Using the PAI-A to Classify Juvenile Offenders by Adjudicated Offenses

Abstract

To improve understanding and treatment of criminal behavior, researchers have developed typologies of juvenile offenders, primarily focusing on personality traits and criminal history to classify according to type of offense committed. Existing literature has examined underlying personality characteristics found in different subcategories of criminal offenses in juveniles; however, few studies have employed the Personality Assessment Inventory-Adolescent (PAI-A), instead choosing the MMPI-A. A typical classification model of juvenile offenses categorizes offenses into: Interpersonal, Property, and Drug/Alcohol-related charges, to further study within-group differences. The current study examines how personality profiles, examined by the PAI-A, can classify offenders into these offense-type groups. Personality profiles of participants were obtained through pre-sentencing psychological evaluations of 142 juvenile offenders ages 14 to 17. Binary logistic regressions were conducted using PAI-A Clinical, Treatment Consideration, and Interpersonal scales to predict offense-type group classifications. Results yielded statistically significant full models for all offense-type groups, with an average overall accuracy rate of 76.3%. Overall, results suggest that the PAI-A has good predictive power to classify juvenile offender types, and may be more effective in classifying certain types of offenders than the MMPI-A. Notably, Interpersonal and Treatment consideration scales were stronger predictors of offense-type than Clinical scales. This model of juvenile offender classification holds promise for more effective treatment, management, and prediction of behavior for juvenile offenders.

Improving the efficacy of interventions for individuals involved in the juvenile justice system is an issue of both economic and pragmatic concern. In 2016, approximately 34,000 youth were incarcerated in juvenile detention centers, 5200 juveniles were incarcerated in adult prisons, and an additional 20,000 youth were held away from home in residential facilities by the juvenile justice system (Wagner and Rabuy 2016). Additionally, current rehabilitation efforts such as jail programs, probation, and rehabilitation have shown limited success in terms of recidivism rates. In Texas alone, 72.9% of juveniles released in 2011 were rearrested within 3 years (Legislative Budget Board 2017). The costs associated with such facilities as well as the long-term concerns associated with subjecting youth to ineffective rehabilitation programs are vast. Understanding the criminal behavior profile of juveniles is the first step to treatment improvement.

Personality and Juvenile Offenders

Past work has examined personality traits in juvenile offenders in order to better understand risk for violence and recidivism (Kennedy, Burnett, & Edmonds, 2011; Mulder et al. 2012; Steiner et al. 1999). Many attempts to understand criminal behavior profiles have utilized an investigation of the relationship of personality characteristics to antisocial behaviors (Jones et al. 2011). Researchers have also utilized broad personality theories, such as the Big Five, to classify offenders by offense type and severity in order to successfully differentiate between violent and non-violent offenders (John, Caspi, Robins, Moffitt, & Stouthamer-Loelber, 1994; Nederlof et al. 2010). The five-factor model has also been utilized to demonstrate that personality domains may moderate treatment responsiveness, further evidence that personality may affect current treatments for juvenile offenders (Asscher et al. 2016).

Classification of Offenders

The ability to classify offenders according to personality type not only has utility in establishing differences between violent and nonviolent offenders, but also in identifying crime motivation and levels of premeditation (Ching et al. 2013; Jones and Harris 1999). These clusters serve to classify offenders accurately for research purposes, but also to potentially match offender personality profiles to targeted interventions best suited to their individual needs (Jones and Harris 1999). The development of offender classification systems based on personality characteristics has aided in the efforts to improve treatment in the criminal justice system in adult offenders; however, there is a general lack of agreement in classification system criteria overall (Davies 1969; Megargee 1994; Panton, 1958; Veneziano & Veneziano, 1986). Further, the majority of prior studies have examined classification of adult offenders; there is a dearth of research in the area of juvenile offender classification. Like adults, juvenile offenders can be classified using either the Minnesota Multiphasic Personality Inventory—2 or the Personality Assessment Inventory, discussed below.

