criminal recidivism among juvenile offenders: testing the incremental and predictive validity of...

17

Click here to load reader

Upload: monica-e-epstein-and-norman-g-poythress

Post on 12-Jan-2017

215 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

Criminal Recidivism among Juvenile Offenders: Testing the Incremental and PredictiveValidity of Three Measures of Psychopathic FeaturesAuthor(s): Kevin S. Douglas, Monica E. Epstein and Norman G. PoythressSource: Law and Human Behavior, Vol. 32, No. 5 (Oct., 2008), pp. 423-438Published by: SpringerStable URL: http://www.jstor.org/stable/25144642 .

Accessed: 12/06/2014 21:13

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

Springer is collaborating with JSTOR to digitize, preserve and extend access to Law and Human Behavior.

http://www.jstor.org

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 2: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

Law Hum Behav (2008) 32:423-438 DOI 10.1007/sl0979-007-9114-8

ORIGINAL ARTICLE

Criminal Recidivism Among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

Kevin S. Douglas * Monica E. Epstein

*

Norman G. Poythress

Published online: 6 December 2007

? American Psychology-Law Society/Division 41 of the American Psychological Association 2007

Abstract We studied the predictive, comparative, and incremental validity of three measures of psychopathic features (Psychopathy Checklist: Youth Version [PCL:YV]; Antisocial Process Screening Device [APSD]; Childhood

Psychopathy Scale [CPS]) vis-a-vis criminal recidivism

among 83 delinquent youth within a truly prospective design. Bivariate and multivariate analyses (Cox propor tional hazard analyses) showed that of the three measures, the CPS was most consistently related to most types of recidivism in comparison to the other measures. However, incremental validity analyses demonstrated that all of the

predictive effects for the measures of psychopathic features

disappeared after conceptually relevant covariates (i.e.,

substance use, conduct disorder, young age, past property crime) were included in multivariate predictive models.

Implications for the limits of these measures in applied juvenile justice assessment are discussed.

Keywords Youth psychopathy Juvenile justice Risk assessment Prediction of recidivism PCL:YV:CPS:APSD

Despite declining rates of most types of criminal behavior

by adolescents in recent years (FBI Uniform Crime Reports

K. S. Douglas (El) Department of Psychology, Simon Fraser University, 8888 University Drive, Burnaby, BC, Canada V5A 1S6 e-mail: [email protected]

M. E. Epstein N. G. Poythress

University of South Florida, Tampa, FL, USA

2004, for rates between 1994 and 2003), it remains a

serious problem. In 2003, there was an estimated 1.3 mil lion arrests of juveniles (persons under 18) for criminal offenses (FBI Uniform Crime Reports 2004). Many of these were violent in nature (250,000+). Further, in recent

years the juvenile justice system has become increasingly punitive (Bonnie and Grisso 2000; Kupchik et al. 2003; Otto and Borum 2004). A concomitant need for risk assessment to differentiate low- and high-risk juvenile offenders has emerged. For this reason, a large amount of research has focused on identifying important risk factors for criminal and violent behavior among adolescents. One of the more commonly studied risk factors for criminal behavior is psychopathy. Most scholarship on psychopathy has been in the context of adult personality functioning, in which it is construed as a personality disorder consisting of behavioral tendencies (e.g., aggression, risk-taking behav

ior), affective deficits (e.g., callousness, shallow affect, lack of empathy), and interpersonal features (e.g., manip

ulativeness, dominance, grandiosity). There is ample evidence from adult-based research

demonstrating a link between psychopathy, typically as measured by the Hare Psychopathy Checklist-Revised

(PCL-R, Hare 1991, 2003) or its screening version, the Hare Psychopathy Checklist: Screening Version

(PCL:SV, Hart et al. 1995), and both general and violent recidivism. Meta-analyses have tended to show average

correlations approximately in the .25-30 range (for reviews and meta-analyses of the PCL-R and PCL:SV, see Douglas et al. 2006; Gendreau et al. 2002 [and Hemphill and Hare 2004, in rebuttal]; Guy et al. 2005; Hare 2003; Hemphill etal. 1998; Salekin etal. 1996; Walters 2003a, b), with some meta-analyses finding smaller effect sizes (rs = .16?.21) for violence (Gendreau et al. 2002; Guy et al. 2005).

? Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 3: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

424 Law Hum Behav (2008) 32:423-438

The downward extension of the psychopathy construct to youth is a recent occurrence that is controversial on

developmental, theoretical, and conceptual bases (see Edens et al. 2001; Frick 2002; Johnstone and Cooke 2004;

Lynam 2002; Seagrave and Grisso 2002; Steinberg 2002).

Despite this, the pressing need in juvenile justice settings to

identify reliable risk factors for adolescent antisocial behavior has led researchers to study the relationship between psychopathic personality features and crime in

this age group. Studies tend to indicate that, regardless of

theoretical or developmental concerns, measures of youth

psychopathic features predict antisocial behavior with

meaningful effect sizes. For instance, Edens et al. (2001) concluded that, on average, there was a moderate associ

ation between various youth psychopathy measures and

relevant external criteria such as conduct disorder,

aggression, and impulsivity. Several contemporary measures of youth psychopathic

features have been developed, most notably the Hare

Psychopathy Checklist: Youth Version (PCL:YV; Forth et al. 1997, 2003), the Antisocial Process Screening Device

(APSD; Frick and Hare 2001), and the Childhood Psy

chopathy Scale (CPS; Lynam 1997). The Youth

Psychopathic Traits Inventory (YPI; Andershed et al.

2002), and the Psychopathy Content Scale (PCS; Murrie

and Cornell 2002) of the Millon Adolescent Clinical

Inventory (Millon 1993) are more recent additions to the

literature. Despite some variation, most studies show an

association between these measures and antisocial conduct.

For instance, many, though not all, studies using the PCL measures with youth (PCL-R, PCL-R modified for adoles

cents, or PCL:YV) indicate an association with criminal

history or institutional infractions (Brandt et al. 1997; Edens et al. 1999; Forth et al. 1990; Kosson et al. 2002; Murrie et al. 2004; Rogers et al. 1997; Salekin et al. 2004),

although there are exceptions (Campbell et al. 2004). In a

meta-analysis of 14 samples, Edens and Campbell (2007)

reported that the Hare psychopathy measures as applied to youth (18 and under) predicted institutional infrac

tions (physically violent and other aggressive incidents; rs =

22-25).

Fewer studies have been conducted of community recidivism. In a retrospective follow-up, youths who scored

high on the PCL: YV were three (violent recidivism) to four

(general recidivism) times more likely to recidivate than

those who scored low (Gretton et al. 2001). Gretton et al.

(2004) reported that the PCL:YV, completed retrospec

tively from files of adolescent offenders, predicted adult

criminally violent recidivism an average of 10 years post release. Further, in one of the few tests of the incremental

validity of the PCL:YV, Gretton et al. (2004) found that

file-only, retrospective PCL:YV ratings (total and Factor 2) remained significant after controlling for criminal history

and conduct disorder symptoms. In a smaller sample (N = 64) based on file-review PCL:YV ratings, O'Neill et al. (2003) reported moderate sized correlations between the PCL:YV and any recidivism. A further file-review

study of 95 adolescent offenders, however, found no rela

tionship between the PCL:YV and community recidivism

(Marczyk et al. 2003). Some studies have used the adult PCL-R modified for youths, and have reported similar

predictive findings (Brandt et al. 1997; Forth et al. 1990; Ridenour et al. 2001), although some have reported asso

ciations that are accounted for by other variables, such as

conduct disorder symptoms (Langstrom and Grann 2002). In one of the few truly prospective studies of the

PCL:YV for the purpose of predicting community recid

ivism, Corrado et al. (2004) followed 161 male

adolescents for 14.5 months after release from a youth detention facility in Canada. They reported that the

PCL:YV predicted any, violent, and non-violent recidi

vism, albeit with small to moderate effects (AUCs ranged from .58 to .68).1 In other analyses based on this sample, Vincent et al. (2003) reported, however, that a cluster

solution that comprised high scorers on all three factors

(interpersonal, affective, behavioral) had a significantly higher percentage of violent recidivists (approximately 50%) compared to cluster solutions consisting of high scorers on only one or two factors (23-27%). In another

community prospective study, Schmidt et al. (2006)

reported significant effects between the PCL:YV and

subsequent violent and general (though not non-violent) recidivism for 127 Canadian court-referred youths. How

ever, Edens and Cahill (2007) failed to observe significant

prediction of the PCL:YV for later community violent or

general recidivism.

Turning to the APSD, several studies?mostly using

parent and/or teacher ratings?have reported associations

between this measure and increased risk for aggression and

antisocial behavior in samples of clinic-referred and com

munity youth (Frick 1995; Frick et al. 2000a, b, 2003; Frick

and Hare 2001). Fewer studies have been reported for

justice-involved youth. In one such study of 69 arrested

youth, the correlations between the APSD and any recidi

vism (measured prospectively) were .32-36 for the self

report version, and .24-40 for the parent-report version

(.40 for total and impulsivity scales, .24 for callous

unemotional scale) (Falkenbach et al. 2003). Two studies

of institutional aggression reported significant correlations

(.25-35) with the APSD self-report version (Murrie et al.

2004; Spain et al. 2004). In the latter study, however, the

1 AUC stands for "area under the curve" of a receiver operating

characteristic analysis, with .50 representing chance prediction, and

1.0 representing perfect prediction.

4y Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 4: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

Law Hum Behav (2008) 32:423-438 425

APSD was no longer significant once the CPS was in the

predictive model. In the Spain et al. (2004) study, a modified version of

the CPS was more strongly and consistently related to

institutional infractions and violence than the APSD or

PCL:YV (being the only one of the three to retain signif icance in multivariate analyses), with bivariate correlations with the rate of any and violent infractions of .43 and .45,

respectively. In Falkenbach et al's (2003) study, the CPS had strong correlations with community recidivism when rated by parents (.40s-.50s), though smaller correlations when rated by self (.20s-.30s).

Across the studies reviewed, there is some support for the relationship between each measure and criminal recidivism. However, viewed as a whole, the literature on

the prediction of adolescent criminal recidivism with measures of psychopathic features remains limited. Few studies are prospective, and some that are (Brandt et al.

