participants was excluded from analyses to provide a reference group.

Socioeconomic status (SES) at baseline was also included in analyses to mitigate bias in estimation that may stem from stratification of offending risk by social class (Craig et al., 2017). SES was measured as a weighted score comprised of participants’ parents’ educational attainment and occupational prestige. If both parents were able to provide SES data at baseline, a mean of the two scores was computed so that each participant was provided a single SES score. This construct was measured using Hollingshead’s index of social position (Hollingshead, 1957).

Exposure to violence has been identified by past research as a risk factor for offending (Mulford et al., 2018), necessitating inclusion of this concept in analyses as another set of control variables. This construct was measured using the Exposure to Violence Inventory (Selner-O’Hagan et al., 1998). Two forms of exposure to violence were examined: witnessed violence and direct victimization. Both of these variables were binary, delineating participants who reported experiencing that specific form of exposure to violence during the previous observation period from those who had not (0 = No; 1 = Yes). The measures from the fifth wave of data collection were utilized in analyses.²

Another variable included as a control on bias in estimation was self-control. This is because low self-control has been identified by past research as a risk factor for offending (Piquero et al., 2016). Self-control was measured using the Weinberger Adjustment Inventory at the fifth wave of data collection (Weinberger & Schwartz, 1990). A series of statements were provided to participants and participants were asked to identify how true the statement was of their own behavior using an ordinal scale. A mean score was derived from the scores on the individual items so that each participant had a single self-control score. Seven of the eight items used were reverse coded so that lower scores on the averaged self-control score corresponded to lower self-control. Good consistency was observed for this variable (Cronbach’s α = .80).

The next variable included in analyses as a control variable was deviant peer association because research has indicated that having more deviant peers is a risk factor for offending (Wojciechowski, 2018). This concept was measured using a series of ordinal items assessing the general number of friends who each participant indicated attempted to get them to engage in seven different antisocial behaviors. This scale comprised items used in the Rochester Youth Development Study (Thornberry et al., 1994). A mean score as computed from the individual items and this score at the fifth wave of data collection was used in analyses.³ Good reliability was observed for this variable (Cronbach’s α = .94).

Social support was also included as a control variable because having a high degree of social support has been found by past research to be a protective factor related to offending (Kurtz & Zavala, 2017). This construct was measured using the Contact with Caring Adults Inventory (Nakkula et al., 1990). Social support was measured at the fifth wave of data collection as a count of the number of caring adults who participants report having in their lives during the previous observation period. This count only included adults who provided social support in two or more domains, eschewing more superficial relationships for those with more depth.⁴

Another control variable included in analyses was negative affect. This mental health variable has been identified within the general strain theory canon as a highly relevant variable predicting offending when aroused (Agnew, 1992). The Brief Symptom Inventory measured the concept of hostility at the fifth wave of data collection using a series of ordinal items assessing symptomatology (Derogatis & Melisaratos, 1983). Each individual item asked participants to rate the degree to with a given symptom had bothered them in the past week. Good internal consistency was observed for the items used in analyses (Cronbach’s α = .76).

Offending variety at baseline was also controlled for in analyses, as previous levels of this variable could lead to continuity influencing the dependent variable in this study. This variable was measured the same way as the main offending variety dependent variable in this study.

The final control variable included in analyses was participants’ age because of the general tendency for offending risk to decline as individuals age through adulthood (Moffitt, 1993). Age was measured simply as an interval variable at the fifth wave of data collection.