Minnesota Multiphasic Personality Inventory in Forensic Populations

The most widely used instrument for personality measurement in forensic settings is the 567-item Minnesota Multiphasic Personality Inventory-2 (MMPI-2; Butcher et al. 1989). Given the ubiquity of the MMPI-2, much of the personality testing conducted in juveniles has utilized the 478-item Minnesota Multiphasic Personality Inventory—Adolescent version (MMPI-A; Butcher 1992). Elevations on the excitatory scales, i.e., Psychopathic Deviate (Pd), Schizophrenia (Sc), and Mania (Ma) distinguish between adolescent delinquents and controls (Peña et al. 1996; Morton et al. 2002). Elevations on the Psychopathic Deviate (Pd) and Paranoia (Pa) scales were found to discriminate between delinquent and normative samples with high levels of sensitivity; however juvenile delinquency and recidivism remains difficult to predict due to low base rates (Morton et al. 2002; Peña et al. 1996). The MMPI-A is also utilized to examine elevated scales to classify offense type (Glaser et al. 2002; Veltri et al. 2014). Scales on the MMPI-A have been shown to discriminate among offense types for male juvenile offenders, i.e., Interpersonal offenses, Property offenses, and Drug/Alcohol-related offenses (Glaser et al. 2002). The authors found that characteristics such as social avoidance (scale Si-2), higher degree of health concerns (scale Hs), and lower proneness for developing a drug or alcohol problem (scale PRO) discriminated Property offenses; higher levels of depressed affect, psychomotor retardation, and lack of hostile or aggressive impulses (scale D-2) as well as greater proneness for developing a drug or alcohol problem (scale PRO) distinguished Drug/Alcohol-related offenses; and manipulative and self-oriented behavior (scale Ma-1) differentiated Interpersonal offenses. Further, Interpersonal and Property offense-types had the most successful rates of classification based on personality characteristics (Glaser et al. 2002). Such offense clusters allow for those at risk for drug and alcohol abuse to receive substance abuse interventions, while individuals committing crimes against persons may require intervention targeting antisocial features or conduct problems. To this end, treatment matching within the justice system is facilitated by application of offender typologies; research has documented that matching treatment to individual’s characteristics optimizes outcomes (Palmer 1994). However, the MMPI-A requires extended periods of time to complete (average: 60 min) and thus application of shorter personality assessments would decrease resources required to classify juvenile offenders, while still gathering important and useful information.

Utilization of the Personality Assessment Inventory with Forensic Populations

The 344-item Personality Assessment Inventory (PAI; Morey, 1991) is a briefer self-report measure of personality and emotional functioning, and it maintains broad base empirical support in clinical forensic assessment (Morey and Quigley 2002). The PAI has demonstrated utility in assessing not only distorted response patterns (i.e., response validity), but also clinical diagnosis, personality pathology, substance abuse, assessment of risk, and treatment matching with adult offenders (Morey and Quigley 2002). Several of the PAI scales parallel concepts relevant to risk assessment, such as Aggression (AGG), a measure of hostility, poor control over anger expression, verbal expressions of anger, and the tendency to have physical displays of anger; Antisocial Features (ANT), a measure of involvement in illegal activities, lack of empathy or remorse, and craving for excitement; and Violence Potential (VPI), an aggregate measure of risk factors of violence including measures of aggression, antisocial behaviors and attitudes, and negative profile distortion (Morey 2003). The link between PAI profiles and violence, recidivism, and misconduct is well-documented (Gardner et al. 2015; Walters 2006), with the AGG and ANT scales most commonly implicated (Reidy et al. 2016). For juvenile populations, the adolescent version of the PAI (PAI-A; Morey 2007) contains only 264 items and requires significantly less time to complete than the MMPI-A. Efficiency and cost-effectiveness, coupled with test features relevant to this population, including a lower reading level and measurement of borderline features, the PAI-A has several advantages for use in the juvenile forensic setting (Blais et al. 2011).

The Present Study

The present study seeks to build upon existing literature that has previously only used the MMPI-A or the PAI for adults in offender classification (Losada-Paisley, 1998; Morey and Quigley 2002). Examining the PAI-A’s discrimination ability to classify juvenile offenders into Interpersonal, Property, and/or Drug/Alcohol-related offense-type groups, as outlined by Glaser et al. (2002) will expand upon its utility in forensic settings and may make existing psychological assessment batteries more broadly applicable and efficient. Further, the current study examines the discrimination validity of specific scales of the PAI-A for offender classification, as few studies have identified specific domains of personality associated with different types of juvenile offending (Morton et al. 2002; Mulder et al. 2012). We expected the data to yield significant classification models for all offense-types, with the Interpersonal and Property offense-types yielding the best level of classification rates, similar to those obtained in Glaser et al. (2002). It was hypothesized that the Dominance (DOM) and Aggression (AGG) scales of the PAI-A would be most predictive of membership into the Interpersonal offense-type. For Property offenses, it was expected that the lower scores on the Warmth (WRM) scale would be a predictor of group membership in line with Glaser et al.’s (2002) findings of higher levels of social avoidance (Si-2) in those who commit Property offenses on the MMPI-A. Lastly, it was predicted that the Drug Problems (DRG) and Depression (DEP) scales would be the most predictive of membership into the Drug/Alcohol-related offense-type, due to the link between internalizing problems, substance abuse, and drug-related offenses. The Alcohol Problems (ALC) scale was not hypothesized to be a predictor of group membership, as the sample used did not report high levels of alcohol use.