1997) have used the (modified) PCL-R rather than the PCL:YV per se. Many PCL:YV studies use file-only rat

ings, a nonfatal methodological characteristic to be sure,

but one that limits the comprehensiveness of the assess ment process and is known to lower scores and restrict

range (Hare 2003). Further, to our knowledge, there has yet to be a prospective comparison ofthe PCL:YV, APSD, and CPS for the prediction of community recidivism among adolescent offenders.

In addition, most studies have not gone beyond reporting the absolute magnitude of association between psychopa thy measures and recidivism to evaluate the incremental contribution of psychopathy measures once other relevant correlates are controlled, a procedure recommended for

testing their utility beyond other factors (Douglas et al.

2006; Edens et al. 2006). In a rigorous example of incre mental validity analysis from the adult psychopathy literature, Skeem and Mulvey (2001) showed a substantial reduction in the predictive validity (from a correlation of .26-. 12) of the PCL:SV Part 2, once numerous relevant

covariates were controlled via propensity analyses. How

ever, there are few such examples from the youth

psychopathy literature. Gretton et al. (2004) reported that the PCL:YV remained associated with later community recidivism after controlling for past criminality. Others have found that the PCL:YV added incrementally to indi ces of externalizing behavior and criminal history (Schmidt et al. 2006). Langstrom and Grann (2002), however, found that the PCL:YV was reduced to statistical nonsignificance once conduct disorder symptoms were entered into the

predictive model. As such, we selected several domains of risk factors

that, based on theory and past research, ought to predict criminal behavior among adolescents (see, generally, Cot tle et al. 2001; Corrado et al. 2002; Dahlberg and Simon

2006; Fried and Reppucci 2002). These domains included

(a) demographics (here, age); (b) past antisocial behavior

(i.e., past crime, antisocial behavior while incarcerated); (c) substance-related problems (i.e., drug and alcohol abuse and dependence); and (d) mental health problems (i.e., anger, conduct disorder). We were interested in whether the various measures of psychopathic features could pre dict recidivism above and beyond these risk factors; that is, do measures of psychopathy show predictive utility incre

mentally beyond common risk factors for antisocial behavior?

The present study was conducted to address these col

lective limitations of the research corpus. As some studies indicate that certain sub-scales are more strongly related to

antisocial behavior than others (in particular, the behav ioral components of the PCL:YV), we conducted tests of each measure's sub-scales as well. Finally, reflecting calls

in the literature to investigate multiple aspects of offend

ing behavior (Hart 1998), we evaluated not only the

relationship between the psychopathy measures and the occurrence of recidivism, but the nature of recidivism (i.e.,

violent, non-violent, weapons-related), time to recidivism,

and volume and diversity of recidivism. Given the value of an incremental validity approach generally (i.e., Sechrest

1963), and the small number of studies that have addres sed this issue for youth psychopathy measures vis-a-vis other risk factors theoretically related to antisocial

behavior, we also tested three measures of youth psy

chopathic features against each other, and alongside other

putative risk factors and demographics (age, previous antisocial behavior, mental health diagnoses and substance related problems).

Method

Participants and Setting

Ninety-six male adolescent offenders between the ages of 11 and 18 (M = 15.77, SD = 1.35) who were remanded to a secure facility located in west central Florida were recruited to participate in the study. Eleven residents declined. Of the 85 participants, 42 were enrolled in a Sex Offender Program (SOP) and 43 residents in a secure res idential Halfway House Program (HHP; e.g., nonsexual

offenders). Data from one SOP and one HHP participant were dropped from analyses due to incomplete measures,

leaving 83 participants for analysis. The sample was pri marily Caucasian (79%), while the remainder identified their race as African American (16%) or Latino (4.9%). Based on the Hollingshead (1975) Index, custodial socio economic status for the majority of the participants was lower to lower-middle class (M = 5.38, SD = 2.0). Length

^ Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 5: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

426 Law Hum Behav (2008) 32:423^38

of incarceration at the facility ranged from 186 to

1,110 days (M = 410.7, SD = 172.1).

Measures of Psychopathic Features in Youth

Hare Psychopathy Checklist: Youth Version (PCL.YV)

The PCL:YV is a 20-item, clinician-administered instru ment patterned after the adult PCL-R but designed to

evaluate psychopathic personality features in youth aged 12-17. Like the PCL-R, the PCL:YV relies on information

gleaned from interview, file review, and collateral infor mation. Scores range from 0 to 40, with each of the 20

items rated on a three-point (0-2) scale. The two-factor, four-facet model encompassing Factor 1 (Interpersonal and

Affective) and Factor 2 (Behavioral and Antisocial Fea

tures) was used in this study. We used the pre-published research version (Forth et al. 1997) although it does not

differ from the published version in terms of administration

procedure or item definitions (Forth et al. 2003).

Adequate psychometric properties of the PCL:YV have

been reported in several studies (see Edens et al. 2001; Forth and Burke 1998; Forth and Mailloux 2000; Kosson

et al. 2002; Vincent and Hart 2002). The PCL:YV manual

(Forth et al. 2003, p. 55) reports excellent interrater

agreement for the Total score (ICC = .90-92) across jus tice-involved youth in institutions, on probation, and in

clinic/community samples. Across various samples, inter

nal consistency for the Total score, as indexed by Cronbach's alpha (.85-94) and item homogeneity as

indexed by mean inter-item correlation (.23-43) are also

good. The PCL:YV manual also reports satisfactory inter nal consistency for the four-facet model, with alphas of .71

(Interpersonal), .72 (Affective), .68 (Behavioral), and .77

(Antisocial) in a sample of 505 incarcerated males (Forth et al. 2003, p. 65). In the present sample, alpha for the total

score, and facets 1-4, respectively, were .71, .53, .55, .43,

and .51.2 Means (SDs) for the total and facets (1-4) were,

respectively, 22.8 (5.9), 3.1 (1.9), 4.3 (2.1), 6.0 (1.9), and

7.0 (2.1). Ten paired ratings of PCL:YV administrations were

collected throughout the data collection period to assess

interrater reliability. Employing absolute agreement intra

class correlation (ICQ procedures for both single ratings

(ICCi) and averaged ratings (ICC2) across the 10 paired

ratings (see Bartko and Carpenter, 1976), acceptable

2 Other investigators have reported reliability based on the three

factor model. Lee et al. (2003) obtained as of .59, .63, and .44,

respectively, for the three factors (which are the same as the first three

of four factors used in the present research), whereas Vincent (2002) obtained as of .71, .67, and .60, respectively. Skeem and Cauffman

(2003) reported as of .57, .56, and .22, respectively.

ratings of ICCX = .82 and ICC2 = .90 were obtained for PCL: YV Total scores. Inter-rater reliabilities (/CCO for the

Interpersonal, Affective, Behavioral, and Antisocial sub scales were 0.86, 0.43, 0.83, and 0.61 respectively. One

highly discordant paired rating strongly affected the reli

ability coefficient of the Affective sub-scale such that its removal increased the coefficient to 0.71.

Antisocial Process Screening Device (APSD)

We employed the 20-item self-report version of the APSD that has been developed for research with youths 12 18 years old (see Caputo et al. 1999). The original measure

(Frick and Hare 2001) was designed for completion by parents or teachers; however, a self-report version was

created by changing the items from third-person ("He

keeps the same friends") to second-person ("You keep the same friends"). Frick et al. (2000a, b) noted "[s]elf-report becomes more reliable and valid as a child enters adoles

cence, especially for assessing antisocial tendencies and

attitudes that may not be observable to parents and other

significant adults" (p. 13).

Except for a study by Rogers et al. (2002), which

obtained an alpha of .58, internal consistency for the self

report Total score has been good, with alphas ranging from

.72 to .82 (mdn = .76) (Cruise et al. 2000; Lee et al. 2003; Murrie and Cornell 2002; Pardini et al. 2003). However, internal consistency has been modest to weak for the three

factor scores, ranging from .56 to .72 for Narcissism

(NAR; mdn = .68), .36-.56 for Callous-Unemotional (C-U; mdn = .52), and .44 to .60 for Impulsivity (IMP; mdn =

.56). We used the official factor structure reported in the

APSD manual (Frick and Hare 2001) for analyses. In the

present sample, alpha for the APSD total, NAR, IMP, and

C-U scales was .76, .69, .56, and .19, respectively.3 Mean

3 The low alphas, particularly for C-U, are consistent with the

literature, and likely not a function of this particular sample. In a

quantitative synthesis of APSD reliability estimates based on 11

samples and 1,253 participants, Poythress et al. (2006) reported median alpha values that were highly consistent with our estimates.

For the APSD NAR scale, the median alpha was .69 (in our sample, a = .69), for IMP it was .53 (in our sample, a = .56), and for the C-U

scale, it was .46 (in our sample, a = .19). Because of concern that low

internal consistency could reduce validity coefficients, we calculated

"reliability enhanced" versions of the APSD scales for some validity

analyses by removing the most poorly performing items from each

scale and sub-scale (validity results summarized in footnote 5). For

the sake of thoroughness, we also attempted to increase the reliability of each PCL:YV and APSD index. The largest increase in alpha was

associated with the C-U scale from the APSD, which increased from

.19 to .50 after removing items 19 and 20. Alpha for other APSD and

PCL:YV indices increased negligibly (range .01-06). We also note

that the removal of items 19 and 20 produces a revised factor model

tested and supported by Poythress, Dembo et al. (2006).

?} Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 6: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

Law Hum Behav (2008) 32:423-438 427

(SD) scores for the APSD total, NAR, IMP, and C-U scales were 16.1 (5.9), 5.1 (2.8), 5.0 (2.1), and 4.2 (1.8),

respectively.