Analytic Strategy

Analyses for the present study proceeded in two phases. The first phase entailed the use of latent profile analysis to identify differential clustering of varying levels of personality dimensions. This involves the identification of these unobserved clusters by examining the intercorrelation that exists between each of the variables included in analyses. Every participant is assigned a probability to each of X-number of latent profiles that the model is fit for and is assigned membership to the profile to which they have the highest probability of assignment. This involves an iterative process of specifying models of varying numbers of profiles and examining model fit to determine the optimum number of profiles that best fits the data. This involves comparison of Bayesian Information Criterion (BIC) statistics for each model to determine this model fit. BIC leverages the maximum likelihood function of the model and the number of data points to determine whether additional parameters added to the model improves fit. This fit metric penalizes additional complexity added to the model in the form of more parameters that do not add additional nuance. In the case of latent profile analysis, these additional parameters refer to the addition of more profiles to the model. In this way, BIC acts as an index of nested model fit, rather than absolute fit to the data. In this way, comparison of BIC statistics indicates whether the addition of another latent profile to the model presents improvement in the fit of the data. Beyond this, groups should also be large enough to be meaningful in analyses. Nagin (2005) indicates that similar latent variable models should have groups that have at least 1% of the total sample assigned membership to be selected as the best fitting model.

Following the identification of the best fitting latent profile model, negative binomial regression was used to estimate the influence of latent profile membership on offending seriousness at the fifth wave of data collection with all control variables included in the model to minimize bias in estimation. Negative binomial regression was chosen because of the heavy right skew to the outcome data. While Poisson models can often be used to account for skewed data like this, the mean and standard deviation of the outcome data varied considerably, necessitating further accounting for the dispersion of the data. The negative binomial model is able to account for this additional consideration, making it the best choice for modeling the data. Regression coefficients indicate that for each one unit increase in a given independent variable coefficient represents the predicted log count change in the dependent variable. A likelihood ratio test was used to assess relative model fit between the two models and determine whether one model provided significantly stronger fit to the data than the other. This involves determining whether the additional parameters added in the full model significantly improve model fit despite the additional degrees of freedom doing so entails. The null hypothesis that the simpler model provides better fit if the obtained test statistic exceeds the necessary threshold. Listwise deletion was used to manage missing data, as preliminary analyses examining correlation between variables included in the model indicated there was little concern regarding nonrandomness in missing data.

While the analytic process involved in identifying personality profiles is inductive in nature, past research on this topic allows for some rough postulations. Given the research cited above regarding common findings of groups characterized by negative correlation between neuroticism and all other personality dimensions, it may be that the individual personality dimensions demonstrated to be linked to offending may manifest in personality profiles that are consistent with this personality profile research. It would then be expected that at least two profiles would be elucidated: a profile low on neuroticism and high on all other personality dimensions and a profile high on neuroticism and low on all other personality dimensions. Past research has described profiles consistent with this high neuroticism and low levels of all other dimensions as being “Undercontrolled”, whereas profiles that are consistent with low neuroticism and high levels of all other personality dimensions are often described as “Resilient” (Claes et al., 2006; Zhang et al., 2015). So, for the sake of this study, the Undercontrolled profile would be understood as high levels of neuroticism and low levels of agreeableness, openness to experience, extraversion, and conscientiousness. Again, for the sake of this study, the Resilient profile would be understood as high levels of agreeableness, openness to experience, extraversion, and conscientiousness and low levels of neuroticism. Furthermore, it would be expected that the profile high on neuroticism and low on other personality dimensions would be at increased risk for offending and/or would engage in more frequent offending. While past research generally indicates that these personality profiles are a common finding, there has yet to be any study which has attempted to identify distinct profiles among juvenile offenders. The present study sought to address these gaps in the literature by testing several hypotheses.

Hypotheses

Hypothesis 1 (H1): Profiles resembling Resilient and Undercontrolled personality patterns will be elucidated.

Hypothesis 2 (H2): Participants assigned to the Undercontrolled profile will demonstrate more serious offending.

Results

The latent profile analytic procedure resulted in the identification of a two-profile solution as the only viable solution for the data. While examination of BIC statistics indicated that three, four, and five profile solutions provided better fit to the data based on this metric, further investigation indicated that the addition of profiles in the model beyond the two-profile solution resulted in the elucidation of a third group with a very small proportion of the sample assigned membership (N = 8).⁵ This violates the Nagin (2005) criteria described above for latent variable models exceeding 1% of the sample assigned to each group. For this reason, the two-profile model was chosen as the only viable option.