Method

Participants

The sample consisted of 142 male and female juvenile offenders incarcerated for various charges in a juvenile delinquent detention center located in Central Texas. The mean age of the sample was 16.07 years (SD = 0.43; ranging from 14 to 17) and majority were male (n = 108). School grade ranged from 7th to applying for the GED, with the modal number of participants in the 10th grade (N = 65). The sample was racially diverse; participants self-identified as African American (39%), Latino/Hispanic (29%), Caucasian (29%), and “Other” (3%). This study was reviewed and designated as exempt by the Baylor University Institutional Review Board in March of 2013.

Measures

Personality Assessment Inventory-Adolescent

The PAI-A (Morey 2007) is a 264-item self-report measure designed to assess personality, psychopathology, and provides information on other client variables such as psychosocial environment. The items comprise 22 non-overlapping scales: 4 Validity scales, 11 Clinical scales, 5 Treatment Consideration scales, and 2 Interpersonal scales. The Clinical scales are designed to assess domains of clinical problems important in diagnosis of psychiatric disorders. The Clinical scales assess constructs related to Somatic Complaints (SOM; preoccupation with physical concerns and vague complaints of bad health), Anxiety (ANX; physical and affective tension, and experiences of worry), Anxiety Related Disorders (ARD; more specific anxiety symptoms such as intrusive thoughts, phobic fears, and traumatic stress), Depression (DEP; thoughts of worthlessness, feelings of sadness, and physical symptoms of depression), Mania (MAN; accelerated thought processes and behavior, grandiosity, and irritability), Paranoia (PAR; suspiciousness, threat monitoring, thoughts of persecution, and a cynical interpersonal style), Schizophrenia (SCZ; psychotic experiences, social detachment, problems with confusion or concentration, and disorganization of thought processes), Borderline Features (BOR; emotional lability, insecure identity, negative relationships, and history of self-harming behaviors), Antisocial Features (ANT; antisocial behaviors and involvement in illegal activities, lack of empathy or remorse, and sensation seeking), Alcohol Problems (ALC; behaviors and consequences related to alcohol abuse), and Drug Related Problems (DRG; behaviors and consequences related to drug abuse). The Treatment Consideration Scales of the PAI-A aim to provide information that could assist in case management and treatment planning such as treatment motivation and potential complications. Treatment Consideration scales assess Aggression (AGG; verbal expressions and physical expressions of anger, and hostile attitudes), Suicidal Ideation (SUI; thoughts about death and contemplation of suicide), Stress (STR; problems in interpersonal relationships, financial stressors, difficulties with employment, and major life changes), Nonsupport (NON; perceived lack of social support), and Treatment Rejection (RXR; attitudes that could hinder motivation for treatment). And the Interpersonal scales measure two factors that affect interpersonal functioning, Dominance (DOM; tendency to act dominant, assertive, and in control of social situations), and Warmth (WRM; tendency to respond to social situations in an empathetic and engaging manner) (Morey 2003). The PAI-A, an adolescent complement to the adult version, is appropriate for use with adolescents ages 12–18. Scores are obtained from the respondents’ rating of items on a 4-point Likert scale (1 = very true, 2 = mainly true, 3 = slightly true, 4 = false). The PAI-A demonstrates similar reliability and validity as the adult instrument, demonstrating alpha coefficients ranging from .66 to .90. Although the PAI-A has not been extensively researched for use with forensic populations, the adult instrument has gained support for use with forensic populations (Newberry and Shuker 2012).

