University of Southern California Dissertations and Theses

Multiple informant-rated psychopathy in young adolescents: a multitrait-multimethod investigation of the Antisocial Process Screening Device

(USC Thesis Other)

Multiple informant-rated psychopathy in young adolescents: a multitrait-multimethod investigation of the Antisocial Process Screening Device

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content MULTIPLE INFORMANT-RATED PSYCHOPATHY IN YOUNG
ADOLESCENTS: A MULTITRAIT-MULTIMETHOD INVESTIGATION OF THE
ANTISOCIAL PROCESS SCREENING DEVICE

by

Michelle Tien-Yee Fung

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PSYCHOLOGY)

August 2007

Copyright 2007 Michelle Tien-Yee Fung
ii
DEDICATION

To Brian, my family, and my cohort, whose love and support helped fill these pages.

iii
ACKNOWLEDGEMENTS
The work within benefited from the efforts of many people and institutions.
My advisor and committee chair, Penelope K. Trickett, gave me a new home at USC
and provided me with invaluable opportunities to learn and develop my skills as a
researcher. I am also grateful to the other members of my dissertation committee,
Franklin R. Manis, Ferol E. Mennen, David Schwartz, and the late John L. Horn, for
their encouragement and guidance. In addition, the other staff, students and
volunteers of the Young Adolescent Project were a source of inspiration,
collaboration, support and, importantly, fun. I am also especially appreciative of the
funding support that I received via Merit Fellowship from the USC College of
Letters, Arts, and Sciences and, in my final year, a Dissertation Grant Award from
the Centers for Disease Control and Prevention, National Center for Injury
Prevention and Control.
1

Finally, this work would not have been possible without the families of the
Young Adolescent Project and their case workers at the Los Angeles Department of
Child and Family Services. I am thankful for their participation in this meaningful,
important study.

1
This dissertation was supported by Grant no. 1 R49 CE000955-01 from the Centers for Disease
Control and Prevention (CDC). The contents of this dissertation are solely the responsibility of the
author and do not necessarily represent the official views of the CDC.
iv
TABLE OF CONTENTS
Dedication

ii

Acknowledgements

iii

List of Tables v

List of Figures

vii

Abstract

viii
Chapter 1: Specific Aims 1

Chapter 2: Background and Significance 5

Chapter 3: Approaches to Evaluating Construct Validity 17

Chapter 4: Research Design and Method 33

Chapter 5: Results 40

Chapter 6: Discussion 88

References 123
Appendices
Appendix A: Antisocial Process Screening Device
Appendix B: Multidimensional Anxiety Scale for Children
Appendix C: Reactive-Proactive Questionnaire
140
v
LIST OF TABLES

Table 1: Hypothetical MTMM Matrix for 3 Traits (A, B, C) and 3 Methods
(1, 2,3) 18

Table 2: Sample Characteristics at Time 3 35

Table 3: Initial two- and three-factor results from factor analysis of the 20-
item ASPD (average scores) 44

Table 4: Successive factor models tested during restricted factor analyses
and invariance analyses 56

Table 5: Standardized factor loadings for parent, teacher and child restricted
factor analyses on final 10-item 3-factor model 57

Table 6: Multitrait-Multimethod correlational matrix of APSD data for three
trait factors and three informants 60

Table 7: Goodness of Fit Indices and Nested Model Comparisons for
General CFA Models 65

Table 8: Fit indices and nested model comparisons for correlated
uniquenesses models 69

Table 9: Parameter estimates and Factor intercorrelations from Model A,
the full correlated uniquenesses model 70

Table 10: Mean proportions of trait variance and method variance estimated
in the full correlated uniqueness model 73

Table 11: Standardized factor loadings from the confirmatory factor
analysis of the Reactive Proactive Aggression Questionnaire
(RPQ) 75

Table 12: Factor intercorrelations and internal consistency reliabilities for
the RPQ 75

Table 13: Standardized factor loadings from the confirmatory factor
analysis of the self-report RPQ 77
vi
LIST OF TABLES (continued)

Table 14: Factor intercorrelations and internal consistency reliabilities for
the MASC 78

Table 15: Subtest intercorrelations and internal consistency reliabilities for
the WJR subtests 78

Table 16: Subsample mean scores and standard deviations on external
correlates 81

Table 17: Correlations between APSD and RPQ subscale scores 82

Table 18: Correlations between APSD and MASC subscale scores 84

Table 19: Correlations between APSD and WJR subtest scores 86

Table 20: Beta coefficients and R
2
statistics from regression analyses
regressing RPQ aggression, MASC anxiety, and WRJ cognitive
subtests on APSD factors and grouping variables 87
vii
LIST OF FIGURES

Figure 1: Hypothetical multitrait-multimethod complete model with three
traits and three methods 22
Figure 2: Hypothetical correlated uniquenesses model with 3 traits 23
Figure 3: General CFA model diagrams 64
Figure 4: Correlated uniquenesses model diagrams 68
viii
ABSTRACT
The Antisocial Process Screening Device (APSD; Frick & Hare, 2001) has
shown promise in identifying psychopathic features in juveniles, but its novel
multi-informant format has received little attention to date. This study investigated
the psychometric properties of the APSD in adolescents participating in a study on
the impact of neglect on development. The first aim of this study was to investigate
the theoretical structure of the psychopathy construct in adolescence and to
establish a universal form of the APSD across caregiver, teacher, and adolescent-
reported data. This was accomplished via restricted structural factor analysis and
invariance analyses. The second aim was to examine the construct validity of the
multi-informant APSD via correlational and confirmatory factor analysis (CFA)
multitrait-multimethod (MTMM) procedures. Traits comprised APSD factors, and
methods comprised the three informants. The third aim was to investigate
relationships between the APSD and aggression, anxiety, and cognitive ability.
Results indicated that several items which showed noninvariance between
maltreatment and comparison individuals were excluded from the model. A 10-
item, 3-factor version of the APSD showed good fit and invariance across sex,
maltreatment, and ethnic groups in all three informants. APSD factors resembled
the Impulsivity, Narcissism and Callous-Unemotional factors from previous
studies, but with fewer items per scale. Factor reliabilities were low, in particular
for self-report factors, calling into question the efficacy of self-report information.
ix
Convergent validity among traits and limited discriminant validity between traits
suggested that the APSD measures a multidimensional construct that appears to tap
interpersonal, affective and behavioral features relevant to psychopathy in youth.
Substantial method effects also emerged; a large proportion of variance was
attributable to the specific informant. Lastly, the APSD showed some significant
relationships with aggression, anxiety and cognitive ability in the hypothesized
directions. Patterns of these relationships varied widely across informants and
across traits, implying discriminant validity and cautioning against treating
informants interchangeably. Overall, this study provides cautious support for the 3-
factor APSD as a measure of psychopathic traits in youth, and offers several
recommendations to improve efficacy. Research on juvenile psychopathy using
measures such as the APSD should proceed with caution until multi-informant
measurement is better understood.
1
CHAPTER 1: SPECIFIC AIMS
Research on psychopathy has long focused on adult male offenders, owing to
the high prevalence of the disorder in such samples. It follows that much of our
knowledge about the correlates and features of psychopathy is the result of
laboratory work with this population, and the applicability of findings from such
research to other samples and methodologies should not be assumed blindly. One
area of much current growth is the development and validation of measurement
tools. Traditionally, the assessment of psychopathy has involved only one informant
– usually the subject themselves, in concert with passive information collected from
a file review. With the advent of juvenile assessment tools, a new era of
measurement has begun, one which includes the report of not only the subject, but
their caregivers and teachers as well. While a growing body of literature attests to
interest in the concomitant efficacy of different measures, little is known about the
efficacy of different informants of the same measure. The simultaneous collection of
self, caregiver, and teacher-report information about personality and behavior has the
potential to make a powerful contribution to our understanding of the fledgling
psychopath, but more needs to be understood about issues of validity and reliability
before its usefulness can be determined.
This study endeavored to examine adolescent, caregiver and teacher
agreement ratings of psychopathy on the Antisocial Personality Screening Device
(APSD; Frick & Hare, 2001; see Appendix A) in a unique sample of adolescents
2
from a large urban area of the United States. Approximately two-thirds of the
sample (which is divided evenly between genders) has experienced substantiated
maltreatment. This sample allowed us to address several important questions about
the invariance of the APSD across gender, ethnic and maltreatment group, the
validity of multi-informant psychopathy ratings, and also how those ratings relate to
other constructs in adolescence.
The specific aims of this study were as follows:
Specific Aim 1. This aim was twofold.
A. To delineate the factor structure of the APSD in the present sample
via restricted structural factor analysis.
B. To test the factorial invariance of the resultant model of the APSD
between gender (males and females), maltreatment group
(maltreated and comparison) and across ethnic group (Caucasian,
Latino, African American and biracial).
Specific Aim 1 was undertaken to both investigate the theoretical structure of
the psychopathy construct and to establish a form of the APSD that was universal
across caregiver, teacher, and child informant data. The latter objective was a
necessary precursor to subsequent study aims. In the current study, procedures
employed were termed structural factor analysis because they were neither purely
exploratory nor confirmatory, and such terms are often too restrictive in describing
factor analytic procedures (McArdle, 1994, 1996). The initial factor analysis
3
conducted here was theoretically informed though essentially exploratory in
nature, and will be referred to as exploratory for simplicity. A subsequent analysis
resembled a confirmatory factor analytic approach, but was here termed restricted
because it was undertaken based on results of the exploratory analysis and thus data-
driven, rather than based on a priori specifications (McArdle, 1996).
Specific Aim 2: This aim was twofold.
A. To examine the reliability, convergent validity and discriminant
validity of the APSD factors as rated by self, caregiver, and teacher
using a correlational analyses in a multitrait-multimethod
framework.
B. To further examine the construct validity of the APSD factors as
rated by self, caregiver, and teacher using confirmatory factor
analysis approaches in a multitrait-multimethod framework.
Specific Aim 2A was accomplished by constructing a multitrait-
multimethod matrix of bivariate correlations and internal consistency reliability
estimates. The matrix was then evaluated in accordance with guidelines set forth by
Campbell & Fiske (1959) for assessing convergent and discriminant validity. Aim
2B expanded on these findings by employing confirmatory factor analysis techniques
to assess convergent validity, discriminant validity, and to allow for partitioning of
the variance of the measures.

4
Specific Aim 3:
To examine the relationship of the self, caregiver, and teacher-reported
APSD total and factor scores to relevant external correlates (aggression, anxiety and
cognitive ability) in the present sample.
Specific Aim 3 first employed correlational analyses to determine bivariate
associations between the APSD and aggression, anxiety and cognitive ability.
Subsequent regression analyses determined the relative contributions of each of the
APSD factors in the prediction of the external correlates.
5
CHAPTER 2: BACKGROUND AND SIGNIFICANCE
Psychopathy is a form of personality disorder characterized by a constellation
of interpersonal, affective and behavioral traits. The psychopath is behaviorally
impulsive, irresponsible, parasitic and promiscuous, interpersonally charming,
manipulative, pathologically dishonest and affectively flat, lacking remorse and
empathy (Cleckley, 1941; Hare, 1998). Often acting ‘without conscience,’ such
individuals present a great challenge to society, inflicting immeasurable physical and
psychological pain on their victims.
The field of psychopathy research has traditionally concentrated its efforts on
adult samples, and only recently has turned its attention to the disorder in childhood
and adolescence. Many features that characterize the adult psychopath can also be
found at the juvenile level, including the key interpersonal and affective traits (e.g.
Frick, O’Brien, Wooton, and McBurnett, 1994), elevations in aggression and violent
offending (e.g. Corrado, Vincent, Hart, & Cohen, 2004), and attenuated
electrodermal arousal in anticipation of punishment (Fung et al., 2005). Behavior
genetics research in both adult (e.g. Blonigen, Hicks, Krueger, Patrick & Iacono,
2005) and child twin (Viding, Blair, Moffitt & Plomin, 2005) samples have found
significant genetic influence on the development of psychopathic characteristics.
Attempts to measure psychopathy in children and adolescents follow from the
measurement instrument used most extensively with adults (see following
discussion), but novel formats present new challenges to the field. Specifically, two
6
of the three principal measures of juvenile psychopathy are written in
questionnaire format and can be administered to various informants. This is in
contrast to adult measures, which rely mostly, if not solely, on self-report
information, raising the question of agreement and validity of the different forms.
Measurement of Psychopathy
The Psychopathy Checklist-Revised
Psychopathy in adults has been most widely assessed with the clinician-rated
Psychopathy Checklist-Revised (PCL-R; Hare, 1991, 2003), based on a semi-
structured interview and file review. The PCL-R has consistently demonstrated a
hierarchical factor structure, with one coherent higher-order factor comprising some
second-order factors. Early analyses delineated a 2-factor structure; F1-
Interpersonal/Affective comprises the callous, manipulative personality traits that are
considered to be key to the construct, while F2-Behavioral comprises reckless,
irresponsible and criminal behaviors (Harpur, Hare & Hakstian, 1989). Extensive
research has been conducted with the two-factor model, and this architecture still
dominates conceptual thinking about the construct today. However, the structure of
the PCL-R has come under debate in recent years, as more rigorous testing has failed
to confirm the two-factor model in several samples (Cooke & Michie, 2001; Darke,
Kaye, Finlay-Jones, and Hall, 1998; Hill et al., 2004; and McDermott et al., 2000).
Notably, Cooke and Michie (2001) employed item response theory methods to better
align the factor structure of the PCL-R with clinical accounts of the diverse
7
characteristics of psychopathy. They developed and validated a hierarchical model
comprising one coherent superodinate trait, Psychopathy, and three correlated
symptom facets: F1– Arrogant and Deceitful interpersonal Style; F2 – Deficient
Affective Experience; and F3 – Impulsive and Irresponsible Behavioral Style. This
model differs from the original two-factor model in that it decomposes the traditional
F1-Interpersonal/Affective factor into two dimensions, emphasizing psychopathic
personality traits (Jackson, Rogers, Neumann & Lambert, 2002; Lilienfeld, 1998)
and downgrading the importance of antisocial behavior by excluding the items that
explicitly assess criminal conduct.
The 3-facet model has been successfully replicated in samples of Hispanic
inmates (Tubb, 2002), female offenders (Jackson et al., 2002), male offenders from
the UK and male offenders and forensic psychiatric patients from the US (Cooke,
Michie, Hart & Clark, 2005). Thus the 3-facet structure shows promise in providing
an alternative organization of psychopathic traits. More recently, however, Hare
(2003) proposed that the Cooke and Michie (2001) model could be incorporated as is
into a 2-factor, 4-facet model, if the items assessing antisocial behavior were
reinstated in a discrete, fourth facet. This 4-facet model of the PCL-R and its variants
(PCL-Youth Version; Forth, Kosson & Hare, 2003 and PCL-Screening Version;
Hart, Cox & Hare, 1995) also shows applicability in samples of adult male offenders
(Vitacco, Rogers, Neumann, Harrison & Vincent, 2005), forensic psychiatric patients
(Hill et al., 2004) and adolescents (Salekin, Neumann, Leistico & DiCicco, 2004).
8
Thus, despite decades of research, our understanding of the factor structure
of adult psychopathy is far from complete. However, the development of methods to
assess psychopathy in children and adolescents has forged ahead in recent years. In
parallel to the adult research, measures of psychopathy in youth are facing similar
challenges with regard to their underlying structures.
The APSD: The Downward Extension of the PCL-R to Youth
The three primary measures of child and adolescent psychopathy used today
are all downward extensions of the PCL-R, where the item content has been retained
but the items and scoring criteria themselves have been reformulated for age-
appropriateness. The PCL-Youth Version (Forth et al., 2001) is a direct modification
of the PCL-R, and follows the same interview and file-review format. The Child
Psychopathy Scale (Lynam, 1997) is a self and caregiver-report rating scale
comprising items from other instruments chosen for their resemblance to PCL-R
items. Finally, the Antisocial Process Screening Device (APSD; Frick & Hare,
2001; see Appendix A) is a rating scale developed to operationalize the PCL-R for
children and adolescents. All three available forms of the APSD - self, caregiver and
teacher-report, were collected in the sample for the present study.
Thus far, the APSD shows promise in identifying a subset of psychopathic
youth who differ from nonpsychopathic youth on a host of maladaptive behaviors.
Studies using the APSD in adjudicated samples have shown that it is related to
various types of offenses (Caputo, Frick, & Brodsky, 1999) and to frequency and
9
variety of delinquency and violence (Kruh, Frick, & Clemens, 2005; Poythress,
Dembo, Wareham & Greenbaum, 2006; Vitacco, Rogers, Neumann, 2003). In
community samples, the APSD is related to higher rates of aggression and
delinquency (Marsee, Silverthorn, & Frick, 2005), deficient passive avoidance
learning in boys (which implies a vulnerability to poor socialization and self-
modulation; Vitale, et al., 2005) and underarousal to affective stimuli (Sharp, van
Goozen, & Goodyer, 2006).
Factor analyses of the APSD have yielded both a 2-factor (Frick, O’Brien,
Wootton & McBurnett, 1994) and 3-factor structure (Frick, Bodin & Barry, 2000).
The two-factor model resembles that originally found in the PCL-R, with one factor
comprising Callous-Unemotional traits and the other comprising
Impulsivity/Conduct Problems. The three-factor solution yielded a similar
Callous/Unemotional traits factor, and in effect split the Impulsivity/Conduct
Problems factor into two, Narcissism, and Impulsivity. These three factors loosely
correspond to the 3 facets outlined by Cooke and Michie (2001). Though Frick et al.
(2000) actually found that both the 2- and 3-factor models were plausible, they
recommend the use of the 3-factor model and provide some convergent evidence in
support. A cluster analysis revealed 5 clusters, one with low scores on all 3 factors,
one with high scores on all 3 factors, and 3 clusters with elevated scores on one
factor and low scores on the other two. Some divergent associations between the
Narcissism and Impulsivity factors, the 5 clusters, and DSM Conduct Disorder,
10
Oppositional Defiant Disorder, and Attention-Deficit Hyperactivity Disorder
symptoms imply differential relations between the three psychopathic trait
dimensions and DSM criteria. However, whether the Callous/Unemotional factor is
truly a strong marker for psychopathy is not entirely confirmed. Clearly, further
investigations must be undertaken in other samples to gain a better understanding of
the structure underlying the APSD.
Some independent studies have been conducted on the 3-factor APSD in
juvenile offender samples. For example, Vitacco, et al. (2003) found that the 3-factor
model demonstrated superior fit (to self-report APSD data) over the 2-factor model
in a sample of maximum-security adolescent offenders. A review of internal
consistency values of the APSD across several studies (Spain, Douglas, Poythress, &
Epstein, 2004) reveals acceptable total score reliability values (0.72-0.82) but lower
factor score reliabilities (.36-.72). A similar review of the self-report APSD
concludes that the Impulsivity and Narcissism scales have good to moderate
reliability but the Callous/Unemotional scale consistently displays poor reliability
(Poythress, Douglas et al., 2006). Overall, the APSD demonstrates moderate
diagnostic accuracy (correctly identifying 67-74% of PCL:Youth Version high-
scorers; Murrie & Cornell, 2002) and item-level correspondence (Lee, Vincent, Hart
& Corrado, 2003) with the PCL:Youth Version, and expected associations with
external criteria such as indices of juvenile diversion program failure, rearrest
11
(Falkenbach, Poythress & Heide, 2003), institutional infractions and poor
treatment progress (Spain et al., 2004).
In sum, the APSD at once shows promise and limitations in its ability to
identify psychopathic traits in children and adolescents. Though it is somewhat
lacking in terms of agreement with the PCL:Youth Version, its relative performance
must be considered in light of its comparative ease of administration. In many study
settings, the time and effort required by the PCL:Youth Version are simply not
feasible. While there may be a difference in terms of amount and possibly quality of
information, in many cases this is a common pragmatic consideration facing
investigators. That the APSD is likely chosen over the PCL:Youth version in such
situations is yet more reason for further investigating its psychometric properties. In
all independent studies cited above, however, used only the self-report version of the
APSD in adolescent offender samples, and thus the picture could be much different
in other, community samples when parent and teacher information is taken into
account (which was the original intent with which the APSD was written; Frick &
Hare, 2001).
The APSD and Cross-Informant Utility
While the APSD was developed as an operationalization of the PCL-R in
youth, it is written in questionnaire format and thus employs a completely different
methodology than the clinician-rated PCL-R. In general, adult measures have
traditionally been administered to one informant via clinical interviews (e.g. the
12
PCL-R) or self-report questionnaires (e.g. the Psychopathic Personality Inventory;
Lilienfeld & Andrews, 1996), whereas the Child Psychopathy Scale and the APSD
gather information from other informants as well, such as caregivers and teachers.
As alluded to above, questionnaire measures are appealing in part because of their
ease of administration. To elaborate, consider the two to four hours required for a
labor-intensive PCL-R (or PCL: Youth Version) interview and comprehensive file
review, in contrast to the few minutes required to complete a 20-item paper-and-
pencil questionnaire. In addition, the latter may also be administered in groups or via
mail rather than the more time-consuming individual, face-to-face administration of
the PCL-R. Often the decision to use one over the other involves many factors,
including variations in the amount and quality of information collected.
While the use of a variety of methods to assess a given construct is widely
recommended, such variety is only beneficial if the methods show relative validity.
To this end, several recent studies have attempted to examine the convergent validity
between different measures of juvenile psychopathy. However, it is not only
important to consider the convergence or discriminability of discrete measures, but
also to evaluate those qualities within different forms (or between different
informants) of a given measure.
There are several ways in which this latter comparison might be carried out;
two of which are utilized in the current study: 1) correlational analysis of a
multitrait-multimethod matrix (MTMM; Campbell & Fiske, 1959); and 2)
13
confirmatory factor analysis of a MTMM matrix (e.g. Kenny & Kashy, 1992;
Widaman, 1985). Prior to turning to a discussion of the technicalities involved with
those MTMM techniques, a discussion of the reasons for using them is offered.
Multiple-Informant Views of Psychopathic Traits
There is somewhat of a paradox in the general expectation of multi-informant
use. That is, this methodological approach is simultaneously prized both for its
convergent and discriminative ability; several informants of the same target should
be able to agree to a certain extent, but informants from different domains should
contribute unique insight regarding the target. Though each measure (or informant)
may be reliable and valid, ratings may be only weakly correlated across informants.
However, disagreement between informants may vary systematically, allowing
interpretation of what is driving the disagreement.
No studies of cross-informant validity have been conducted for either the
APSD or the Child Psychopathy Scale to date, so studies examining internalizing and
externalizing problems in children are of particular relevance here. Previous studies
have found modest agreement across self, caregiver, and teacher-report ratings of
(internalizing and externalizing) problem behaviors (Achenbach, et al., 1987;
Biederman et al., 1993; Meyer et al., 2001; Sawyer, Baghurst & Clark, 1992) and
wide variation in patterns of disagreement (Achenbach, 1991), leading some to
caution against treating informants as interchangeable (Rubio-Stipec, Fitzmaurice,
Murphy & Walker, 2003). Several adolescent studies indicate that levels of self-
14
reported problem behaviors exceed parent-rated levels which in turn exceed
teacher-rated levels (Epkins, 1996; Sawyer, Baghurst & Mattias, 1992; Stanger &
Lewis, 1993; Verhulst & Van der Ende, 1992). As might be expected, agreement is
generally higher for externalizing than internalizing problems (Achenbach, et al.,
1987). However, some variation is seen among informants; parents may be more
sensitive to covert states such as depression and anxiety, while teachers notice more
overt disruptive behavior such as attention problems and report less internalizing
problems (Abikoff, Courtney, Pelham & Koplewicz, 1993; Youngstrom, Loeber,
Stouthamer-Loeber, 2000) though teachers have been found to report more
internalizing problems than parents (Mesman & Koot, 2001). Contrary to the above,
one study found that a sample of physically abused adolescents reported fewer
internalizing problems than their caregivers or teachers (Kaplan, Labruna, et al.,
1999). Clinicians generally place greater value in self-report of internalizing
symptoms, since patients have direct access to their internal feelings (Loeber,
Brinthaupt, & Green, 1990). However, lack of insight (Dell’Osso et al., 2002) and
the externalizing manifestation of internalizing problems, which are common in
children (e.g. Biederman, 1998), can decrease self-report accuracy.
These conditions and concerns likely apply to the measurement of
psychopathic traits as well, especially given that psychopathy involves both overt
behaviors (e.g. impulsivity, lying, aggression) and also more covert internal states
(e.g., guilt, shallow affect). Since these two domains of traits reliably separate out
15
into at least two factors within psychopathy, an examination of cross-informant
agreement by factor (i.e., a MTMM analysis) would elucidate which factor(s) of
psychopathy might be less reliably or validly measured by different informants.
Self-Report Information on Psychopathy
A MTMM approach to psychopathy assessment may also inform the question
of how much confidence can be placed in psychopathic individuals’ responses to
questions about themselves. Recall that the disorder is hallmarked in part by features
of pathological lying and manipulativeness. Self presentation – and preservation – is
important to the psychopath, and thus it is plausible that self-ratings or interview
responses are subject to response distortions and impression management (Lee et al.,
2003). The extent to which this is the case is difficult to estimate, as is the role of
motivational factors. For example, biases in responses to a personality inventory
used in decisions of detentional release may differ greatly from those of a screening
instrument used to determine eligibility for a research experiment. Age and context
may also intervene; incarcerated adult offenders may have more or different reasons
to lie or fake a response style than non-referred school-aged children. In addition,
adolescents may not have full command of the more abstract and complex
processing required to accurately gauge interpersonal and affective phenomena, as
opposed to more readily observable, concrete behavioral traits (Lee et al., 2003).
Importantly, response biases may not be the result of intentional misleading or
conscious self-presentation, but may simply follow from the psychopath’s lack of
16
insight and awareness of their internal state (Hare, Cook & Hart, 1999; Haviland,
Sonne, & Kowert, 2004). This may in part explain why self-report measures have
been found to more accurately measure the behavioral but not the more important
interpersonal and affective characteristics of psychopathy (Edens, Hart, Johnson,
Johnson & Olver, 2000; Harpur, Hare, & Hakstian, 1989; Hare, 1985).
Thus, if self-report information gathered from psychopathic individuals is
found unreliable or invalid, the underlying reasons may be difficult to discern.
However, the advantages of tapping other informants become clear: information
from multiple informants should aid not only in detection of invalid or unreliable
measurement, but also in overcoming it or diminishing its effects. The detection of
invalid responding requires a systematic comparison of the information collected
from multiple informants on multiple traits, which is well-served by different
applications of the multitrait-multimethod matrix.
17
CHAPTER 3: APPROACHES TO EVALUATING CONSTRUCT VALIDITY
The Multitrait-Multimethod Matrix
Campbell and Fiske (1959) described an influential empirical technique for
examining construct validity for method-trait combinations. This approach requires
the simultaneous measurement of at least two traits by at least two methods. Entering
the resulting intercorrelations into a table, along with internal consistency reliability
estimates for each measure, results in a multitrait-multimethod matrix. Campbell and
Fiske’s (1959) use of the term “trait” was rather broad. It could refer equally to, for
example, independently measured constructs or different subscales of a measure.
“Method” is a similarly expansive term, and may refer to different instruments,
informants on the same instrument or multiple occasions of measurement, among
other operationalizations.
Table 1 illustrates a hypothetical MTMM matrix for 3 traits and 3 methods.
The upper triangle of the matrix is redundant and thus omitted. The main diagonal
(here indicated by values in italics) are monotrait monomethod estimates of internal
consistency (0 reliability). The values just below the main diagonal (in the solid
triangles) are heterotrait-monomethod estimates of correlations between traits
measured by the same method. Together with the corresponding reliability estimates,
these values constitute monomethod blocks (solid rectangles).
18
The blocks below the monomethod blocks are called heteromethod blocks
(dashed rectangles) and contain heterotrait-heteromethod estimates, correlations
among traits measured by two of the methods (in the dashed triangles). The main
diagonals within heteromethod blocks are termed validity diagonals, containing
monotrait-heteromethod estimates, correlations of the same trait measured by two
different methods (underscored values).
Referring to this matrix, Campbell and Fiske (1959) outlined a validational
process with several guidelines for evaluating construct validity.
1. At the outset, it is important to realize that reliability is a necessary
condition of validity. The values of the reliability diagonal should not only be
significant and high enough to indicate that there is sufficient agreement between
Table 1
Hypothetical MTMM Matrix for 3 Traits (A, B, C) and 3 Methods (1, 2, 3)
Method 1 Method 2 Method 3
Traits A B C A B C A B C
A 0
A1