The two profiles in the chosen model differed substantially in all personality dimensions. The first profile demonstrated elevated levels of agreeableness, openness to experience, extraversion, and conscientiousness compared to the second profile in the model. This profile is described as “Resilient” profile (N = 410). Furthermore, the second profile in the model demonstrated elevated levels of neuroticism, relative to Group 1. This profile is described as the “Undercontrolled” profile (N = 821). These differences between the profiles on all dimensions of personality are stark and Table 1 provides mean levels of all of these personality dimensions for each profile.⁶ A dummy variable delineating participants assigned to each profile was coded so that it could be included as an independent variable in the second phase of analyses.⁷

Table 1.

Mean Levels of Personality Dimensions for Latent Personality Profiles.

Personality Dimensions	Resilient	Undercontrolled
Neuroticism*	1.869	2.597
Extraversion*	3.956	3.438
Openness to Experience*	3.327	2.967
Agreeableness*	3.701	3.070
Conscientiousness*	4.145	3.445

Between-group differences are significant at the p ≤ .001 level.

The second phase of analyses involved the estimation of the impact of personality dimension profile assignment on offending seriousness using negative binomial regression. Table 2 provides descriptive statistics of variables included in the regression models. Females made up 13.8% of the sample analyzed, whereas males made up the other 86.2% of this sample. The mean age observed in this sample was 18.018. Black participants made up 40.3% of this sample, White participants comprised 20.9% of the sample, Hispanic participants made up 34.2% of the sample, and participants assigned to the Other Race category comprised 4.6% of the sample. Table 3 describes the direct effect of profile assignment on offending seriousness in the absence of covariates (Model 1), whereas Table 4 includes control variables to mitigate bias in estimation (Model 2). Model 1 estimates indicated that participants assigned to the Undercontrolled profile had higher offending seriousness than the Resilient profile (Coefficient = .831). Model 2 examined this same effect with all control variables included in the model to mitigate the risk of bias in estimation. Results indicated that the greater offending seriousness exhibited by the Undercontrolled profile observed in Model 1 was robust upon inclusion of control variables, albeit attenuated somewhat (Coefficient = .541). Being male, lower SES, experiencing victimization, witnessing violence, lower self-control, greater negative affect, and greater offending seriousness at baseline were also all associated with greater offending seriousness at the fifth wave of data collection (Gender coefficient = −.550; SES coefficient = −.011; Direct victimization coefficient = .565; Witnessed violence coefficient = .541; Self-control coefficient = −.162; Negative affect coefficient = .191; Baseline offending seriousness coefficient = .011).

Table 2.

Descriptive Statistics.

Covariates	M/proportion	SD	Minimum	Maximum
Offending seriousness at wave 5	6.154	10.893	0	86
Personality profile assignment (0 = Resilient; 1 = Undercontrolled)	0.667	0.472	0	1
Gender (0 = Male; 1 = Female)	0.138	0.345	0	1
Race
White	0.209	0.407	0	1
Black	0.403	0.491	0	1
Hispanic	0.342	0.475	0	1
Other race	0.046	0.210	0	1
Socioeconomic status	51.437	12.200	11	77
Direct victimization (0 = No; 1 = Yes)	0.109	0.312	0	1
Witnessed violence (0 = No; 1 = Yes)	0.383	0.486	0	1
Self-control	3.063	0.948	1	5
Deviant peer association	0.722	0.799	0	4
Social support	0.830	0.524	0	2
Negative affect	0.570	0.689	0	4
Age	18.018	1.143	16	21
Offending seriousness at baseline	15.034	15.388	0	91

Table 3.