Procedure

Personality profiles of participants were obtained through pre-sentencing psychological evaluations ordered by the court between September 2009 to August 2016. This evaluation consisted of standard clinical administration of the PAI-A, a measure of intellectual functioning, a measure of achievement, a measure of impulsivity, and a clinical interview for purposes of informing adjudication placement and treatment recommendations. Informed consent was obtained from legal guardians, and assent was obtained from participants prior to the start of the psychological evaluation.

Offense and demographic data were readily available and obtained through participant files and court records. Classification of offense types, into categories was as described in Glaser et al. (2002) and based upon a hierarchy of offense seriousness defined by the Office of Juvenile Justice and Delinquency Prevention (OJJDP). Classification efforts were based on both index and historical offenses incurred by the participants of the sample. Three offense-type groups characterized this sample: Interpersonal offenses (n = 88), Property offenses (n = 59), or Drug/Alcohol-related offenses (n = 34). Examples of Interpersonal offenses included, but were not limited to, physical assault, organized crime, terroristic threat, harassment, and indecent exposure with a minor. Examples of Property offenses included offenses such as burglary, arson, graffiti, and unauthorized use of a vehicle. The Drug/Alcohol-related offenses were confined to drug possession charges, as these were the only Drug/Alcohol-related offenses in this sample. Membership to these offense-type groups were not mutually exclusive, as most participants (86.1%) were chronic offenders who committed crimes across multiple categories. Charges were classified into the offense-type groups prior to examination of PAI-A profiles.

Analyses

The purpose of the current study was to identify personality characteristics that predicted group membership for three groups of offenses. Invalid profiles were excluded from the sample based on scores >92 T on the Negative Impression Management (NIM) scale and > 68 T on the Positive Impression Management (PIM) scale, as these scores fall two standard deviations above the mean and suggest that a participant attempted to distort the personality profile (Morey 2003). The NIM scale is intended to measure a participant’s attempt to exaggerate symptoms and make the profile appear more pathological, while the PIM scale is intended to measure an attempt to appear more psychologically adjusted and deny minor faults. These particular validity cut-off scores were chosen for their sensitivity and specificity in detecting malingering in former research with adult subjects (Morey, 1991). In total, nine profiles were excluded from the original sample (N = 151) based on validity concerns. Descriptive analyses determined the mean PAI scores for the sample, the mean PAI scores for each of the offense-type groups, and to determine any mean differences amongst racial and gender groups. After examining descriptive data, the Clinical, Treatment Consideration, and Interpersonal scales of the PAI-A were entered into a stepwise binary logistic regression to predict membership in each offense-type group. The four Validity scales of the PAI were excluded from analyses as response style was not theorized to contribute to between-group differences. A stepwise binary logistic regression was utilized due to the dichotomous nature of the criterion variable (group membership or non-membership). The relationship between the scales of the PAI-A and group membership was assumed to be non-linear. Cohen’s kappa (κ) statistic was also calculated in order to provide a measure of magnitude of agreement between the predicted classification model and observed group membership, as suggested by Glaser et al. (2002). Interpretation of kappa coefficients were based on the following suggested cut-off scores: <0.00 poor agreement; 0.00–0.20 slight agreement; 0.21–0.40 fair agreement; 0.41–0.60 moderate agreement; 0.61–0.80 substantial agreement; and 0.81–1.00 almost perfect agreement (Landis and Koch 1977).

Results

Descriptive Statistics

The Clinical and Interpersonal Scale T-score descriptive statistics for the total sample and each of the three offense-type groups are displayed in Table 1. Notably, the highest mean score across all scales of the PAI-A was the DRG scale, which holds true for the total sample (M = 62.06, SD = 13.54) as well as across all offense-type groups (Interpersonal M = 60.22, SD = 13.42; Property M = 62.61, SD = 12.02; Drug/Alcohol-related M = 69.50, SD = 12.10). Interestingly, none of the means of the PAI-A scales met the level for clinical significance (T = 70), however, the DRG scale in the Drug/Alcohol-related offense-type group was close to clinically significant (M = 69.50, SD = 12.09).

Table 1.