B r
A1
,
B1
0
B1

Method
1
C r
A1,C1
r
B1,C1
0
C1

A
r
A1,A2
r
A2,B1
r
A2,C1

0
A2

B r
A1,B2
r
B1,B2
r
B2,C1
r
A2
,
B2
0
B2

Method
2
C r
A1,C2
r
B1,C2
r
C1,C2
r
A2,C2
r
B2,C2
0
C2

A r
A1,A3
r
A3,B1
r
A3,C1
r
A2,A3
r
A3,B2
r
A3,C2

0
A3

B r
A1,B3
r
B1,B3
r
B3,C1
r
A2,B3
r
B2,B3
r
B2,C3
r
A3,B3
0
B3

Method
3
C r
A1,C3
r
B1,C3
r
C1,C3
r
A2,C3
r
B2,C3
r
C2,C3
r
A3,C3
r
B3,C3
M
C3

Note. Main diagonal, italics = internal consistency reliability estimates. Solid triangle =
monotrait-monomethod values. Solid rectangles = monomethod blocks. Underscored =
monotrait-heteromethod values (validity diagonals). Dashed triangles = heterotrait
heteromethod values. Dashed rectangles = heteromethod blocks.
19
similar methods attempting to measure the same trait, but they should present the
highest values in the whole matrix.
2. Convergent validity is then assessed via the validity diagonals – these
correlations must also be significant and high, but smaller than the monotrait-
monomethod reliability correlations.
3. Subsequently, discriminant validity is considered in three ways. First,
values in the heterotrait-monomethod triangles should be lower than values in the
corresponding validity diagonal, since the former represent relationships between
instruments which use the same method to measure different traits. We might expect
elevations in these correlations due to shared method variance.
4. Second, the lowest correlations (ideally, approaching zero) should be
found in the heterotrait-heteromethod triangles, which represent relationships in
which no shared trait or method variance is assumed.
5. Finally, patterns of trait interrelationships should be similar in all
heterotrait-monomethod and heterotrait-heteromethod triangles. Similar patterns
would indicate that trait interrelationships remain consistent across different
measurement methods, providing further support for the validity of the measures.
The methods outlined by Campbell & Fiske (1959) are appealing in their
elegance, but are subject to several limitations which restrict their utility (Byrne &
Goffin, 1993; Schmitt and Stults, 1986). First, judgments are made based on
qualitative visual inspections of the MTMM matrix, making it difficult to quantify
20
the degree to which the criteria are met. Second, the matrix is constructed with
observed, manifest data, which are then used to make inferences about underlying
latent traits. Given that correlations are subject to sampling error and unreliability of
the measures, the accuracy of such inferences may be questionable. In particular, if
the size of the sample from which the correlations are drawn is small, the influence
of error factors can be quite large. Third, this method of judgment makes no attempt
to differentiate method variance from trait variance, which could result in spuriously
high monomethod correlations. Lastly, the criterion which requires negligible
heterotrait-heteromethod correlations assumes independence of measures. This
assumption is untenable in practice, since method and trait factors, as well as
constructs of interest, are often inherently related. Moreover, independence of
measures is virtually impossible within the framework of questions one might ask in
applying the MTMM method. Of particular relevance to the current study,
information gathered from different forms of a single measure is fraught with
method variance given that the items in each form are similar, if not identical, by
design.
In sum, though the method proposed by Campbell & Fiske (1959) has been
instrumental in shaping our understanding of issues of construct validity, there are
some intrinsic problems which limit its usefulness. As such it is rarely used alone
and more often, as recommended by several researchers (e.g. Byrne & Goffin, 1993),
it functions as a preliminary rather than substantive analysis. Various other methods
21
have been sought which strive to overcome these limitations and more soundly
accomplish the goals of MTMM analysis. A number of methods apply confirmatory
factor analysis to the MTMM matrix and offer reasonable alternatives.
Confirmatory Factor Analysis (CFA) and the MTMM
Several CFA approaches have been proposed for the evaluation of the MTMM
matrix (see Kenny & Kashy, 1992; Widaman, 1985). The general CFA approach
may be conducted in two stages, first at the matrix level, with broadband tests of
convergent and discriminant validity, and then more precisely at the parameter level.
Broadband analysis requires the estimation of various nested models; generally,
comparisons are made between a model which posits freely correlated
methods/freely correlated traits (the complete or full model; see Figure 1 for a
hypothetical model with 3 traits and 3 methods) and other models which specify
varying numbers of traits, methods, and intercorrelations between methods, and
traits. For example, discriminant validity can be ascertained by comparing the full
general model to a model that is identical except that one general trait is posited,
onto which all 9 indicators load. If the full model demonstrates better fit than the
general trait model, this would indicate that the data is better described by a
multidimensional construct than a single psychopathy factor.

22
Figure 1. Hypothetical multitrait-multimethod complete model with three traits
and three methods.
The general CFA approach is valued because it is a faithful, quantitative
representation of the conceptualization laid down by Campbell & Fiske (1959), and
also because it allows for objective, empirical analyses of the more subjective tests
of construct validity that they described. However, a very serious problem with the
complete model (the core of the general CFA approach) is that, more often than not,
it results in ill-defined solutions and serious problems with model identification. To
overcome this, several alternative approaches have been developed. At present, the
most useful and thus the most widely recommended alternative is the Correlated
Uniquenesses (CU) model (Becker & Cote, 1994; Kenny & Kashy, 1992; Marsh &
Bailey, 1991), where method effects are inferred from CUs between indicators
measured by the same method, rather than method variables as in the general CFA
model (see Figure 2 for a hypothetical CU model). Variance of the indicators is
decomposed into trait variance and error variance, the latter of which includes
method and unique variance.
Trait 1 Trait 2 Trait 3
A1 A2 A3 B1 B2 B3 C1 C2 C3
Method B Method C Method A
23
Figure 2. Hypothetical Correlated Uniquenesses model with 3 traits.
A major assumption of the CU approach is that methods are uncorrelated,
which is reflective of the Campbell & Fiske (1959) guidelines for the MTMM
matrix. However, if methods are in fact correlated, which is most likely the case
here, that covariance may be relegated to other relationships, such as trait
intercorrelations (resulting in decreased discriminant validity) and trait factor
loadings (resulting in inflated convergent validity). Some studies have suggested that
these biases have minimal impact on model results (Saris, 1990; Scherpenzeel,
1995). Even in cases where the CU model fails to specify existing heteromethod
error covariances, parameter estimates are more accurate than those of the correctly
specified general CFA complete model (Marsh, 1989; Marsh & Bailey, 1991).
Despite the severe estimation problems referred to above, the general method is
sometimes recommended over the CU approach because of the ease of variance
partitioning it affords (e.g. trait variance and method variance are calculated as
squared trait and method factor loadings, respectively). However, this advantage no
longer holds since a simple method for estimating method variance (i.e. computing
O
A1
O
A2
O
A3
O
B1
O
B2
O
B3
O
C1
O
C2
O
C3
Trait 1 Trait 2 Trait 3
A1 A2 A3 B1 B2 B3 C1 C2 C3
24
the average of CUs within indicators measured by a given method) was described
by Conway (1998) and further explicated by Scullen (1999). Moreover, the most
important advantage of the CU model over the general CFA model is its statistical
soundness – a proper, estimable solution is obviously preferable to a solution which
does not converge, or returns out of range parameter estimates. Here, the general
CFA model was attempted but did not converge, and thus the CU model was
subsequently estimated and interpreted.
Once a best-fitting model has been identified, narrowband analyses can then
be performed on individual parameter estimates to further assess validity.
Convergent validity is addressed by examining the magnitude and significance of
factor loadings. Estimates of trait variance can be obtained by computing the square
of trait factor loadings, while estimates of method variance can be obtained by
computing the average of CUs. Comparing proportions of trait, method and error
variance will allow us to evaluate the efficacy of the measures across informants.
Discriminant validity involves an examination of the trait factor correlations.
Conceptually, trait intercorrelations should approach zero, but again this is highly
unlikely and unrealistic in psychological data. Marsh and Hocevar (1983) suggested
that concern should only be raised when correlations approach unity, and other more
conservative recommendations have also been made (e.g. Widaman, 1985; Goffin &
Jackson, 1992).

25
Relating the APSD to External Correlates
Another important step in assessing the construct validity of a measure is to
examine its relationship with external correlates. Three correlates of interest in the
current study are: aggression, anxiety, and cognitive ability, all self-report measures.
Many different relationships can be examined among these correlates and the APSD
(e.g. associations between each informant’s APSD rating and each correlate, between
each factor of the APSD and the correlate, etc.). While aggression is clearly
predicted to correlate positively with the APSD scores, anxiety and cognitive ability
were chosen for this study with the objective of clarifying their somewhat ambiguous
relationships to psychopathy in the extant literature. The selection of solely self-
report measures was deliberate, so that analyses and interpretation was not
complicated by variation in informant of external variables in addition to APSD
factors.
Aggression
Aggression is a robust correlate of psychopathy in juveniles (e.g. Frick,
Cornell, Barry, Bodin & Dane, 2003; Soderstrom, Nilsson, Sjodin, Carlstedt, &
Forsman, 2005; Stafford & Cornell, 2003) and adults (e.g. Hare, 1985; Patrick &
Zempolich, 1998), regardless of the measure used to assess psychopathy. As such, it
should be relatively consistent in its relation to both factors of the APSD, though the
magnitude of those correlations may differ between informants.
26
Differential association could be expected between the APSD factors and
different kinds of aggression. Two types of aggression are measured by the Reactive-
Proactive Aggression Questionnaire (RPQ; Raine et al., 2006, see Appendix B).
Reactive aggression is hostile and often emotional, proactive aggression is goal-
directed and instrumental. The manipulative, cold character of the psychopath is
well-suited to instrumental aggression, implemented for personal gain rather than as
an affective response to an anger-provoking event. Indeed, adult and adolescent
psychopathic offenders show greater amounts of this type of aggression than
nonpsychopathic offenders (Caputo, Frick & Brodsky, 1999; Cornell et. al., 1996;
Kruh, Frick, & Clements, 2005; Williamson, Hare & Wong, 1987). Thus it might be
predicted that APSD CU traits and APSD Narcissism would be more highly
correlated with instrumental aggression than reactive aggression. It might also be
expected then that APSD Impulsivity would show a greater correlation with reactive
aggression than instrumental aggression.
Anxiety
Anxiety is an important concept in the study of psychopathy, as a lack of
anxiety is largely regarded as a cardinal feature (Cleckley, 1941; Hare, 1978;
Newman, Kosson & Patterson, 1992) at the crux of some etiological theories
(Fowles, 1980; Lykken, 1957). However, despite widespread agreement on the
theoretical importance of low anxiety to the psychopathic construct, the empirical
picture is rather ambiguous. Some research suggests that comparisons between
27
psychopathic and nonpsychopathic individuals may only yield group differences if
both groups are also characterized by low anxiety. For example, reward dominant
responding, the tendency to continue producing a previously rewarded response even
in the face of diminishing reward probability, has long been posited as a central
deficit in psychopathy. A series of studies by Newman and colleagues (Newman et
al., 1992; Newman, Widom & Nathan, 1985; O’Brien & Frick, 1996; Smith, Arnett,
& Newman, 1992) found that this response style differentiated psychopaths from
nonpsychopaths only when both groups were low in anxiety, and not when both were
high in anxiety. Additionally, psychopathic individuals consistently show reduced
electrodermal activity in anticipation of an aversive event a finding that is taken as
strong evidence for a lack of anxiety in the face of impending punishment.
Less clarity is found in studies which use psychological or clinical measures
of anxiety in adult psychopaths. Some show the expected inverse relationship
between anxiety and the personality aspects of psychopathy (e.g. Harpur, Hare &
Hakstian, 1989), but others find no correlation (Lilienfeld & Penna, 2001; Schmitt &
Newman, 1999; Stalenheim & von Knorring). Part of this ambiguity may lie in the
associations between psychopathy and different types of anxiety. Consider the
subtypes of anxiety purportedly measured by the (self-report) Multidimensional
Anxiety Scale for Children (MASC; March, James, Sullivan, Stallings, & Conners,
1997; see Appendix C), which was collected in the present study. Harm Avoidance,
generally thought of as fearfulness, is consistently negatively related to psychopathy
28
(e.g. Herpertz & Sass, 2000; Hicks, Markon, Patrick, Krueger & Newman, 2004)
and is expected to show the same negative relationship in the present sample. Social
Anxiety, another MASC subscale, is seldom examined in psychopaths, but given that
superficially charming, manipulative and conning behaviors are central to the
interpersonal aspects of the disorder, it is rather unlikely that psychopaths experience
much of this type of anxiety. Separation Anxiety has not been studied in
psychopaths, so hypotheses regarding their relation must be based on theoretical
grounds. Items such as “many short-term marital relationships” (PCL-R) and “keeps
the same friends” (APSD, reversed) are found in psychopathy measurements,
designed to capture the inability of the psychopath to form longterm bonds and
relationships with others (Cleckley, 1941). If they are incapable of such attachments,
it is likely that they will not be more concerned than the norm if separated from their
parents, and thus a neutral or inverse relationship might be expected here. Lastly, in
light of the robust psychophysiological evidence for low anticipatory arousal in
psychopaths, an inverse relation with the Somatic Anxiety subscale might be
expected. However, some have found that psychopathy scores and self-rated somatic
anxiety are nonsignificantly (Harpur, Hare, & Hakstian, 1989) or even positively
related in adults (af Klinteberg, Humble, & Schalling, 1992; Fowles & Missel,
1994). It may be that psychopaths are reporting somatic experiences from anxious
situations that do not involve anticipation of an aversive event (which are not as
consistently related to low anxiety in psychopaths) or that their somatic experiences
29
are not, to themselves, aberrant. What little shakiness, sweating, or clamminess of
hands they do experience is enough for them to endorse the corresponding item on a
rating scale, even though that experience may be at an abnormally low level
according to more objective physiological measures. Furthermore, it is also possible
that endorsement of somatic items does not necessarily reflect anxiety, but a more
general state of excitement or arousal. Thus we might hypothesize a neutral or
positive relationship between psychopathy and MASC Somatic Anxiety in this
sample.
Cognitive Ability
A fairly undisputed finding is that antisocial individuals consistently score
lower than nonantisocial individuals on tests of intelligence (Moffitt, 1993; Henry &
Moffitt, 1997). However, traditional conceptualizations of psychopathy include
“good intelligence” (Cleckley, 1941); arguably, a certain level of intellectual
functioning is required in order to successfully con, manipulate and charm others.
Accordingly, some studies have found psychopathy to be associated with average or
above average intelligence levels (e.g. Holland, Beckett & Levi, 1981; Walsh, Beyer,
& Petee, 1987). Children with conduct problems that are also high on APSD CU
traits have shown no difference in IQ scores when compared to other clinic referred
children, whereas children with conduct problems and low CU traits showed lower
IQ scores than all other children (Frick et al., 1994; Christian et al, 1997). However,
other work has suggested that psychopaths evidence lower intelligence (Solomon,
30
1939; Van Vorst, 1943), neurological deficits (Gorenstein, 1982; Raine, O’Brien,
Smiley, Scerbo, & Chan, 1990) and learning difficulties (Newman, 1998).
Methodological and theoretical issues should be considered in light of these
discrepant findings. Early studies may have used tests of intelligence that are less
reliable or normatively inconsistent with intelligence measures in use today.
Furthermore, neurological deficits do not necessarily result in lowered IQ, just as
learning difficulties may leave IQ intact but instead affect scores on achievement
tests in specific academic domains.
These discrepancies in the literature may also reflect the fact that there is no
strong overall relationship between IQ and psychopathy (Hare, 2003), but that
various factors of intelligence associate with the different aspects of psychopathy.
For instance, Vitacco, Neumann & Jackson (2005) tested Hare’s (2003) recent four-
facet solution of psychopathy against a latent verbal IQ variable, and found that IQ
was positively related to the interpersonal factor, but negatively related to the
affective and lifestyle factors. That the interpersonal factor alone was positively
related to verbal IQ provides support for the idea that psychopaths’ ability to charm
and manipulate others is underpinned by a competent level of (verbal) intellect.
Though previous studies of cognitive ability in psychopaths have used
various measures of IQ (e.g. the Wechsler Adult Intelligence Scale-Revised;
Wechsler, 1981), another model of intellectual processing (Horn-Cattell theory;
Cattell, 1963; Horn, 1976, 1985; Horn & Cattell, 1966) may be useful in defining the
31
cognitive profile of the psychopath. The Woodcock Johnson Psycho-Educational
Battery-Revised (Woodcock & Johnson, 1989) includes subtests which assess fluid
reasoning, comprehension-knowledge, short-term memory and long-term retrieval
abilities (among others). There is no specific previous empirical research to draw
upon when making hypotheses about the performance of psychopathic individuals on
these subtests. Thus, the following hypotheses are presented more as an exploratory
discussion of theory and findings from related areas than a set of clearly defined
predictions.
Fluid reasoning is the ability to educe relations and associations, manifested
in inferential thinking and understanding of implications mostly independent of
previously learned or acquired knowledge (Woodcock & Maher, 1989). Psychopaths
are not likely to show elevations on this ability, given that they have difficulty
thinking through things, considering implications, and shifting response set (O’Brien
& Frick, 1996; Newman & Kosson, 1986; Scerbo, et al., 1990). Comprehension-
knowledge, or “crystallized intelligence,” is based on accumulated knowledge and
experience, and also includes communication and verbal abilities (Woodcock &
Maher, 1989). Psychopaths might be expected to do well here, especially to the
extent that the task measures verbal ability. There is little reason to expect previously
learned concrete information to be less than intact. Short-term memory is the ability
to immediately access stored information, exemplified by the ability to hold a new
telephone number in memory long enough to dial it successfully (Woodcock &
32
Maher, 1989). There is no real reason to think that psychopaths differ from
nonpsychopaths on tasks of short-term memory. Their appearance of being
intelligent may be driven in part by a working memory flexibility and ability to
manipulate recently stored information. However, the frontal lobes, which are often
thought to be abnormal in psychopaths (Fedora and Fedora, 1983; Gorenstein, 1982;
Gorenstein and Newman, 1980; Yeudall, Fedora, & Fromm, 1987), are implicated in
attention which in turn may affect short term memory performance. So to the extent
that frontal lobe functioning is impaired in the psychopath, short term-memory may
also be impaired. Long-term retrieval involves the storage of information and its
associative retrieval after a sustained period of time (Woodcock & Maher, 1989).
Psychopaths may perform somewhat badly here, as they seem unable to link
outcomes with antecedents. On the other hand, for psychopaths, this deficit is most
obvious for learning which results from fear conditioning and punishment, and thus
may not be obvious under conditions of normal encoding and retrieval.
33
CHAPTER 4: RESEARCH DESIGN AND METHOD
Participants
Participants in the present study were adolescents and their caregivers
enrolled in the Young Adolescent Project (YAP), an ongoing longitudinal
investigation of the impact of neglect on development. Data used in the current
project was collected at the third timepoint of the larger YAP study, which has not
yet been completed. The third timepoint is conducted approximately 1.5 years after
the second timepoint, which is conducted about 1 year after the initial timepoint and
entry into the study.
Recruitment at Time 1
A total of 454 male and female adolescents (and primary caregivers) were
recruited into the study. Maltreated young adolescents were recruited from active
substantiated cases of neglect and abuse presenting to the Los Angeles County
Department of Child and Family Services (DCFS). To control for some of the
diversity inherent in the large area serviced by the DCFS, participants were drawn
only from 10 zip codes in LA County. Based on DCFS data and census tract
information, these zip codes were chosen because of their easy travel distance to
USC (to facilitate family participation) and large numbers of the 3 most prevalent
ethnicities of the urban LA area.
Time 1 inclusion criteria for the neglect sample (N=303) were: (1) a new
substantiated report of any type of maltreatment to the DCFS in the preceding
34
month; (2) child age of 9 to 12 years; (3) child identified as African-American,
Latino or Caucasian (non-Latino); and (4) at time of referral, child resided in one of
10 zipcodes as described above. After receiving approval from DCFS, the courts and
USC IRB, families meeting the above criteria were contacted via mail, and asked to
return a postcard to indicate their willingness to participate. Recruitment of this
subsample began in Spring, 2002 and ended in Fall, 2004.
The comparison sample (N=151) was recruited from the same 10 zipcodes
via school lists of families with 9-12 year olds. Caregivers of potential comparison
adolescents were also contacted via mail and asked to return a postcard to indicate
their willingness to participate. Recruitment of the comparison participants began
after that of the neglect participants, and was completed in Fall, 2005.
Demographics at Time 1
Upon entry into the study, the neglect and comparison samples were similar
in age, gender and ethnic distribution, but, not unexpectedly, differed in terms of
living arrangements. A greater proportion of the comparison group lived with a
parent than the maltreated group (94 vs. 54% at Time 1, respectively), the latter
being more likely to reside in foster care or with an extended family. Extensive
comparisons between census block groups of the two study subsamples revealed that
overall, neighborhoods of both study groups were very comparable on social,
educational, economic and demographic census variables for the year 2000.

35
Sample for the Current Study at Time 3
Thus far, 249 participants, 155 maltreated and 94 comparison, have returned
for Time 3 (all having participated at Time 2). Due to the fact that Time 3 data
collection has not been completed, an attrition rate cannot be accurately reported.
However, attrition between Time 1 and Time 2 was low; 86% of the sample returned
to complete Time 2 and it is expected that a similar rate of return will characterize
Time 3.Table 2 presents the demographic information for participants at Time 3. The
actual sample size carried forth in analyses was 214, as 35 teachers did not return
data questionnaires.