Negative Binomial Regression Estimating Covariate Effects on Offending Seriousness: Model 1.

Covariate	Coefficient	p value	95% CI
Personality profile assignment (0 = resilient; 1 = undercontrolled)	.831	p ≤ .001	[.558, 1.103]
Constant	.363	.132	[–.110, .837]

Table 4.

Negative Binomial Regression Estimating Covariate Effects on Offending Seriousness: Model 2.

Covariates	Coefficient	p value	95% CI
Personality profile assignment (0 = resilient; 1 = undercontrolled)	.541	p ≤ .001	[.267, .815]
Gender (0 = male; 1 = female)	−.550	.002	[–.892, –.209]
Race (reference = White)
Black	−.055	.732	[–.368, .259]
Hispanic	.009	.959	[–.326, .343]
Other race	.189	.511	[–.375, .754]
Socioeconomic status	−.011	.043	[–.021, –.000]
Direct victimization (0 = No; 1 = Yes)	.565	.001	[.217, .913]
Witnessed violence (0 = No; 1 = Yes)	.541	p ≤ .001	[.293, .790]
Self-control	−.162	.026	[–.304, –.019]
Deviant peer association	.602	p ≤ .001	[.437, .767]
Social support	−.076	.528	[–.311, .159]
Negative affect	.191	.045	[.004, .378]
Age	−.058	.267	[–.160, .044]
Offending seriousness at baseline	.011	.006	[.003, .019]
Constant	2.272	.033	[.178, 4.366]

A likelihood ratio test was conducted to determine whether there existed significantly greater nested model fit between the two models. A test statistic of .609 was computed from the log-likelihoods of each model (−2,170.369; −2,942.315), indicating that addition of all covariates did not result in improved model fit, thus the null hypothesis was not rejected.

Discussion

Results indicated support for hypotheses posited by this study. A two-profile model best fit the personality dimension data. Participants assigned to the Undercontrolled profile were characterized by elevated neuroticism and lower levels of all other personality dimensions. Furthermore, analyses indicated that individuals assigned to the Resilient profile reported less serious offending, thus Hypothesis 1 was supported, and participants assigned to the Undercontrolled profile demonstrated elevated offending seriousness, providing support for Hypothesis 2. This support was indicated by regression analyses indicating a positive coefficient of .541 net of control variables. This indicates that there is indeed evidence that a particular personality cluster is associated with offending risk. This study was novel in identifying these profiles among juvenile offenders, as past research had not explored the existence of distinct profiles among this population. Considering that the present study was also the first to examine the relevance of full personality profiles as they pertain to offending risk, identification of personality profiles among this indicated subpopulation may be even more relevant. There are numerous implications of the results observed in this study.

Individuals high in neuroticism generally were low in extraversion, agreeableness, openness to experience, and conscientiousness. The clustering of these personality dimensions indicate that it is highly relevant to examine the full personality profile, rather than just one dimension, when assessing risk for offending. All of the mechanisms by which personality impacts offending risk appear to be intertwined. For example, it is not just the negative affect associated with neuroticism that increases offending risk but also how this consistent negative affect influences the social situations that may result in offending that also intersect with one’s agreeableness. Because these dimensions are negatively correlated in the manner described in the observed results, it appears likely that it is the entire personality constellation that is influencing offending, rather than one distinct dimension. The intertwined nature of personality dimensions indicates the importance of determining whether programming focused on one dimension may also influence other dimensions as well. If this is the case, then a holistic approach to treatment may not be as necessary. While personality can be rigid, some research has indicated that treatment may be able to elicit some change and that individual treatment protocols may influence multiple personality dimensions (Johansson et al., 2013; Tang et al., 2009), though this evidence remains limited. Beyond this, juvenile offenders presenting a personality profile that is more consistent with the Resilient profile may not be as important to target with treatment while under criminal justice supervision. The robust relationship between personality and offending even after controlling for numerous other variables would seem to indicate as such. Future research should continue to investigate the relevance of developing treatment protocols for influencing potentially criminogenic personality profiles and the role that the criminal justice system may play in this regard.