Mean T-scores and standard deviations obtained for juvenile offenders on the personality assessment inventory-adolescent (PAI-A) selected scales

PAI-A Scale	Person (n = 88)		Property (n = 59)		Drug/Alcohol (n = 34)
	M	SD	M	SD	M	SD
SOM	52.64	12.39	53.42	13.69	55.03	14.14
ANX	51.81	11.31	53.22	13.22	52.12	15.01
ARD	53.58	10.03	54.14	12.55	53.85	9.47
DEP	52.03	10.72	54.42	12.06	56.59	11.10
MAN	52.58	11.67	50.41	11.05	50.79	10.49
PAR	52.72	10.31	51.92	9.35	53.79	10.77
SCZ	50.26	11.51	51.54	10.62	53.26	12.96
BOR	52.59	10.20	53.80	10.84	53.94	11.60
ANT	50.24	7.81	53.34	8.26	54.29	8.12
ALC	47.41	6.65	49.07	9.72	49.35	9.70
DRG	60.22	13.41	62.61	12.02	69.50	12.10
AGG	56.16	10.71	54.53	10.91	54.62	12.44
SUI	46.65	7.28	50.44	12.03	48.47	8.20
STR	49.34	10.06	49.88	9.68	51.74	11.29
NON	48.49	9.73	51.78	10.88	54.00	12.61
RXR	42.55	10.48	43.54	10.85	42.53	9.66
DOM	53.49	7.41	52.15	7.32	54.35	8.41
WRM	46.68	9.29	43.63	10.69	44.97	10.01

SOM Somatic Complaints, ANX Anxiety, ARD Anxiety Related Disorders, DEP Depression, MAN Mania, PAR Paranoia, SCZ Schizophrenia, BOR Borderline Features, ANT Antisocial Features, ALC Alcohol Problems, DRG Drug Problems, AGG Aggression, SUI Suicidal Ideation, STR Stress, NON Nonsupport, RXR Treatment Rejection, DOM Dominance, WRM Warmth

Table 1 shows average mean T-scores and standard deviations for the Clinical, Treatment Consideration, and Interpersonal scales for each of the offense-types

Inferential Statistics

One-way ANOVAs were run for each of the offense-type groups to examine mean differences across groups. In the Interpersonal offense-type, mean T-scores on DEP, SCZ, ANT, ALC, DRG, SUI, and NON, were significantly lower for members versus nonmembers (mean difference = 4.36 t). For the Property offense-type group, mean scores on the WRM scale were significantly lower for group members (mean difference = 3.63 t). For the Drug/Alcohol-related offense-type group, mean scores on the ANT and DRG scales were significantly higher for members and nonmembers of this offense-type group, with an average mean difference score of 6.02 T-score points. Chi-square analyses did not suggest disproportionate frequencies among race or gender for the three offense-type groups (p > .05).

One-way ANOVAs were performed to examine differences in mean PAI-A scores for male and female participants (Table 2), and amongst racial groups (Table 3). Significant differences in mean scores were found between male and female participants on the BOR scale [F (1,141) = 4.06, p < .05] with females scoring higher than males (mean difference = 4. 27 t), and males scoring higher than females (mean difference = 4.5 t) on the RXR scale [F (1,141) = 4.71, p < .05] scales. African American participants scored higher on the ARD [F (3, 127) = 3.23, p < .05] and MAN [F (3,127) = 5.48, p = .001] scales than Caucasian and Latino-Hispanic participants. Given the nature of the convenience sample, differences among groups on the basis of gender and race/ethnicity are interpreted with caution and necessitate further study. Demographic information was not included in the binary logistic regression analyses due to limitations in respective group sizes preventing adequate representativeness of the sample.

Table 2.

Analysis of variance for gender amongst juvenile offenders on the personality assessment inventory-adolescent (PAI-A) selected scales