Maltreatment Status. Though the YAP study was designed specifically with
the intent of examining the effects of neglect on development, participants
experiencing all types of maltreatment reported to DCFS were recruited. This is in
part because it would have been very difficult to amass a large sample with only
Table 2
Sample characteristics at Time 3

Group
Demographic Variable

Maltreated
N = 155

Comparison
N = 94
Age 13.32 (1.15) 13.99 (1.35)
Gender (%)
Male
Female

44
56

54
46
Ethnicity (%)
African American
Latino
White
Mixed Biracial

46
31
10
13

28
46
13
13
Living Arrangement (%)
With Parent
Foster Care/Extended Family

54
46

97
3
36
experiences of neglect, especially given evidence which indicates that most
maltreated children are likely to experience more than one type of maltreatment
(Clausen & Crittenden, 1991; McGee, Wolfe, Yuen, Wilson, & Carnochan, 1994).
Indeed, 90% of the maltreatment sample participating in this study have experienced
multiple forms of maltreatment. Though efforts are underway to describe
participants’ maltreatment experiences in more accurate detail, the present study
defined maltreatment group as a dichotomous variable.
Procedure
Approximately 1.5 years after their Time 2 appointment, families were
contacted and scheduled for their Time 3 appointment. Upon arrival, informed
consent and permission to contact the adolescent’s school and main (usually
language arts) teacher were obtained from adolescent and caregiver together.
Interviews were then conducted separately.
Caregivers completed questionnaires addressing various attitudes, beliefs and
behaviors of themselves, the adolescent participant, and the family environment.
Adolescents underwent a thorough testing battery which was divided into four
modules. Module 1 included a social stressor task with salivary cortisol collection.
Module 2 comprised a health assessment, where information about physical and
pubertal development was collected. In Module 3, the adolescents’ cognitive abilities
were evaluated via Woodcock-Johnson subtests. Module 4 consisted of all remaining
psychological measures which assessed behavior problems, psychiatric symptoms,
37
family functioning and trauma history. All modules were completed in one same-
day session, approximately 4 hours in duration. At the close of the interview, a
debriefing session was followed by the disbursement of payments to the family
based on rates specified by the NIH Normal Volunteer Program.
After the interview, the participant’s teacher was mailed a questionnaire
comprising questions about classroom behavior, behavior problems, and
relationships with the teacher and other classmates. Upon return of the completed
questionnaire, a gift certificate was sent to the teacher in a small amount for the store
of their choosing. The teacher response rate in the current study was quite high; 86%
of teachers contacted returned completed questionnaires (thus bringing the total
sample size to 214). A request for grade, standardized testing and attendance records
was also mailed separately to the school principal.
Measures
Psychopathy
Psychopathy was measured via the Antisocial Process Screening Device
(APSD; Frick & Hare, 2001, Appendix A). The APSD comprises 20 items rated on a
3-point scale (0 = Not true, 1 = Somewhat True and 2 = Definitely True), and thus
has a possible range of 0-40. Previous studies have mostly used the self-report form
(e.g. Vitacco et al., 2003; Spain et al., 2004) or a composite of caregiver and teacher
ratings (taking the highest of the two item ratings). Previous factor analyses have
found 2-factor (F1-Impulsivity/Conduct Problems, and F2-Callous/Unemotional
38
traits; Frick, O’Brien, Wooton, & McBurnett, 1994) and 3-factor (F1-Narcissism,
F2-Impulsivity and F3-Callous-Unemotional; Frick, Bodin & Barry, 2000) structures
in the APSD. Internal consistency estimates reported in these studies are adequate for
total score (.72-.82) but low for factor scores (.32-.62).
As yet, there is no standard cutoff score recommended for the APSD which
would allow for categorical comparisons of psychopathic traits. Rather than
imposing a categorical distribution on the sample, the whole range of the APSD was
used in the present study in order to retain as much information from the scale as
possible.
Anxiety
Self-report anxiety was measured with the Multidimensional Anxiety Scale
for Children (MASC; March et al., 1997, Appendix B), a 39-item scale designed for
use with 8-17 year olds. Items are written in first person (e.g. “I get scared when my
parents go away”) and are scored on a 4-point scale (0 = Never true about me, 1 =
Rarely true about me, 2 = Sometimes true about me, and 3 = Often true about me).
The MASC comprises four subscales: Somatic Anxiety, Social Anxiety, Harm
Avoidance, and Separation Anxiety, which have evidenced high internal consistency
(0 =.72-.90) and test-retest reliability (intraclass correlations = .65 - .93), invariant
factor structure across age and gender, and discriminant validity (March et al., 1997).
Prior to its use in substantive analyses, confirmatory factor analysis was performed
39
on the MASC in this sample, to determine whether the 4-factor structure found by
March et al. was evident in the present sample.
Aggression
Self-report aggression was measured in the adolescents via the Reactive-
Proactive Aggression Questionnaire (RPQ; Raine et al., 2006; see Appendix B),
which consists of 23 items which are scored on a 3-point scale (0 = Never, 1 =
Sometimes, or 2 = Often) for frequency of occurrence. In the original paper (Raine et
al., 2006), confirmatory factor analysis revealed a two-factor proactive-reactive
structure, with high internal consistencies (in excess of .83) for each subscale and the
total. Prior to its use in substantive analyses, confirmatory factor analysis was
performed on the RPQ in this sample, to determine whether the 2-factor structure
found by Raine et al. is evident in the present sample-0The Concept Formation,
Picture Vocabulary, Memory for Names, and Numbers Reversed subtests of the
Woodcock Johnson-Revised Tests of Cognitive Ability and Achievement (WJ-R;
Woodcock & Johnson, 1989) were used in the YAP to assess fluid reasoning,
crystallized knowledge, short-term memory and long-term retrieval, respectively
(Horn & Noll, 1996). Developed for use with ages three and above, the WJ-R scales
have internal consistency and test-retest reliabilities of .75 or greater (McGrew et al.,
1991; Woodcock, 1990). For the present study, raw total scores were computed and
used in analyses.
40
CHAPTER 5: RESULTS
Owing to the serial and nested nature of the specific aims heretofore
described, analyses were conducted with several overarching objectives in mind. In
particular, as the major goal of this study was to conduct a MTMM investigation
with informants representing the methods, it was imperative that the APSD factor
structure (i.e., the traits) that entered into the MTMM analysis was identical across
informants. Thus, factor analysis and invariance analyses were performed iteratively
– once a satisfactory solution was reached within one informant, the same solution
was then investigated in the other informants. If at any time, an alteration was made
to the factor structure (e.g., an item was deleted due to a nonsignificant loading or
noninvariance), the model was respecified. Analyses on the initial model were
discontinued and the respecified model was treated as a new starting point; all
preceding analyses were re-performed to ensure the universality of the new factor
structure for each informant. Another consideration was that, though two traits or
methods can be successful under certain circumstances and with certain restraints on
the model tested, MTMM analyses perform best when there are at least three traits
and three methods. Since both a two-factor and a three-factor structure have
precedents in the psychopathy literature in general and in previous research on
juvenile psychopathy and the APSD in particular, efforts were made to maintain the
three-factor structure as long as statistically and theoretically plausible. During this
process, emphasis was placed on parsimony, simple structure and theoretical
41
soundness. The requirement of retaining a minimum of three items per factor (for
identification purposes, (Brown, 2006) was also necessarily upheld.
Preliminary Analyses
Tests for Normality and Transformations
All variables were tested for univariate normality prior to their use in
substantive analyses. Some items of the APSD (caregiver-report: Items 2, 13;
teacher-report: Items 2, 5, 16, self-report: 2, 5, 10), and the RPQ (Items 4, 6, 9, 10,
12, 15, 17, 20, 21, 23) showed positive skew and leptokurtosis. Both the APSD and
RPQ comprise response scales of 0, 1 or 2, and given their content, it was not
unexpected that many of the items had substantial positive skew in this sample.
Transformations were applied (e.g. square root plus constant, and plus constant
inverse; Tabachnick & Fidell, 2000) but had little or no effect on the item
distributions. With the exception of the correlational and regression analyses of
Specific Aim 3, analyses employed statistics and estimators that are known to be
quite robust to normality, and thus raw, untransformed scores were used in all
analyses.
Missing Data Imputation
Due to diligent data collection practices, very little data was missing from the
measures used in the present study. The RPQ, WJR Subtests and the APSD were all
complete, however the MASC contained a very small amount of missing data (5
42
items across all participants, no two missing items within the same participant).
2
To maximize the MASC data, missing data was imputed with the NORM software
program (Schafer, 1999) employing an Estimation-Maximization algorithm
(Dempster, Lair, & Rubin, 1977; Orchard & Woodbury, 1972). This algorithm uses a
process that involves multiple imputations that converge over iterations to a
minimum-error estimation to generate “best guess” data.
Substantive Analyses

Specific Aim 1A. To delineate the factor structure of the APSD in the present sample
via restricted structural factor analysis.
Exploratory factor analysis (EFA) was first conducted on average APSD
scores in order to maximize the information used to determine a preliminary factor
structure that would then be refined for each informant. An average item score was
computed for each of the 20 items by summing the child, caregiver, and teacher-
rated item scores and dividing the total by three. Using the MPLUS v.4.0 software
program (Muthén & Muthén, 1998-2006), unweighted least squares (ULS) extraction
with oblique (Promax, k = 4) rotation was conducted on the 20 APSD average items.
One- through four-factor solutions were extracted as theoretically possible structures.
Results indicated that 5 factors had eigenvalues greater than 1.0. Many statistical
packages retain factors with eigenvalues > 1.0 by default, and some have suggested

2
Teacher-report APSD was complete for those questionnaires that were returned. Of 249 total
participants, 214 teachers complied with questionnaire requests. Several teacher questionnaires were
returned with missing items, however in those cases, teachers were contacted via mail, email and/or
phone and the missing items were recovered.
43
this as a rule of thumb for identifying factor solution (cf. Kaiser, 1960). However,
this method is widely regarded as inaccurate and prone to overfactoring (Velicer &
Jackson, 1990). Standardized root mean square residual (SRMR) values were high
for all four solutions (1.20 to .24), but decreased as the number of factors increased.
Examination of the scree plot resulting from the factor analysis suggested two- or
three-factor solutions. Though the four-factor structure had the lowest SRMR value,
it included one factor which had only 2 items, and thus was discarded. The extreme
misfit and lack of theoretical rationale for the 1-factor model led to its exclusion
from further study. The two- and three factor structures (2F and 3F, respectively)
were thus retained for further examination (see Table 3).
The 2F and 3F structures resulting from the preceding EFA were subjected to
restricted factor analysis (RFA) with maximum likelihood estimator to ensure their
viability in subsequent analyses. Items with loadings of .4 or greater on only one
factor were retained. Model fit was evaluated on the basis of several fit indices,
including the 8
2
statistic, Comparative Fit Index (CFI; Bentler, 1990), the Tucker-
Lewis Index (TLI; Tucker & Lewis, 1973), and root mean squared error of
approximation (RMSEA; Steiger & Lind, 1980). Values of greater than .90 for the
CFI and TLI (Bentler & Bonnett, 1980) and less than .08 for the RMSEA (Brown &
Cudeck, 1993; Steiger & Lind, 1980) indicate good fit. Though 8
2
is of historical
value in the SEM literature (being the first fit index developed), several criticisms
such as sensitivity to sample size and a tendency toward not being 8
2
distributed in
44
small samples or with non-normal data preclude its use as a sole indicator of
model fit. In addition, 8
2
p-values are not meaningful in restricted factor analysis,
without a priori restrictions. Therefore, while 8
2
is used for other purposes such as
nested model comparisons, it is used only in concert with other fit indices for
evaluations of model fit.
Table 3
Initial two- and three-factor results from factor analysis of the 20-item APSD
(average scores)
2 Factor Solution 3 Factor Solution
Item F1 F2 F1 F2 F3
1. Blames Others 0.72 -0.05* 0.57 -0.09* 0.26
2. Illegal Activities 0.49 0.06 0.60 -0.04* -0.01*
4. Acts without thinking 0.59 0.20 0.63 0.11 0.08
5. Emotions shallow 0.60 0.07 0.49 0.03 0.21
6. Lies easily, skillfully 0.70 0.08 0.64 0.01 0.18
9. Gets bored easily 0.47 0.06 0.33 0.05 0.21
10. Uses/cons others 0.67 0.05 0.45 0.05 0.32
13. Risky activities 0.56 0.13 0.67 0.02 0.02
15. Angry when corrected 0.62 -0.03* 0.40 -0.01* 0.35
17. Does not plan ahead 0.44 0.08 0.46 0.02 0.07
8. Brags excessively 0.56 -0.16* -0.01* 0.65 -0.04*
11. Teases others 0.67 -0.04* 0.28 0.48 0.02
14. Can be charming 0.67 -0.03* 0.25 0.51 0.03
16. Thinks more important 0.57 -0.15* 0.13 0.51 -0.07*
3. Concern for school 0.03 0.80 0.22 -0.16* 0.69
7. Keeps promises -0.06* 0.70 0.08 -0.07* 0.63
12. Feels bad or guilty -0.03* 0.52 0.06 0.07 0.54
18. Concerned for others 0.19 0.46 0.23 0.44 0.65
19. Does not show emotions 0.21 0.21 0.32 0.14 -0.04*
20. Keeps same friends 0.08 0.27 0.17 0.23 -0.05*
Note. Factor analysis was conducted using ULS estimation for factor extraction and Promax
rotation. Standardized rotated factor scores of .30 or greater are in boldface.

The 18-item 2F structure showed good fit to the data, 8
2
(134) = 240.13, p <
.001, CFI = .92, TLI = .90, RMSEA = .06. All loadings were significant. Factor 1
included twelve items, and Factor 2 included four items. The items making up Factor
2 corresponded to items comprising the Callous-Unemotional factor reported by
45
Frick et al. (2000), while the remaining items that made up Factor 1 corresponded
to the Narcissism and Impulsivity factors reported by Frick et al. (2000). The
correlation between Factor 1 and Factor 2 was .48.
The 14-item 3F structure also showed good fit to the data, 8
2
(74) = 116.38, p
<.001, CFI = .95, TLI = .94, RMSEA = .05, and a direct nested model comparison
indicated that this fit was significantly better than that of the 2-factor structure (U8
2
=
123.75, Udf = 60, p < .001). All loadings were significant. Factor 1 included 7 items
(four of which corresponded to Frick et al.’s [2000] Impulsivity factor, 2 of which
were dropped from their solution, and one of which loaded onto their Narcissism
factor), Factor 2 included four items (all of which corresponded to Frick et al.’s
Narcissism factor) and Factor 3 included 3 items (all of which corresponded to Frick
et al.’s Callous-Unemotional factor). As such, the factors will herein be labeled F1-
IMP, F2-NAR and F3-CU. F1-IMP and F2-NAR correlated .82, Factor 1-IMP and
F3-CU correlated .47, and F2-NAR and F3-CU correlated .26.
Summary of Average score EFA and RFA Results. The EFA on APSD item
scores averaged across caregiver, teacher and child ratings suggested 2F and 3F
models. Though both evidenced good fit to the data, a nested model comparison
indicated that the 14-item 3F model, with F1-IMP (7 items), F2-NAR (4 items) and
F3-CU (3 items) scales showed significantly better fit to the data than the 2F model.
Results of the EFA indicated that items 9, 15, 19 and 20 failed to load significantly
on any factors, and thus they were excluded from the measure. Items 19 and 20 have
46
often performed poorly across many types of studies, leading to suggestions that
they be excluded from the APSD measure (see Poythress et al., 2006).
Thus, given the desirability, theoretical meaningfulness, and statistically
superior fit of the three-factor structure, it was retained for use in subsequent
analyses.
Restricted Factor Analysis by Informant
The 3F model derived in the EFA was then investigated for each informant
via RFA. Models were first tested in the caregiver-report data and then in the teacher
and self-report data. As described above, the following analyses were performed in a
stepwise fashion and are presented in the order in which they were performed. Table
4 provides a chronology of successive models tested. In contrast to the average
scores used in the EFA, the raw score variables for each informant are considered to
be ordered categorical, and thus the robust weighted least squares mean and variance
adjusted (WLSMV) estimator was used in the individual informant RFAs.
RFA of 14-item 3F model in caregiver-report data. The initial 3F model
showed acceptable fit to the caregiver data, 8
2
(74) = 140.13, p <.001, CFI = .91, TLI
= .89, RMSEA = .07. All factor loadings were significant. The factor correlations
showed a very high correlation between F1-IMP and F2-NAR (r = .99), and more
modest correlations between F1-IMP and F3-CU (r = .36) and between F2-NAR and
F3-CU (r = .32).
47
RFA of 14-item 3F model in teacher-report data. The initial 3F model
also showed good fit to the teacher data, 8
2
(74) = 128.24, p <.001, CFI = .94, TLI =
.92, RMSEA = .06. All factor loadings were significant. F1-IMP and F2-NAR
correlated .92, F1-IMP and F3-CU correlated .62 and F2-NAR and F3-CU correlated
.36.
RFA of 14-item 3F model in self-report data. The initial 3-factor model did
not converge in the self-report data, implying misspecification of the model. The
original EFA results were revisited, and it was noted that Item 18 was the strongest-
loading item of those that were initially excluded from the model. Item 18 (Is
concerned about the feelings of others) loaded firstly onto F3-CU and secondly onto
F2-NAR. Though it loaded simply on the F3-CU factor in Frick et al.’s (2000) study,
allowing it to crossload on the two factors here was both statistically and
theoretically justifiable. Statistically, factor loadings were substantial; .44 on F2-
NAR and .65 on F3-CU. Theoretically, a lack of concern for others’ feelings could
be interpreted as a narcissistic trait, along with teasing others and thinking one is
more important than others, but it could certainly also be thought of as a callous-
unemotional feature, when considered alongside not keeping ones promises and not
feeling bad or guilty after committing a wrongdoing. Thus, the 3F model was
respecified, allowing Item 18 to load on both F2-NAR and F3-CU. This model was
then re-evaluated for fit in the caregiver- and teacher-report data.
48
RFA of 3F model with Item 18 cross-loading in caregiver-report data.
This model showed good fit to the caregiver data, 8
2
(40) = 70.26, p <.001, CFI =
.97, TLI = .98, RMSEA = .06. Remarkably, all factor loadings were significant with
the exception of the loading of item 18 on F2-NAR. Item 18 loaded significantly on
F3-CU, however. Thus, Item 18 was dropped from F2-NAR, and allowed only to
load onto F3-CU. This model was then re-tested in the caregiver-report data.
RFA of 15-item 3F model with item 18 (no cross-loading) in caregiver-report
data. This model showed very good fit to the caregiver data, 8
2
(41) = 78.52, p <.001,
CFI = .96, TLI = .98, RMSEA = .06. All loadings were significant. F1-IMP and F2-
NAR again correlated highly (r = .96), while correlations were more moderate for
F1-IMP and F3-CU (r = .48), and F2-NAR and F3-CU (r = .48).
RFA of 15-item 3F model with item 18 (no cross-loading) in teacher-report
data. This model also showed good fit to the teacher data, 8
2
(41) = 110.74, p <.001,
CFI = .94, TLI = .96, RMSEA = .09. All loadings were significant. Factor
correlations were similar to those in the caregiver data for F1-IMP and F2-NAR (r =
.94), F1-IMP and F3-CU (r =.77) and F2-NAR and F3-CU (r = .61).
RFA of 15-item 3F model with item 18 (no cross-loading) in self-report data.
This model showed moderate fit to the child data, 8
2
(41) = 93.70, p <.001, CFI =
.87, TLI = .89, RMSEA = .07. Though fit indices were slightly lower than desired,
no localized sources of misfit were identified (e.g. all loadings were significant, all
residuals were positive and modification indices did not offer any suggestions for
49
improvement of fit). F1-IMP and F2-NAR were correlated .70, while correlations
between F1-IMP and F3-CU (r = .33), and F2-NAR and F3-CU (r = .16) were lower.
Summary of RFA-by-Informant Results
The left half of Table 4 displays the succession of the 3F models tested by
RFA in each informant dataset. The resultant 15-item 3F model presented acceptable
to very good fit across all three informants. Item 18 was initially excluded after the
EFA because it appeared to crossload on F2-NAR and F3-CU, but its inclusion (and
later relegation to loading on F3-CU only) allowed for better fit of the model,
especially in the self-report data. Each factor and its items corresponded to the
factors delineated by Frick et al., (2000), with the single exception of Item 5
(Emotions seem shallow) loading not on the Narcissism factor (as in Frick et al.’s
model) but instead on the Impulsivity factor. The factor correlation between F1-IMP
and F2-NAR was very high for caregiver- and teacher-report data, but more
reasonable in the self-report data and thus the 3F model was retained for further
analyses. Subsequent evaluations of discriminant validity were especially pertinent
to this correlation.

Specific Aim 1B. To test the factorial invariance of the resultant model of the APSD
between gender (males vs. females), maltreatment group (maltreated vs.
nonmaltreated) and across ethnic group (Caucasian, Latino, African American and
Biracial).
50
Testing for invariance across groups involves comparing the fit of two
nested models (i.e., determining whether the difference in chi-square for the two
models is significant given the change in degrees of freedom). As previously
mentioned, when using the software program Mplus with ordered categorical
variables (and a polychoric correlation matrix) as was the case here, the WLSMV
estimator is used. Unlike most other situations, neither the difference in chi-square
(U8
2
) nor the difference in degrees of freedom (Udf) between two models using the
WLSMV estimator are calculated as straightforward subtractions. The resulting U8
2
does not follow a chi-square distribution and thus a special, two-step difference test
procedure is employed via Mplus. Of importance is the p-value of the difference test
generated, which is adjusted for the WLSMV estimator and is the relevant indicator
of significant difference between two models (see Muthén, duToit, & Spisic, 1997,
and Muthén & Muthén, 1998-2006 for more detail).
With ordered categorical outcomes, measurement parameters comprise
thresholds (the minimum level of the latent trait needed to endorse a given response
on the observed measure) and factor loadings. As suggested by Muthén & Muthén
(1998-2006), measurement invariance was tested by first specifying a model in
which thresholds and factor loadings were free to vary across groups, while residual
variances were fixed to 1 and factor means were fixed to zero in all groups. This
model was then compared to a second, nested model where thresholds and factor
loadings were constrained to be equal across groups. If the chi-square difference test
51
is significant, this indicates that there is some noninvariance across groups. In this
case, to identify sources of noninvariance, constraints on individual item thresholds
and factor loadings were relaxed one at a time. Again, recall that it was necessary to
delineate a stable factor structure that was identical across each informant.
Accordingly, when a specific item was found to be noninvariant across sex,
maltreatment, or ethnic groups, that item was eliminated from the factor structure
(and the resulting ‘new baseline model’ was subjected to all prior analyses).
Invariance of the 15-item 3F model across Sex. The 15-item model showed
invariance across male and female groups in the caregiver data, U8
2
= 10.88, p = .82.
Invariance across sex was also shown in the teacher data, U8
2
= 16.18, p = .37 as
well as in the child data, U8
2
= 20.71, p = .24.
Invariance of the 15-item 3F model across Maltreatment Group. The 15-
item model showed noninvariance across maltreated and comparison groups in the
caregiver data, U8
2
= 34.72, p < .01. Subsequent individual item analyses indicated
that items 2 (U8
2
= 7.55, p < .05), 6 (U8
2
= 10.97, p < .001), 14 (U8
2
= 4.69, p < .05)
and 17 (U8
2
= 8.39, p = .01) were noninvariant. Originally, items 2, 6, and 17 loaded
on F1-IMP, while Item 14 loaded on F2-NAR. These items were dropped and the
model was re-specified with the exclusion of these four items. To ensure that this
new factor structure fit well in each of the informant datasets, the new 11-item 3F
model was tested via RFA for each informant prior to continuing with invariance
analyses (see Table 4).
52
RFAs testing the 11-item 3F model. The 11-item model fit the caregiver
data very well, 8
2
(26) = 41.18, p = .03, CFI = .98, TLI = .98, RMSEA = .05. Factor
correlations remained similar to those of previous models for F1-IMP and F2-NAR
(r = .93), F1-IMP and F3-CU (r = .47) and F2-NAR and F3-CU (r = .41). The 11-
item model also fit the teacher data well, 8
2
(26) = 71.33, p < .001, CFI = .94, TLI =
.96, RMSEA = .09. F1-IMP and F2-NAR were correlated .88, F1-IMP and F3-CU
correlated .62, and F2-NAR and F3-CU correlated .41. Consistent with previous
models, fit indices indicated modest fit for the 11-item model in the child data, 8
2
(26) = 44.49, p < .001, CFI = .87, TLI = .86, RMSEA = .07, though all loadings
were significant and no evidence of localized strain was found. F1-IMP and F2-NAR
correlated .74, F1-IMP and F3-CU correlated .27 and F2-NAR and F3-CU correlated
.27 as well. Having found acceptable fit in the informant datasets, invariance
analyses were once again attempted.
Invariance of the 11-item 3F Model Across Sex. The 11-item model showed
invariance between males and females in the caregiver data, U8
2
= 11.29, p = .59, in
the teacher data, U8
2
= 8.11, p = .70 and in the child data, U8
2
= 17.64, p = .28.
Invariance of the 11-item 3F Model Across Maltreatment Group. The 11-
item model was also invariant between maltreated and comparison groups in the
caregiver-report data, U8
2
= 14.38, p = .28. However, noninvariance was found in the
teacher data, U8
2
= 19.12, p = .04. Individual invariance analyses identified item 7
(which loaded on F3-CU) as the source of noninvariance. Item 7 was thus excluded
53
from the measure, and a new 10-item 3F model was specified. Once again, this
new model was tested via RFA for each informant prior to continuing with
invariance analyses (see Table 4).
RFAs testing the 10-item 3F model. RFA on the 10-item model showed good
fit in the caregiver data, 8
2
(22) = 37.08, p <.05, CFI = .98, TLI = .98, RMSEA = .05,
in the teacher data, 8
2
(22) = 46.25, p <.01, CFI = .96, TLI = .97, RMSEA = .07 and
modest fit in the child data, 8
2
(22) = 54.41, p <.001, CFI = .86, TLI = .84, RMSEA =
.08. Factor correlations remained high for F1-IMP and F2-NAR (.93 in caregivers,
.88 in teachers, and .74 in children) and moderate for F1-IMP and F3-CU (.46, .67,
and .27 respectively) and F2-NAR and F3-CU (.44, .65 and .27 respectively).
Invariance of the 10-item 3F model across Sex. Model 5 was invariant across
males and females in the caregiver data, U8
2
= 7.19, p = .78, teacher data, U8
2
= 9.39,
p = .50, and child data U8
2
= 15.81, p = .15.
Invariance of the 10-item 3F model across Maltreatment Group. The 10-item
3F model was also invariant across maltreatment and comparison groups in the
caregiver data, U8
2
= 13.75, p = .18, teacher data, U8
2
= 12.36, p = .14 and child data,
U8
2
= 59.40, p = .87.
Invariance of the 10-item 3F model across Ethnicity. The 10-item 3F model
was invariant across Black, White, Latino and Biracial ethnic groups in the caregiver
data, U8
2
= 21.88, p = .41, teacher data, U8
2
= 11.54, p = .83, and child data U8
2
=
21.77, p = .47.
54
Summary of Invariance Analyses. The latter half of Table 4 displays the
succession of the 3F models tested in the invariance analyses. Following a
conservative procedure that required items to load significantly and invariantly on
the three factors, a universal factor structure was found for the APSD across the three
informants. Items 2 (Engages in illegal activities), 6 (Lies easily and skillfully), 7
(Keeps promises), 14 (Can be charming) and 17 (Does not plan ahead) were
excluded from the models due to invariance across maltreatment group. In the
caregiver data, Item 2 showed greater nonnormality in the comparison group (mean
= .04, SD = .24, kurtosis = 55.60, skewness = 7.31) than in the maltreated group
(mean = .08, SD = .27, kurtosis = 8.451, skewness = 3.21), which may reflect the
noninvariance between groups. The remaining noninvariant items did not show
differences in distribution between groups, and may instead have evinced
noninvariance for reasons relating to their interpretation within the different groups.
For instance, items 6 and 14 address behaviors that are by definition somewhat
difficult to identify, perhaps causing caregivers in the two groups to interpret the
item meaning differently. Differences in threshold parameters indicated that it was
‘easier’ for caregivers in the maltreated group to rate Item 6 as being somewhat true
of their child than for caregivers in the comparison group. That is, lower levels of the
latent trait (in this case, Impulsivity) were sufficient for a positive endorsement (a
score above 0) on Item 6 in maltreated adolescents than were sufficient in
comparison adolescents. The reverse was true of Items 14 and 17 (Does not plan
55
ahead, leaves things until the last minute), where it was ‘easier’ for caregivers of
comparison adolescents to endorse the ‘somewhat true’ value than caregivers of
maltreated adolescents. For teachers, it was easier to endorse the ‘somewhat true’
value for Item 7 (Is good at keeping promises) in maltreated adolescents than
comparison adolescents, but easier to endorse the ‘often true’ response in
comparison adolescents vs. maltreated adolescents.
Factor loadings from the RFAs-by-informant for this model are shown in
Table 5. The 10-item 3F model, which showed modest to very good fit across
caregiver, teacher and self-report data, and invariance between sex, maltreatment
group and ethnic group, was thus carried forth and applied to the subsequent
analyses. F1-Impulsive comprised Items 1 (Blames others for own mistakes), 4 (Acts
without thinking), 5 (Emotions seem shallow or fake), and 13 (Engages in risky
activities). F2-Narcissistic comprised Items 8 (Brags excessively), 11 (Teases or
makes fun of others) and 16 (Thinks self is more important than others). F3-Callous-
Unemotional comprised Items 3 (Is concerned about schoolwork – reversed), 12
(Feels bad or guilty after doing something wrong – reversed), and 18 (Is concerned
about the feelings of others – reversed).