Child Psychopathy Scale

The original version of the CPS had 41 items and was

intended to be rated by parents (Lynam 1997). It was

modified to include 55 items in a self-report format

(Lynam, personal communication). Each item is rated on a

2-point scale (0 = no, 1 = yes). Lynam (1997) reported excellent internal consistency (Cronbach's a = .91) for the

original CPS. Evidence for the construct validity of the measure was reflected through positive correlations (range .19-39) with various measures of delinquency and exter

nalizing problems, and negative correlations with indices

of internalizing problems. As the CPS does not have an

empirically-derived factor structure, we derived three fac tors rationally in consultation with the CPS author in order to facilitate comparisons with the APSD and PCL:YV. These rationally derived factors have been used in previous studies (Spain et al. 2004). In the present sample, alpha for the total, interpersonal, affective, and behavioral CPS scales were .87, .73, .68, and .71, respectively. Mean (SD)

scores for the CPS total, interpersonal, affective, and behavioral CPS scales were 5.31 (1.85), 1.85 (.63), 1.27

(.63), and 1.55 (.84), respectively.4

Measure Intercorrelations

The total scores for the psychopathy measures correlated with one another as follows: PCL:YV and APSD (r = .35, p < .01); PCL:YV and CPS (r = .27, p < .02); CPS and APSD (r = .79, p < .001). PCL:YV sub-scales produced correlations with the sub-scales on the APSD ranging from .04 (ns) between Facet 1 (Interpersonal) and APSD

Impulsivity, to .29 (p < .001) between Facet 2 (Affective) and APSD Narcissism. PCL:YV sub-scales correlated with the sub-scales of the CPS with a range of .05 (ns) between Facet 3 (Impulsivity) and CPS Interpersonal, to .34

(p < .001) between Facet 4 (Antisocial) and CPS Impul sivity. Correlations between APSD and CPS sub-scales

ranged from .35 (p < .001) between APSD C-U and CPS

Interpersonal to .62 (p < .001) between APSD Narcissism and CPS Affective.

As the CPS sub-scales were rationally derived, we

provide additional detail on the correlations between its three 'scales' and the corresponding scales on the APSD

Scale scores on the CPS are the means of item scores, hence their low magnitudes.

and PCL:YV. CPS Interpersonal correlated with the cor

responding PCL:YV scale (Facet 1) at .22 (p < .05), and the APSD scale (Narcissism) at .54 (p < .001). Its Affec tive scale correlated with PCL:YV Facet 2 at .25 (p < .03) and APSD C-U at .38 (p < .001). Finally, its Impulsivity scale correlated with the PCL:YV Facet 3 (Impulsivity) at

.24 (p < .05), PCL:YV Facet 4 (Antisocial) at .34

(p < .01), and APSD Impulsivity at .60 (p < .001). These cross-measure correlations indicate at least preliminary evidence of concurrent validity of the rationally derived scales of the CPS.

Covariates

Mini-International Neuropsychiatric Interview

for Children and Adolescents (MINI-Kid)

The MINI-Kid (Sheehan et al. 2001) was used to assess

selected DSM-IV Axis I disorders (e.g., generalized anxiety, major depressive episode, conduct disorder), including alcohol and substance abuse/dependence. A full day of

training was provided to the authors by one of the authors of the MINI-Kid. The instrument is a brief diagnostic struc tured interview designed to meet the need for a short but accurate psychiatric assessment. The MINI-Kid is a

downward version of the adult Mini-International Neuro

psychiatric Interview (MINI; Sheehan et al. 1997) that frames questions in language that is easy for children and adolescents to understand. Most prompts can be responded to by "yes" or "no," and limited clinical judgment is

required to complete the assessment. The adult MINI diag nostic concordance with the Structured Clinical Interview for DSM-III-R Diagnosis (SCID) has reported kappas ranging between .50 and .90 for each disorder assessed (with one exception: Current drug dependence kappa

= .43).

Sensitivity was greater than .70 for all but three values

(dysthymia, OCD, and current drug dependence); specificity and negative predictive values (NPV) were 0.85 or higher across all diagnoses; and positive predictive values (PPV) ranged from 0.45 to 0.75, suggesting that it is a valid measure of eliciting symptom criteria used to make DSM-IV diag noses. The MINI-Kid has been used in studies of Tourette's

syndrome (Silver et al. 2001) and the impact of residential

placement on pharmacotherapy with youths (Bastiaens 2004). Although the MINI-Kid is modeled directly after

MINI, to date, no studies of its psychometric properties have been published. Unfortunately, we do not have interrater

reliability data from this sample. For incremental validity analyses, we used indices from

the MINI-Kid with a conceptual relationship to crime and violence?alcohol and drug-related disorders (diagnoses for alcohol abuse and dependence, drug abuse and

4y Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 7: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

428 Law Hum Behav (2008) 32:423^138

dependence, and substance abuse and dependence), con

duct disorder, and attention-deficit hyperactivity disorder. When possible (i.e., for those disorders for which the MINI-Kid permits it), we used symptom counts rather than dichotomous diagnostic indications (conduct disorder;

ADHD).

The Massachusetts Youth Screening Instrument-Second

Version (MAYSI-2)

The MAYSI-2 (Grisso et al. 2001) is a self-report inventory designed to screen 12-17 year-old youth in juvenile justice facilities who may have special mental health needs (Grisso and Barnum 2000). Youths respond "yes" or "no" to each of 52 items regarding their thoughts, feelings, and behaviors "within the past few months." These items contribute to seven factor analytically-derived scales representing prob lem areas. Although the entire measure was administered,

only two scales were included in the present analyses: Alcohol/Drug Use (8 items: e.g., "Have your parents or

friends thought you drink too much?"), and Angry-Irritable (9 items: e.g., "Have you felt angry a lot?").

Norms for the MAYSI-2 were developed on a sample of

1,279 youths, aged 12-17 inclusive, recruited through various juvenile justice agencies in Massachusetts during 1997. Coefficient alpha values were >.70 for all scales in

the normative sample except for two (Thought Disturbance a = .61, and Trauma Experience (boys) a = .63). Internal consistencies for the MAYSI-2 factors used in this

study were as follows: Alcohol/Drug Use, a = .88; Angry Irritable, a = .82.

Past Antisocial Behavior

We used two indices of past crime?crimes against persons

and crimes against property (each coded dichotomously).

Fifty-three participants (62.4%) had previous crimes

against persons, and 54 (63.5%) had previous crimes

against property. We also calculated two rate variables:

prior offending (number of crimes divided by age at

admission) and institutional infractions (number of insti

tutional infractions divided by time in program). Past crime was coded from official juvenile justice records maintained

by the Florida Department of Juvenile Justice.

We categorized the covariates above into four catego

ries: (a) mental health problems (conduct disorder; MAYSI-2 Angry-Irritable; ADHD), (b) substance-related

problems (substance-related diagnostic variables; MAYSI

2 Alcohol/Drug Use), (c) prior antisocial behavior (past

property crimes; past person crimes; rates of past crime and

institutional infractions), and (d) age at discharge.

Outcome Recidivism Data

Recidivism was coded from official Florida DJJ records. We classified formal contacts (charges and convictions) into the following dichotomously scored recidivism cate

gories: (a) any; (b) violent; (c) non-violent; and (d) weapons. A violent offense was defined as any that involved actual or potential harm to persons, including assault, assault with a weapon, or robbery. Most violent

offenses were assaults. Note that these categories are not

mutually exclusive, but are intended to describe different

types of recidivism. Any recidivism is a summary category that was coded positive for any post-release formal contact.

This index was divided into (mutually exclusive) catego ries of violent and non-violent recidivism. Weapons-related offenses are not independent of the other offense catego ries, but could appear in either the violent or non-violent

category depending on the nature of the charge (i.e., simple possession was considered non-violent; assault with a

weapon was considered violent). These more detailed

categories were created to address the specificity of recidivism in which participants might be involved. We observed the following recidivism base rates (all/83): (a)

any (n = 37; 44.6%); (b) violent (n = 19; 22.9%); (c) non

violent (n = 32; 38.6%); and (d) weapons (n = 6; 7.2%). In order to test whether the psychopathy measures pre

dicted not only recidivism per se, but serious repeat recidivism, or "high volume" offending, we first calculated

the total number of re-offenses committed by the cohort of

participants, which was 179. Approximately 10% of the

sample (8 youths) committed well over half of these offenses (n = 109; 60.9%; range = 8-21 offenses). We

then created a dichotomous variable (participant was [1] or was not [0] among the 10% of high-volume offenders) and used this as the dependent variable for a subset of analyses.

We also calculated the number of offense types that

participants engaged in as an index of their diversity of

offending. For this one analysis, we summed across four

independent types of offenses: violent offenses, property offenses, breaches, and drug-related offenses. Hence this

index could range from 0 to 4. Because it was a highly skewed variable, analyses used a natural logarithmic transformation. Note that for all other analyses, property, breach, and drug-related offenses were subsumed under the

'non-violent offense' category.

We attempted to allow approximately 1 year post

discharge prior to collecting recidivism data to increase the

probability that crimes that were actually committed had

time to be processed and entered into the official DJJ

database. The minimum follow-up time was 310 days; the

maximum was 1,282 days. The mean number of days

between discharge and the date that we requested recidi

vism data was 874.77 (SD = 223.27; Mdn = 912).

4y Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 8: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

Law Hum Behav (2008) 32:423-438 429

Procedure

Graduate level clinical research assistants were trained on

protocol administration in a 2-day workshop. Training for PCL:YV administration included didactic presentations on

the psychopathy construct and a review of differences between the adult and youth versions of the PCL. Research assistants also completed scoring of videotaped PCL-R

interviews with feedback from supervisors. There were two

graduate students who collected most of the data (65

cases). One graduate student had to stop involvement with the project after two cases. The remaining 17 cases were

collected by the two faculty involved in the project (8 and 9

cases, respectively). Participants were recruited using methods approved by the University of South Florida's Institutional Review Board and the Florida Department of Juvenile Justice Institutional Review Board. Assent from

each participating adolescent was obtained and parents or

guardians were informed of the study by a letter that

explained that they could exclude their child from the study by contacting the investigators (passive consent). Assent for voluntary participation was obtained face-to-face dur

ing recruiting interviews with youths who were provided a

description of the study and afforded an opportunity to ask

questions. These same research assistants returned to the

facility at a later date to administer the single-session, 2-3 h protocol to those residents who had given prior assent. The PCL:YV ratings were based on data from institutional and juvenile justice files as well as the PCL:YV semi-structured interview.