The results described herein highlight the importance of personality for criminal justice. Assessment of personality dimensions among juvenile offenders may help to guide reentry and treatment initiatives that may impact recidivism risk. Such an assessment could also be conducted upon entry into the criminal justice system to similarly guide treatment mandates while under supervision. This, of course, necessitates the identification of risk factors which may help explain the personality-offending relationship that treatment initiatives should target. Post hoc analyses revealed that each risk/protective factor accounted for a small amount of the effect exhibited by personality profile on offending seriousness, but no risk factor was identified that was particularly salient for accounting for this effect. Rather, it was mainly a cumulative effect of all of these control variables that accounted for this partial mediation effect. This means that there is no distinct criminological risk factor that these results could suggest could be targeted to address offending risk. Furthermore, despite this partial accounting of the direct effect of personality dimensions on offending seriousness, a strong direct effect did indeed remain. This suggests that future research should continue to investigate the mechanism by which differences in these latent personality profiles impact offending.

While the present study has provided unique insight into the role of personality for predicting offending, there remain noteworthy limitations of this research. First, the present study only used one personality scale for assessing the dimensionality of this construct. While the “big five” are a prominent conceptualization of personality, there remain other scales which differ in their dimensions. Future research should seek to utilize these other scales to test for the presence of different types of personality profiles. Another limitation of this study relates to the latent profile analysis method used to identify personality profiles. While this is a highly useful tool for understanding the clustering of continuous variables at relatively high and low levels of a scale, it is important not to reify these results as something concrete. Rather, these profiles should be viewed as rough approximations of how personality dimensions tend to be correlated. As such, interpretation should be done with the utmost caution. Another limitation of this study relates to the potential for social desirability bias pertaining to the NEO five-factor personality scale, as this instrument does not control for this. There exists the potential that participants’ responses on the scales reflected a latent urge to demonstrate some of the more positive personality dimensions and suppress more negative dimensions. This may be particularly concerning for the population of interest, that is, juvenile offenders. Research has indicated that youth generally demonstrate greater social desirability bias in their responses to some prompts (Krumpal, 2013). This may be additionally concerning because of the justice-involved nature of the population, as there may be concerns responses may be biased in a manner that reflects a more respectable demeanor if participants are concerned that their responses may be used against them in legal proceedings in some way. For these reasons, there should be some concern about the validity of obtained personality profile results. A final limitation pertains to the negative binomial model chosen for analyses. This modeling strategy is generally used for count data and the dependent variable of interest to this study was not a count variable. However, the distribution of the data necessitated a modeling strategy that count account for a heavy right skew and over-dispersion. The negative binomial model presented the best option in this regard, even if the data were non-count data. However, for this reason, interpretation of results also must be carried out with caution in terms of making sense of the actual meaning of the obtained findings.

This study was novel in identifying personality profiles among juvenile offenders. While past research has identified such profiles among other populations, juvenile offenders had been excluded from distinct analyses prior to this study. This subpopulation may be of particular importance because of the relationship that has been observed between personality and offending. A growing literature in the field of criminology has focused on personality as a predictor of offending indicates the importance of this psychological construct. The results of this study indicated that the intercorrelation between personality dimensions was important for understanding offending seriousness. Highly neurotic individuals also tended to demonstrate lower levels of all other personality dimensions and the opposite was observed as well. These individuals fitting the Undercontrolled profile demonstrated higher offending seriousness. While past criminological research has focused on individual personality dimensions as predictors of offending, the results of this study indicate the importance of a holistic perspective on personality profiles. Future research should continue to investigate this relationship, particularly in terms of examining personality as a predictor of distinct forms of offending; like drug use and violent offending.