56
Table 4
Successive factor models tested during restricted factor analyses and invariance analyses
Models tested during RFA analyses (Specific Aim 1A)
Models tested during Invariance analyses (Specific Aim 1B)
14-item 3F model from
EFA
3F model, Item 18
cross-loading allowed
15-item 3F Model, no
cross-loading
11-item 3F Model 10-item 3F model
Item F1
IMP
F2
NAR
F3
CU
F1
IMP
F2
NAR
F3
CU
F1
IMP
F2
NAR
F3
CU
F1
IMP
F2
NAR
F3
CU
F1
IMP
F2
NAR
F3
CU
1. Blames others X - - X - - X - - X - - X - -
2. Illegal Activities X - - X - - X - - - - - - - -
4. Acts without thinking X - - X - - X - - X - - X - -
5. Shallow emotions X - - X - - X - - X - - X - -
6. Lies easily, skillfully X - - X - - X - - - - - - - -
13. Risky activities X - - X - - X - - X - - X - -
17. Does not plan ahead X - - X - - X - - - - - - - -
8. Brags excessively - X - - X - - X - - X - - X -
11. Teases others - X - - X - - X - - X - - X -
14. Can be charming - X - - X - - X - - - - - - -
16. Thinks more important - X - - X - - X - - X - - X -
3. Concern for school
a
- - X - - X - - X - - X - - X
7. Keeps promises
a
- - X - - X - - X - - X - - -
12. Feels bad or guilty
a
- - X - - X - - X - - X - - X
18. Concern for others
a
- - - - X X - - X - - X - - X
Results of analyses:
RFA: Good fit in
caregiver and teacher
data, no convergence
in self-report data
»
RFA: Item 18 did not
load significantly on
F2-NAR in caregiver
data
»
RFA: Good fit in
caretiver, teacher, and
self-report data
INV: Items 2, 4, 14,
and 17 noninvariant
across maltreatment
group in caregiver data
»
INV: Item 7
noninvariant across
maltreatment group
in teacher data
»
Final model; good fit
in caregiver, teacher
and self-report data
Note. X indicates specified factor loading. F1-IMP = Factor 1 Impulsivity, F2-NAR = Factor 2 Narcissistic, F3-CI = Factor 3 Callous-
Unemotional. RFA = restricted factor analysis. INV = invariance analysis.
a
Reverse-scored.
57

Specific Aim 2A: To employ correlational analyses to examine the reliability,
convergent and discriminant validity of the APSD factors as rated by self, caregiver,
and teacher using a multitrait-multimethod matrix.
Using the universal model delineated in Specific Aim 1, correlations
between the three factor scores were entered into a 3 by 3 trait-method matrix (see
Table 6). Since the areas above and below the main diagonal are redundant, only
values in the lower area are presented. The five Campbell and Fiske (1959)
guidelines outlined previously were evaluated by comparing values in the matrix.
1. Reliability. The reliability estimates in the main diagonal ranged from low
to moderate. The caregiver F1-IMP reliability value was the highest (0 = .76), and
the child F3-CU value was the lowest (0 = .39). F1-IMP yielded the highest M values
Table 5
Standardized factor loadings for caregiver, teacher and child restricted factor
analyses on final 10-item 3-factor model
Caregiver-Report Teacher-Report Self-Report
Item
F1
IMP
F2
NAR
F3
CU
F1
IMP
F2
NAR
F3
CU
F1
IMP
F2
NAR
F3
CU
1. Blames others .82 .72 .66
4. Acts without thinking .81 .84 .68
5. Emotions shallow .79 .74 .51
13. Risky activities .68 .80 .54
8. Brags excessively .54 .66 .60
11. Teases or makes fun .73 .88 .73
16. Thinks more important .71 .76 .52
3. Concerned about school (R) .49 .55 .43
12. Feels bad or guilty (R) .58 .60 .42
18. Concerned for others (R) .91 .88 .82
Note. RFAs computed with robust weighted least squares means and variances adjusted estimator.
F1-IMP = Factor 1 Impulsivity. F2-NAR = F2 Narcissism. F3-CU = Callous/Unemotional.
All loadings significant at p < .05.
58

across all three informants, while teacher ratings yielded the highest average M value
across all three trait factors. Though 0s in general are lower than the .70 that is
desirable for internal consistency (Cicchetti, 1994; Nunnally & Bernstein, 1994),
they are based on very short scales. For the most part they are the highest values in
the matrix. Mean 0 is .60.
2. Convergent Validity. The correlations in the validity diagonals
(underlined values in Table 6) are monotrait-heteromethod values, indexing the
extent to which different informants agree on the same trait. To provide evidence of
convergent validity, these correlations should be high and significant. In this case,
they are all significant, and overall are the next highest items after the reliability
coefficients in the matrix following the pattern described by Campbell and Fiske.
The correlations represent small to medium associations (Cohen, 1988), and values
are typical of multi-informant data on child problem behavior (e.g. Achenbach, et al.,
1987). Caregiver and teacher F1-IMP yields the highest correlation (r = .34, p <
.001) and caregiver and teacher F1-NAR yields the lowest (r = .18, p < .01). Self-
report correlations with both caregiver and teacher data are fairly consistent, ranging
from r = .21 to r = .26. Mean MTHM value is .25.
3. Discriminant Validity and Method Effects – heterotrait, monomethod
(HTMM) values. Inspection of the HTMM values (solid triangles in Table 6) provide
little evidence for discriminant validity and substantial evidence for method effects
among the three traits. The raw score factor correlations for F1-IMP and F2-NAR are
59

fairly high (r = .62 for caregiver data, r = .63 for teacher data, and r = .36 for child
data), reflecting the high RFA-derived factor correlations. Campbell and Fiske
recommended that these HTMM values should be lower than the corresponding
validity diagonal (MTHM) estimates. Failures occur for the caregiver and child F1-
IMP/F2-NAR correlation, and all of the teacher HTMM correlations. Elevations in
HTMM values may also be attributed to method effects (variance common to two
measurements that is not due to trait covariance). In the case of F1-IMP and F2-NAR
here, it was known from the previously conducted RFAs that these two constructs
are indeed highly correlated. However, it is impossible to distinguish the effects of
trait covariance from method variance in the context of the MTMM matrix. Variance
can be partitioned and calculated via RFA techniques which will be presented later.
Mean HTMM value was .34.
4. Discriminant Validity – heterotrait-heteromethod (HTHM) triangles. To
provide evidence for discriminability among traits, these values should be the lowest
in the matrix. The Campbell and Fiske formulation assumed that traits and methods
were not correlated, and thus HTHM values, which represent relationships between
different traits measured by different methods, should be negligible. In practice,
however, this assumption is untenable, in particular when methods are informants
using essentially the same form, and traits are factors of a larger construct, as is the
case here. In general, these values conform to the Campbell & Fiske guideline; most
are relatively low and nonsignificant. The mean HTHM correlation was .13.
60

Table 6
Multitrait-Multimethod correlational matrix of APSD data for three trait factors and three informants
Caregiver Teacher Self
Traits F1
IMP
F2
NAR
F3
CU
F1
IMP
F2
NAR
F3
CU
F1
IMP
F2
NAR
F3
CU
Mean SD
F1-IMP .76
1.71 1.75
F2-NAR
.62
***
.51 1.11 1.22
Caregiver
F3-CU .25
***
.15
*
.63 2.83 1.64
F1-IMP
.34
***
.17
*
.12
†
.74
1.65 1.76
F2-NAR .14
*
.18
**
.06 .64
***
.69 1.11 1.43
Teacher
F3-CU .29
***
.20
**
.32
***
.43
***
.33
***
.63 2.23 1.47
F1-IMP
.24
***
.25
***
.05
.21
**
.12
†
.21
**
.55
1.93 1.57
F2-NAR .05 .26
***
.08
†
.12
†
.21
**
-.01 .36
***
.49 1.25 1.27 Self
F3-CU .11 .10 .25
***
.16
*
.11 .26
***
.11 .12
†
.39 1.85 1.32
Note. Main diagonal, italics = internal consistency reliability estimates. Solid triangles = heterotrait-monomethod values. Solid rectangles =
monomethod blocks. Underscored = monotrait-heteromethod values (validity diagonals). Dashed triangles = heterotrait heteromethod values.
Dashed rectangles = heteromethod blocks. F1-IMP = Factor 1 Impulsivity, F2-NAR = Factor 2 Narcissistic, F3-CU = Factor 3 Callous-
Unemotional.
*
p < .05,
**
p < .01,
***
p < .001,
†
p < .10.
61
5. Trait interrelationships. Patterns among the correlations varied across the
different blocks. The HTMM triangles have a similar pattern across informants,
where the F1-IMP/F2-NAR correlation is greater than the F2-NAR/F3-CU
correlation, which in turn is greater than the F1-IMP/F3-CU correlation. The patterns
among the HTHM values are less consistent. Three out of 6 method pairs (i.e.
teacher/caregiver, self/teacher, and teacher/self) showed patterns where the F1-
IMP/F3-CU correlation was greater than the F1-IMP/F2-NAR correlation which was
greater than the F2-NAR /F3-CU correlation. The other 3 method pairs each have
different patterns. The lack of agreement in trait pattern indicates that relationships
among the traits may differ across informants.
Summary of MTMM Matrix Analysis
Overall, the data presented here follow the rules suggested by Campbell and
Fiske, and imply convergent validity among the three traits. There is mixed evidence
for discriminant validity, in that HTHM values are low and nonsignificant, but
HTMM correlations are substantial. Method effects are also implied by these HTMM
correlations, though it is difficult to determine whether the magnitude of the
correlations is driven by high factor correlations (low discriminant validity), high
method variance, or both in the context of the matrix. Reliability coefficients for the
measures are low but respectable given the small numbers of items per scale.

62
Specific Aim 2B: To employ confirmatory factor analysis to further examine the
construct validity of the APSD factors as rated by self, caregiver, and teacher using
a multitrait-multimethod approach.
Confirmatory Factor Analysis of the MTMM
The solution for the full model (Model 1; see Figure 3a), which posits
oblique traits and oblique methods, did not converge. Failure of this (three-method,
three-trait, full) model is well documented in the MTMM literature (see e.g. Kenny
& Kashy, 1992; Marsh, 1989) and thus was not unexpected in the present study.
Indeed, even when the model is estimable, improper or nonsensical solutions often
result, leading some to recommend that interpretations and conclusions be based on
correlated uniqueness (CU) models. In CU models, method variance is modeled not
via method factors but by correlating the unique variances of indicators measured by
the same method (Marsh, Byrne, & Craven, 1993). The general and CU models are
very similar and the same conclusions can essentially be drawn from both. One
drawback of the CU model is that it does not allow for correlations between
methods. Consequences for this generally manifest as elevated trait factor loadings
and factor intercorrelations. These effects have been shown to be minimal (Saris,
1990; Scherpenzeel, 1995), but will be discussed in context. Accordingly, correlated
uniqueness models were also estimated and employed in nested model comparisons.
However, prior to employing the correlated uniqueness approach, and guided
by the taxonomy of models previously presented in the literature (Marsh, 1989;
Widaman, 1985), several models were tested which differed in terms of the number
63
of traits or method factors posited and whether trait or method factors were
intercorrelated. These models will be termed “General CFA models” and labeled
numerically, to distinguish them from the CU models which will be labeled
alphabetically. Model diagrams for the general CFA models are presented in Figure
3, and fit indices and nested model comparisons are presented in Table 7.
64
(a) Model 1: Oblique traits and oblique methods (b) Model 2: 1 Trait and no methods

(c) Model 3: Oblique traits and no methods (d) Model 4: Orthogonal traits and no
Methods

(e) Model 5: No traits and oblique methods (f) Model 6: No traits and orthogonal methods
Figure 3. General confirmatory factor analysis model diagrams.
Oval objects represent latent variables, rectangular objects represent measured indicators (e.g. TF1 =
observed teacher-report F1-IMP score;), two-headed arrows represent bi-drectional (correlational)
effects, and one-headed arrows represent direct effects. ‘Parent’ is used here instead of ‘caregiver’ to
distinguish between caregiver (P) and child (C) variables. Thus, 'Parent = caregiver-report method
factor. Teacher = teacher report method factor. Child = self-report method factor. F1-
IMP=Impulsivity Trait factor. F2-NAR = Narcissistic trait factor. F3-CU=Callous/unemotional trait
factor. Psychopathy=general psychopathy trait factor.
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
Psychopathy
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
Teacher Child Parent
0
0
0
F1-IMP
F2-NAR F3-CU
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
Teacher Child Parent
F1-IMP
F2-NAR F3-CU
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
0
0
0
F1-IMP F2-NAR F3-CU
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
Teacher Child Parent
65
Model 2 specified a single trait factor, defined by all 9 informant-trait
indicators. This model did not fit the data well, 8
2
(27)

= 207.67, p < .001, providing
some preliminary support for a multidimensional construct (discriminant validity)
comprising more than one trait.
Model 3 specified oblique (freely correlated) trait factors (F1-IMP, F2-NAR,
and F3-CU) and no method effects. This model did not fit the data well, 8
2
(24) =
184.24, p < .001. In contrast, Model 4 comprised orthogonal trait factors and no
method effects. Fit indices indicated that Model 4 provided a very poor fit to the
data, \
2
(27) = 323.23, p < .001. Since Model 4 is nested in Model 3, the difference in
their \
2
values can be tested for significance. Model 3 provides a better fit to the data
(although fit is still quite poor), indicating that modeling correlations between APSD
factors more accurately describes the observed data than modeling uncorrelated
factors, providing some evidence for convergent validity.
Table 7
Goodness of Fit Indices and Nested Model Comparisons for General CFA
Models
8
2
df RMSEA CFI TLI U8
2
Udf p
Model 1 - Full model Did not converge
Model 2 - One trait 207.67 27 .18 .55 .40
Model 3 - 3 Oblique Traits 184.24 24 .18 .60 .40
Model 4 - 3 Orthogonal Traits
(vs. Model 3)
323.23 27 .23 .26 .02
138.50

3 < .001
Model 5 - 3 Oblique Methods 88.57 24 .11 .84 .75
Model 6 - 3 Orthogonal Methods
(vs. Model 5)
133.25 27 .14 .73 .65
43.69

3 < .001
Note. RMSEA = Root mean squared error of approximation. CFI = Comparative Fit index. TLI
= Tucker-Lewis Index.
66
Model 5 specified oblique method factors (caregiver, teacher, and child
ratings) and no trait factors. Fit for this model appeared to improve over previously
tested models, but was still unsatisfactory, 8
2
(24) = 88.57, p < .001. Model 6
constrained the method factors to be orthogonal, which resulted in a significant
worsening of fit, 8
2
(27) =133.25, p < .001 (U8
2
= 44.69, Udf = 3, p < .001). As with
trait factors, method factors appear to be correlated.
Next, models including correlated uniquenesses were tested (see model
diagrams in Figure 4). Results of the models and nested model comparisons are
presented in Table 8. Model A, which would serve as the “full” model, posited three
oblique trait factors and correlated uniquenesses of indicators measured with the
same method (rated by the same informant). Fit for this model was very good, 8
2
(15)
= 22.27, p = .11.
Model B removed the presence of trait factors, including only correlated
uniquenesses among the indicators. This model showed very poor fit to the data,
significantly worse than Model A (U8
2
= 112.04, Udf = 12, p < .001). The results of
this model comparison provide support for the convergent validity of the modified
APSD factors.
Model C replaced the 3 oblique traits of Model A with one trait which was
defined by all 9 informant-trait indicators. This model showed a significant decrease
in fit when compared to Model A, U8
2
= 38.01, Udf = 3, p < .001, providing evidence
of discriminant validity. This comparison indicates that three correlated trait factors
fit the data better than one general trait factor.
67
Given the high correlation between F1-IMP and F2-NAR seen in the CFAs, a
further test of discriminant validity was performed. Model D tested the fit of a 2
factor model in which the F1-IMP and F2-NAR indicators loaded onto one trait
factor (F1-IMP/NAR) and the F3-CU factor remained the same. Model fit was poor,
8
2
(17) = 46.17, p < .001, and was significantly worse than that of the full model, U8
2
= 24.43,Udf = 2, p < .001, indicating that in fact 3 factors describe the data more
accurately than 2 factors.
Returning to the three-trait model, Model E posited orthogonal relationships
among F1-IMP, F2-NAR and F3-CU. A significant decrease in fit resulted from the
model comparison, U8
2
= 28.67, Udf = 3, p < .001, providing evidence for
intercorrelations among the traits (and thus, evidence for convergent validity).
Having established that 3 correlated traits factors provide the best fit to the
data, Model F again posited 3 traits and removed all correlations between
uniquenesses. A comparison between Model A and Model F tested the presence of
method effects. A significant decrease in fit between Model A and Model F provided
evidence that there were significant method effects within informant in the data U8
2
= 112.04,Udf = 9, p < .001.
68
(a) Model A: Oblique traits and correlated (b) Model B: No traits and correlated
uniquenesses uniquenesses

(c) Model C: One trait and correlated (d) Model D: Two traits and correlated
uniquenesses uniquenesses

(e) Model E: Three orthogonal traits and (f) Model F: Three oblique traits and
correlated uniquenesses uncorrelated uniquenesses
Figure 4. Correlated uniquenesses CFA
model diagrams.
Note. Oval objects represent latent variables,
rectangular objects represent measured indicators,
two-headed arrows represent bi-directional
(correlational) effects, and one-headed arrows
represent direct effects. P = caregiver-report.
Teacher = teacher-report. Child = self-report. F1-
IMP = Impulsivity Trait factor. F2-NAR =
Narcissism trait factor. F3-CU =
Callous/Unemotional trait factor. F1-IMP/F2 AR=
Combined Impulsivity and Narcissism trait factor.
Psychopathy = general psychopathy trait factor.

F1-IMP
F2-NAR
F3-CU
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
O
PF1
O
PF2
O
PF3 O
TF1
O
TF2
O
TF3
O
CF1
O
CF2
O
CF3
0 0
0
O
PF1
O
PF2
O
PF3
O
TF1
O
TF2
O
TF3
O
CF1 O
CF2
O
CF3
F1-IMP F2-NAR F3-CU
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
F1-IMP F2-NAR F3-CU
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
O
PF1
O
PF2
O
PF3 O
TF1
O
TF2 O
TF3
O
CF1
O
CF2
O
CF3
F1-IMP F2-NAR F3-CU
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
O
PF1
O
PF2
O
PF3
O
TF1
O
TF2
O
TF3
O
CF1
O
CF2
O
CF3
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
O
PF1
O
PF2
O
PF3
O
TF1
O
TF2
O
TF3
O
CF1
O
CF2 O
CF3
O
CF3
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
F1-IMP/
NAR
F3-CU
O
PF1
O
PF2
O
PF3
O
TF1
O
TF2
O
TF3
O
CF1
O
CF2
PF1 PF2 PF3 TF1 TF2 TF3 CF1 CF2 CF3
Psychopathy
O
PF1
O
PF2
O
PF3
O
TF1
O
TF2
O
TF3
O
CF1
O
CF2
O
CF3
(g) Model G: Three oblique traits, freely
correlated uniquenesses
69
.
Table 8
Fit indices and nested model comparisons for correlated uniquenesses models
8
2
df RMSEA CFI TLI U8
2
Udf p
Model A - 3 Oblique traits, CUs 22.27 15 .05 .98 .96 .11
Model B - No traits, CUs
(vs. Model A)
133.7827 .14 .73 .65
112.04 12

<.001
Model C - 1 Oblique Trait, CUs
(vs. Model A)
59.75 18 .11 .90 .79
38.01 3 <.001
Model D - 2 Oblique Traits, CUs
(vs. Model A)
46.17 17 .09 .93 .85
24.43 2 <.001
Model E - 3 Orthogonal Traits, CUs
(vs. Model A)
50.41 18 .09 .92 .84
28.67 3 <.001
Model F - 3 Oblique Traits,
Uncorrelated Uniquenesses
(vs. Model A)
134.1924 .18 .60 .40

112.04 9 <.001
Model G - 3 Oblique traits, Perfectly
Correlated Uniquenesses Did not converge
Note. F1-IMP = Impulsivity Trait factor, F2-NAR = Narcissism trait factor. F3-CU =
Callous/Unemotional trait factor. CU = Correlated Uniquenesses. RMSEA = Root mean squared
error of approximation. CFI = Comparative Fit index. TLI = Tucker-Lewis Index.
Lastly, Model G specified all of the uniquenesses to be perfectly correlated,
which would describe a single method factor. This model did not converge,
indicative of major model misspecification.
The full CU model was deemed the best-fitting model, and independently,
provided very good fit to the data. Subsequently, examination of the parameters
allowed for a more fine-grained analysis of convergent validity, discriminant validity
and method effects. Factor loadings, factor intercorrelations, and correlations of the
error terms are presented in Table 9.

70
All factor loadings (C) were significant and moderate, ranging from .36 to
.68, mean C = .50. Overall, caregiver loadings were highest (mean C = .57) although
the single highest loading was for teacher ratings of F3-CU on F3 (C = .68).
Interestingly, the single lowest loading was also a teacher loading, of F2-NAR (C =
Table 9
Parameter estimates and Factor intercorrelations from Model A, the full
correlated uniquenesses model
APSD Factor
Parameter Estimate F1-IMP F2-NAR F3-CU
Factor loadings
Parent (Caregiver)
F1-IMP .59*
F2-NAR .65*
F3-CU .47*
Teacher
F1-IMP .53*
F2-NAR .36*
F3-CU .68*
Child
F1-IMP .43*
F2-NAR .40*
F3-CU .41*
Correlated Uniquenesses
Parent (Caregiver)
F1-IMP
F2-NAR .39*
F3-CU .09 .01
Teacher
F1-IMP
F2-NAR .52*
F3-CU .20* .23*
Child
F1-IMP
F2-NAR .27*
F3-CU .01 .08
Factor Correlations
F1-IMP 1.00
F2-NAR .59* 1.00
F3-CU .61* .39* 1.00
Note. F1-IMP = Factor 1 Impulsivity, F2-NAR = Factor 2 Narcissistic, F3-CU = Factor 3
Callous-Unemotional.
*z-value > 1.96, p < .05.
71
.36). Child ratings had the lowest overall mean loading (mean C = .41) but were more
consistent than the caregiver or teacher report loadings (range .40-.43). These
parameters offer mild support for the convergent validity of the three factors – the
indicators appear significantly related to the latent traits of Impulsivity, Narcissisism
and Callous/Unemotionality, although the strength of those relationships is
conservative. In the absence of method factors, however, variance that may be
common to different methods may be relegated to trait factor loadings (upwardly
biasing convergent validity).
Trait factor correlations are substantial, r = .59 for F1-IMP and F2-NAR, r =
.61 for F1-IMP and F3-CU, and r = .39 for F2-NAR and F3-CU. While significant,
they are not so high as to preclude notion of discriminant validity, especially
between F2-NAR and F3-CU. In addition, a limitation of the CU model is that it
does not allow for correlations between methods. The limitation of uncorrelated
methods may also be expressed here in the form of increased factor correlations
(hence downwardly biased discriminant validity).
Method effects are represented as correlations among the uniquenesses for
variables that share the same method. As can be seen from Table 9, method effects
were strongest for the Teacher-report data. Uniquenesses were correlated for all three
trait factor combinations in the teacher-report data, indicating that there is some
method and unique variance common to the traits when measured by the teachers.
For caregiver and child data, there is no evidence of method effects across the F1-
IMP and F3-CU factors, or across the F2-NAR and F3-CU factors. There does
72
appear to be some common method variance across the F1-IMP and F2-NAR factors
in both child and caregiver data. In contrast to the MTMM matrix results discussed
earlier, the method effects here are independent of shared trait variance, as trait
covariances are independently estimated. Following suggestions by Conway (1998)
which showed the average CU to be a good estimate of method variance, mean CU
was calculated within each informant and for the overall matrix (see Table 10).
Results indicate that overall, the proportion of trait variance was greater than the
proportion of method variance for the entire matrix. Trait variance accounted for
little over a quarter of the total variance, and method variance accounted for a fifth,
suggesting that error variance was quite prominent. Within each informant, results
are consistent with conclusions from the MTMM matrix and CU parameter estimate
analyses. Caregiver ratings fared the best, with trait variance accounting for 33% of
the variance, twice as much as method variance (16%). Child ratings also showed a
greater proportion of trait variance (17%) to method variance (12%) although the
ratio was much lower, and altogether the variance accounted for was quite small
(indicating a large amount of error variance). Teacher ratings showed poor
measurement properties. While trait variance was relatively high (29%), method
variance was even higher (32%), indicating that the majority of variance in the
APSD teacher measures was due not to trait-related information, but simply to the
fact that the ratings were made by teachers.

73
Table 10
Mean proportions of trait variance and method variance estimated in the full
correlated uniqueness model
Mean Trait Variance Mean Method Variance
Parent (Caregiver) .33
a
.16
b
Teacher .29
a
.32
b
Child .17
a
.12
b
Overall matrix .26
c
.20
d
Note.
a
Average squared trait factor loading for each informant.
b
Average of the correlated uniquenesses for each informant.
c
Average squared trait factor loading for all informants.
d
Average of the correlated uniquenesses for all informants.
Summary of MTMM Analysis via CFA. In sum, the full CU model is most
representative of the observed data. Convergent validity is implied by the better fit of
the model which specifies correlated trait factors, and therefore accounts for trait
variance, in comparison to models which do not include trait factors, or posit
orthogonal trait factors. Discriminant validity is evident in the superiority of
specified trait factors over a general trait factor model, and significant trait factor
loadings. Further validity of the 3-factor model over the 2-factor model was shown
in a direct comparison between these two models. Method effects were modeled in
the full CU model, and then more closely investigated by comparing the proportion
of trait variance to the proportion of method variance in measures rated by the same
informant. Findings indicate that, overall, the APSD factors used here represent a
modest amount of trait variance, and in fact include a substantial amount of error and
method variance. In particular, the teacher-rated factors are dominated by method
variance.
74
Specific Aim 3: To examine the relationship of the self, caregiver, and teacher-
reported APSD total and factor scores to relevant external correlates (aggression,
anxiety and cognitive ability) in the present sample.
CFA of RPQ Aggression and MASC Anxiety
The aggression and anxiety measures were first subjected to confirmatory
factor analysis to establish and confirm their factor structures in this sample.
Aggression (RPQ). The 2-factor solution of the RPQ reported by Raine et al.
(2006) was tested in the RPQ data in this study. Twelve of 23 items load on the
Proactive aggression factor, while the remaining 11 items load onto the Reactive
Aggression factor. Results for this two-factor structure were favorable; 8
2
(44) =
79.96, p < .001, CFI - .96, TLI = .96, and RMSEA = .06. All loadings were
significant, all parameters were properly estimated (e.g. there were no negative
residual variances, or out of range values for loadings). Thus, the 2-factor solution
from Raine et al. (2006) was deemed an acceptable model of the present RPQ data.
Factor loadings are presented in Table 11, factor intercorrelations and internal
consistency reliability estimates are presented in Table 12.
75
Table 11
Standardized factor loadings from the confirmatory factor analysis of the RPQ
RPQ Subscale
Item
Factor 1
Proactive
Factor 2
Reactive
2. Fight for status .74
4. Take things from others .82
6. Damage or break things for fun .81
9. Fight to be cool .71
10. Hurt others to win a game .63
12. Use force to manipulate others .78
15. Use force to get money .61
17. Threaten and bully others .75
18. Make prank phone calls for fun .52
20. Manipulate others to gang up .72
21. Carry weapon to use in a fight .75
23. Yell to manipulate others .78
1. Yell when annoyed .83
3. Get angry when annoyed .74
5. Get angry when frustrated .66
7. Have temper tantrums .71
8. Damage things when mad .80
11. Get angry when don’t get way .64
13. Get angry when lose a game .44
14. Get angry when others threaten .63
16. Feel better after yelling or hitting .69
19. Hit others in self-defense .53
22. Get mad or hit others when teased .61
Note. Model fit: 8
2
(44) = 79.96, p < .001, CFI = .96, TLI = .96, RMSEA = .06. RPQ = Reactive
Proactive Aggression Questionnaire.
All loadings significant at p < .05.
Table 12
Factor intercorrelations and internal consistency reliabilities for the RPQ
RPQ Subscale
Proactive Aggression Reactive Aggression Total RPQ Score
Proactive Aggression .79
Reactive Aggression .51 .83
Total RPQ Score .79 .93 .86
Note. RPQ = Reactive Proactive Aggression Questionnaire. Main diagonal contains internal
consistency estimates (Cronbach’s 0). All correlations significant at p < .001.