Analyses

Bivariate Analyses

We used Receiver Operating Characteristic (ROC) anal

yses as the primary index of predictive accuracy at the bivariate level. ROC analysis can be used to estimate

predictive accuracy when there is a continuous predictor and a dichotomous outcome. The effect size that it yields, called the Area under the Curve (AUC), is not greatly attenuated by deviations from base rates of 50% in the outcome variable of interest (here, recidivism) (Mossman 1994). "Receiver operating characteristic" analysis is so

named because the "receiver" of the data (the researcher or clinician) can "operate" (make decisions) at any given point on the curve (Metz 1978), hence understanding its

predictive "characteristics" across the whole range of

scores. The curve produced by ROC analysis is a func tion of the plotting of sensitivity (true positive rate

[TPR]) of the predictor against the false positive rate

(FPR [1-specificity]) (Mossman and Somoza 1991). At

any and all levels of sensitivity, the "receiver" knows the

specificity. The AUC, an index of overall predictive accuracy,

ranges from 0 (perfect negative prediction), to .50 (chance

prediction), to 1.0 (perfect positive prediction). The AUC is the probability that a randomly chosen person who scores

positive on the dependent measure (in this case, is actually violent) will score higher than a non-violent person on the

predictor measure (Mossman and Somoza 1991). Although there would seem to be no general agreement in the area of risk assessment regarding what values of AUC represent small, moderate, or large effect sizes, formulae from

Dunlap (1999) indicate that an AUC value of .71 corre

sponds to a standardized difference score (d) of .80, which was suggested by Cohen (1992, 1988) to represent an effect of large magnitude. An AUC value of .64 corresponds to a

d of .50 (moderately sized effect). Because follow-up times varied greatly across partici

pants, we chose a uniform follow-up period for these

analyses. The minimum follow-up time (310 days) was

met by the entire sample. However, limiting the measure ment of recidivism to this time period resulted in low base rates for violent recidivism (<10%), one of the main out come indices, and hence would not have provided a

reasonable test of the measures' predictive accuracy. We

were able to retain approximately 90% of the sample (n = 71) with a uniform follow-up of 620 days. Using this

follow-up period, the base rates of the main outcome cri

teria were high enough to justify analyses: any (40.9%); violent (25.4%); and non-violent (36.6%). The base rate for

weapons-related recidivism was low (5.6%), and hence we

recommend caution in interpreting those particular analyses.

Multivariate Analyses

Because we had uneven follow-up periods across partici

pants, survival analysis was used as the primary multivariate procedure. "Survival analysis" is the term

applied to a cluster of analyses that uses time to an event as the dependent measure, models the time to an event, and

controls for unequal follow-up times between participants (Luke and Homan 1998). Survival analysis also permits the evaluation of changes in the hazard function (rate of fail

ure) as a function of relevant covariates. Specifically, the Cox proportional hazards model?a semi-parametric pro

cedure that allows evaluation of how numerous categorical or continuous covariates affect the hazard rate of the

dependent variable?was used. The procedure is semi

parametric because it does not model the shape of the hazard function, but it does model the effects of covariates on the hazard function (Luke and Homan 1998). The Cox

? Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 9: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

430 Law Hum Behav (2008) 32:423-438

proportional hazards model also provides estimates of the

magnitude of effect of covariates, an advantage over other

survival procedures, such as the Kaplan-Meier method.

Results

Bivariate Relationship between Youth Psychopathy Measures and Criminal Recidivism

For informational purposes, we first report AUCs that were

significant at a traditional alpha level of .05 (the details of

non-significant AUCs are available upon request). However,

as there was a large number of significance tests conducted, we applied an alpha correction procedure for interpretive purposes. Each psychopathy index was tested against four outcome criteria, and hence we set alpha for interpretive purposes to .05/4 = .0125. A summary ofthe indices which met this more conservative alpha level is presented after a

report of those that met the traditional alpha level.

PCL:YV total score did not significantly predict the any outcome criterion (AUC = .55, SE = .07, 95% CI = .42

.69, ns) or the non-violent outcome criterion (AUC = .50,

SE = .07, 95% CI = .36-.63, ns). However, it was signif

icantly related to both violent recidivism (AUC = .66, SE = .07, 95% CI = .52-/79,/? < .05) and weapons-related recidivism (AUC = .81, SE = .05, 95% CI = .71-.91,

p < .05). The APSD total score followed the same pattern, in that

it did not significantly predict the any outcome criterion

(AUC = .63, SE = .07, 95% CI = .50-.76, ns) or non

violent recidivism (AUC = .58, SE = .07, 95% CI = .45

.72, ns). However, it was significantly predictive of violent

recidivism (AUC = .69, SE = .07, 95% CI = .55-.83,

p < .02) and weapons-related recidivism (AUC = .90, SE = .04, 95% CI = .82-.98, p < .01).

The CPS total score predicted any recidivism (AUC =

.67, SE = .06, 95% CI = .55-.80, p

= .01), non-violent

recidivism (AUC = .64, SE = .07, 95% CI = .51-/77,

p < .05), violent recidivism (AUC = .70, SE = .07, 95%

CI = .57-84, p =

.01), and weapons-related recidivism

(AUC = .83, SE = .13, 95% CI = .58-1.1, p < .05).

Turning to the measures' sub-scales, none of the

PCL:YV facets were predictive of the any outcome crite

rion (AUC range = .52-62) or non-violent recidivism

(AUC range = .45-62). For violent recidivism, Facet 4

(AUC = .65, SE = .07, 95% CI = .51-.80, p = .05) was

predictive, and for weapons-related recidivism, only Facet

2 (Affective) was predictive (AUC = .84, SE = .05, 95%

CI = .74-.95, p =

.02).

Of the APSD sub-scales, only IMP predicted the any outcome criterion (AUC = .64, SE = .07, 95% CI = .51

.77, p < .05). None of the sub-scales were significantly

predictive of non-violent recidivism (AUC range = .54

.61). For violent recidivism, both NAR (AUC = .70, SE = .07, 95% CI = .56-.84, p = .01) and IMP (AUC =

.70, SE = .07, 95% CI = .57-.83, p = .01) were signifi cantly predictive. Similarly, both NAR (AUC = .80, SE = .09, 95% CI = .63-.98, p < .05) and IMP (AUC =

.89, SE = .05, 95% CI = .81-.98, p < .01) were signifi cantly predictive of weapons-related recidivism.

Finally, for CPS sub-scales, both the Interpersonal and

Behavioral/Impulsivity factors were predictive of each type of outcome, whereas the Affective factor did not predict any of the outcomes. The Interpersonal (AUC = .66, SE = .07, 95% CI = .53-.79, p = .02) and Behavioral/Impulsivity factors (AUC = .68, SE = .07, 95% CI = .55-.81, p = .01) were predictive of the any outcome criterion. Similarly, they were predictive of violent recidivism (Interpersonal AUC = .67, SE = .07, 95% CI = .53-.81, p < .05; Behav

ioral AUC = .73, SE = .07, 95% CI = .58-.87, p < .005), non-violent recidivism (Interpersonal AUC = .66, SE =

.07, 95% CI = .53-.79, p < .05; Behavioral AUC = .66, SE =

.07, 95% CI = .53-.80, p < .05), and weapons

related recidivism (Interpersonal AUC = .80, SE = .12, 95% CI = .56-1.0, p < .05; Behavioral AUC = .83, SE =

.07, 95% CI = .68-.97, p < .05). At the more conservative alpha level of .0125, the

PCL:YV total score was not significantly related to any outcome. The APSD total score was significantly predic tive only of weapons-related recidivism. The CPS total score was significantly predictive of any and violent

recidivism. None of the PCL: YV sub-scales met this more

conservative alpha level. APSD NAR and IMP were sig nificantly predictive of violent recidivism, and NAR was

predictive of weapons-related recidivism as well. CPS

Behavioral/Impulsivity was predictive of any recidivism and violent recidivism.5

High-Volume and Diverse Offending

As it was not possible to use a uniform follow-up period for

these analyses (we did not record the date of every offense,

just that of the first offense), instead of AUC analyses, we

used partial correlational analyses, controlling for time at

5 Owing to the concern that the low reliability of some of the scales

could have reduced the predictive validity of these scales, we

conducted additional ROC analyses (for any, violent, and non-violent

re-offenses) for the APSD and PCL:YV using the "reliability enhanced" sub-scales described in the Methods section. For the

PCL:YV tests, AUC values increased, on average, by .007. For the

APSD, the mean change was -.001. Overall, our evaluation of these

changes was that they were minimal and did not warrant further

investigation (i.e., in the multivariate analyses) or substitution of the

reliability enhanced scales for the original scales throughout all

analyses.

?) Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 10: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

Law Hum Behav (2008) 32:423-438 431

risk. As with the previous bivariate analyses, we first report

analyses based on alpha = .05. We then provided a sum

mary of which indices remained significant using a more

conservative alpha to correct for multiple tests. For this set

of analyses, each index was used to predict two outcomes, and hence we set this more conservative alpha at .025.

Of the total scores, the PCL: YV predicted neither high volume nor diverse recidivism, the APSD predicted

diversity of offending (partial r - .24, p = .03), and the

CPS predicted both high-volume (partial rpb = .25,

p = .02) and diverse offending (partial r = .28, p = .01). The only PCL:YV sub-scale that was predictive of either

index was Facet 4 (Antisocial) (partial rpb = .23, p = .04)

with respect to high-volume offending. The only APSD

sub-scale that was predictive was IMP with respect to

diverse offending (partial r = .25, p = .02). Finally, for the

CPS, both the Interpersonal and Behavioral sub-scales

predicted both high-volume (Interpersonal partial rpb =

.27, p = .0V, Behavioral partial r = .24, p

= .03) and

diverse recidivism (Interpersonal partial rpb = .29, p < .01;

Behavioral partial r - .29, p < .01).

The analyses that met the more conservative alpha level of .025 included the following: CPS total score for both

high-volume and diverse offending; APSD IMP for diverse

offending; CPS Interpersonal for both high-volume and

diverse offending; and CPS Behavioral for diverse

offending.

Multivariate Relationship between Youth Psychopathy Measures and Criminal Recidivism

We conducted multivariate analyses to discern which of the three psychopathy measures was independently predictive of each type of outcome. We used Cox proportional hazard

analysis, which models time to event as the outcome. Two

primary sets of analyses were conducted?first for the measures' total scores, and second for the measures' sub

scale scores. For the sub-scale analyses, we first conducted

analyses within-scale to determine which of each mea sures' sub-scales was independently related to recidivism.