PAI-A Scale	Source	SS	df (1, 140)	MS	F
SOM	Between Groups	162.95		163.95	.98
	Within Groups	23317.17		166.55
	Total	23480.12	141
ANX	Between Groups	422.78		422.78	2.72
	Within Groups	21746.64		155.33
	Total	22169.42	141
ARD	Between Groups	266.31		266.31	2.34
	Within Groups	15958.06		113.99
	Total	16224.37	141
DEP	Between Groups	193.17		193.17	1.60
	Within Groups	16865.80		120.47
	Total	17058.97	141
MAN	Between Groups	91.68		91.68	.63
	Within Groups	20308.92		145.06
	Total	20400.60	141
PAR	Between Groups	9.88		9.88	.09
	Within Groups	14761.84		105.44
	Total	14771.72	141
SCZ	Between Groups	238.88		238.88	1.58
	Within Groups	21158.11		151.13
	Total	21396.99	141
BOR	Between Groups	462.39		462.39	4.06*
	Within Groups	15941.98		113.87
	Total	16404.37	141
ANT	Between Groups	174.06		176.06	2.52
	Within Groups	9665.30		69.04
	Total	9839.36	141
ALC	Between Groups	197.35		197.35	2.73
	Within Groups	11643.38		83.17
	Total	11840.73	141
DRG	Between Groups	123.74		123.74	.67
	Within Groups	25738.18		183.44
	Total	25861.92	141
AGG	Between Groups	37.47		17.47	.32
	Within Groups	16215.13		115.82
	Total	16252.60	141
SUI	Between Groups	42.86		42.86	.41
	Within Groups	14764.10		105.46
	Total	14806.97	141
STR	Between Groups	244.10		244.10	2.33
	Within Groups	14694.77		104.46
	Total	14938.87	141
NON	Between Groups	108.29		108.29	.88
	Within Groups	17154.37		122.53
	Total	17262.66	141
RXR	Between Groups	513.30		513.30	4.70*
	Within Groups	15268.36		109.06
	Total	15781.66	141
DOM	Between Groups	4.91		4.91	.08
	Within Groups	8712.86		62.24
	Total	8717.78	141
WRM	Between Groups	341.42		341.42	3.44
	Within Groups	13949.95		99.64
	Total	14292.37	141

Table 2 shows mean difference calculations between male and female offenders for each of the Clinical, Treatment Consideration, and Interpersonal scales. The BOR and RXR scales exhibited significant differences between the two gender groups

*p < .05, ** p < .01, ***p < .001

Table 3.

Analysis of variance for race amongst juvenile offenders on the personality assessment inventory-adolescent (PAI-A) selected scales

PAI-A scale	Source	SS	df (3127)	MS	F
SOM	Between Groups	250.47		83.49	.47
	Within Groups	22436.07		176.66
	Total	22686.53	130
ANX	Between Groups	134.78		44.93	.27
	Within Groups	21176.82		166.75
	Total	21311.60	130
ARD	Between Groups	1081.21		360.40	3.23*
	Within Groups	14172.51		111.60
	Total	15253.73	130
DEP	Between Groups	20.06		7.69	.06
	Within Groups	16328.66		128.57
	Total	166351.71	130
MAN	Between Groups	2071.33		691.44	5.58**
	Within Groups	16,038.27		126.29
	Total	18112.60	130
PAR	Between Groups	171.33		57.11	.53
	Within Groups	13728.41		108.10
	Total	13899.74	130
SCZ	Between Groups	1034.10		344.70	2.24
	Within Groups	19548.84		153.93
	Total	20582.93	130
BOR	Between Groups	348.00		116.00	.97
	Within Groups	15212.69		119.79
	Total	15560.69	130
ANT	Between Groups	119.36		39.79	.58
	Within Groups	8701.22		68.51
	Total	8820.58	130
ALC	Between Groups	637.79		212.60	2.50
	Within Groups	10800.84		85.05
	Total	114438.63	130
DRG	Between Groups	1158.82		386.27	2.09
	Within Groups	23524.07		185.23
	Total	24682.89	130
AGG	Between Groups	73.62		24.54	.21
	Within Groups	15209.33		119.76
	Total	15282.95	130
SUI	Between Groups	39.03		13.01	.11
	Within Groups	14508.19		114.24
	Total	14547.22	130
STR	Between Groups	59.31		19.77	.18
	Within Groups	13924.34		109.64
	Total	13983.65	130
NON	Between Groups	101.37		33.79	.26
	Within Groups	16259.17		128.03
	Total	16360.53	130
RXR	Between Groups	585.82		195.27	1.84
	Within Groups	13500.84		106.31
	Total	14086.66	130
DOM	Between Groups	413.43		137.81	2.40
	Within Groups	7292.62		57.42
	Total	7706.05	130
WRM	Between Groups	582.49		194.16	2.01
	Within Groups	12289.82		96.77
	Total	12872.31	130

Table 3 shows average mean differences between racial/ethnicity groups of offenders (e.g. Caucasian, African American, Hispanic/Latino, Asian, and Other) for each of the Clinical, Treatment Consideration, and Interpersonal scales. The ARD and MAN scales yielded statistically significant differences among racial groups