76
Anxiety (MASC). As described previously, the MASC has been found to
have 4 main factors (Somatic Anxiety, Harm Avoidance, Social Anxiety and
Separation Anxiety/Panic; March et al. 1997). This four-factor model was tested in
the present data. Fit indices were satisfactory, 8
2
(106) = 270.82, p < .001, CFI = .87,
TLI = .94, and RMSEA = .08. All loadings were significant, all parameters were
within range and thus, the 4-factor solution was deemed an acceptable model of the
present MASC data. Factor loadings are presented in Table 13, factor
intercorrelations and internal consistency reliability estimates are presented in Table
14.
77
Table 13
Standardized factor loadings from the confirmatory factor analysis of the self-
report MASC
MASC Subscale
Item
Somatic
Anxiety
Harm
Avoidance
Social
Anxiety
Separation
Anxiety
1. Feel tense or uptight .52
5. Trouble getting breath .61
8. Shaky or jittery .80
12. Dizzy or faint .75
15. Jumpy .60
18. Chest pains .66
20. Feel strange, weird or unreal .69
24. Heart races, skips beats .72
27. Restless, on edge .70
31. Sick to stomach .78
35. Hands shake .78
38. Hands sweaty or cold .68
2. Ask permission .45
6. Keep eyes open for danger .65
11. Try to obey parents and teachers .48
13. Check things out .72
21. Try to do things others like .64
25. Stay away from upsetting things .54
28. Try to do exactly right .65
32. Let someone know if upset or scared .65
36. Check to make sure safe .76
3. Worry about others laughing .75
10. Afraid others will make fun .82
14. Worry about getting called on .58
16. Afraid people think stupid .85
22. Worry about what others think .80
29. Worry about doing exactly right .60
33. Nervous performing in public .60
37. Trouble asking kids to play .76
39. Feel shy .57
4. Scared when parents go away .72
7. Idea of camp scary .77
9. Try to stay near mom or dad .66
17. Keep light on at night .55
19. Avoid going places without family .54
23. Avoid watching scary movies and TV .39
26. Sleep next to family .45
30. Get scared in car or bus .72
34. Scared of weather, dark, heights, animals, bugs .49
Note. All loadings significant at p < .05. MASC = Multidimensional Anxiety Scale for
Children.
78
Cognitive Ability (WJR Subtests). Factor analysis was not performed on the
WJR subtests as each constitutes its own unidimensional measure. Subtest
intercorrelations and internal consistency reliability estimates are presented in Table
15.
Table 15
Subtest intercorrelations and internal consistency reliabilities for the WJR
subtests
WJR Subtest
Concept
Formation
Picture
Vocabulary
Memory for
Names
Numbers
Reversed
Concept Formation .93
Picture Vocabulary .49 .77
Memory for Names .32 .41 .88
Numbers Reversed .45 .33 .35 .84
Note. WJR = Woodcock Johnson-Revised. Main diagonal contains internal consistency
estimates (Cronbach’s 0). All correlations significant at p < .001.

Determining Covariates
To determine whether sex, maltreatment group or ethnicity should be
controlled in the following analyses, t-tests were run between these variables and the
aggression, cognitive and anxiety variables. Results are presented in Table 14. For
Table 14
Factor intercorrelations and internal consistency reliabilities for the MASC
MASC Subscale
Physical
Anxiety
Harm
Avoidance
Social
Anxiety
Separation
Anxiety
MASC
Total
Physical Anxiety .86
Harm Avoidance .28 .80
Social Anxiety .59 .39 .84
Separation Anxiety .57 .46 .55 .74
Total score .80 .67 .83 .80 .91
Note. MASC = Multidimensional Anxiety Scale for Children. Main diagonal contains
internal consistency estimates (Cronbach’s 0). All correlations significant at p < .001.

79
analyses including ethnicity, the sample was divided into White and non-White
groups (n = 26 and n = 188, respectively) for t-tests. If a difference was found
between subgroups (e.g. males and females) on a given outcome variable, the
grouping variable was entered as a covariate into the subsequent partial correlations
and regression analyses for that outcome variable.
WJR Cognitive Ability. Males performed significantly better than females on
Memory for Names total score, t(214) = 5.211, p < .001 and females performed
significantly better than males on the Concept Formation raw score, t(214) = -2.028,
p = .044. Comparison participants outperformed maltreated participants on all four
cognitive tests: Memory for Names, t(214) = 2.839, p = .005; Concept Formation:
t(214) = 4.806, p < .001; Picture Vocabulary: t(214) = 4.390, p < .001; and Numbers
Reversed: t(214) = 2.839, p = .005. Also, in comparison to non-white participants,
white participants performed more poorly on the Picture Vocabulary subtest, t(214)
= -2.957, p = .004.
RPQ Aggression. Males and females did not differ on RPQ aggression
subscales. Comparison participants scored lower than maltreated participants, and
non-white participants scored lower than white participants on the Reactive
Subscale, t(214) = -2.146, p = .033 and t (214) = 2.517, p = .012.
MASC Anxiety. Comparison and maltreatment groups did not show any mean
differences on the MASC anxiety subscales. Females consistently showed elevations
on the Separation Anxiety scale, t(214) = 2.860, p = .047, the Harm Avoidance scale,
t(214) = -2.351, p = .02, and the Social Anxiety scale, t(214) = -2.123, p = .035.
80
White participants did not differ from non-white participants on MASC subscale
scores, ts = -.535 to .396, ps = .593 to .896.
Correlational and Regression Analyses
Relationships between each APSD factor and Aggression, Cognitive Ability
and Anxiety were explored through correlational analyses. Partial correlations were
employed, controlling for sex, ethnicity or maltreatment group if deemed necessary
by the preceding t-tests. Further, to examine the relative contributions of the 3 APSD
factors in the prediction of each of the external correlates, all factors and relevant
covariates (ethnicity, sex or maltreatment group) were entered simultaneously into a
multiple regression analysis.
81

Table 16
Subsample mean scores and standard deviations on external correlates
MASC Subscale WJR Subtests RPQ Subscale
Group SOM HA SOC SEP MFN PV CF NR PRO REACT
Male
6.39
(6.49)
13.51
(5.46)
8.95
(6.04)
5.18
(4.39)
63.23
(7.97)
32.45
(2.99)
20.27
(8.28)
13.02
(3.92)
1.80
(2.65)
7.96
(4.60)
Female
7.25
(6.49)
15.46
a
(6.24)
10.83
a
(6.62)
7.10
a
(5.19)
57.21
(10.35)
31.91
a
(3.81)
22.26
a
(7.43)
13.04
(3.73)
1.45
(2.37)
8.00
(4.45)
Comparison
6.95
(6.27)
14.01
(5.50)
9.91
(5.64)
6.00
(4.68)
62.33
(8.94)
33.66
(3.40)
24.22
(7.42)
14.22
(4.19)
1.52
(2.74)
7.17
(4.57)
Maltreated
6.79
(6.65)
14.87
(6.28)
9.97
(6.88)
6.33
(5.07)
58.82
b
(9.98)
31.43
b
(3.20)
19.53
b
(7.67)
12.32
b
(3.39)
1.67
(2.35)
8.52
b
(4.41)
White
7.65
(4.89)
14.15
(6.02)
9.92
(5.20)
5.92
(4.06)
59.84
(9.66)
31.95
(3.21)
21.06
(7.71)
12.90
(3.68)
1.73
(2.72)
8.21
(4.63)
Non-White
7.10
(6.88)
14.43
(5.87)
10.15
(6.62)
6.48
(5.12)
62.59
(10.18)
34.47
c
(4.61)
23.33
(9.31)
14.15
(4.75)
1.70
(2.55)
5.89
c
(3.63)
Note. MASC = Multidimensional Anxiety Scale for Children; SOM = Somatic Anxiety; HA = Harm Avoidance; SOC = Social Anxiety; SEP
= Separation Anxiety. WJR = Woodcock Johnson-Revised; MFN = Memory for Names; PV = Picture Vocabulary; CF = Concept
Formation; NR = Numbers Reversed. RPQ = Reactive-Proactive Aggression Questionnaire; PRO = Proactive; REACT = Reactive.
a
Significant difference between males and females.
b
Significant difference between comparison and maltreatment groups.
c
Significant difference between white and non-white groups.
82
Aggression – RPQ. Correlations between caregiver, teacher and self-reported
APSD factor and RPQ factor scores are presented in Table 17. Results indicate that
overall, RPQ scores and APSD scores are significantly positively correlated.
Table 17
Correlations between APSD and RPQ factor scores.
APSD
Caregiver Teacher Child
RPQ Scale
F1
IMP
F2
NAR
F3
CU
Total
F1
IMP
F2
NAR
F3
CU
Total
F1
IMP
F2
NAR
F3
CU
Total
Proactive .26
***
.21
**
.15
*
.28
***
.23
**
.12 .21
**
.24
***
.57
***
.44
***
.29
***
.64
***
Reactive
a
.17
*
.22
**
.11
*
.22
**
.24
***
.24
**
.21
**
.29
*
.49
***
.44
***
.14
**
.53
***
a
Partial correlations controlling for ethnicity and maltreatment group. RPQ = Reactive-Proactive
Aggression Questionnaire.
*
p < .05,
**
p < .01,
***
p < .001.
Table 20 presents beta weight estimates and R
2
values (which indicate the
amount of variance in the outcome variable accounted for by the predictor variables),
results of the analyses regressing RPQ subscales, MASC subscales, and Cognitive
Ability subtest scores (outcome variables) and grouping variable covariates (if
necessary) on the APSD factors. When caregiver APSD scores were entered as
predictor variables, Proactive Aggression was not predicted by APSD. For teacher-
rated APSD scores, F1-IMP significantly predicted Proactive aggression after
controlling for F2-NAR and F3-CU. In contrast, all self-reported APSD scores
significantly and positively predicted Proactive aggression, accounting for 42.5% of
the variance. Caregiver F2-NAR predicted Reactive aggression after controlling for
ethnicity, maltreatment group and the other APSD scores. Despite significant partial
83
correlations, teacher-report APSD did not predict Reactive aggression after
controlling for Ethnicity and Maltreatment group and the other APSD factors.
However, self-reported F1-IMP and F2-NAR scores positively predicted Reactive
Aggression. The caregiver and child APSD score predictions provide evidence of
discriminant validity of APSD factors; in particular, the caregiver relationships
indicate that the F2-NAR is related to RPQ scores in cases where F1-IMP is not, and
the child relationships indicate that all three APSD factors are independently related
to RPQ scores. The variation in pattern among APSD informants indicates that
caregivers, teachers and children ascribe different relationships between RPQ
aggression and Impulsive, Narcissistic, and Callous/Unemotional traits. Not
surprisingly, total score APSD is positively correlated with both RPQ subscales in all
informant datasets, indicating an overall relationship between the APSD factors and
aggression.
Anxiety – MASC. Correlations between MASC and APSD factor and total
scores are presented in Table 18. Caregiver APSD scores were essentially unrelated
to MASC subscale scores, with the exception of a negative relationship between F3-
CU and the Harm Avoidance scale. Teacher APSD scores also showed few
relationships with MASC scores, though several correlations were marginally
significant. In particular, negative relationships appeared emergent between the
Harm Avoidance subscale and F1-IMP and F2-NAR scores, and the Social Anxiety
and F2-NAR factor scores. Self-reported APSD scores showed some significant
correlations with the MASC. In particular, F1-IMP showed significant positive
84
relationships to Somatic Anxiety, while F3-CU was significantly negatively
related to Harm Avoidance and Social Anxiety score. Regression analyses indicated
that Somatic anxiety was predicted by teacher-rated F1-IMP and F2-NAR, and self-
reported F1-IMP (see Table 20). In line with hypotheses, Harm Avoidance was
independently negatively predicted by caregiver- and self-reported F3-CU. Social
Anxiety was negatively predicted by teacher-rated F2-NAR, and self-reported F3-
CU. Separation anxiety was only predicted by self-reported F2-NAR.
For the most part, caregiver-rated APSD was unrelated to MASC anxiety
scores except for the expected relationship between Harm Avoidance and F3-CU.
Teacher-report of Narcissistic traits correlated negatively with some Anxiety
subscales, and self-rated Callous/Unemotional traits were negatively related to
MASC anxiety. Total APSD also reflected the hypothesized relationship between
F3-CU and Harm Avoidance, which seems to be fairly robust across the informants.

Table 18
Correlations between APSD and MASC subscale scores
APSD
Caregiver Teacher Child
MASC Scale
F1
IMP
F2
NAR
F3
CU
Total
F1
IMP
F2
NAR
F3
CU
Total
F1
IMP
F2
NAR
F3
CU
Total
Somatic
Anxiety
.06 .04 -.08
*
.01 .08 -.07
*
.08 .04 .29
*
.08 -.04
**
.18
*
Harm
Avoidance
a
-.02 -.01 -.14
*
-.08 -.13
†
-.11
†
-.07
*
-.13
†
-.09
**
-.05
*
-.41
***
-.26
***
Social
a
.04 .06 -.07
*
.07 -.02
*
-.13
†
-.02
*
-.05 .10
*
.03 -.16
**
.00
Separation
a
.04 -.02 -.09
*
-.03 .01 -.05
*
-.04
*
-.03 .04
*
.20 -.07
**
.08
a
Partial correlations controlling for gender. MASC = Multidimensional Anxiety Scale for
Children
*
p < .05,
**
p < .01,
***
p < .001,
†
p < .10
85
Cognitive Ability – WJR Subtests
Correlations between WJR Cognitive subtests and APSD scores are presented
in Table 19. Overall, results indicate that APSD is not correlated with cognitive
ability (though significant relationships were generally negative). For caregiver-rated
APSD scores, significant, negative relationships were found between Picture
Vocabulary and F2-NAR scores, as well as Numbers Reversed and F1-IMP scores.
Teacher-report data indicated significant negative correlations between F2-NAR, F3-
CU, Total APSD score and Picture Vocabulary. Consistent with the caregiver- and
teacher-rated data, self-report F3-CU was significantly negatively related to Picture
Vocabulary. Contrary to hypotheses, self-report F1-IMP also correlated positively
with Concept Formation. F1-IMP and Numbers Reversed appeared to be related
across informants, with a significant relationship in the caregiver-report data, and
marginally significant relationships emerging in the teacher and self-report data. The
only significant relationship for total score APSD was a negative one with Picture
Vocabulary, which mirrors the similar relationship involving the F3-CU factor and
Picture Vocabulary.
Regression analyses are fairly consistent with correlational results (see Table
20). Interestingly, Numbers Reversed (a measure of short-term retrieval) negatively
predicts caregiver F1-IMP but positively predicts caregiver F2-NAR. As with the
correlation analyses, the negative relationship between F3-CU and Picture
Vocabulary was consistent across all three informants.

86
Table 19
Correlations between APSD factor and total scores and Cognitive Abilities
APSD
Caregiver Teacher Child
Cognitive
Ability
F1
IMP
F2
NAR
F3
CU
Total
F1
IMP
F2
NAR
F3
CU
Total
F1
IMP
F2
NAR
F3
CU
Total
Concept
Formation
ab

-.07 -.02 -.11 -.08 -.04 -.07 -.06 -.07 .14* .07 -.01 .10
Picture
Vocabulary
abc
.01 .06 -.25
**
-.09 -.12 -.17
*
-.28
**
-.23
**
.06 .04 -.18* -.03
Numbers
Reversed
b
-.14
*
.02 -.01 -.07 -.12
†
-.08 -.07 -.11 .13† -.03 .02 .06
Memory for
Names
b
-.00 -.01 -.00 -.00 -.07 -.14
†
.01 -.08 .04 .02 .08 .07
Note. F1-IMP = Factor 1 Impulsivity, F2-NAR = Factor 2 Narcissistic, F3-CU = Factor 3
Callous-Unemotional.
a
Partial correlation controlling for gender
b
Partial correlation controlling for maltreatment group
c
Partial correlation controlling for ethnic group
*
p < .05,
**
p < .01,
†
p < .10

Summary of Analyses of Relationships Between the APSD and External Correlates
In sum, correlational analyses showed that the APSD factors were generally
positively related to reactive and proactive aggression. Most significant correlations
between APSD and MASC anxiety subscales were negative, except for the
hypothesized positive relationship with Somatic anxiety. Most of the correlations
with cognitive subtests involved negative relationships between F3-CU and Picture
Vocabulary score. Overall, there were relatively few relationships, especially after
accounting for sex, maltreatment group status and ethnicity. Those relationships that
did emerge, however, were for the most part in line with hypotheses.
87
Table 20
Beta coefficients and R
2
statistics from regression analyses regressing RPQ aggression, MASC anxiety, and WRJ
Cognitive Subtests on APSD factors and grouping variables
Criterion Variable
RPQ MASC WJR
Predictor Variable Proactive
Aggression
Reactive
Aggression
Somatic
Anxiety
Harm
Avoidance
Social
Anxiety
Separation
Anxiety
Concept
Formation
Picture
Vocabulary
Numbers
Reversed
Memory
for Names
Ethnicity -
aaa
-.157
*a*a
-
aaa
-
aaa
-
aaa
-
aaa
-
aaa
.194
aaa
-
aaa
-
aaa
Maltreatment Group -
aaa
.078
aaa
-
aaa
-
aaa
-
aaa
-
aaa
-.245
** aa
-.273
aaa*
-.220
****
-.175
* aa
Gender -
aaa
-
aaa
-
aaa
.150
*aa
.137
†**
.189
* aa
.091
aaa
-.151
aaa*
-
aaa
-
aaa
F1-IMP .194
*aa
.032
aaa
.088
aaa
.013
aaa
.026
aaa
.105
aaa
-.094
aaa*
.013
aaa
-.252
** aa
-.016
aaa*
F2-NAR .076
aaa
.192
* aa
.003
aaa
.004
aaa
.054
aaa
-.063
aaa*
.082
aaa
.067
aaa
.173
* a
.018
aaa
F3-CU .105
aaa
.077
aaa
-.103
aaa*
-.145
*aa*
-.085
aaa
-.106
aaa*
-.097
aaa*
-.241
** a
.019
aaa
.005
aaa
Caregiver-Report
R
2
.079
**a
.104
***
.014
aaa
.046
* a*
.031
aaa
.053
* aa
.097
** a
.208
***
.092
** a
.030
aaa
Ethnicity -
aaa
-.154
* a*
-
aaa
-
aaa
-
aaa
-
aaa
-
aaa
.154
† a
-
aaa
-
aaa
Maltreatment Group -
aaa
.078
aaa
-
aaa
-
aaa
-
aaa
-
aaa
-.255
****
-.236
** a
-.210
** a
-.163
* a*
Gender -
aaa
-
aaa
-
aaa
.152
* a
.154
* aa
.190
* aa
.093
aaa
-.170
* a*
-
aaa
-
aaa
F1-IMP .210
***
.106
aaa
.189
*aa
-.091
aaa
.087
aaa
.084
aaa
.022
aaa
.084
aaa
-.100
aaa*
.001
aa
F2-NAR -.060
***
.132
aaa
-.215
* aa
-.048
aaa
-.200
* aa
-.088
aaa*
-.062
aaa*
-.136
aaa
-.003
aaa*
-.146
† aa
F3-CU .144
†**
.120
†aaa
.073
aaa
-.016
aaa
.046
aaa
-.049
aaa*
-.054
aaa*
-.264
* aa
-.027
aaa*
.052
aa
Teacher-Report
R
2
.072
***
.129
***
.036
† aa
.045
† a
.046
* aa
.045
† aa
.086
** a
.227
***
.067
** a
.092
** a
Ethnicity -
aaa
-.163
**
-
aaa
-
aaa
-
aaa
-
aaa
-
aaa
.066
aaa
-
aaa
-
aaa
Maltreatment Group -
aaa
.183
** a
-
aaa
-
aaa
-
aaa
-
aaa
-.265
***a
-.142
* aaa
-.231
** a
-.172
* a a
Gender -
aaa
-
aaa
-
aaa
.045
aaa
.105
aaa
.210
** a
.107
aaa
-.348
****
-
aaa
-
aaa
F1-IMP .451
***
.365
***
.309
***
-.044
aa*a
.121
† aa
-.025
aaa*
.128
† a*
.039
aaa
.155
* a
.033
aaa
F2-NAR .249
***
.289
***
-.016
aaa
.000
aaa
.003
aaa
.214
** a
.023
aaa
-.060
aaa?
-.092
aa*
-.005
aa**
F3-CU .211
***
.058
aaa
-.079
aaa
-.414
****
-.177
* aa
-.086
aaa*
-.030
aaa*
-.012
aaa*
.013
aaa
.075
aaa Self-Report
R
2
.425
***
.355
***
.188
***
.190
***
.061
* a
.083
** a
.099
** a
.153
***
.076
** a
.037
† a a
Note. RPQ = Reactive Proactive Aggression Questionnaire. MASC = Multidimensional Anxiety Scale for Children. WJR = Woodcock Johnson
III-Revised Cognitive Ability subtests.
*
p < .05,
**
p < .01,
***
p < .001,
†
p < .10.
88

CHAPTER 6: DISCUSSION
This study examined the psychometric properties of the Antisocial
Screening Process Device (Frick & Hare, 2001) in a diverse sample of
adolescents that included males and females, maltreated and comparison
individuals, and African American, Latino, White, and biracial ethnicities. The
main objective was to evaluate the construct validity of the trait factors of the
APSD as rated by caregivers, teachers and the adolescents themselves. Multiple-
informant assessment of psychopathy is a significant methodological departure
from previous measurement approaches. This novelty, in addition to the relative
infancy of research and application in the area of juvenile psychopathy and the
serious implications of such endeavors, requires that measures such as the APSD
receive much attention and scrutiny. In examining caregiver, teacher and self-
report APSD in an MTMM framework, this study aimed to ascertain the value of
collecting information from various informants, and to evaluate whether those
informants vary in their ability to reliably and validly assess different aspects of
psychopathic personality.
Prior to these multitrait-multimethod analyses, restricted structural factor
analysis and invariance procedures were conducted. These analyses were
essentially undertaken to delineate the factor structure of the APSD, but with the
ultimate goal of establishing a version of the measure that was universal and thus
comparable across informants which would enter into the subsequent MTMM
analyses. Results from the factor analysis led to a structure that much resembled
89

that of the initial 3-factor model found by Frick et al. (2000), although with fewer
items per factor and a high correlation between two of the factors. Subsequently,
the multitrait-multimethod analyses (using both correlational and latent variable
analyses) found evidence for convergent validity, limited evidence for
discriminant validity, and significant method effects, in particular for teacher-
report data. Finally, investigations of the relationship between the APSD and
external correlates provided evidence for convergent validity and some
discriminant validity between the trait factors, and presented evidence against
treating caregiver, teacher and self-report information as interchangeable.
Factor Structure and Invariance
The results of the structural factor analysis yielded a 15-item 3-factor
structure which provided a better fit to the data than a 2-factor structure, though
the correlation between two of factors (those with impulsivity and narcissism
items) was very high in the caregiver and teacher-report data. The three factors
reported by Frick et al. (2000) were intercorrelated more modestly than those
found here, but other researchers have found a similarly strong relationship to that
found here for the Impulsivity and Narcissism factors (e.g. Vitacco et al., 2003).
As a result of the tests for invariance, the number of items decreased to 10 due to
the policy of dropping items which showed noninvariance across gender,
maltreatment or ethnic group. Items 2, 6, 14 and 17 were noninvariant between
the maltreated and comparison groups in the caregiver-report data, and item 7
showed noninvariance between the maltreatment and comparison groups in the
90

teacher-report data. Interestingly, Items 2 (Engages in illegal activities) and 6
(Lies easily and skillfully) were excluded from Frick et al.’s (2000) original 3F
solution because they did not load onto any of the factors in their data. In the
present study, caregiver-reported Item 2 had a very nonnormal distribution in the
comparison group (positive skew and leptokurtosis), but less nonnormality in the
maltreated group. Though there was no mean-level difference, caregivers of
maltreated children may be somewhat more willing to report their child’s illegal
activity (or may presume more negative behavior in their child) than comparison
caregivers. As mentioned previously, Items 6 (Lies easily and skillfully) and 14
(Can be charming in ways that seem superficial or insincere) are probably
difficult to report on accurately in general (e.g. if an adolescent is a skillful liar,
caregivers may not be aware that they are being lied to) and as such, sensitive to
any differences between groups.
It is difficult to determine why these noninvariant differences might have
arisen for caregiver and teacher ratings between the maltreated and comparison
groups, and it should be kept in mind that the sample size was fairly small so
these findings could be capitalizing on chance within this sample. In this study,
noninvariance could mean that the groups of adolescents themselves are different,
or that the raters of each group (caregivers or teachers in this case) rated them
differently. Thus for example, one possibility for Item 6 (Lies) could possibly be
that lying was more strongly indicative of a maltreated adolescents’ impulsive
behavior than was the case in comparison adolescents. On the other hand (or in
91

addition), caregivers of maltreated youth may view their child’s lying as
impulsive behavior moreso than comparison caregivers, and may have been more
likely to report it. The converse may be true for Items 14 and 17. Regarding Item
14, superficial charm may be a greater indicator of narcissism in comparison
youth than in maltreated youth, or the caregivers of either group might presume
this, or both explanations may be true as well. For Item 17, comparison
adolescents may simply exhibit more impulsivity through a greater lack of
planning than their maltreated counterparts. Another possible explanation may be
that comparison caregivers have a higher level of organizational skill than
maltreated caregivers, and thus any irresponsibility on their child’s part provokes
a stronger reaction in comparison caregivers. In the maltreated group, parents may
be less demanding of planfulness in their adolescents given a lower amount of
organization in their own lives.
3
Once these noninvariant items were identified, it
was prudent to exclude them from the measure, rather than attempting to use
partial invariance procedures to retain them. Allowing partial invariance of these
items would have involved permitting the noninvariant items to remain in the
measure, but allowing them to vary freely between maltreatment and comparison
groups. Since the multitrait-multimethod analyses were not performed using