Then, we pitted these sub-scales against one another in a

further series of analyses. We performed this two-step

procedure with the sub-scales because with an N of 83 and 10 sub-scales across measures, entering all sub-scales into

one Cox regression analysis would have underpowered the

analysis. All analyses used the forward conditional variable

entry method.

In terms of the total score analyses, Table 1 presents the measures that were significantly and independently pre dictive of the different categories of re-offending. As can be

seen, three recidivism variables were significantly pre dicted: any recidivism (CPS), violent recidivism (APSD),

and weapons-related offenses (APSD). Full model details

are presented in the Table 1. Notably, the PCL:YV was not

predictive of any outcome. With respect to sub-scale anal

yses, the results are presented in Table 2.6 As is evident, the

CPS (sub-scales) were most consistently related to the

various recidivism indices, with the APSD (Impulsivity)

being related only to weapons-related offenses. Again, the

PCL:YV was not predictive of any outcome.

Incremental Validity Analyses

Ofthe 15 covariates that we hypothesized would be related to recidivism, only 2 were unrelated to any of the recidi vism outcomes (rate of institutional infractions; rate of pre confinement offending), and hence these were dropped from all subsequent analyses. Given our moderate sample size, we adopted a sequential incremental validity analysis

procedure that minimized the number of covariates in any

given analysis. In Step 1, we identified each covariate that

had a bivariate relationship with any of the types of

recidivism that were significant in multivariate analyses (any; violent; non-violent; weapon-related). In Step 2, we

tested these covariates within their categories (past anti

social behavior; substance-related problems; mental health

problems) in a series of multivariate analyses (Cox

regression) using each relevant recidivism outcome as the

dependent measure. We did this to identify the covariates within categories that were independently related to the various types of recidivism. On Step 3, we used only those

covariates from Step 2 in incremental validity analyses with the psychopathy indices that we previously identified as being independently related to recidivism (i.e., those

reported in Tables 1 and 2). As young age was related to

each type of recidivism, it was included in all incremental

validity analyses as well. Most incremental validity anal

yses involved four covariates, which is a reasonable

number in terms of power, given our sample size. Typi

cally, one covariate from each category was represented in

these analyses, as described below.

The following sets of covariates, then, were used in incremental validity analyses, for the following recidivism outcomes: any and violent (young age; past property offenses; alcohol abuse diagnosis; conduct disorder

6 The following sub-scales were tested against one another for the

following types of recidivism, depending on whether they were

significantly predictive of recidivism in the first series of sub-scale

Cox regression analyses (i.e., those conducted within each measure

with just its own sub-scales): any (APSD Impulsivity; CPS Impul

sivity); violent (APSD Narcissism; PCL:YV Behavioral [Facet 3,

p < .10]; CPS Impulsivity); non-violent (CPS Interpersonal); and

weapons-related (APSD Impulsivity; PCL:YV Affective; CPS Impul sivity). Full details of this series of analyses are available upon request.

^ Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 11: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

432 Law Hum Behav (2008) 32:423^38

Table 1 Final Cox proportional hazard analysis models for total scores of psychopathy measures

Criterion/Predictor B SE (B) Wald eB (95% CI) Sig. Model cp

Any recidivism3

CPS .207 .088 5.498 1.231 (1.035-1.464) .019 .26

Violent recidivism5

APSD .088 .038 5.502 1.092 (1.050-1.425) .019 .26

Weapons-related recidivism0

APSD .165 .067 5.996 1.180(1.034-1.346) .014 .28

Note: Only the variables that were significant in final models are presented. The Model (p in the right-most column is a phi coefficient derived

from the overall model X2 calculated from the formula cp = yJ(X2/N) provided by Rosenthal (1991). It is provided as an effect size estimate of the

overall strength of the predictive model a Model -2LL = 291.51, X2 (df

= 1, N = 82) = 5.57, p < .05 b Model -2LL = 156.63, X2 (df= I, N = 82) = 5.56, p < .05

c Model -2LL = 46.02, x2 (df= \,N = 82) = 6.46, p < .01

Table 2 Final Cox proportional hazard analysis models for sub-scale scores of psychopathy measures

Criterion/Predictor B SE (B) Wald eB (95% CI) Sig. Model cp

Any recidivism3

CPS impulsivity .469 .187 6.317 1.599 (1.109-2.306) .012 .28

Violent recidivism13

CPS impulsivity .654 .252 6.723 1.924(1.173-3.155) .010 .29

Non-violent recidivism0

CPS interpersonal .674 .295 5.236 1.962 (1.102-3.495) .022 .25

Weapons-related recidivism41

APSD impulsivity .672 .254 6.985 1.957(1.190-3.221) .008 .32

Note: Only the variables that were significant in final models are presented. The Model cp in the right-most column is a phi coefficient derived

from the overall model X2 calculated from the formula (p = j(X2/N) provided by Rosenthal (1991). It is provided as an effect size estimate of the

overall strength of the predictive model a Model -2LL = 291.05, x2(df=l,N

= 82) = 6.37, p < .05 b Model -2LL = 155.32, X2 (df= 1, N = 81) = 6.90, p < .01

0 Model -2LL = 251.60, X2 (df= 1, N = 82) = 5.27, p < .05

d Model -2LL = 34.73, x2 (df= 1, N = 79) = 8.14, p < .005

symptom count); non-violent (young age; past property offenses; alcohol abuse diagnosis; MAYSI alcohol/drug use factor; conduct disorder symptom count); weapons

related (young age; past property offenses; substance

dependence diagnosis; conduct disorder symptom count). As is evident, some covariates were consistently related to

recidivism?young age, conduct disorder symptoms, alcohol abuse diagnosis, and past property offenses.

Our strategy was to extend the Cox regression analyses summarized in Tables 1 and 2 by adding the sets of

covariates, as single blocks, described in the previous

paragraph, in the first step of each Cox regression analysis (one for each recidivism outcome), and then entering the

relevant psychopathy scale or sub-scale. Incremental

validity was evaluated by whether the psychopathy index

added significantly to the covariates.

As is evident from Tables 3 (analyses involving total scores of psychopathy measures) and 4 (analyses involving

sub-scale scores of psychopathy measures), none of the

psychopathy indices added incrementally to the first step of

analyses involving the covariates only. Further, with

respect to the weapons-related recidivism outcome, despite

there being a significant overall model, neither of the

variables in the models retained individual levels of sig nificance. For the total score analysis, these variables were

conduct disorder symptoms (p = .09) and APSD Total

(p =

.23). For the sub-scale analyses, these variables were

conduct disorder symptoms (p = .07) and APSD Impul

sivity (p = .19).

Effect of Sub-Sample

As noted in the Methods section, our sample was approx

imately evenly divided between sexual offenders and non

sexual (general) offenders. This raises the question of

?} Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 12: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

Law Hum Behav (2008) 32:423-438 433

Table 3 Final incremental Cox proportional hazard analysis models for total scores of psychopathy measures

Criterion/Predictor B SE (B) Wald eB (95% CI) Sig. Model q>

Any recidivism3

Young age -.462 .144 10.298 0.630(0.475-0.835) .001 .65

Past property offenses 1.210 .460 6.920 3.353(1.361-8.261) .009

Alcohol abuse 1.187 .564 4.428 3.277(1.085-9.898) .035

Violent recidivism5

Young age -.506 .204 6.131 0.603(0.404-0.900) .013 .60

Alcohol abuse 2.142 .590 13.178 8.513 (2.679-27.057) .001

Note: Only the variables that were significant in final models are presented. The Model <p in the right-most column is a phi coefficient derived

from the overall model #2 calculated from the formula (p = ^/(/2/N) provided by Rosenthal (1991). It is provided as an effect size estimate ofthe

overall strength of the predictive model a Model -2LL = 264.57, x2 (df= 4, N = 80) = 33.46, p < .001

b Model -2LL = 140.31, %2 (df= 3, N = 82) = 29.93, p < .01

whether offender type influenced findings, in terms of the

relationship between predictors and outcomes. There is some reason to suspect that it might have, in that offender

type correlated with some of the predictor and outcome measures. In terms of psychopathy measures, offender

status (general offender) correlated with PCL:YV Facet 2

(rpb = .24, p < .05), Facet 3 (rpb

= .34, p < .01), and CPS

Total (rpb = .23, p < .05) and Behavioral (rpb

= .31,

p < .01) scores. It correlated with any (cp = .24, p < .05), violent (cp = .34, p < .01), and non-violent recidivism

(cp = .28, p < .05).

It is also important to note that in a logistic regression analysis we could account for 66% of the variance in offender status with the set of covariates used in the incremental validity analyses (-2LL = 57.83, df=6,

N - 82, Nagelkerke R2 = .66). Given this high correspon dence between offender status and the covariates, we ran

sets of hierarchical Cox regression analyses to test whether offender status added anything beyond the covariates in

predicting outcomes. We used the covariate sets shown in

the Tables (psychopathy total score analyses) for any, violent, and non-violent recidivism, entering the covariates

listed as the first block, and offender status as the second block. Offender status failed to enter any of the analyses, suggesting that it did not add to the covariates' predictive power vis-a-vis recidivism (details available upon request). For this reason, we did not run additional analyses with offender status as a separate covariate in addition to the more substantively meaningful covariates.

Nonetheless, it is of interest to test whether offender status (essentially a proxy for a more serious history of

offending, substance use problems, and conduct disorder) moderates the association between the psychopathy indices and outcomes. That is, does psychopathy predict recidi vism at the same strength for sex offenders and general offenders? We tested whether the psychopathy-outcome

relationships depicted in Table 1 (psychopathy total score

analyses) were moderated by offender status in a series of

hierarchical Cox regression analyses. On the first block of each analysis, we entered the relevant psychopathy index

plus offender status, and on the second block we entered the interaction between the two. We were interested in whether the interaction term was able to enter the model, based on a statistically significant improvement to model fit. Results indicated that only the CPS-Any recidivism

relationship was moderated by offender status, such that the strength of association between the CPS and any recidivism was stronger for general offenders (details available upon request).

Discussion

This was the first truly prospective study that we are aware

of that has contrasted the predictive validity of the

PCL.YV, APSD, and CPS, and one of the few studies within the youth psychopathy area to include analyses of the incremental validity of psychopathy beyond other

putative risk factors. Given the increasing use of these instruments in clinical decision-making, and the heightened debate about youth psychopathic features generally, this

type of research is necessary to understand the potential

predictive utility?and limits?of these measures.