*p < .05, ** p < .01, *** p < .001

Ability of the PAI-A to Predict Offense-Type Group Membership

Binary logistic regression analyses were employed to test the utility of the PAI-A in classifying offenders by offense-type (Table 4). This analysis examines how a set of predictor variables, in this case the Clinical, Treatment Consideration, and Interpersonal scales of the PAI-A, is related to a dichotomous variable, in this case offense-type group membership (Harrell 2015). Each offense-type group warranted separate evaluation, as offense-type groups were not mutually exclusive. Regressions revealed good utility of the PAI-A for offender classification, as each model was statistically significant with an average overall accuracy rate for classification of 76.3%, which is higher than would be expected by chance (50%). Cohen’s kappa coefficient was calculated between predicted membership and observed group membership to provide information regarding the magnitude of agreement between the models (κ = .51). The coefficient produced by the overall classification model represents moderate and acceptable agreement between predicted and observed group membership (Landis and Koch 1977; Viera and Garrett 2005).

Table 4.

Binary logistic regression analyses for juvenile offenders on the personality assessment inventory-adolescent (PAI-A) selected scales

	Person (n = 88)		Property (n = 59)		Drug/Alcohol (n = 34)
Predictor Scale	O.R.	S.E.	O.R.	S.E.	O.R.	S.E.
SOM	1.03	.03	1.01	.02	1.02	.03
ANX	1.00	.03	1.01	.03	.95	.04
ARD	.95	.03	.97	.03	.99	.03
DEP	1.02	.04	1.01	.03	1.10	.05
MAN	1.01	.03	1.01	.03	.98	.03
PAR	1.09*	.04	.95	.03	1.02	.04
SCZ	.98	.04	.96	.04	1.00	.04
BOR	.95	.05	1.08	.04	.98	.05
ANT	.93*	.04	1.11**	.03	1.08	.04
ALC	.96	.03	1.00	.03	.98	.03
DRG	.98	.02	.98	.02	1.07**	.02
AGG	1.08**	.03	.93*	.03	.95	.03
SUI	.93*	.03	1.04	.03	.94	.04
STR	1.05	.04	.99	.03	1.01	.03
NON	.92	.03	.98	.03	1.00	.03
RXR	.97	.03	1.02	.03	1.02	.03
DOM	.92**	.03	.99	.03	1.03	.04
WRM	1.00	.03	.93*	.03	1.01	.03

Table 4 shows odds ratios and the standard error of estimate from the results of each of the binary logistic regression analyses conducted for the offense types. For the Interpersonal offense type, the AGG, SUI, and DOM scales yielded significant partial effects and accounted for 28.7% of the classification variance. For the Property offenses, the ANT, AGG, and WRM scales were significant predictors of offense type and explained 19.0% of the classification variance. And for the Drug/Alcohol-related offenses, the DRG scale predicted group membership and explained 20.9% of classification variance

*p < .05, ** p < .01, ***p < .001

Interpersonal Offense-Type

For the Interpersonal offense-type group, a test of the full model versus the intercept-only model was statistically significant, X²(18, N = 142) = 47.96, p < .001. This model was able to correctly classify 86.4% of those who committed Interpersonal offenses and 57.4% of those who did not, for an overall success rate of 75.4%. The kappa coefficient (κ = .46) examining agreement between predicted group membership and observed group membership suggests moderate agreement between the two models. Employing a .05 criterion of statistical significance, PAR, ANT, AGG, SUI, NON, and DOM had significant partial effects. Odds ratios indicated that higher levels of PAR and AGG, and lower levels of ANT, SUI, NON, and DOM, were more likely to predict group membership in those who committed Interpersonal offenses.

Property Offense-Type

For the Property offense-type group, a test of the full model versus the intercept-only model was statistically significant X²(18, N = 142) = 29.98, p < .05. This model yielded an overall classification success rate of 72.5%, correctly classifying 57.6% of those who committed Property-type offenses, and 83.1% of those who did not. Cohen’s kappa between predicted and observed group membership for the Property offense type was .42, which suggests an acceptable level of agreement between the two models. The ANT, AGG, and WRM scales had significant partial effects at a .05 level of statistical significance. Interpretation of odds ratios indicated that higher levels of ANT, and lower scores on the AGG and WRM scales were more likely to predict group membership in those who committed Property offenses.