Footnotes

Declaration of Conflicting Interests

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

Funding

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

ORCID iD

Thomas Wojciechowski

Notes

Author Biography

Thomas Wojciechowski is currently an assistant professor with the Michigan State University School of Criminal Justice. His research mainly focuses on mental health, substance use, and human development.

References

Agnew

(1992). Foundation for a general strain theory of crime and delinquency. Criminology, 30(1), 47–88.

Brame

Fagan

Piquero

A. R.

Schubert

C. A.

Steinberg

(2004). Criminal careers of serious delinquents in two cities. Youth Violence and Juvenile Justice, 2(3), 256–272.

Carvalho

Nobre

P. J.

(2019). Five-factor model of personality and sexual aggression. International Journal of Offender Therapy and Comparative Criminology, 63(5), 797–814.

Chen

Xie

X. C.

Ping

Wang

M. Z.

(2017). Personality differences in online and offline self-disclosure preference among adolescents: A person-oriented approach. Personality and Individual Differences, 105, 175–178.

Claes

Tavernier

Roose

Bijttebier

Smith

S. F.

Lilienfeld

S. O.

(2014). Identifying personality subtypes based on the five-factor model dimensions in male prisoners: Implications for psychopathy and criminal offending. International Journal of Offender Therapy and Comparative Criminology, 58(1), 41–58.

Claes

Vandereycken

Luyten

Soenens

Pieters

Vertommen

(2006). Personality prototypes in eating disorders based on the big five model. Journal of Personality Disorders, 20(4), 401–416.

Claes

Vandereycken

Vandeputte

Braet

(2013). Personality subtypes in female pre-bariatric obese patients: Do they differ in eating disorder symptoms, psychological complaints and coping behaviour? European Eating Disorders Review, 21(1), 72–77.

Cloninger

C. R.

Przybeck

T. R.

Svrakic

D. M.

Wetzel

R. D.

(1994). The Temperament and Character Inventory (TCI): A guide to its development and use. Washington University.

Costa

P. T.

McCrae

R. R.

(1992). Professional manual: Revised NEO Personality Inventory (NEO-PI-R) and NEO Five-Factor Inventory (NEO-FFI). Psychological Assessment Resources.

10.

Craig

J. M.

Piquero

A. R.

Farrington

D. P.

(2017). The economic maturity gap encourages continuity in offending. Journal of Developmental and Life-Course Criminology, 3(4), 380–396.

11.

Cudmore

R. M.

Cuevas

C. A.

Sabina

(2017). The impact of polyvictimization on delinquency among Latino adolescents: A general strain theory perspective. Journal of Interpersonal Violence, 32(17), 2647–2667.

12.

Decuyper

Colins

O. F.

De Clercq

Vermeiren

Broekaert

Bijttebier

Roose

De Fruyt

(2013). Latent personality profiles and the relations with psychopathology and psychopathic traits in detained adolescents. Child Psychiatry & Human Development, 44(2), 217–232.

13.

Derogatis

Melisaratos

(1983). The Brief Symptom Inventory: An introductory report. Psychological Medicine, 13(3), 595–605.

14.

Hay

Meldrum

(2015). Self-control and crime over the life course. SAGE.

15.

Hogan

(1995). Hogan Personality Inventory. Hogan Assessment Systems.

16.

Hollingshead

A. B.

(1957). Two factor index of social position: Mimeo. Yale University.

17.

Johansson

Lyssarides

Andersson

Rousseau

(2013). Personality change after Internet-delivered cognitive behavior therapy for depression. PeerJ: The Journal of Life and Environmental Sciences, 1, Article e39.

18.

Jones

S. E.

Miller

J. D.

Lynam

D. R.

(2011). Personality, antisocial behavior, and aggression: A meta-analytic review. Journal of Criminal Justice, 39(4), 329–337.

19.