Findings for the PCL: YV were perhaps surprising, given past studies that have supported its relationship to violence and antisocial behavior among delinquent samples (Gretton et al. 2001, 2004), albeit some with small or moderate effects (Corrado et al. 2004; Vincent et al. 2003), and others with nonsignificant results (Edens and Cahill 2007;

Marczyk et al. 2003). In the present study, even in bivar iate analyses the PCL:YV (total score and sub-scales) was not significantly related to recidivism when judged against

^ Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 13: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

434 Law Hum Behav (2008) 32:423^138

Table 4 Final incremental Cox proportional hazard analysis models for sub-scale scores of psychopathy measures

Criterion/Predictor B SE (B) Wald eB (95% CI) Sig. Model <p

Any recidivism3

Young age -.445 .143 9.720 0.641 (0.485-0.848) .002 .65

Past property offenses 1.190 .459 6.717 3.287(1.337-8.086) .010

Alcohol abuse 1.249 .562 4.944 3.488(1.16-10.490) .026

Violent recidivism15

Young age -.504 .208 5.884 0.604(0.402-0.908) .015 .61

Alcohol abuse 1.938 .582 11.102 6.946(2.221-21.722) .001

Non-violent recidivism0

Young age -.413 .143 8.291 0.662(0.499-0.876) .004 .63

Past property offenses .994 .505 3.879 2.703(1.005-7.272) .049

Alcohol abuse 1.354 .573 5.573 3.872(1.258-11.915) .018

Note: Only the variables that were significant in final models are presented. The Model (p in the right-most column is a phi coefficient derived from the overall model y1 calculated from the formula cp = ^/(^/N) provided by Rosenthal (1991). It is provided as an effect size estimate ofthe overall strength of the predictive model a Model -2LL = 264.10, f (df= 4, N = 80) = 34.08, p < .001

b Model -2LL = 139.74, x2 (df= 3, N = 80) = 29.88, p < .001

c Model -2LL = 228.77, jr2 (df= 4, N = 79) = 31.02, p < .001

a more conservative alpha level of .0125. In one of the other prospective studies of the PCL:YV, Vincent et al.

(2003) observed small to moderate effect sizes (AUCs =

.58-64), which overlap with and only slightly surpass those in the present study. Edens et al. (2007) observed AUCs that (save one) did not differ from chance in another

prospective study. The predictive validity of the self-report measures was

better than for the PCL:YV. In particular, the CPS was the most consistent predictor of recidivism in bivariate and non

incremental multivariate analyses. For instance, the CPS total score and Behavioral sub-scale remained predictive of

any and violent recidivism in bivariate analyses when

judged against the more conservative alpha. The CPS and its sub-scales also were the most consistently related to

high-volume and diverse offending. Both APSD NAR and IMP were predictive of violent recidivism in these bivariate

analyses. The APSD total score remained significantly related to weapons-related offenses when judged against the more conservative alpha. These findings build upon other

research using self-report methods that have shown rela

tionships with antisocial behavior, much (i.e., Frick et al.

2003; see Frick and Hare 2001, generally) though not all

(i.e., Falkenbach et al. 2003) of which has been with

community or clinic samples, by using a prospective design, a juvenile justice sample, and the CPS as well as the APSD.

It is of some interest to note which specific sub-scales of measures were related to recidivism in bivariate and non

incremental multivariate analyses. Behavioral features of

psychopathy tend to outperform the interpersonal and

affective features in terms of their relationship to general

criminal recidivism, although for violent recidivism, with some exceptions (Skeem and Mulvey 2001), interpersonal and affective features also tend to predict violence (for a

summary, see Douglas et al. 2006). Consistent with past research, in the present study, it was primarily behavioral sub-scales (impulsivity) that predicted recidivism (includ

ing violent recidivism), although the CPS Interpersonal scale entered one analysis.

Despite some promise of the self-report measures at the bivariate and non-incremental multivariate levels, these measures fared poorly in incremental validity analyses. Commentators in the psychopathy field (Douglas et al.

2006; Edens et al. 2006), drawing from more general, classic sources (Sechrest 1963), have highlighted the

importance of incremental validity analyses. It is important to know whether a construct adds anything beyond other

theoretically or empirically important variables in terms of

predicting outcomes. The present incremental analyses

provided no support for the incremental validity of the CPS or APSD in terms of their unique predictive strength vis-a vis recidivism once a small set of covariates was included

in predictive models (and the PCL:YV did not make it into

the incremental validity stage of analyses). This poor per formance in incremental validity analyses should temper any enthusiasm for such measures as providing unique information that is useful for prediction in the juvenile

justice system.

The covariates that reduced the predictive strength of the psychopathy measures in most analyses generally are

well known in the criminological and juvenile delinquency literatures. The current findings simply confirm that factors

^ Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 14: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

Law Hum Behav (2008) 32:423-438 435

such as serious substance related problems and previous crime are robust risk factors (Farrington 2002). The novelty of the present research was to test measures of another

putatively robust risk factor?psychopathy?against these

in incremental analyses. Although some studies have

evaluated the incremental validity of the APSD and/or CPS

vis-a-vis past or concurrent antisocial behavior (Lynam

1997; Poythress et al. 2006), research has not yet addressed

the incremental validity of the CPS and APSD in juvenile

justice settings in terms of predicting future recidivism.

Data on the PCL:YV are inconsistent, with some studies

reporting incremental validity (Gretton et al. 2004; Schmidt et al. 2006) and others not (Langstrom and Grann

2002). Further, only one of these studies (Schmidt et al.

2006) used a prospective design. Lack of incremental

validity suggests that, for predictive purposes, the measures

may have been related to outcome because they contain

information about these other risk factors (i.e., past crime

and substance related problems both enter into the assess

ment of psychopathy). This study has four primary limitations. First, the sample

size is moderate. Although large samples are always pref

erable to modest ones, in our view the sample size was

justifiable given the research questions. That is, these mea

sures had never been compared to one another in a

prospective study using justice-involved youth. Further, the

APSD and CPS had rarely been used to predict criminal

recidivism within delinquent youth. Hence, given that our

specific research questions had not been addressed previ ously, we reasoned that a moderate sized study7 was

justifiable and appropriate under the circumstances. None

theless, especially for multivariate analyses, we urge caution in interpreting the findings, in that power for such analyses is affected by a host of factors (i.e., base rate; beta; correlation between covariates). This is especially so for analyses of

weapons-related recidivism, given its base rates (7.2%). Second, our sample included both sexual and general

offenders, which in principle could have limited its gen

eralizability if results had differed dramatically between

sub-groups. However, we found that offender type mod

erated just one psychopathy-recidivism relationship. This

suggests that the current findings were not unduly influ enced by sample mix. In fact, sample mix could be viewed as adding to the generalizability of the findings.

Third, to measure recidivism, we relied upon official recidivism data solely. While this is a common method both in psychopathy research and the risk assessment field more broadly, it underestimates the true base rate of recidivism (a well-known phenomenon in criminology

7 We note that with the current sample size, statistical power was

adequate for analyses conducted, with moderate size effects (Cohen

1992) being detected as significant.

referred to as the "dark figure" of crime?see Coleman and

Moynihan 1996). This might have been exacerbated by limiting the source of recidivism data to Florida DJJ

records. Any recidivism that occurred outside of Florida

would not have been detected. As might be the case with

other studies of psychopathic features and recidivism

among adolescents (Corrado et al. 2004; Gretton et al.

2001, 2004; Vincent et al. 2003), the effect of this meth

odological feature would likely be to underestimate the

validity estimates derived for the psychopathy (or any

other) measures. We considered this design feature war

ranted given that our research was the first to compare these three measures in a prospective design. Nonetheless,

we would recommend that more resource-intensive follow

up procedures be employed that include self- and collat

eral-report as sources of recidivism information. We also

note that we were able to derive strong predictive models

from the covariates, suggesting that the method of mea

suring recidivism may not have unduly affected the

predictive validity of measures.

Fourth, there was lower than preferable internal con

sistency for some measures (Nunnally and Bernstein

1994). These values, despite being low, actually are quite consistent with others reported in the literature (for the

APSD, see Cruise et al. 2000; Falkenbach et al. 2003; Lee et al. 2003; Murrie and Cornell 2002; Pardini et al.

2003; Poythress et al. 2006; for the PCL:YV, see Lee et al. 2003; Skeem and Cauffman 2003; Vincent 2002),

suggesting that a feature of the measures, rather than a

deficit of the current study, contributed to the low reli

ability of some of the indices. Given this, low reliability has substantive meaning vis-a-vis validity, which is lim

ited by this feature of the measures. If lower than

preferred reliability is a feature of the measures them selves rather than a feature of the particular study, it is

important to understand the predictive and incremental

validity of the measures as they currently exist, whether

reliability is low or high. In fact, this and other research

(Poythress et al. 2006) can be taken in part as evidence that some of the youth measures of psychopathic features

likely are in need of improved reliability. Further, we note that the base rate of weapons-related

recidivism was small. Hence, results based on this outcome

criterion should be considered with perhaps additional

skepticism, and are in particular need of replication in

independent samples. Finally, we wish to issue a caveat. Given that this was

one of the first comparative studies of these three psy chopathy measures in a prospective research design among

justice-involved youth, we do not consider the issue settled

regarding which measure might have the most utility for

predictive purposes. Although the PCL:YV received very little support in the current study, others have indeed

^ Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 15: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

436 Law Hum Behav (2008) 32:423-438

observed meaningful associations with criminal conduct in either retrospective follow-up (Gretton et al. 2001, 2004) or prospective designs (Vincent et al. 2003), although the latter with small to moderate effects. Concerning the APSD and CPS, there is simply not yet enough research on pre dictive validity, stemming from prospective studies, within

juvenile justice settings to conclude that our observed

findings reflect a stable, generalizable pattern of predictive effects. For this reason, we strongly encourage other

researchers to attempt to replicate and improve upon our

research using larger sample sizes, more comprehensive

follow-up procedures (i.e., including self-report of recidi

vism) and attending to issues of diversity (i.e., does gender or race moderate the predictive effect of the measures?).