3
Mennen & Trickett (2006) have shown some evidence of a difference in level of organization
(using the organization/control subscale of the Family Environment Scale; Moos & Moos, 1981)
between parents at the first wave of date collection of the present study. However, they found a
lower level of family organization/control in biological maltreating mothers vs. other caretakers
(foster mothers of maltreated children, relatives of maltreated children and biological mothers of
control children) rather than a difference between maltreated and comparison families as is
conjectured here.
92

multiple group procedures, but instead required use of the full sample collapsed
across groups (in large part to maximize the sample size), partial invariance was
not considered. In addition, it was most desirable to establish a ‘pure’ measure
which was applicable to all subsample groups and informants for multitrait-
multimethod analysis; allowing items to perform differently across groups would
have compromised the direct comparison between informants and complicated
interpretation of the results.
The result of the factor analysis and invariance analyses was a 10-item
measure with 3 factors, F1-Impulsivity (4 items), F2-Narcissism (3 items), and
F3-Callous/Unemotional (3 items). Though justification for the exclusion of items
has been provided (e.g. to facilitate direct comparability), it is acknowledged that
the decrease in scale volume is not a small matter and has implications for
generalizability and reliability. However, consider that previous research has been
conducted with versions that did not contain all 20 items (e.g. even the original
Frick et al., 2000 article reported on a three-factor measure that excluded two
items). In the present study, the exclusion of some items that have performed
poorly and have been eliminated from past studies (e.g. 2, 6, 19, and 20) may in
fact bring the measure used here more in line with what has been used previously.
Ultimately, the items that were retained formed three factors that essentially
mirror those that have been used in previous studies, but using a subset of the
items. F1-IMP comprised items that imply thoughtless or impulsive action (not
considering consequences and doing risky or dangerous activities), and a lack of
93

responsibility or accountability for ones actions (blaming others for one’s own
mistakes). The lone diversion from previous studies appeared here, in that Item 5
(Emotions seem shallow) loaded unequivocally on this factor. While at first
glance, this item might seem incongruent with the construct of impulsivity, a
possible explanation might be that the appearance of shallow emotions is
interpreted as an apparent lack of care, investment or consideration given to one’s
behavior, and thus a greater tendency toward impulsive acts. F2-NAR was
defined by items which address the interpersonal aspects of psychopathic
personality, including the features of grandiosity (bragging, thinking of oneself as
more important than others) and malevolence (teasing others) that are typical. F3-
CU comprised reversed items which assess a level of engagement (concern about
schoolwork), involvement with others (concern about others’ feelings), and
culpability (feels bad or guilty).
Reliability
The results of the first set of multitrait-multimethod analyses revealed that
there were several areas of weakness for the version of the APSD used here. First
and foremost, some of the factors showed rather low internal consistency
reliabilities. With the exception of caregiver- and teacher-rated F1-IMP, the
remaining reliabilities were lower than the recommended value of .70 (Cicchetti,
1994; Nunnally & Bernstein, 1994). The relatively respectable reliabilities of F1-
IMP are not surprising, as caregivers and teachers tend to more accurately report
on externalizing behaviors such as impulsivity (Achenbach, et al. 1987;
94

Youngstrom, Loeber, & Stouthamer Loeber, 2000). The Narcissistic and
Callous/Unemotional traits contain items that require more of an inference about
the child’s internal state, and thus lower reliabilities might be expected of other-
informants. In the present study, self-report F3-IMP reliability was higher than
caregiver-report F2-NAR reliability. It should be acknowledged, however, that the
F1-IMP factor had one more item than the other two factors, which constitutes an
advantage in the computation of reliability in comparison to the other two factors
which only have 3 items each. The fact that reliability was generally low for self-
ratings and lowest for F3-CU in particular is notable but was not truly unexpected.
Poythress et al. (2006) reviewed the internal consistency reliability of the self-
report APSD across 11 studies and found similarly low 0s across studies when
examining the full 6-item F3-CU scale (items 3, 7, 12, 18, 19 and 20); median
reliability of the F3-CU across studies was .46, range .22 to .60. However, on the
basis of mean corrected item-to-total correlations, Poythress et al. (2006)
recalculated 0s with the exclusion of items 19 and 20 from the F3-CU scale,
which resulted in a large improvement of reliabilities. In the present study, items
19 and 20 had already been excluded during the preliminary EFA, and thus
reliability of F3-CU (0 = .39) was comparatively low. The reliability of a F3-CU
scale including items 19 and 20 was not computed in this study, and thus it is
unknown whether the reliability of the 3-item scale used here represents a relative
improvement over an even lower estimate that might result from a factor which
includes more (but poor-performing) items. It should be noted that the reliability
95

of the self-report F1-NAR factor was also lower than the mean estimate computed
by Poythress et al. (although theirs was calculated on the original 7-item scale). In
contrast, the self-report 4-item F2-IMP factor in this study yielded an estimate
which was in fact slightly higher than the mean 0 reported by Poythress et al. (.55
vs. .53) on the full 5-item scale.
The lower reliabilities seen in the present study (in all three reporter
datasets) are likely a function of the truncated scales used in their calculation.
Cronbach’s 0 is strongly influenced by number of items in a scale, and thus the
exclusion of items at the structural factor analysis and invariance steps may have
served as a penalty against scale reliability. This was most probably the main
reason for lower reliabilities in the caregiver and teacher data, and thus the writing
of additional items which tap these traits is recommended. Regarding the self-
report data, it seems prudent to consider more substantive threats to the reliability
of these scales. In general, the transparency of the items (e.g. I use or con others
to get things I want) may evoke a defensive response from some adolescents who
are in the midst of establishing their identity and may have heightened awareness
of self-presentation issues. Additionally, as has been previously suggested (Lee et
al. 2003), youth may simply have difficulty interpreting some of the items of a
more abstract nature (e.g. My emotions are shallow or fake, I hide my emotions or
feelings from others). Adolescents may interpret items in different ways than
caregivers or teachers might. For instance, they may answer Item 20 (I keep the
same friends) in terms of frequency of friends retained between grades, or at
96

summer camp vs. school, whereas caregivers and teachers may see the question as
more of a judgment of social skills and adaptability. Though adolescents and
caregivers may have a greater shared knowledge of the adolescent’s behavior in
various domains, it is likely that caregivers and teachers, given their closer
similarity in development and experience, would make more similar
interpretations of item meanings. The results and implications of the self-report
data will be discussed in more detail shortly.
Low factor score reliabilities (and also differences in levels of reliability
across measures) may also influence the accuracy of conclusions drawn from the
multitrait-multimethod matrix. For example, if a trait is measured more reliably
by one method than another, correlations among traits assessed with the more
reliable method will be higher, which may be interpreted as a method effect
(Marsh & Hocevar, 1983). Thus when reliabilities are low, interpretation of the
matrix should be made cautiously (Campbell & Fiske, 1959; Marsh & Hocevar,
1983).
Convergent Validity
Convergent validity is a major consideration in determining the validity of
a construct or measure. Measures designed to assess theoretically related traits
should be related to each other (indicating common factor trait variance), and a
given measure should perform similarly across various samples and studies
(indicating that it is measuring true score variance). The results of the factor
analyses indicated a 3-factor structure of the APSD here that for the most part
97

resembled the APSD used in previous studies. Further, analysis of the multitrait-
multimethod matrix provided evidence for convergent validity of the three traits
of the APSD used here. In the matrix, correlations between different informants
on the same factors were all significant, showing that different informants agreed
on a particular trait to a significant degree. Though these values were modest,
they were in fact very much within the range of correlations that many previous
studies have found for caregiver/teacher, self/teacher and self/caregiver agreement
(e.g. Achenbach et al., 1987; Gagnon, Vitaro, & Tremblay, 2000). Thus, the
convergent validity values here are consistent with previous literature showing
modest correlation of ratings on juvenile behavior problems across informants.
Inasmuch as those studies used reliable, valid measures of internalizing and
externalizing behavior problems, the results in the multitrait-multimethod matrix
of the present study present evidence for levels of convergent validity consistent
with the extant literature. From a general point of view, such low agreement
between informants might seem inauspicious given that informants are all
reporting on the same target individual. However, though personality is fairly
stable over time (Rutter, 2005), it can vary substantially in different domains,
such as at school vs. at home, and around others vs. alone. So to the extent that
various informants have limited interaction with the individual in limited
domains, ratings may vary widely. In addition, each informant brings their own
knowledge and biases to the evaluative situation. In many cases, such breadth is
considered to be an advantage, allowing for diverse observations that provide a
98

more comprehensive view of the individual. If informants’ reports are considered
separately, and incongruent information is considered incremental, there is no
need for alarm. However, the danger lies in situations where ratings from multiple
informants are combined or used interchangeably (as has historically been the
case with the APSD). In such cases it is important to better understand the
contributions made by each informant, and to ensure that there is acceptable
convergent validity.
The results of the general and correlated uniquenesses confirmatory factor
analyses (CFA) also provide evidence for the convergent validity of the three
factors in this version of the APSD. Whereas evaluation of the matrix involves
making inferences about underlying traits based on the inspection of correlations
between observed variables, CFA allows a priori specification of latent trait and
method factors. CFA methods are thus an important extension of the multitrait-
multimethod matrix and allow for a more quantitative approach to the guidelines
suggested by Campbell and Fiske. The nested model comparisons indicated that a
substantial amount of trait variance existed in the data that cannot be accounted
for solely by the methods (or correlated uniquenesses), and that APSD trait
factors were correlated (and not orthogonal).
At the parameter level, significant trait factor loadings indicated that the
indicators appear significantly related to the latent traits of Impulsivity,
Narcissisism and Callous/Unemotionality, although the strength of those
relationships was conservative. Even so, the correlated uniquenesses approach
99

does not posit method factors, and thus variance that may be common to different
methods may be relegated to trait factor loadings (upwardly biasing convergent
validity).
Overall, the results of the multitrait-multimethod analyses suggest that the
three factors of the APSD used in the present study possess a moderate amount of
convergent validity. The trait factors appeared to index a construct that accounts
for a significant amount of variance in the data, with items assessing impulsivity,
narcissism and callousness. In this way, the measure used here resembles the
APSD that has been used fairly successfully in other studies. However, whether
the trait factors were at least partially independent was a question of discriminant
validity.
Discriminant Validity – Multitrait-Multimethod analyses
In general, the three APSD trait factors lacked definitive evidence of
discriminant validity, but were also not completely indistinct. In the multitrait-
multimethod matrix, correlations between the F1-IMP and F2-NAR factors rated
by the same informant were consistently higher than correlations between
monotrait-heteromethod values (e.g. caregiver-rated F1-IMP and teacher-rated
F1-IMP). This reflected the high correlation between these two factors that
emerged after the structural factor analysis. However, given that these are
monomethod values (the informant is the common element that enters into the
correlation), this finding could also be interpreted as method effect (which shall
be discussed forthwith). The second guideline for discriminant validity in the
100

matrix was upheld, in that heterotrait heteromethod values, those that have the
least commonality in the whole matrix, were the most weakly correlated.
Importantly, even F1-IMP and F2-NAR showed low correlations when rated by
different informants.
The results of the latent variable model approaches offered additional
evidence for discriminant validity. While discriminant validity in this study is
defined as differentiation between traits and indicated by low correlations
between factors, it is impractical and in fact atheoretical to expect nonsignificant
or near-zero correlations between the three APSD factors. These factors are
believed to be associated and, taken together, to offer some measurement of
psychopathic features. Thus in reality, moderate correlations between them should
be expected. In the general CFA models, neither one general trait factor nor three
orthogonal traits fit the data as well as three oblique traits, implying that the latent
trait was multidimensional though the dimensions were not independent. In
addition, the correlated uniquenesses CFA models allowed for a further test of
discriminant validity by comparing the 3F full model to a 2F model that reflected
the high correlation between F1-IMP and F2-NAR by collapsing them into one
factor. The 2F model demonstrated significantly worse fit than that of the 3F
model, despite the high correlation between F1-IMP and F2-NAR. However it
seems clear that a significant portion of the correlation between traits seen
initially is due to method effects, which will be discussed more fully in the
following section. For now, it should be reiterated that, while the full correlated
101

uniquenesses model is able to account for the within-informant method variance,
it is unable to address any variance that may arise between methods. That methods
themselves were correlated was implied by one of the model comparisons in the
general CFA approach. Thus, the displacement of method variance resulting from
correlations between methods may be contributing to some elevation in the trait
intercorrelations as well.
Method Effects
As mentioned above, method effects could in part be inferred from the
inflated correlations among the monomethod values in the multitrait-multimethod
matrix. In particular, correlations among teacher-rated APSD factors were all
highly significant. All caregiver-reported APSD factor intercorrelations were also
significant, although they were somewhat lower in magnitude than the teacher-
rated factors. Method effects in the matrix were less evident for self-reported
APSD factors, as only the correlation between F1-IMP and F2-NAR was
significant. Of particular interest from the preceding discussion of discriminant
validity, is that the correlation between F1-IMP and F2-NAR was known to be
high, especially in the caregiver- and teacher-ratings. As was discussed
previously, one of the limitations of the Campbell and Fiske matrix approach is
that it is difficult to determine whether these inflated monomethod correlations
are representative of method variance, trait covariance, or some combination of
the two.
102

The latent variable analyses provided further investigations of method
effects. Foremost, the full correlated uniquenesses model (with correlated
uniquenesses among indicators measured by the same informant) was superior to
the model which posited uncorrelated uniquenesses. The parameter analysis of the
full correlated uniquenesses model reiterated and expanded this finding; all
teacher-reported APSD factors shared method and unique variance whereas for
both caregiver- and self-report, method and unique variance was only common
across F1-IMP and F2-NAR. Here especially, CFA offers an improvement over
the multitrait-multimethod matrix, because in the correlated uniquenesses models,
trait intercorrelations are modeled and thus shared trait variance is separated from
method variance. Therefore it can be retrospectively concluded that the inflated
monomethod correlations between F1-IMP and F2-NAR in the MTMM matrix
across all three informants (as well as the other trait intercorrelations in the
teacher-reported data) were evidence of both high trait correlations and shared
method variance.
Calculations of proportions of method variance and trait variance in the
correlated uniqueness model identified caregiver ratings as the most effective (in
terms of trait variance to method variance ratio), though still displaying a
substantial amount of method and error variance. Self ratings accounted for more
trait variance than method variance, but included a large proportion of unique
error variance. Teacher ratings had less unique variance (higher proportions of
103

trait and method variance), but the proportion of method variance in their
measures was greater than the proportion of trait variance.
Method effects are thus strongest for teacher-reports of the APSD factors.
Several possible explanations might be posited for this result. One is that the
setting in which the measurement was taken was similar for caregivers and
adolescents, but quite different for teachers. Caregivers and adolescents
completed the APSD questionnaire during a visit to our on-campus offices. The
APSD was one paper and pencil measure embedded in several others, and thus
there may have been some priming effects, that is, respondents may have
achieved a certain mindset as a result of having to consider features of the
adolescent’s personality on multiple occasions. Specifically in terms of reduced
method variance, over the course of the testing session caregivers and adolescents
may have become ‘trained’ in considering different dimensions of personality or
behavior and thus had a heightened ability to differentiate between such
dimensions. In contrast, teachers received the questionnaire via mail and filled it
out at their convenience, likely at a time that was less hectic than most in their
classroom. Though the APSD was embedded in a questionnaire packet that
contained a handful of other measures, the entire packet only required
approximately 15 minutes to fill out, and thus was likely completed under very
different conditions than the controlled, multi-hour session attended by caregivers
and adolescents. Even without distraction, the comparatively small amount of
time required for the questionnaire may have meant that teachers had less
104

‘training’ and were unable to achieve the same mindset as caregivers and
adolescents, and thus they may have been less successful in differentiating
between personality dimensions. Another possible explanation for the increased
method variance for teacher ratings over caregiver and adolescent ratings is that
teachers observe and interact with the subject in a more limited domain than
caregivers and the adolescents themselves. Generally, teachers only see the
adolescent at school in the classroom interacting with their peers, whereas
caregivers are with them at home and elsewhere while they are at work, at play,
interacting with friends and family. Thus in comparison to teachers, caregivers
and adolescents have a greater wealth of knowledge and experience to draw upon
when making judgments about the adolescent’s personality which may allow for
greater differentiation of traits. Finally, all informants were asked to focus on one
target individual, which for the adolescent themselves is of course a special case.
For the adult informants, the pool from which the target adolescent was drawn is
much larger in general for teachers than for caregivers, as teachers have a
classroom full of students of similar age (and to an extent, behavior), as well as
previous classes to draw upon. Caregivers may have other children, but certainly
not as many on average as teachers have students, and interactions with their
child’s peers are likely not as extensive as a teacher’s interactions. Thus it may be
more difficult for teachers to even differentiate between adolescents, much less
different aspects of one adolescent’s personality.

105

Discriminant Validity – External Correlates
The third specific aim of this study was undertaken to further examine the
construct validity of the APSD measure delineated here by examining its
relationship with other relevant variables. Preliminary tests of mean differences
were performed to first evaluate whether the external variables were related to
grouping variables relevant to this sample (gender, maltreatment, and ethnic
group). Cognitive ability tests were most diversely related to these grouping
variables; males scored lower than females on Concept Formation but higher on
Picture Vocabulary, maltreated participants scored lower than comparison
participants on all four subtests (Memory for Names, Concept Formation Picture
Vocabulary and Numbers Reversed), which is a common finding in the literature
(Crozier, & Barth, 2005; Veltman & Browne, 2001 noninvariant) and non-white
participants scored higher than white participants on Picture Vocabulary. Group
differences on MASC Anxiety were limited to males and females; females scored
higher than males on the Harm Avoidance, Social and Separation anxiety
subscales, (also a common finding in the literature, see e.g. Dierker et al., 2001).
Of the RPQ Aggression scales, only the Reactive Aggression subscale showed
differences across groups; maltreated adolescents had a slightly higher score than
nonmaltreated adolescents, while white participants scored higher than non-white
participants. Unexpectedly, neither RPQ aggression subscale differed between
males and females. Given the extensive history of robust findings of increased
aggression in adolescent males (Maccoby & Jacklin, 1980) and that differences
106

between sexes have been found using different measures of reactive and proactive
aggression (e.g. Connor, Steingard, Anderson, & Melloni, 2003), it is surprising
that there is no such elevation for males in this sample. Reactive scale scores for
males and females were similar to those reported by Raine et al. (2006), while
proactive scores were slightly lower. The RPQ was developed in a sample of boys
at high-risk for antisocial behavior, and the findings reported here call attention to
its generalizability and accuracy in females.
As expected, the APSD factors in the total sample were significantly
correlated with reactive, proactive and overall aggression levels. In general,
correlations between self-reported APSD traits and aggression were larger than
correlations with teacher-reported APSD, which were larger than correlations
with caregiver-reported APSD. In adolescents, this strong correlation between
aggression and APSD traits may indicate that they were more likely to respond to
the APSD items as though they were tapping external behaviors, which is
reflected in their more reliable ratings of F1-IMP, the APSD trait that is manifest
the most externally. Alternatively, the high correlations may be another example
of shared method variance, since both the APSD and RPQ, which purportedly
measure two closely related constructs, were self-report measures. The results of
the regression analyses indicate that self-reported APSD trait factors were
independently related to aggression scores, with the exception of F3-CU and
Reactive Aggression. Congruent with hypotheses, the Callous-Unemotional factor
did not predict Reactive aggression regardless of informant, though it did predict
107

Proactive regression when rated by adolescents, and nearly so for teachers (p =
.054). Contrary to hypotheses, the Impulsive APSD factor predicted Proactive
aggression across informants. A closer look at the F1-IMP factor used here
reveals that at least 2 of the 4 items could be substantively related to proactive
aggression (Item 1 – Blames others and Item 5 – Emotions seem shallow)
although theoretically, impulsivity itself should not correlate positively with
Proactive aggression. The consistent relationship between F1-IMP and Proactive
aggression found here could be driven by these items. In other words, perhaps the
F1-IMP factor used here is not as strong a measure of impulsivity in previous
versions, which included the Items 9 (Gets bored easily), and 17 (Does not plan
ahead). Nevertheless, the APSD factors showed different predictive patterns with
regard to the aggression subscales, with F1-IMP and F3-CU being somewhat
more predictive of Proactive aggression and F2-NAR being more (though not
significantly) predictive of Reactive aggression. All of the relationships between
Teacher-rated APSD and Reactive aggression were attenuated after controlling
for maltreatment and ethnicity, suggesting that APSD traits did not offer any
incremental explanation for reactive aggressive behavior. Lastly, these results
indicate that caregiver, teacher and self-ratings of APSD yield widely varying
relationships with RPQ aggression scales. There is little overlap in the pattern of
predictive relationships across informants, cautioning against using APSD
informants interchangeably when examining relationships between psychopathic
traits and aggressive behaviors.
108

Few correlational relationships between the APSD and MASC anxiety
scores emerged. As with RPQ Aggression scores, the majority of relationships
arose for self-reported APSD whereas no significant relationships were found in
the teacher-report data (though several were marginally significant), and only F3-
CU and Harm Avoidance were negatively related in the caregiver-rated data.
Some relationships were positive, i.e. F1-IMP and Somatic anxiety, as well as F2-
NAR and Separation anxiety in the self-report data. The latter relationship is of
interest, as it implies that the same adolescents who admit that they consider
themselves more important than others, tease others and brag about themselves
also admit to being frightened when they are not near family members and trying
to stay near their parents. The positive relationship involving Somatic Anxiety is
in line with hypotheses and is congruent with some previous studies which have
found elevated levels of physical symptoms of anxiety in psychopathic adults (af
Klinteberg, et al., 1992; Fowles & Missel, 1994). The present findings imply that
the more impulsive features of psychopathy are driving these findings, although,
as speculated above, it is possible that the F1-IMP factor here is not a very pure
measure of impulsivity. As expected, Harm Avoidance was negatively related to
the Callous/Unemotional factor, suggesting that those who show a lack of concern
for their work and for others also show a similar lack of concern for dangerous or
frightening situations. Regression analyses actually reveal more significant
relationships than the correlational analyses, implying that gender differences may
have served to suppress some of the associations between APSD factors and
109

Anxiety. In particular, teacher-reported Narcissism negatively predicted Somatic
and Social Anxiety, and self-reported Callous-Unemotional traits negatively
predicted Social Anxiety. The latter two associations are in line with hypotheses.
Lack of affect, a feature central to the psychopathy construct, could in theory
facilitate a lack of concern regarding other people’s opinions of oneself, while
narcissistic tendencies could in theory be a reflection of such disregard.
Again, regression analyses with MASC Anxiety subscales indicate
varying relationships between APSD factors and MASC Anxiety scores.
Evidence for discriminant validity between the three APSD factors is provided in
the teacher and self-report data, as patterns of prediction of MASC subscales
differs widely for each factor. In the teacher data, the significant relationships
involve only the F1-IMP and F2-NAR factors, whereas in the child data, all three
factors are differentially related to MASC subscales. These results imply that each
APSD informant contributes information that is at least partly independent of the
others and uniquely contributes to the prediction of anxiety.
Finally, the findings for the relationship between cognitive abilities and
the APSD factors show somewhat greater correspondence across informants. In
the correlational analyses, the majority of significant relationships are negative
with the exception of self-reported F1-IMP and the Concept Formation (Fluid
Reasoning) subtest. Otherwise, F1-IMP appears to be negatively related to
Numbers Reversed (Short-Term retrieval, significant in caregiver-reported data
and marginally significant in teacher and self-reported data) and Picture
110