Kinnunen

M. L.

Metsäpelto

R. L.

Feldt

Kokko

Tolvanen

Kinnunen

Lappänen

Pulkkinen

(2012). Personality profiles and health: Longitudinal evidence among Finnish adults. Scandinavian Journal of Psychology, 53(6), 512–522.

20.

Krumpal

(2013). Determinants of social desirability bias in sensitive surveys: A literature review. Quality & Quantity, 47(4), 2025–2047.

21.

Kurtz

D. L.

Zavala

(2017). The importance of social support and coercion to risk of impulsivity and juvenile offending. Crime & Delinquency, 63(14), 1838–1860.

22.

Lee

Ashton

M. C.

(2004). Psychometric properties of the HEXACO Personality Inventory. Multivariate Behavioral Research, 39(2), 329–358.

23.

Loeber

Jennings

W. G.

Ahonen

Piquero

A. R.

Farrington

D. P.

(2017). Gender differences: Comparisons with males in the Pittsburgh youth study. In Loeber

Jennings

W. G.

Ahonen

Piquero

A. R.

Farrington

D. P.

(Eds.), Female delinquency from childhood to young adulthood (pp. 29–35). Springer.

24.

Lynam

D. R.

Leukefeld

Clayton

R. R.

(2003). The contribution of personality to the overlap between antisocial behavior and substance use/misuse. Aggressive Behavior: Official Journal of the International Society for Research on Aggression, 29(4), 316–331.

25.

McCrae

R. R.

Costa

P. T.

(1987). Validation of the five-factor model of personality across instruments and observers. Journal of Personality and Social Psychology, 52(1), 81–90.

26.

Meier

B. P.

Robinson

M. D.

Wilkowski

B. M.

(2006). Turning the other cheek: Agreeableness and the regulation of aggression-related primes. Psychological Science, 17(2), 136–142.

27.

Miller

J. D.

Lynam

(2001). Structural models of personality and their relation to antisocial behavior: A meta-analytic review. Criminology, 39(4), 765–798.

28.

Moffitt

T. E.

(1993). Adolescence-limited and life-course-persistent antisocial behavior: A developmental taxonomy. Psychological Review, 100(4), 674–701.

29.

Moon

Morash

(2017). Gender and general strain theory: A comparison of strains, mediating, and moderating effects explaining three types of delinquency. Youth & Society, 49(4), 484–504.

30.

Morgan

Cooper

Paul

Hammer

M. J.

Conley

Y. P.

Levine

J. D.

Miaskowski

Dunn

L. B.

(2017). Association of personality profiles with depressive, anxiety, and cancer-related symptoms in patients undergoing chemotherapy. Personality and Individual Differences, 117, 130–138.

31.

Mulford

C. F.

Blachman-Demner

D. R.

Pitzer

Schubert

C. A.

Piquero

A. R.

Mulvey

E. P.

(2018). Victim offender overlap: Dual trajectory examination of victimization and offending among young felony offenders over seven years. Victims & Offenders, 13(1), 1–27.

32.

Nagin

(2005). Group-based modeling of development. Harvard University Press.

33.

Nakkula

M. J.

Way

Stauber

H. Y.

London

(1990). Teenage Risk Prevention Questionnaire and interview: An integrative assessment of adolescent high-risk behavior. Graduate School of Applied and Professional Psychology, Rutgers University.

34.

O’Reilly

J. M.

Hubbard

M. L.

Lessler

J. T.

Biemer

P. P.

Turner

C. F.

(1994). Audio and video computer-assisted self interviewing: Preliminary tests of new technologies for data collection. Journal of Official Statistics, 10(2), 197–214.

35.

Piquero

A. R.

(2015). Understanding race/ethnicity differences in offending across the life course: Gaps and opportunities. Journal of Developmental and Life-Course Criminology, 1(1), 21–32.

36.