Acknowledgment Kevin S. Douglas gratefully acknowledges the

support of the Michael Smith Foundation for Health Research for

providing him with a Career Scholar Award. Douglas also is affiliated

with Mid-Sweden University, Sundsvall, Sweden, as a Guest Pro

fessor of Applied Criminology.

References

Andershed, H., Gustafson, S. B., Kerr, M., & Stattin, H. (2002). The

usefulness of self-reported psychopathy-like traits in the study of

antisocial behaviour among non-referred adolescents. European Journal of Personality, 16, 383-402.

Bartko, J. J., & Carpenter, W. T. (1976). On the methods and theory of reliability. Journal of Nervous and Mental Disease, 163, 307

317.

Bastiaens, L. (2004). Does residential treatment impact pharmaco

therapy in children and adolescents? Psychiatric Times, XXI,

(10), 34-35.

Bonnie, R., & Grisso, T. (2000). Adolescents' capacities as trial

defendants. In T. Grisso, & R. Schwartz (Eds.), Youth on trial

(pp. 67-204). Chicago: University of Chicago Press.

Brandt, J. R., Kennedy, W. A., Patrick, C. J., & Curtin, J. J. (1997). Assessment of psychopathy in a population of incarcerated

adolescent offenders. Psychological Assessment, 9, 429-435.

Campbell, M. A., Porter, S., & Santor, D. (2004). Psychopathic traits

in adolescent offenders: An evaluation of criminal history, clinical, and psychosocial correlates. Behavioral Sciences and

the Law, 22, 23-47.

Caputo, A. A., Frick, P. J., & Brodsky, S. L. (1999). Family violence

and juvenile sex offending: The potential mediating role of

psychopathic traits and negative attitudes toward women.

Criminal Justice and Behavior, 26, 338-356.

Cohen, J. (1992). A power primer. Psychological Bulletin, 112, 155

159.

Cohen, J. (1988). Statistical power analysis for the behavioral

sciences (2nd ed.). Hillsdale: Lawrence Erlbaum Associates.

Coleman, C, & Moynihan, J. (1996). Understanding crime data:

Haunted by the dark figure. Buckingham: Open University Press.

Cottle, C. C, Lee, R. J., & Heilbrun, K. (2001). The prediction of

criminal recidivism in juveniles: A meta-analysis. Criminal

Justice and Behavior, 28, 367-394.

Corrado, R. R., Roesch, R., Hart, S. D., & Gierowski, J. K. (Eds.).

(2002). Multi-problem violent youth: A foundation for compar ative research on needs, interventions, and outcomes.

Amsterdam: IOS Press.

Corrado, R. R., Vincent, G. M., Hart, S. D., & Cohen, I. M. (2004). Predictive validity of the Psychopathy Checklist: Youth Version

for general and violent recidivism. Behavioral Sciences and the

Law, 22, 5-22.

Cruise, K. R., Rogers, R., Neumann, C. S., & Sewell, K. W. (2000,

March). Measurement of adolescent psychopathy: Construct

validation of the two-factor model. Paper presented at the

Biennial Conference of the American Psychology-Law Society, New Orleans.

Dahlberg, L. L., & Simon, T. (2006). Predicting and preventing youth violence. In J. R. Lutzker (Ed.), Preventing violence: Research

and evidence based intervention strategies (pp. 97-124).

Washington, DC: American Psychological Association.

Douglas, K. S., Vincent, G. M., & Edens, J. F. (2006). Risk for

criminal recidivism: The role of psychopathy. In C. Patrick

(Ed.), Handbook of psychopathy. New York: Guilford.

Dunlap, W. P. (1999). A program to compute McGraw and Wong's common language effect size indicator. Behavior Research

Methods, Instruments, & Computers, 31, 706-709.

Edens, J. F., & Cahill, M. A. (2007). Psychopathy in adolescence and

criminal recidivism in young adulthood: Longitudinal results

from a multiethnic sample of youthful offenders. Assessment, 14, 57-64.

Edens, J. F., & Campbell, J. S. (2007). Identifying youths at risk for

institutional misconduct: A meta-analytic investigation of the

Psychopathy Checklist measures. Psychological Services, 4, 13

27.

Edens, J. F., Poythress, N. G., & Lilienfeld, S. O. (1999). Identifying inmates at risk for disciplinary infractions: A comparison of two

measures of psychopathy. Behavioral Sciences and the Law, 17, 435^43.

Edens, J. F., Skeem, J. L., Cruise, K. R., & Cauffman, E. (2001). Assessment of "juvenile psychopathy" and it's association with

violence: A critical review. Behavioral Sciences and the Law,

19, 53-80.

Edens, J. F., Skeem, J. L., & Douglas, K. S. (2006). Generalizability of the VRAG or generalizability of the PCL:SV? Incremental

validity analyses of the MacArthur violence risk assessment

data. Assessment, 13, 368-374.

Falkenbach, D. M., Poythress, N. G., & Heide, K. M. (2003).

Psychopathic features in a juvenile diversion population: Reli

ability and predictive validity of two self-report measures.

Behavioral Sciences and the Law, 21, 787-805.

Farrington, D. P. (2002). Multiple risk factors for multiple problem violent boys. In R. R. Corrado, R. Roesch, S. D. Hart, & J. K.

Gierowski (Eds.), Multi-problem violent youth: A foundation for

comparative research on needs, interventions, and outcomes (pp. 23-34). Washington, DC: IOS Press.

Federal Bureau of Investigation. (2004). Uniform crime reports: Crime in the United States: 2003. Washington, DC: U.S.

Government Printing Office.

Forth, A. E., & Burke, H. C. (1998). Psychopathy in adolescence:

Assessment, violence, developmental precursors. In D. Cooke, A. Forth, & R. Hare (Eds.), Psychopathy: Theory, research and

implications for society (pp. 205-230). Dordrecht: Kluwer

Academic Publishers.

Forth, A. E., Hart, S. D., & Hare, R. D. (1990). Assessment of

psychopathy in male young offenders. Psychological Assessment:

A Journal of Consulting and Clinical Psychology, 2, 342-344.

Forth, A. E., Kosson, D., & Hare, R. D. (1997). The Hare

Psychopathy Checklist-Youth (Research) Version (PCL-YV). Toronto: Multi-Health Systems, Inc.

Forth, A. E., Kosson, D. S., & Hare, R. D. (2003). Hare Psychopathy Checklist: Youth Version. Toronto: Multi-Health Systems.

Forth, A. E., & Mailloux, D. L. (2000). Psychopathy in youth: What

do we know? In C. B. Gacono, (Ed.), The clinical and forensic

4y Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 16: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

Law Hum Behav (2008) 32:423-438 437

assessment of psychopathy: A practitioner's guide. The LEA

series in personality and clinical psychology (pp. 25-54).

Mahwah: Lawrence Erlbaum Associates, Inc.

Frick, P. J. (1995). Callous-unemotional traits and conduct problems: A 2?factor model of psychopathy in children. Issues in

Criminological and Legal Psychology, 24, 47-51.

Frick, P. J. (2002). Juvenile psychopathy from a developmental

perspective: Implications for construct development and use in

forensic assessments. Law and Human Behavior, 26, 247-253.

Frick, P. J., Barry, C. T., & Bodin, S. D. (2000a). Applying the

concept of psychopathy to children: Implications for the

assessment of antisocial youth. In C. B. Gacono (Ed.), The

clinical and forensic assessment of psychopathy: A practitioner's

guide (pp. 3-24). Mahwah: Lawrence Erlbaum Associates.

Frick, P. J., Bodin, S. D., & Barry, C. T. (2000b). Psychopathic traits

and conduct problems in community and clinic-referred samples of children: Further development of the Psychopathy Screening Device. Psychological Assessment, 12, 382-393.

Frick, P. J., Cornell, A. H., Barry, C. T., Bodin, S. D., & Dane, H. A.

(2003). Callous-unemotional traits and conduct problems in the

prediction of conduct problem severity, aggression, and self

report of delinquency. Journal of Abnormal Child Psychology,

31 451-410.

Frick, P. J., & Hare, R. D. (2001). The Antisocial Process Screening Device. Toronto: Multi-Health Systems.

Fried, C. S., & Reppucci, N. D. (2002). Youth violence: Correlates,

interventions, and legal implications. In B. L. Bottoms, M. Bull

Kovera, & B. D. McAuliff (Eds.), Children, social science, and

law (pp. 233-269). New York: Cambridge University Press.

Gendreau, P., Goggin, C, & Smith, P. (2002). Is the PCL-R really the

"unparalleled" measure of offender risk? A lesson in knowledge cumulation. Criminal Justice and Behavior, 29, 397-426.

Gretton, H. M., Hare, R. D., & Catchpole, R. E. H. (2004).

Psychopathy and offending from adolescence to adulthood: A

10-year follow-up. Journal of Consulting & Clinical Psychology,

72, 636-647.

Gretton, H. M., McBride, M., Hare, R. D., O'Shaughnessy, R., &

Kumka, G. (2001). Psychopathy and recidivism in adolescent

sex offenders. Criminal Justice and Behavior, 28, 427-449.

Grisso, T., & Barnum, R., (2000). Massachusetts Youth Screening Instrument-2: User's Manual and Technical Report. Sarasota:

Professional Resource Exchange. Grisso, T., Barnum, R., Fletcher, K., Cauffman, E., & Peuschold, D.

(2001). Massachusetts Youth Screening Instrument for mental

health needs of juvenile justice youths. Journal of the American

Academy of Child and Adolescent Psychiatry, 40, 541-548.

Guy, L. S., Edens, J. F., Anthony, C, & Douglas, K. S. (2005). Does

psychopathy predict institutional misconduct among adults? A

meta-analytic investigation. Journal of Consulting and Clinical

Psychology, 73, 1056-1064.

Hare, R. D. (1991). The Hare Psychopathy Checklist-Revised.

Toronto: Multi-Health Systems. Hare, R. D. (2003). The Hare Psychopathy Checklist-Revised (2nd

ed.). Toronto: Multi-Health Systems. Hart, S. D. (1998). The role of psychopathy in assessing risk for

violence: Conceptual and methodological issues. Legal and

Criminological Psychology, 3, 121-137.

Hart, S. D., Cox, D. N., & Hare, R. D. (1995). Manual for the Hare

Psychopathy Checklist-Revised: Screening Version (PCL-SV). Toronto: Multi-Health Systems.