Vocabulary (crystallized knowledge) appears to be negatively related to F3-CU
across informants. For the most part, followup regression analyses show this same
pattern.
In general, the analyses between APSD factors and WJR subtests were
fairly inconclusive about the relationship between psychopathic features and
cognitive abilities. This actually reflects the current state of knowledge about
psychopathy and intelligence, which is rather undetermined. In fact, the clearest
relationship to emerge from these analyses was a negative association between
maltreatment and all of the subtest scores, which replicates a consistent finding
from the extant maltreatment literature (e.g. Crozier & Barth, 2005). Contrary to
hypotheses, findings here indicated that F3-CU and Picture Vocabulary are
negatively related in the caregiver- and teacher-report data. A possible
explanation my lie in the item content of the F3-CU factor. The Picture
Vocabulary subtest assesses crystallized knowledge, which is often tested in
schools. The F3-CU factor includes the reversed Item 3 (Cares about how well
he/she does at schoolwork), and it is possible that caregivers and teachers
interpret low test scores and class grades as a lack of conscientious effort, which
could be reflected in this negative correlation.
The findings of this study indicate that the APSD is a measure with some
convergent validity that appears to assess a multidimensional construct. Item
content appears to map onto psychopathy-related traits of impulsivity, narcissism
and callous/unemotionality. The three factor structure demonstrated better fit than
111

the 2F model, and indeed showed some discriminant validity (both within the
measure itself and when relationships to external variables were examined)
despite high overlap between F1-IMP and F2-NAR factors in caregiver and
teacher ratings. Reliabilities for the three factors were not very high, but were
well within the range of estimates that have been found in previous APSD studies.
Cross-informant agreement correlations fared similarly – they were not very high
but were entirely consistent with the findings of the extant literature on cross-
informant ratings of child problem behavior. Also in line with previous findings,
self-reported APSD scores yielded lower reliabilities than caregiver- and teacher-
reported information. It is the recommendation of the author that more valid,
invariant items be written for all factors (though self-report F3-CU in particular
may require more items). In general, significant relationships between the APSD
factors and aggression, anxiety and cognitive ability were as hypothesized, though
only a handful of these emerged. A key result of the external correlate analyses
was that the caregiver, teacher and self-report information all appeared to have
different patterns of relationships with the various outcome variables. The
importance of this finding lies not necessarily in the contradictory nature of these
patterns, but because it indicates that each informant may contribute independent
and incremental information to the APSD measurement. This has obvious
implications for investigators’ decisions about which informant(s) to include in
their studies. For instance, collecting only self-report and not caregiver-report
APSD in this study would have provided information that suggested elevated
112

aggression in the psychopathic individual but would not have provided much
information about the relationships with cognitive abilities that could be seen in
the caregiver data. It would thus seem ideal to attempt to collect as much
information from as many varied raters as possible, and to at least initially treat
the information independently. However if time and funds are limited, which is
the case in most research endeavors, informants should be carefully chosen.
Overall, findings from this study would recommend the use of caregiver-report
information, based on acceptable reliabilities of APSD factor scores, and a
favorable ratio for trait variance to method variance – however that
recommendation is tempered by the fact that caregiver APSD showed little
convergence with the self-reported external correlates (especially aggression).
Thus if the research question focuses on investigating the relationships among
psychopathic traits in an individual and perhaps other caregiver-rated variables,
the caregiver-report data may be very suitable. In studies that involve other self-
reported measures, caregiver-report information should be used in concert with
(although not amalgamated with) information from another reporter. Based on
this study, it is difficult to recommend the use of the self-report version of the
measure, given the unacceptably low reliabilities of the factor scores. A
discussion of the utility of self-report psychopathy in youth will follow shortly.
Teacher–reported information should be interpreted cautiously given the greater
proportion of method variance than trait variance found here. While teacher-
reported trait factor scores showed high reliabilities, much of the covariance
113

between traits was driven by method effects (the fact that the teacher was doing
both ratings) rather than by true trait covariance. Thus teacher-report information
is not recommended when evaluating relationships among the different APSD
factors. The important caveat to bear in mind is that these recommendations are
specific to the version of the measure that was used here, which is a very
restricted form of the APSD. It is possible that with a more inclusive version,
reliabilities will improve, and relationships with external correlates will present
differently.
Self-Reported Psychopathy in Youth
An important question is whether psychopathy can be adequately
measured in youth. At this point in time, questions about measurement are
perhaps even more pertinent in youth than in adults. The difficulty in treating the
disorder in adulthood (Ogloff, Wong, & Greenwood, 1990; Seto & Barbaree,
1999; Skeem, Monahan & Mulvey, 2002) lends gravity to the perjorativeness of
labeling an adolescent as a psychopath. In examining the construct validity of the
multi-informant APSD, this study was able to evaluate the performance of the
self-report version in comparison to caregiver and teacher-reported versions.
As has been discussed, the self-report scales had low reliabilities. Though
the small number of items per scale is likely somewhat responsible, it does not
explain the comparative lack of reliability between self-report and the other
informants. Reliability is key to sound and valid measurement. Thus, specific to
the particular APSD scales used in this study, the low reliability of adolescent
114

self-report of psychopathy factor scores fosters little confidence in their use in
research and practice. However, the self-reported scales were clearly positively
related to RPQ aggression, and for the most part, negatively related to MASC
anxiety, which lends some credence to their validity (though validity is
fundamentally undermined by low reliability). The negative relationship between
F3-CU and MASC Harm Avoidance, and positive relationship between F3-CU
and RPQ Proactive Aggression are particularly encouraging, given that they
reflect the importance of fearlessness and instrumental aggression to the
psychopathy construct. An alternative strategy to relying on the individual factors
is to use total score self-reported APSD in research, and to frame research
questions in terms of overall psychopathic features, rather than breaking down the
factors. Here, the total score was more reliable (0 = .59) and showed essentially
the same relationships to external correlates as the individual scales. This
approach would limit the ability to parse a group of psychopathic individuals into
subgroups, or to determine whether certain correlates or sequelae are specifically
related to one or another aspect of psychopathy. Nevertheless, some researchers in
the field, including Hare (2005) himself have questioned the utility of devoting so
much effort toward factorizing the psychopathy construct, when it is classically
defined as a constellation of traits which co-occur in psychopathic individuals.
Thus perhaps the contribution that the APSD could make to self-report
psychopathic measurement lies in its use as a total score measure, rather than a 3-
factor measure. The caregiver and teacher reported information fared somewhat
115

better in terms of reliability among the factor scores and their use could be evoked
when a further breakdown of the individual’s characteristics is desired.
The question still remains as to whether the low correspondence between
the self-reported information and caregiver and teacher-reported information is of
concern. As mentioned previously, this low rate of agreement is common to the
child problem behavior literature, and the APSD may simply behave as the other
instruments do. In any case, some further discussion of the difficulties of self-
report psychopathy seems appropriate.
In adults, a major concern regarding self-report psychopathy is that
individuals may be extrinsically motivated to reply untruthfully to psychological
evaluation and risk assessment measures to present themselves favorably. This is
especially pertinent in the context of parole hearings, treatment evaluation, and so
on. In the present study, however, this sample of community adolescents had little
such extrinsic motivation to lie on the measures collected. They were assured of
confidentiality, and told that the study had no bearing on them outside of the
research study. In the absence of such extrinsic motivation, intrinsic reasons could
be in operation. One such possibility that comes from the emotion literature, is
that psychopathic individuals may experience “duping delight” which is to say
that they simply derive pleasure from misleading others (Ekman, 2002; Porter &
Woodworth, 2007). Also, if dishonesty is truly pathological for the psychopath
(Hare, 1991), lying may just be more natural to them than telling the truth. These
effects would be virtually impossible to discern in the present study, but should be
116

considered as possible explanations for untruthful responding. In addition,
developmental issues such as a lack of cognitive maturity and defensiveness (as
discussed previously) may lead to an inaccuracy in self-evaluation.
Alternatively, the items may just be poorly suited for gleaning accurate
self-report information. Given that the APSD was originally downwardly
translated from the adult Psychopathy Checklist Revised (Hare, 1991) for
caregivers and teachers, there was less concern about reducing the face validity of
the items than there might have been had the measure first been designed as a
self-report instrument. In converting the caregiver and teacher version to the self-
report version, the only changes made were of pronouns from my child/student to
the first person. While ensuring direct comparability of the three informant forms,
this lack of alteration for self-report purposes may well have meant that items
were too transparent (e.g. My emotions are shallow or fake) or confusing (e.g. I
keep the same friends) for children and adolescents.
In sum, whether the APSD is a valuable measure of self-report
psychopathy is difficult to conclude from the results of this study. In comparison
to the caregiver and teacher versions, the low reliabilities would suggest not (and
these results are reinforced by the Poythress et al. [2000] review). However,
agreement with teacher and caregiver versions, and the association of
psychopathy as measured by both the factor and total scores of the APSD with
anxiety and depression, in the hypothesized direction, lend some tentative support
to its usefulness. Regardless, it seems clear that in order to answer the larger
117

question of whether psychopathy can truly be measured via self-report,
improvements must be made to the instruments used. More, better performing,
and more developmentally appropriate items should be written to first establish a
reliable measure. Once suitable reliability is achieved, then validity can be
evaluated in greater detail.
Sample demographics: The maltreated majority
The sample included in this study is also quite unique, primarily in that it
is oversampled for known maltreatment experience. Though adolescents’
maltreatment experience was not central to the objectives of the present study, it
is acknowledged that the high proportion of maltreated individuals (in fact, the
majority) in the sample limits the interpretability of the findings in terms of
implications for the construct of psychopathy. The invariance analyses were in
part an attempt to control for a history of maltreatment, and indeed several items
were identified as being noninvariant between maltreatment and comparison
groups. There is a strong association between maltreatment (physical abuse in
particular) and behavior problems (Kaplan, Pelcovitz & LaBruna, 1999; Lansford
et al., 2002), which could serve to obscure or moderate the relationship between
psychopathy and various external correlates. For example, consider the
association between cognitive ability and psychopathy, which is very unclear in
the extant literature. The classic definition of psychopathy (Cleckley, 1941)
specifically posits that psychopaths possess at least normal levels of intelligence
(if not higher), but most research has not found a relationship (Johansson & Kerr,
118

2005). It is often thought that the confusion may be somewhat attributable to the
negative association between antisocial individuals and intelligence, given the
high overlap between antisocial behavior (e.g. as defined by Antisocial
Personality Disorder or delinquency) and psychopathy. However, it seems
plausible that maltreatment may also contribute to the uncertainty. Research
indicates that both the negative association between cognitive ability and
maltreatment (e.g. Veltman & Browne, 2001) and the positive association
between maltreatment and antisocial behavior (Cicchetti & Manly, 2001; Jaffee,
Caspi, Moffitt, & Taylor, 2004) are rather robust. In the present study in
particular, despite efforts to control for maltreatment experience, perhaps these
strong correlations operated to mimic the expected relation between psychopathic
traits and aggression but attenuate the expected positive association between
psychopathic traits (F3-CU in particular) and intelligence. More broadly, it is
possible that previous studies of juvenile psychopathy in at-risk community or
young offender samples have included a significant number of individuals with
maltreatment histories, where maltreatment has not been addressed. Similar
overlap between correlates of maltreatment experience and psychopathic features
might be seen in symptoms of blunted affect or autonomic underarousal. Few
studies of psychopathy have addressed childhood maltreatment experience as
more than a just covariate to control for, though some theoretical endeavors have
considered its possible role in etiological processes (e.g. Lykken, 1995; Porter,
1996). Future research on psychopathy within this sample will endeavor to
119

investigate more substantive questions about the influence of maltreatment on
psychopathic traits.
Limitations of the Current Study
The findings of this study should be considered in light of several
limitations. Firstly, the nature of the statistical procedures followed here presents
major limits to the generalizability of the findings. The policy of dropping items
that showed noninvariance, coupled with the sequential process of testing factor
models and invariance in each informant data set likely meant that some part of
the final model was very sample-specific, and some decisions may have been
made on the basis of sampling error. Also, the conservativeness of the procedure
itself led to the reduction in items (which was expected and accepted in the
present study, but perhaps a different approach, such as attempting to achieve
partial invariance of noninvariant items) could have resulted in a more inclusive
measure. As discussed previously, scale reliability suffers from small numbers of
items and thus the procedure followed here may have contributed to the low
reliabilities of the factors found here. However, while these procedures may have
limited the generalizability of the findings, they in fact facilitated the analyses
conducted here. In order for the multitrait-multimethod analyses to make direct
comparisons across informants (and trait factors), the traits needed to be identical.
Another limitation was the size of the sample. Several practical reasons precluded
the delay of analyses until the full sample was collected. Though the sample
120

included here was not prohibitively small, the full sample size (and possible
change in sample characteristics) may see substantial change in the findings.
Conclusion
Previous studies have shown psychopathy to be a useful construct in
youths, as it is characterizes a subset of youths who demonstrate increased rates of
offending (Corrado et al., 2004), a selective difficulty for recognition of emotional
stimuli (Blair, Budhani Colledge, & Scott, 2005; Stevens, Charman & Blair,
2001), impaired empathy and moral reasoning, (Pardini, Lochman, & Frick,
2003), and decreased autonomic arousal in anticipation of punishment (Fung et
al., 2006). In addressing some fundamental aspects of the psychometric properties
of the APSD in a diverse sample, this study attempts to inform the empirical
literature on psychopathic traits in youth, and to provide a caution regarding
measurement approaches to its more applied usage. The importance of the latter
lies in the increasing application of the construct of psychopathy to, for example,
the legal justice system. The Psychopathy Checklist-Revised has a long history in
risk assessment and parole consideration procedures in the prison system
(Webster, Muller-Isberner, & Fransson, 2002) and youth measures such as the
PCL:Youth Version and the APSD are trickling into the juvenile justice system as
well. Specifically, this study should provide some confidence in the initial 3-
factor model presented by Frick et al. (2000), but it should also motivate APSD
users to strive to better understand the implications of multi-informant
information, and to choose informants without the assumption that they are
121

interchangeable. This study should also prompt researchers to investigate the
invariance of the measure across their relevant groups of interest. Doing so here
resulted in the identification of 5 noninvariant items – which may be an
underestimate given that some items were dropped prior to the invariance
analyses. In particular, all 5 identified items were noninvariant between
maltreated and comparison groups, indicating that they were differentially related
to the latent traits in the two groups. To the author’s knowledge, this is not only
the first MTMM analysis of the APSD, but also one of the first investigations of
its invariance. While the reduction of the number of items in the scales due to
noninvariance may have decreased the generalizability of the model analyzed here
and its interpretability in light of previous findings based on more inclusive APSD
measures, it is possible that similar levels of noninvariance existed in those data
but were simply not addressed.
Future research using the existing ASPD measure should continue to
investigate its factor structure. Two and 3-factor structures both have theoretical
and statistical precedents in the literature. Limited discriminant validity for the 3
factor structure was found here, and requires replication. As mentioned above,
invariance analyses should be applied to subsequent research with the APSD to
ensure that the measurement has the same meaning and structure in pertinent
groups. The results of this study suggest a number of procedures which might
serve to improve informant agreement and general performance of the APSD. As
has already been suggested, the most effective would likely be to write more
122

items designed to tap the constructs in a less transparent manner. Another
possibility might be to try to administer the measure to all informants in the same
environment or, perhaps more practically, to only utilize different informants who
share environmental characteristics, e.g. caregiver and child, two different
teachers.
Psychopathy in juveniles is a controversial and sensitive subject, and thus
it should be studied with caution. Research has implications for many applied
situations, and thus it must be supported by reliable and valid measurement.
Though the Antisocial Process Screening Device is widely used in the literature
today, findings from this study suggest that the measure itself is in need of further
study and development. Efforts should be made to increase the reliability of the
factors and to decrease method variance across informants. In particular, self-
report applications require greater attention.
123

References
Abikoff, H., Courtney, M., Pelham, W.E., & Koplewicz, H.S. (1993). Teachers'
ratings of disruptive behaviors: The influence of halo effects. Journal of
Abnormal Child Psychology, 21, 519-533.

Achenbach, T.M. (1991). Manual for the Child Behavior Checklist and 1991
Profile. Burlington, VT: University of Vermont, Department of
Psychiatry.

Achenbach, T.M., McConaughy, S.H., & Howell, C.T. (1987). Child/adolescent
behavioral and emotional problems: Implications of cross-informant
correlations for situational specificity. Psychological Bulletin, 101, 213-
232.

Af Klinteberg, B., Humble, K., & Schalling, D. (1992). Personality and
psychopathy of males with a history of early criminal behaviour.
European Journal of Personality, 6, 4, 245-266.

Becker, T.E., & Cote, J.A. (1994). Additive and multiplicative method effects in
applied psychological research: An empirical assessment of three models.
Journal of Management, 20, 625-641.

Bentler, P.M. (1990). Comparative fit indexes in structural models. Psychological
Bulletin, 107, 2, 238-246.

Bentler, P. M., & Bonnett, D. G. (1980). Significance tests and goodness-of-fit in
the analysis of covariance structures. Psychological Bulletin, 88, 588-606.

Biederman, J. (1998). Resolved: mania is mistaken for ADHD in prepubertal
children. Journal of the American Academy of Child and Adolescent
Psychiatry, 37, 1091-1093.

Blair, R.J.R, Budhani, S., Colledge, E., & Scott, S. (2005). Deafness to fear in
boys with psychopathic tendencies. Journal of Child Psychology and
Psychiatry, 46(3), 327-336.

Blonigen, D.M., Hicks, B.M., Krueger, R.F., Patrick, C.J., & Iacono, W.G.
(2005). Psychopathic personality traits: heritability and genetic overlap
with internalizing and externalizing psychopathology. Psychological
Medicine, 35, 637-648.

124

Brown, T.A. (2006). Confirmatory factor analysis for applied research. New
York: Guildford Press.

Browne, M.A., & Codec, R. (1993). Alternative ways of assessing model fit. In
K.A. Bollen, & J.S. Long (Eds.), Testing Structural Equation Models (pp.
136-162). Newbury Park, CA:Sage Publications.

Byrne, B.M., & Goffin, R.D. (1993). Modeling MTMM data from additive and
multiplicative covariance structures: An audit of construct validity
concordance. Multivariate Behavioral Research, 28, 67-96.

Campbell, D.T., & Fiske, D.W. (1959). Convergent and discriminant validation
by the multitrait-multimethod matrix. Psychological Bulletin, 56, 2, 81-
105.

Caputo, A.A., Frick, P.J., & Brodsky, S.L. (1999). Family violence and juvenile
sex offending: Potential mediating roles of psychopathic traits and
negative attitudes toward women. Criminal Justice and Behavior, 26,
338-356.

Cattell, R.B. (1963). Theory for fluid and crystallized intelligence: A critical
experiment. Journal of Educational Psychology, 54, 1-22.

Cattell, R.B. (1966). The scree test for the number of factors. Multivariate
Behavioral Research, 1, 245-276.

Cicchetti, D. V. (1994). Guidelines, criteria, and rules of thumb for evaluating
normed and standardized assessment instruments in psychology.
Psychological Assessment, 6, 284-290.

Cicchetti, D.V., & Manly, J.T. (Eds.) (2001). Operationalizing child
maltreatment: Developmental processes and outcomes [Special issue].
Development and Psychopathology, 13(4).

Clausen, A.H. & Crittenden, P.M. (1991). Physical and psychological
maltreatment: Relations among types of maltreatment. Child Abuse &
Neglect, 15, 1-2, 5-18.

Cleckley, H. (1941). The mask of sanity. St. Louis, MO: C.V. Mosby.

Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlational analysis
for the behavioral sciences. Hillsdale, NJ: Erlbaum.

125

Cooke, D.J., & Michie, C. (2001). Refining the construct of psychopathy:
Towards a hierarchical model. Psychological Assessment, 13, 171-188.

Cooke, D.J., Michie, C., Hart, S.D., & Clark, D. (2005). Assessing psychopathy
in the UK: concerns about cross-cultural generalisability. British Journal
of Psychiatry, 186, 335-341.

Darke, S., Kaye, S., Finlay-Jones, R., & Hall, S. (1998). Antisocial personality
disorder, psychopathy and injecting heroin use. Drug and Alcohol
Dependence, 52, 63-69.

Christian, R.E., Frick, P.J., Hill, N.L., Tyler, L., & Frazer, D.R. (1997).
Psychopathy and conduct problems in children: II. Implications for
subtyping children with conduct problems. Journal of the American
Academy of Child and Adolescent Psychiatry, 36, 233-241.

Cornell, D.G., Warren, J., Hawk, G., Stafford, E., Oram, G., & Pine, D. (1996).
Psychopathy in instrumental and reactive violent offenders. Journal of
Consulting and Clinical Psychology, 64, 783-790.

Connor, D.F., Steingard, R.J., Anderson, J.J., & Melloni, R.H. (2003).Gender
differences in reactive and proactive aggression. Child History and
Humany Development, 33, 4, 279-194.

Conway, J.M. (1998). Estimation and uses of the proportion of method variance
for multitrait-multimethod data. Organizational Research Methods, 1,
209-222.

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.

Cronbach, L.J. (1951). Coefficient alpha and the internal structure of tests.
Psychometrika, 16, 297-334.

Crozier, J.C., & Barth, R.P. (2005). Cognitive and academic functioning in
maltreated children. Schools and Children, 27, 4, 197-206.

Dell'Osso, L., Pini, S., Cassano, G.B., Mastrocinque, C., Seckinger, R.A.,
Saettoni, M., Papasogli, A., Yale, S.A., & Amador, X.F. (2002). Insight
into illness in patients with mania, mixed mania, bipolar depression and
major depression with psychotic features. Bipolar Disorders, 4, 315-322.

126

Dempster, A.P., Laird, N.M., & Rubin, D.B. (1977). Maximum likelihood
estimation from incomplete data via the EM algorithm (with discussion).
Journal of Royal Statistical Society, B39, 1-38.

Dierker, L.C., Albano, A.M., Clarke, G.N., Heimberg, R.G., Kendall, P.C.,
Merikangas, K.R., Lewinsohn, P.M., Offord, D.R., Kessler, R., & Kupfer,
D.J. (2001). Screening for anxiety and depression in early adolescence.
Journal of the American Academy of Child and Adolescent Psychiatry, 40,
8, 929-936.

Edens, J. F., Hart, S. D., Johnson, D. W., Johnson, J., & Olver, M. E. (2000). Use
of the Personality Assessment Inventory to assess psychopathy in offender
populations. Psychological Assessment, 12, 132–139.
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, 435-443.

Ekman, P. (2002). Telling lies: clues to deceit in the marketplace, politics, and
marriage (2nd ed.). New York, NY: Norton.

Epkins, C.C. (1996). Parent ratings of children's depression, anxiety, and
aggression: A cross-sample analysis of agreement and differences with
child and teacher ratings. Journal of Clinical Psychology, 52, 6, 599-608.

Falkenbach, D.M., Poythress, N., & Heide, K.M. (2003). Psychopathic features in
a juvenile diversion population: Reliability and predictive validity of two
self-report measures. Behavioral Sciences and the Law, 21, 787-805.

Fedora, O., & Fedora, S. (1983). Some neuropsychological and
psychophysiological aspects of psychopathic and nonpsychopathic
criminals. In P. Flor-Henry, & J. Gruzelier (Eds.), Laterality and
psychopathology (pp.41-58). Amsterdam: Elsevier Science Publishers.

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

Fowles, D. (1980). The three arousal model: Implications of Gray's two-factor
learning theory for heart rate, electrodermal activity, and psychopathy.
Psychophysiology, 17, 87-104.

127

Fowles, D.C. (2000). Electrodermal hyporeactivity and antisocial behavior: does
anxiety mediate the relationship? Journal of Affective Disorders, 61, 177-
189.

Fowles, D., & Missel, K. (1994). Electrodermal hyporeactivity, motivation and
psychopathy: Theoretical issues. In D. Fowles, P. Sutker, & S. Goodman
(Eds.), Special focus on psychopathy and antisocial personality : A
development perspective (pp. 263-284). New York, NY: Springer.

Frick, P.J., Lilienfeld, S.O., Ellis, M., Loney, B.R., & Silverthorn, P. (1999). The
association between anxiety and psychopathy dimensions in children.
Journal of Abnormal Child Psychology, 27, 381-390.

Frick, P.J., O'Brien, B.S., Wooton, J.M., & McBurnett, K. (1994). Psychopathy
and conduct problems in children. Journal of Abnormal Psychology, 103,
700-707.

Frick, P.J., Bodin, S.D., & Barry, C.T. (2000). 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 delinquency. Journal
of Abnormal Child Psychology, 31, 457-470.

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

Frick, P.J., O'Brien, B.S., Wooton, J.M., & McBurnett, K. (1994). Psychopathy
and conduct problems in children. Journal of Abnormal Psychology, 103,
700-707.

Fung, M.T., Raine, A., Loeber, R., Lynam, D.R., Steinhauer, S., Venables, P.V.,
Stouthamer-Loeber, M. (2005). Reduced autonomic functioning in
psychopathy-prone adolescents. Journal of Abnormal Psychology, 114, 2,
187-196.

Gagnon, C., Vitaro, F., & Tremblay, R.E. (2000). (1992). Parent-teacher
agreement on kindergartners’ behavior problems. Journal of Child
Psychology and Psychiatry, 33, 1255-1261.
128

Goffin, R.D., & Jackson, D.N. (1992). Analysis of multitrait-multirater
performance appraisal data: Composite direct product method versus
confirmatory factor analysis. Multivariate Behavioral Research, 27, 363-
385.

Gorenstein, E.E. (1982). Frontal lobe functions in psychopaths. Journal of
Abnormal Psychology, 91, 368-379.

Gorenstein, E.E., & Newman, J.P. (1980). Disinhibitory psychopathology: A new
perspective and a model for research. Psychological Review, 87, 301-315.

Hare, R.D. (1978). Electrodermal and cardiovascular correlates of psychopathy.
In R.D. Hare and D. Schalling (Eds.), Psychopathic Behavior: Approaches
to Research (pp. 107-144). New York: Wiley.

Hare, R.D. (1985). Comparison of procedures for the assessment of psychopathy.
Journal of Consulting and Clinical Psychology, 53, 1, 7-16.

Hare, R.D. (1991). The Hare Psychopathy Checklist-Revised. Toronto, Canada:
Multi-Health Systems.

Hare, R.D. (1998). Psychopathy, affect and behavior. In D.J. Cooke, A.E. Forth,
and R.D. Hare (Eds.), Psychopathy: Theory, research and implications for
society. Dordrecht: Kluwer.

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

Hare, R.D. (2005, July). Lifetime achievement award acceptance address. Paper
presented at the meeting of the Society for the Scientific Study of
Psychopathy, Vancouver, BC, Canada.

Hare, R. D., Cooke, D.J., & Hart, S.D. (1999). Psychopathy and sadistic
personality disorder. In T. Millon, P.H. Blaney, & R.D. Davis (Eds.),
Oxford textbook of psychopathology (pp. 555-584). New York, NY:
Oxford University Press.

Harpur, T.J., Hare, R.D., & Hakstian, A.R. (1989). Two-factor conceptualization
of psychopathy: Construct validity and assessment implications.
Psychological Assessment, 1, 6-17.

Hart, S.D., Cox, D.N., & Hare, R. D. (1995). The Hare Psychopathy Checklist:
Screening Version. Toronto, ON, Canada: Multi-Health Systems.
129

Haviland, M.G., Sonne, J.L., & Kowert, P.A. (2004). Alexithymia and
psychopathy: comparison and application of California Q-set Prototypes.
Journal of Personality Assessment, 82, 3, 306-316.

Henry, B., & Moffitt, T.E. (1997). Neuropsychological and neuroimaging studies
on juvenile delinquency and adult criminal behavior. In D. M. Stoff, J.
Breiling, & J. D. Maser (Eds.), Handbook of Antisocial Behaviour (pp.
280-288). New York, NY: Wiley.

Herpertz, S.C., & Sass, H. (2000). Emotional deficiency and psychopathy.
Behavioral Sciences and the Law, 18, 567-580.

Hicks, B.M., Markon, K.E., Patrick, C.J., Krueger, R.F., & Newman, J.P. (2004).
Identifying psychopathy subtypes on the basis of personality structure.
Psychological Assessment, 16, 3, 276-288.

Hill, C., Neumann, C.S., & Rogers, R. (2004). Confirmatory factor analysis of the
Psychopathy Checklist: Screening Version (PCL:SV) in offenders with
Axis I disorders. Psychological Assessment, 16, 90-95.

Holland, T.R., Beckett, G.E., & Levi, M. (1981). Intelligence, personality, and
criminal violence: A multivariate analysis. Journal of Consulting and
Clinical Psychology, 49, 106-111.

Horn, J.L. (1976). Human abilities: A review of research and theory in the early
1970s. Annual Review of Psychology, 27, 437-485.

Horn, J.L. (1985). Remodeling old models of intelligence. In B.B. Wolman (Ed.),
Handbook of intelligence (pp. 267-300). New York, NY: Wiley.

Horn, J.L., & Cattell, R.B. (1966). Refinement and test of the theory of fluid and
crystallized intelligence. Journal of Educational Psychology, 57, 253-270.

Jackson, R.L., Rogers, R., Neumann, C.S., & Lambert, P.L. (2002). Psychopathy
in female offenders: An investigation of its underlying dimensions.
Criminal Justice and Behavior, 29, 692-704.

Jaffee, S.R., Caspi, A., Moffitt, T.E., & Taylor, A. (2004). Physical maltreatment
victim to antisocial child: Evidence of an environmentally mediated
process. Journal of Abnormal Psychology, 113, 1, 44-55.

130

Johansson, P., & Kerr, M. (2005). Psychopathy and intelligence: a second look.
Journal of Personality Disorders, 19, 4, 357-369.

Kaiser, H.F. (1960). The application of electronic computers to factor analysis.
Educational and Psychological Measurement, 20, 141-151.

Kaplan, S.J., Labruna V., Pelcovitz D., Salzinger S., Mandel F., & Weiner M.
(1999). Physically abused adolescents: behavior problems, functional
impairment, and comparison of informants' reports. Pediatrics, 104, 43-
49.

Kaplan, S.J., Pelcovitz, D., & LaBruna, V. (1999). Child and Adolescent Abuse
and Neglect Research: A Review of the Past 10 Years. Part I: Physical and
Emotional Abuse and Neglect. Journal of the American Academy of Child
& Adolescent Psychiatry, 38, 10, 1214-1222.