Drug Offense-Type

For the Drug offense-type group, a test of the full model versus the intercept-model was also statistically significant X²(18, N = 142) = 33.21, p < .05. The overall classification success rate for this model was 81.0%, correctly classifying 35.3% of those who committed Drug-type offenses, and 95.4% of those who did not. The kappa coefficient yielded for this prediction model was .42, which suggests a moderate agreement between predicted and observed group membership. With a .05 criterion of statistical significance, the DRG scale yielded significant partial effects. Significant odds ratios indicated that levels of DRG were most likely to predict group membership for this offense-type.

Discussion

Overall, the results suggest the PAI-A may have utility in classifying offender types in juveniles, building upon literature demonstrating the predictive power of the MMPI-A (Glaser et al. 2002). Notably, the Interpersonal and Treatment Consideration scales, rather than Clinical scales were better able to differentiate among classification for all types of offenses. This suggests the possibility that patterns of interpersonal interaction and dispositional traits, rather than psychopathology/syndromal characteristics may carry more weight in offender classification. These considerations may be especially important due to the nature of the personality profiles obtained from the participants, where mean scores were largely in the subclinical range. Personality profiles produced by the participants in our sample are consistent with subclinical scores yielded for juvenile offenders on other personality measures such as the MMPI-A, and similar to PAI profiles yielded in adult offenders (Glaser et al. 2002; Newberry and Shuker 2012). These results suggest that an under-reporting bias may be present in juvenile offenders, especially on a face valid measure such as the PAI-A when considering that clinical symptoms are expected given their placement in detention. Previous investigations have suggested the importance of risk assessment and treatment matching for juvenile offenders, especially in considering dynamic risk factors such as attitudinal factors and dispositional traits as treatment targets (Borum, 2003). Treatments aimed at developing effective interpersonal and general coping skills may be especially relevant for this population due to the level of clinical impairment and risk associated with incarceration.

Our results suggest the PAI-A may be as, or in some cases more, effective in classifying certain types of offenders than the MMPI-A. The PAI-A yielded greater overall success in classification of Interpersonal and Drug/Alcohol offense-types than previous research with the MMPI-A (Glaser et al. 2002). In the present study, the PAI-A produced the highest sensitivity rate for the Interpersonal offense-type, which may be due to the multiple predictors that had significant partial effects in the classification model. Notably, the Drug offense-type had the highest level of overall classification; this can be attributed to the high rate of specificity of this group. For this specific offense-type group, the PAI-A scales may be better suited to discriminate those who did not commit these types of offenses, which may be due to the nature of the sample—such as the high degree of multiple offenses committed by participants. The present study supports the construct validity of the PAI-A scale relate to Drug Problems, as it was the only scale that significantly differentiated between groups for this offense type. While interpersonal interactions may be relevant and implicated in treatment for Interpersonal and Property offenses, clinical symptoms and personality traits were not relevant for determining who has committed a Drug/Alcohol offense, suggesting only targeted substance abuse intervention may be more effective for these offenders.

Due to the relatively comparable ability of the PAI-A to classify offenders relative to the MMPI-A, it may be considered for use in classifying juvenile delinquent populations. Further, the PAI-A may hold some advantages in justice settings due to its shorter administration time and lower reading level (Blais et al. 2011). These modifications reduce costs and make the PAI-A uniquely applicable to a population disproportionately affected by reading difficulties and low academic achievement, such as juvenile offenders (Baltodano et al. 2005; Hammond 2007). Previous research has demonstrated support for the use of offender classification and treatment matching in rehabilitation due to lower costs and better outcomes through tailoring of intervention to the individual (Vieira et al. 2009). Limitations of the current study include the nature of the sample, namely convenience sampling and the overlapping of offenses for participants, as a majority of the sample had charges for multiple types of offenses. The large number of scales with potential for utilization in offender classification necessitates further narrowing of the approach, utilizing larger samples to increase generalizability. Future directions may include examining personality variables that can be detectable as precursors to commission of criminal offenses. Future examination of differences among gender and racial/ethnic groups is recommended to allow for greater understanding of the PAI-A’s performance in offender classification. Early detection of youth at risk for particular types of offenses can aid in prevention efforts targeted at specific risk factors, rather than applying interventions when individuals are already involved in the justice system and thus exposed to further risk factors.

Compliance with Ethical Standards

Conflict of Interest

On behalf of all authors, the corresponding author states that there is no conflict of interest.