Hemphill, J. F., & Hare, R. D. (2004). Some misconceptions about the

Hare PCL-R and risk assessment: A reply to Gendreau, Goggin, and Smith. Criminal Justice and Behavior, 31, 203-243.

Hemphill, J. F., Hare, R. D., & Wong, S. (1998). Psychopathy and

recidivism: A review. Legal and Criminological Psychology, 3, 139-170.

Hollingshead, A. B. (1975). Four factor index of social status. New

Haven, CT: Department of Sociology, Yale University.

Johnstone, L., & Cooke, D. J. (2004). Psychopathic-like traits in

childhood: Conceptual and measurement concerns. Behavioral

Sciences and the Law, 22, 103-125.

Kosson, D. S., Cyterski, T. D., Steuerwald, B. L., Neumann, C. S., &

Walker-Matthews, S. (2002). The reliability and validity of the

Psychopathy Checklist: Youth Version (PCL:YV) in nonincar

cerated adolescent males. Psychological Assessment, 14, 97-109.

Kupchik, A., Fagan, J., & Liberman, A. (2003). Punishment,

proportionality, and jurisdictional transfer of adolescent offend

ers: A test of the leniency gap hypothesis. Stanford Law and

Policy Review, 14, 57-83.

Langstrom, N., & Grann, M. (2002). Psychopathy and violent

recidivism among young criminal offenders. Acta Psychiatrica Scandinavica, 706(Suppl. 412), 86-92.

Lee, Z., Vincent, G., Hart, D. S., & Corrado, R. R. (2003). The

validity of the Antisocial Process Screening Device as a self

report measure of psychopathy in adolescent offenders. Behav

ioral Sciences and the Law, 21, 771-786.

Luke, D. A., & Homan, S. M. (1998). Time and change: Using survival analysis in clinical assessment and treatment evaluation.

Psychological Assessment, 10, 360-378.

Lynam, D. R. (1997). Pursuing the psychopath: Capturing the

fledgling psychopath in a nomological net. Journal of Abnormal

Psychology, 106, 425-438.

Lynam, D. R. (2002). Fledgling psychopathy: A view from person

ality theory. Law & Human Behavior, 26, 255-259.

Marczyk, G. R., Heilbrun, K., Lander, T., & DeMatteo, D. (2003).

Predicting juvenile recidivism with the PCL:YV, MAYSI, and

YLS/CMI. International Journal of Forensic Mental Health, 2, 7-18.

Metz, C. E. (1978). Basic principles of ROC analysis. Seminars in

Nuclear Medicine, 8, 283-298.

Millon, T. (1993). Millon Adolescent Clinical Inventory manual.

Minneapolis: National Computer Systems. Mossman, D. (1994). Assessing predictions of violence: Being

accurate about accuracy. Journal of Consulting and Clinical

Psychology, 62, 783-792.

Mossman, D., & Somoza, E. (1991). ROC curves, test accuracy, and

the description of diagnostic tests. Journal of Neuropsychiatry and Clinical Neurosciences, 3, 330-333.

Murrie, D. C, & Cornell, D. G. (2002). Psychopathy screening of

incarcerated juveniles: A comparison of measures. Psycholog ical Assessment, 14, 390-396.

Murrie, D., Cornell, D. G., Kaplan, S., McConville, D., & Levy Elkon, A. (2004). Psychopathy scores and violence among

juvenile offenders: A multi-measure study. Behavioral Sciences

and the Law, 22, 49-67.

Nunnally, J., & Bernstein, I. (1994). Psychometric theory (3rd ed.). New York: McGraw-Hill.

O'Neill, M. L., Lidz, V., & Heilbrun, K. (2003). Adolescents with

psychopathic characteristics in a substance abusing cohort:

Treatment process and outcomes. Law and Human Behavior,

27(3), 299-313.

Otto, R. K., & Borum, R. (2004). Evaluation of youth in the juvenile

justice system. In W. T. O'Donohue, & E. R. Levensky (Eds.), Handbook of forensic psychology: Resource for mental health

and legal professionals (pp. 873-895). New York: Elsevier

Science.

Pardini, D. A., Lochman, J. E., & Frick, P. J. (2003). Callous/

unemotional traits and social-cognitive processes in adjudicated

youth. Journal of the American Academy of Child & Adolescent

Psychiatry, 42, 364-371.

Poythress, N. G., Dembo, R., Wareham, J., & Greenbaum, P. (2006). Construct validity of the Youth Psychopathic features Inventory

5) Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions

Page 17: Criminal Recidivism among Juvenile Offenders: Testing the Incremental and Predictive Validity of Three Measures of Psychopathic Features

438 Law Hum Behav (2008) 32:423^*38

(YPI) and the Antisocial Process Screening Device (APSD) with

justice involved adolescents. Criminal Justice and Behavior, 33, 26-55.

Poythress, N. G., Douglas, K. S., Falkenbach, D., Cruise, K., Lee, Z.,

Murrie, D. C, & Vitacco, M. (2006). Internal consistency of the

self-report Antisocial Process Screening Device. Assessment, 13, 107-113.

Ridenour, T. A., Marchant, G. J., & Dean, R. S. (2001). Is the revised

psychopathy checklist clinically useful for adolescents? Journal

of Psychoeducational Assessment, 19, 227-238.

Rogers, R., Johnson, J., Chang, J. J., & Salekin, R. T. (1997). Predictors of adolescent psychopathy: Oppositional and conduct

disordered symptoms. Journal of the American Academy of

Psychiatry and the Law, 25, 261-271.

Rogers, R., Vitacco, M. J., Jackson, R. L., Martin, M., Collins, M., &

Sewell, K. W. (2002). Faking psychopathy? An examination of

response styles with antisocial youth. Journal of Personality Assessment, 78, 31^-6.

Rosenthal, R. (1991). Meta-analytic procedures for social research

(rev. ed.). Newbury Park: Sage. Salekin, R. T., Neumann, C. S., Leistico, A. R., DiDicco, T. M., &

Duros, R. L. (2004). Psychopathy and comorbidity in a young offender sample: Taking a closer look at psychopathy's potential

importance over disruptive behavior disorders. Journal of Abnormal Psychology, 113, 416-427.

Salekin, R., Rogers, R., & Sewell, K. (1996). A review and meta

analysis of the Psychopathy Checklist and the Psychopathy Checklist-Revised: predictive validity of dangerousness. Clinical

Psychology: Science and Practice, 3, 203-215.

Schmidt, F., McKinnon, L., Chattha, H. K., & Brownlee, K. (2006). Concurrent and predictive validity of the Psychopathy Checklist:

Youth Version across gender and ethnicity. Psychological Assessment, 18, 393-^401.

Seagrave, D., & Grisso, T. (2002). Adolescent development and the

measure of juvenile psychopathy. Law and Human Behavior, 26, 219-239.

Sechrest, L. (1963). Incremental validity: A recommendation. Edu

cational and Psychological Measurement, 23, 153-158.

Sheehan, D. V., Lecrubier, Y., Sheehan, K. H., Amorim, P., Janavs, J.,

Weiller, E., Hergueta, T., Baker, R., & Dunbar, G. C. (1997). The Mini-International Neuropsychiatric Interview (M.I.N.I): The development and validation of a structured diagnostic

psychiatric interview for DSM-IV and ICD-10. Journal of Clinical Psychiatry, 59, 22-33.

Sheehan, D. V., Shytle, D., Milo, K., Lecrubier, Y., & Hergueta, T.

(2001). MINI International Neuropsychiatry Interview for Children and Adolescents, Unpublished test manual.

Silver, A. A., Shytle, R. D., Sheehan, K. H., Sheehan, D. V., Ramos,

A., & Sanberg, P. R. (2001). Multi-center, double blind, placebo controlled study of mecamylamine monotherapy for Tourettes'

disorder. Journal of the American Academy of Child &

Adolescent Psychiatry, 40, 1103-1110.

Skeem, J. L., & Cauffman, E. (2003). Views of the downward

extension: Comparing the Youth Version of the Psychopathy Checklist with the Youth Psychopathic Traits Inventory. Behav

ioral Sciences and the Law, 21, 737-770.

Skeem, J., & Mulvey, E. P. (2001). Psychopathy and community violence among civil psychiatric patients: Results from the

MacArthur Violence Risk Assessment Study. Journal of Con

sulting and Clinical Psychology, 69, 358-374.

Spain, S. E., Douglas, K. S., Poythress, N. G., & Epstein, M. (2004). The relationship between psychopathic features, violence and

treatment outcome: The comparison of three youth measures of

psychopathic features. Behavioral Sciences and the Law, 22, 85

102.

Steinberg, L. (2002). The juvenile psychopath: Fads, fictions and

facts. National Institute of Justice Perspectives on Crime and

Justice: 2001 Lecture Series, Vol. V, pp. 35-64.

Vincent, G. M. (2002). Investigating the legitimacy of adolescent

psychopathy assessments: Contributions of item response theory.

Unpublished dissertation, Department of Psychology, Simon

Fraser University, Burnaby, BC.

Vincent, G. M., & Hart, S. D. (2002). Psychopathy in childhood, adolescence: Implications for the assessment and management of

multi-problem youths. In R. R. Corrado, R. Roesch, S. D. Hart, & J. K. Gierowski (Eds.), Multi-problem violent youth: A

foundation for comparative research on needs, interventions and

outcomes (pp. 150-163). Burke: IOS Press, Inc.

Vincent, G. M., Vitacco, M. J., Grisso, T., & Corrado, R. R. (2003).

Subtypes of adolescent offenders: Affective traits and antisocial

behavior patterns. Behavioral Sciences and the Law, 21, 695-712.

Walters, G. D. (2003a). Predicting criminal justice outcomes with the

Psychopathy Checklist and Lifestyle Criminality Screening Form: A meta-analytic comparison. Behavioral Sciences and

the Law, 21, 89-102.

Walters, G. D. (2003b). Predicting institutional adjustment and

recidivism with the Psychopathy Checklist factor scores: A

meta-analysis. Law and Human Behavior, 27, 541-558.

^ Springer

This content downloaded from 185.44.78.113 on Thu, 12 Jun 2014 21:13:56 PMAll use subject to JSTOR Terms and Conditions