Piquero

A. R.

Jennings

W. G.

Farrington

D. P.

Diamond

Gonzalez

J. M. R.

(2016). A meta-analysis update on the effectiveness of early self-control improvement programs to improve self-control and reduce delinquency. Journal of Experimental Criminology, 12(2), 249–264.

37.

Robinson

M. D.

(2007). Personality, affective processing, and self-regulation: Toward process-based views of extraversion, neuroticism, and agreeableness. Social and Personality Psychology Compass, 1(1), 223–235.

38.

Selner-O’Hagan

Kindlon

Buka

Raudenbush

Earls

(1998). Assessing exposure to violence in urban youth. Journal of Child Psychology and Psychiatry and Allied Disciplines, 39(2), 215–224.

39.

Snyder

S. M.

Hartinger-Saunders

Brezina

Beck

Wright

E. R.

Forge

Bride

B. E.

(2016). Homeless youth, strain, and justice system involvement: An application of general strain theory. Children and Youth Services Review, 62, 90–96.

40.

Specht

Luhmann

Geiser

(2014). On the consistency of personality types across adulthood: Latent profile analyses in two large-scale panel studies. Journal of Personality and Social Psychology, 107(3), 540–556.

41.

Steinberg

L. D.

Cauffman

Monahan

(2015). Psychosocial maturity and desistance from crime in a sample of serious juvenile offenders. U.S. Department of Justice, Office of Justice Programs, Office of Juvenile Justice and Delinquency Prevention.

42.

Tang

T. Z.

DeRubeis

R. J.

Hollon

S. D.

Amsterdam

Shelton

Schalet

(2009). Personality change during depression treatment: A placebo-controlled trial. Archives of General Psychiatry, 66(12), 1322–1330.

43.

Thomas

K. M.

Hopwood

C. J.

Donnellan

M. B.

Wright

A. G.

Sanislow

C. A.

McDevitt-Murphy

M. E.

Ansell

E. B.

Grilo

C. M.

McGlashan

T. H.

Shea

M. T.

Markowitz

J. C.

Skodol

A. E.

Zanarini

M. C.

Morey

L. C.

(2014). Personality heterogeneity in PTSD: Distinct temperament and interpersonal typologies. Psychological Assessment, 26(1), 23–34.

44.

Thornberry

T. P.

Lizotte

A. J.

Krohn

M. D.

Farnworth

Jang

S. J.

(1994). Delinquent peers, beliefs, and delinquent behavior: A longitudinal test of interactional theory. Criminology, 32, 47–83.

45.

Turanovic

J. J.

Reisig

M. D.

Pratt

T. C.

(2015). Risky lifestyles, low self-control, and violent victimization across gendered pathways to crime. Journal of Quantitative Criminology, 31(2), 183–206.

46.

Van Dam

Janssens

J. M.

De Bruyn

E. E

. (2005). PEN, big five, juvenile delinquency and criminal recidivism. Personality and Individual Differences, 39(1), 7–19.

47.

Vaughn

M. G.

Howard

M. O.

DeLisi

(2008). Psychopathic personality traits and delinquent careers: An empirical examination. International Journal of Law and Psychiatry, 31(5), 407–416.

48.

Vize

C. E.

Miller

J. D.

Lynam

D. R.

(2018). FFM facets and their relations with different forms of antisocial behavior: An expanded meta-analysis. Journal of Criminal Justice, 57, 67–75.

49.

Weinberger

D. A.

Schwartz

G. E.

(1990). Distress and restraint as superordinate dimensions of self-reported adjustment: A typological perspective. Journal of Personality, 58(2), 381–417.

50.

Wojciechowski

T. W.

(2018). The development of deviant peer association across the life-course and its relevance for predicting offending in early adulthood. Journal of Developmental and Life-Course Criminology, 4(1), 73–91.

51.

Zhang

Bray

B. C.

Zhang

Lanza

S. T.

(2015). Personality profiles and frequent heavy drinking in young adulthood. Personality and Individual Differences, 80, 18–21.