Kenny, D.A. & Kashy, D.A. (1992). Analysis of the multitrait-multimethod
matrix by confirmatory factor analysis. Psychological Bulletin, 112, 1,
165-172.
Kruh, I.P., Frick, P.J., & Clements, C.B. (2005). Historical and personality
correlates to the violence patterns of juveniles tried as adults. Criminal
Justice and Behavior, 32, 1, 69-96.

Lansford, J.E., Dodge, K.A., Pettit, G.S., Bates, J.E., Crozier, J., & Kaplow, J.
(2002). A 12-year prospective study of the Long-term Effects of Early
Child Physical Maltreatment on Psychological, Behavioral, and Academic
Problems in Adolescence. Archives of Pediatric and Adolescent Medicine,
156, 824-830.

Lee, Z., Vincent, G.M., Hart, S.D., & Corrado, R.R. (2003). The validity of the
Antisocial Process Screening Device as a self-report measure of
psychopathy in adolescent offenders. Behavioral Sciences and the Law,
21, 771-786.

Lilienfeld, S.O. (1998). Methodological advances and developments in the
assessment of psychopathy. Behavior Research and Therapy, 36, 99-125.

Lilienfeld, S.O., & Andrews, B.P. (1996). Development and preliminary
validation of a self-report measure of psychopathic personality traits in
noncriminal populations. Journal of Personality Assessment, 66, 488-524.

131

Lilienfeld, S.O., & Penna, S. (2001). Anxiety sensitivity: Relations to
psychopathy, DSM-IV personality disorder features and personality traits.
Journal of Anxiety Disorders, 15, 367-383.

Loeber, R., Brinthaupt, V.P., & Green, S.M. (1990). Attention deficits,
impulsivity, and hyperactivity with or without conduct problems:
Relationships to delinuency and unique contextual factors. In R.J.
McMahon, R.D. Peters (Eds.), Behavior disorders of adolescence:
Research, intervention, and policy in clinical and school settings (pp. 39-
61). New York, NY: Plenum Press.

Lynam, D.R. (1997). Pursuing the psychopath: Capturing the fledgling
psychopath in a nomological net. Journal of Abnormal Psychology, 106,
3, 425-438.

Lykken, D.T. (1957). A study of anxiety in the sociopathic personality. Journal
of Abnormal and Social Psychology, 55, 6-10.

Lykken, D.T. (1995). The Antisocial Personalities. New Jersey: Erlbaum.

Maccoby, E.E., & Jacklin, C.N. (1980). Sex differences in aggression: A rejoinder
and reprise. Child Development, 51, 4, 964-980.

March, J.S., James, D.A., Sullivan, K., Stallings, P., & Conners, C.K. (1997). The
Multidimensional Anxiety Scale for Children (MASC): factor structure,
reliability, and validity. Journal of the American Academy of Child and
Adolescent Psychiatry, 36, 4, 554-565.
Marsee, M.A., Silverthorn, P., & Frick, P.J. (2005). The association of
psychopathic traits with aggression and delinquency in non-referred boys
and girls. Behavioral Sciences and the Law, 23, 6, 803 – 817.
Marsh, H.W. (1989). Confirmatory factor analyses of multitrait-multimethod
data: Many problems and a few solutions. Applied Psychological
Measurement, 13, 335-361.

Marsh, H.W., & Bailey, M. (1991). Confirmatory Factor Analyses of Multitrait-
Multimethod Data: A Comparison of Alternative Models. Applied
Psychological Measurement, 15, 47-70.

132

Marsh, H.W., Byrne, B. M., & Craven, R. (1993). Overcoming problems in
confirmatory factor analyses of MTM data: The correlated uniqueness
model and factorial invariance. Multivariate Behavioral Research, 27,
489-507.

Marsh, H.W., & Hocevar, D. (1983). Confirmatory factor analysis of multitrait-
multimethod matrices. Journal of Educational Measurement, 20, 231-248.

McArdle, J.J. (1994). Factor analysis. In R.J. Sternberg (Ed.)., Encyclopedia of
Human Intelligence, (pp. 422-430). New York: MacMillan Publishing
Company.

McArdle, J.J. (1996). Current directions in structural factor analysis. Current
Directions in Psychological Science, 5, 1, 10-17.

McDermott, P.A., Alterman, A.I., Cacciola, J.S., Rutherford, M.J., Newman, J.P.,
& Mullholland, E.M. (2000). Generality of psychopathy checklist-revised
factors over prisoners and substance-abuse patients. Journal of Consulting
and Clinical Psychology, 68, 1, 181-186.

McGee, R., Wolfe, D.A., Yuen, S.A., Wilson, S.K., & Carnochan, J. (1995). The
measurement of maltreatment: A comparison of approaches. Child Abuse
& Neglect, 19, 2, 233-249.

McGrew, K.S., Werder, J.K., & Woodcock, R.W. (1991). Woodcock-Johnson
Technical Manual. Allen, TX: DLM Teaching Resources.

Mennen, F. & Trickett, P.K. (2006). Parenting attitudes, family environments, and
psychopathology in maltreating and non maltreating urban parents. Paper
presented at the annual meeting of the Society for Social Work and
Research, San Antonio, TX.

Mesman, J., & Koot, H.M. (2000). Self-reported depression and anxiety in pre-
adolescence I: Association with parent- and teacher-reported problems.
Journal of the American Academy of Child and Adolescent Psychiatry, 39,
1371-1378.

Meyer, G.J., Finn, S.E., Eyde, L.D., Kay, G.G., Moreland, K.L., Dies, R.R.,
Eisman, E.J. Kubiszyn, T.W., & Reed, G.M. (2001). Psychological
testing and psychological assessment. A review of evidence and issues.
American Psychologist, 56, 128-165.

133

Moffitt, T.E. (1993). Adolescence-limited and life-course persistent antisocial
behavior: A developmental taxonomy. Psychological Review, 100, 4, 674-
101.

Murrie, D., & Cornell, D.G. (2002). Psychopathy screening of incarcerated
juveniles: A comparison of measures. Psychological Assessment, 14, 390-
396.

Muthén, B., du Toit, S.H.C., & Spisic, D. (1997). Robust inference using weighted
least squares and quadratic estimating equations in latent variable modeling
with categorical and continuous outcomes. Accepted for publication in
Psychometrika.
Muthén, L.K. and Muthén, B.O. (1998-2006). Mplus User’s Guide. Fourth
Edition. Los Angeles, CA: Muthén & Muthén.
Newman, J.P. (1998). Psychopathic behavior: An information processing
perspective. In D.J. Cooke (Ed.), Psychopathy: Theory, Research and
Implications for Society (pp. 81-104). The Netherlands: Kluwer Academic
Publishers.

Newman, J.P., & Kosson, D. (1986). Passive avoidance learning in psychopathic
and nonpsychopathic offenders. Journal of Abnormal Psychology, 95,
257-263.

Newman, J.P., Kosson, D., & Patterson, C.M. (1992). Delay of gratification in
psychopathic and nonpsychopathic offenders. Journal of Abnormal
Psychology, 101, 630-636.

Newman, J.P., Widom, C.S., & Nathan, S. (1985). Passive-avoidance in
syndromes of disinhibition: Psychopathy and extraversion. Journal of
Personality and Social Psychology, 48, 1316-1327.

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

O'Brien, B.S., & Frick, P.J. (1996). Reward dominance: Associations with
anxiety, conduct problems, and psychopathy in children. Journal of
Abnormal Child Psychology, 24, 233-240.

Ogloff, J.R.P., Wong, S., & Greenwood, A. (1990). Treating criminal
psychopaths in a therapeutic community program. Behavioral Sciences
and the Law, 8, 181-190.
134

Orchard, T., & Woodbury, M.A. (1972). A missing information principle: Theory
and applications. In Proceedings of the 6th Berkeley Symposium on
Mathematical Statistics and Probability (pp. 697-729). Washington, DC:
University of California Press.

Patrick, C.J., & Zempolich, K.A. (1998). Emotion and aggression in the
psychopathic personality. Aggression and Violent Behavior, 3, 303-383.

Porter, S., & Woodworth, M. (2007). “I’m sorry I did it…but he started it”: A
comparison of the official and self-reported homicide descriptions of
psychopaths and non-psychopaths. Law and Human Behavior, 31, 1, 91-
107.

Poythress, N.G., Douglas, K.G., Falkenbach, D., Cruise, K., Lee, Z., Murrie,
D.C., & Vitacco, M. (2006). Internal consistency reliability of the self-
report Antisocial Process Screening Device. Assessment, 13, 107.
Poythress, N.G., Dembo, R., Wareham, J., Greenbaum, P.E. (2006). Construct
validity of the Youth Psychopathic Traits Inventory (YPI) and the
Antisocial Process Screening Device (APSD) with justice-involved
adolescents. Criminal Justice Behavior, 33, 1, 26-55.

Raine, A., Dodge, K.A., Loeber, R., Gatze-Kopp, L., Lynam, D.R., Reynolds, C.,
Stouthamer-Loeber, M., & Liu, Jianghong. (2006). The Reactive-
Proactive Aggression Questionnaire: Differential correlates of reactive and
proactive aggression in adolescent boys. Aggressive Behavior, 32, 159-
171.

Raine, A., O'Brien, M., Smiley, N., Scerbo, A., & Chan, C.J. (1990). Reduced
lateralization in verbal dichotic listening in adolescent psychopaths.
Journal of Abnormal Psychology, 99, 272-277.

Rubio-Stipec, M., Fitzmaurice, G., Murphy, J., & Walker, A. (2003). The use of
multiple informants in identifying the risk factors of depressive and
disruptive disorders – are they interchangeable? Social Psychiatry and
Psychiatric Epidemiology, 38, 51-58.

Rutter, M. (2005). Commentary: What is the meaning and utility of the
psychopathy concept? Journal of Abnormal Child Psychology, 33(4), 499-
503.

135

Salekin, R.T., Neumann, C.S., Leistico, A.M., & DiCicco, T.M. (2004). Construct
validity of psychopathy in a youth offender sample: Taking a closer look
at psychopathy's importance over disruptive behavior disorders. Journal of
Abnormal Psychology, 113, 416-427.

Saris, W.E. (1990). Models for evaluation of measurement instruments. In W.E.
Saris & van Meurs (Eds.), Evaluation of measurement instruments by
meta-analysis of multitrait-multimethod matrices (pp. 520-80).
Amsterdam: North-Holland.

Sawyer, M.G., Baghurst, P., & Clark, J. (1992). Differences between reports from
children, parents and teachers: Implications for epidemiological studies.
Australian and New Zealand Journal of Psychiatry, 26, 652-660.

Sawyer, M.G., Baghurst, P., & Mattias, J. (1992). Differences between informants
reports describing emotional and behavioral problems in a community and
clinic-referred children: A research note. Journal of Child Psychology
and Psychiatry, 33, 441-449.

Scerbo, A., Raine, A., O'Brien, M., Chan, C.J., Rhee, C., & Smiley, N. (1990).
Reward dominance and passive avoidance learning in adolescent
psychopaths. Journal of Abnormal Child Psychology, 18, 4, 451-463.

Schafer, J.L. (1999). NORM: Multiple imputation of incomplete multivariate data
under a normal model (Version 2.0), Computer statistical software
available from http://www.stat.psu.edu/~jls/misoftwa.html.

Scherpenzeel, A.C. (1995). A question of quality. Evaluating survey questions by
multitrait-multimethod studies. Doctoral dissertation, University of
Amsterdam, Leidschendarn: The Netherlands.

Schmitt, W.A., & Newman, J.P. (1999). Are all psychopaths low-anxious?
Journal of Abnormal Psychology, 108, 2, 353-358.

Schmitt, N., & Stults, D.M. (1986). Methodological review: Analysis of MTMM
matrices. Applied Psychological Measurement, 10, 1-22.

Scullen, S.E., (1999). Using confirmatory factor analysis if correlated
uniquenesses to estimate method variance in multitrait-multimethod
matrices. Organizational Research Methods, 2, 275-292.

Seto, M.C., & Barbaree, H.E. (1999). Psychopathy, treatment behavior, and sex
offender recidivism. Journal of Interpersonal Violence, 14, 1235-1248.
136

Sharp, C., van Goozen, S., Goodyer, I. (2006). Children's subjective emotional
reactivity to affective pictures: gender differences and their antisocial
correlates in an unselected sample of 7–11-year-olds. Journal of child
psychology and psychiatry, 47, 2, 113-224.

Skeem, J.L., Monahan, J., & Mulvey, E.P. (2002). Psychopathy, treatment
involvement, and subsequent violence among civil psychiatric patients.
Law and Human Behavior, 26, 577-603.

Smith, S.S., Arnett, P.A., & Newman, J.P. (1992). Neuropsychological
differentiation of psychopathic and nonpsychopathic criminal offenders.
Personality and Individual Differences, 13, 1233-1243.

Soderstrom, H., Nillson, T., Sjodin, A.K., Carlstedt, A., & Forsman, A. (2005).
The childhood-onset neuropsychiatric background to adulthood
psychopathic traits and personality disorders. Comprehensive Psychiatry,
46, 2, 111-116.

Solomon, R.S. (1939). Cognitive aspects of deterioration in mental patients.
Journal of Abnormal Social Psychology, 34, 497-517.

Spain, S.E., Douglas, K.S., Poythress, N., & 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.

Stafford, E.,& Cornell, A.H. (2003). Psychopathy scores predict adolescent
inpatient aggression. Assessment, 10, 1, 101-112.

Stalenheim, E.G., & von Knorring, L. (1996). Psychopathy and Axis I and II
psychiatric disorders in a forensic psychiatric population in Sweden. Acta
Psychiatrica Scandinavica, 94, 217-233.

Stanger, C., & Lewis, M. (1993). Agreement among parents, teachers and
children on internalizing and externalizing behavior problems. Journal of
Clinical Child Psychology, 22, 107-115.

Steiger, J.H., & Lind, J.C. (1980). Structural equation modeling and modification:
An interval estimation approach. Multivariate Behavioral Research, 25,
173-180.

137

Stevens, D., Charman, T., & Blair, R.J.R (2001). Recognition of emotion in facial
expressions and vocal tones in children with psychopathic tendencies.
Journal of Genetic Psychology, 162, 201-211.

Tabachnick, B.G., & Fidell, L.S. (2001). Using Multivariate Statistics (4
th
ed.).
Boston, MA: Allyn and Bacon.

Tucker, L.R., & Lewis, C. (1973). A reliability coefficient for maximum
likelihood factor analysis. Psychometrika.

Van Vorst, R.B. (1943). An evaluation of test performances of a group of
psychopathic delinquents. Psychological Bulletin, 40, 583.
Vasey, M.W., Kotov, R., Frick P.J., & Loney B.R. (2005). The latent structure of
psychopathy in youth: a taxometric investigation. Journal of Abnormal
Child Psychology, 33, 4, 411-429.

Velicer, W.F. & Jackson, D.N. (1990). Component analysis versus common-
factor-analysis – Some further observations. Multivariate Behavioral
Research, 25(1), 89-95.

Veltman, M.W.M., & Browne, K.D. (2001). Three decades of child maltreatment
research: Implications for the school years. Trauma, Violence, & Abuse, 2,
3, 215-239.

Verhulst, F.C., & Van der Ende, J. (1992). Agreement between parents' reports
and adolescents' self-reports of problem behavior. Journal of Child
Psychology and Psychiatry, 33, 1011-1023.

Viding, E., Blair, R.J.R., Moffitt, T.E., & Plomin, R. (2005). Strong genetic risk
for psychopathic syndrome in children. Journal of Child Psychology and
Psychiatry, 46, 592-597.

Vitacco, M.J., Neumann, C.S., & Jackson, R.L. (2005). Testing a four-factor
model of psychopathy and its association with ethnicity, gender,
intelligence and violence. Journal of Consulting and Clinical Psychology,
73, 3, 466-437.

Vitacco, M.J., Rogers, R., & Neumann, C.S. (2003). The Antisocial Process
Screening Device: An examination of its construct validity and criterion-
related validity. Assessment, 10, 2, 143-150.
138

Vitacco, M.J., Rogers, R., Neumann, C.S., Harrison, K., & Vincent, G.M. (2005).
A comparison of factor models on the PCL-R with mentally disordered
offenders: The development of a four-factor model. Criminal Justice and
Behavior, 32, 5, 526-545.

Vitale, J.E., Newman, J.P., Bates, J.E., Goodnight, J., Dodge, K.A., & Pettit, G.S.
(2005). Deficient behavioral inhibition and anomalous selective attention
in a community sample of adolescents with psychopathic traits and low-
anxiety traits. Journal of Abnormal Child Psychology, 33, 4, 461-470.

Walsh, A., Beyer, J.A., & Petee, R.A. (1987). Violent delinquency: An
examination of psychopathic typologies. Journal of Genetic Psychology,
148, 385-392.

Webster, C.D., Muller-Isberner, R., & Fransson, G. (2002). Violence risk
assessment: Using strucxtured clinical guids professionally. International
Journal of Forensic Mental Health, 1,43-51.

Wechlser, D. (1981). Manual for the Wechsler Adult Intelligence Scale – Revised.
New York: Psychological Corporation.

Widaman, K.F. (1985). Hierarchically nested covariance structure models for
multitrait–multimethod data. Applied Psychological Measurement, 9, 1-
26.

Williamson, S., Hare, R. D., & Wong, S. (1987). Violence: Criminal psychopaths
and their victims. Canadian Journal of Behavioral Science, 19, 454-462.

Woodcock, R.W. (1990). Theoretical foundations of the WJ-R measures of
cognitive ability. Journal of Psychoeducational Assessment, 8, 231-258.

Woodcock, R.W., & Johnson, M.B. (1989). Woodcock-Johnson Psycho-
Educational Battery – Revised. Chicago, IL: Riverside.

Woodcock, R.W., & Mather, N. (1989). WJ-R Tests of Cognitive Ability-
Standard and supplemental batteries: Examiners' manual. In R.W.
Woodcock & M.B. Johnson (Eds.), Woodcock-Johnson Psycho-
Educational Battery – Revised. Allen, TX: DLM Teaching Resources.

Yeudall, L.T., Fedora, O., & Fromm, D. (1987). A neuropsychological theory of
persistent criminality: Implications for assessment and treatment.
Advances in Forensic Psychology and Psychiatry, 2, 119-191.

139

Youngstrom, E., Loeber, R., & Stouthamer-Loeber, M. (2000). Patterns and
correlates of agreement between parent, teacher, and male adolescent
ratings of externalizing and internalizing problems. Journal of Consulting
and Clinical Psychology, 68, 6, 1038-1050.
140

APPENDIX A

The Antisocial Process Screening Device (Self-Report Version)

Instructions: Please read each statement and decide how well it describes you. Mark your
answer by circling the appropriate number (0-2) for each statement. Do not leave any
statement unrated.

Not at
all
True
Sometimes
True
Definitely
True
1. You blame others for your mistakes. 0 1 2
2. You engage in illegal activities. 0 1 2
3. You care about how well you do at
school/work.
0 1 2
4. You act without thinking of the consequences. 0 1 2
5. Your emotions are shallow and fake. 0 1 2
6. You lie easily and skillfully. 0 1 2
7. You are good at keeping promises. 0 1 2
8. You brag a lot about your abilities,
accomplishments, or possessions.
0 1 2
9. You get bored easily. 0 1 2
10. You use or “con” other people to get what you
want.
0 1 2
11. You tease or make fun of other people. 0 1 2
12. You feel bad or guilty when you do something
wrong.
0 1 2
13. You do risky or dangerous things. 0 1 2
14. You act charming and nice to get things you
want.
0 1 2
15. You get angry when corrected or punished. 0 1 2
16. You think you are better or more important
than other people.
0 1 2
17. You do not plan ahead, or you leave things
until the “last minute.”
0 1 2
18. You are concerned about the feelings of others. 0 1 2
19. You hide your feelings or emotions from
others.
0 1 2
20. You keep the same friends. 0 1 2
141

APPENDIX B
The RPQ Aggression Scale (Child Version)
There are times when most of us feel angry, or have done things we should not have
done. Rate each of the items below by putting a circle around either 0 (never), 1
(sometimes), or 2 (often). Don't spend a lot of time thinking about the items - just give
your first response. Make sure you answer all the items.
0 =NEVER
1 = SOMETIMES
2 = OFTEN
1. I yell at others when they annoy me 0 1 2
2. I fight others to show who is on top 0 1 2
3. I get angry when others annoy me 0 1 2
4. I take things from other kids 0 1 2
5. I get angry when frustrated 0 1 2
6. I damage or break things for fun 0 1 2
7. I have temper tantrums 0 1 2
8. I damage things when I am mad 0 1 2
9. I get into fights to be cool 0 1 2
10. I hurt others to win a game 0 1 2
11. I get angry or mad when I don’t get my way 0 1 2
12. I use force to get others to do what I want 0 1 2
13. I get angry or mad when I lose a game 0 1 2
14. I get angry when others threaten me 0 1 2
15. I use force to get money or things from others 0 1 2
16. I feel better after hitting or yelling at someone 0 1 2
17. I threaten and bully other kids 0 1 2
18. I make prank phone calls just for fun 0 1 2
19. I hit others to defend myself 0 1 2
20. I get others to gang up on other kids 0 1 2
21. I carry a weapon to use in a fight 0 1 2
22. I get mad or hit others when they tease me 0 1 2
23. I yell at others so they will do things for me 0 1 2
142

APPENDIX C

The Multidimensional Anxiety Scale for Children
This questionnaire asks you how you have been thinking, feeling, or acting recently. For each item, please circle
the number that shows how often the statement is true for you. If a sentence is true about you a lot of the time,
circle 3. If it is true about you some of the time, circle 2. If it is true about you once in a while, circle 1. If a
sentence is hardly ever true about you, circle 0. Remember, there are no right or wrong answers, just answer how
you have been feeling recently.
Here are two examples to show you how to complete the Never Rarely Sometimes Often
questionnaire. In Example A, if you were hardly ever scared of true true true true
dogs, you would circle 1,that the statement is rarely true about about about about about
you. In Example B, if thunderstorms sometimes upset you, you me me me me
would circle 2, meaning that the statement is sometimes true.
Example A I’m scared of dogs ............……………………………… 0 1 2 3
Example B Thunderstorms upset me ………………………………... 0 1 2 3
1. I feel tense or uptight ……………………………………… 0 1 2 3
2. I usually ask permission …………………………………… 0 1 2 3
3. I worry about other people laughing at me ………………... 0 1 2 3
4. I get scared when my parents go away ……………………. 0 1 2 3
5. I have trouble getting my breath …………………………... 0 1 2 3
6. I keep my eyes open for danger …………………………… 0 1 2 3
7. The idea of going away to camp scares me ……………….. 0 1 2 3
8. I get shaky or jittery ……………………………………….. 0 1 2 3
9. I try to stay near my mom or dad ………………………….. 0 1 2 3
10. I’m afraid that other kids will make fun of me ……………. 0 1 2 3
11. I try hard to obey my parents and teachers ………………... 0 1 2 3
12. I get dizzy or faint feelings ………………………………... 0 1 2 3
13. I check things out first …………………………………….. 0 1 2 3
14. I worry about getting called on in class …………………… 0 1 2 3
15. I’m jumpy …………………………………………………. 0 1 2 3
16. I’m afraid other people will think I’m stupid……………… 0 1 2 3
17. I keep the light on at night…………………………………. 0 1 2 3
18. I have pains in my chest …………………………………... 0 1 2 3
19. I avoid going to places without my family ………………... 0 1 2 3
20. I feel strange, weird, or unreal …………………………….. 0 1 2 3
21. I try to do things other people will like …………………… 0 1 2 3
22. I worry about what other people think of me…………….... 0 1 2 3
23. I avoid watching scary movies and TV shows ……………. 0 1 2 3
24. My heart races or skips beats ……………………………... 0 1 2 3
25. I stay away from things that upset me …………………….. 0 1 2 3
26. I sleep next to someone from my family ………………….. 0 1 2 3
27. I feel restless and on edge …………………………………. 0 1 2 3
28. I try to do everything exactly right ………………………... 0 1 2 3
29. I worry about doing everything exactly right……………..... 0 1 2 3
30. I get scared riding in the car or on the bus ……………….... 0 1 2 3
31. I feel sick to my stomach ………………………………….. 0 1 2 3
32. If I get upset or scared, I let someone know right away …... 0 1 2 3
33. I get nervous if I have to perform in public ……………….. 0 1 2 3
34. Bad weather, the dark, heights, animals, or bugs scare me... 0 1 2 3
35. My hands shake …………………………………………… 0 1 2 3
36. I check to make sure things are safe……………………….. 0 1 2 3
37. I have trouble asking other kids to play with me ………….. 0 1 2 3
38. My hands feel sweaty or cold ……………………………... 0 1 2 3
39. I feel shy …………………………………………………… 0 1 2 3

Abstract (if available)

Abstract The Antisocial Process Screening Device (APSD

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The relationship between pubertal timing and delinquent behavior in maltreated male and female adolescents

PDF

Examining the longitudinal relationships between community violence exposure and aggressive behavior among a sample of maltreated and non-maltreated adolescents

PDF

Attachment, maltreatment and autonomic nervous system responsivity as predictors of adolescent anxiety and depression

PDF

Construct validity of psychopathic personality traits in a cohort of young twins

PDF

The impact of child maltreatment on the mental health of adolescents: a longitudinal study of social anxiety and self-perception

PDF

Parents' attunement: relation to adolescent problem behavior and the moderating role of marital conflict

PDF

The development of electrodermal fear conditioning from ages 3 to 8 years and relationships with antisocial behavior at age 8 years

PDF

Shyness, attachment security, and cognitive abilities: measurement and relationships

PDF

Psychometric and biometric evaluation of psychopathic personality traits in 9 and 10 year old twins

PDF

Friendship as a protective factor in adolescence

PDF

Theoretical, experimental, device fabrication, and degradation studies of materials for optoelectronic devices

PDF

Evaluating a cultural process model of depression and suicidality in Latino adolescents

PDF

Numerical weak approximation of stochastic differential equations driven by Levy processes

PDF

Pushmi-pullyu and little c: a search for the structure of personal creativity in a general population

PDF

A multiple systems approach to examining physiological stress and its association with internalizing disorders in adolescence

PDF

Search, synthesis, and analysis for semiconductor device design

PDF

New materials and fabrication techniques and their utilization in organic photovoltaic devices

PDF

Patterns of EEG spectral power in 9-10 year old twins and their relationships with aggressive and nonaggressive antisocial behavior in childhood and adolescence

PDF

Trauma-related treatment gains among women with histories of interpersonal violence and co-occurring mental health and substance abuse disorders

PDF

The association between peer victimization and school functioning in Latino children

Asset Metadata

Creator Fung, Michelle Tien-Yee (author)

Core Title Multiple informant-rated psychopathy in young adolescents: a multitrait-multimethod investigation of the Antisocial Process Screening Device

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Psychology

Publication Date 06/19/2007

Defense Date 05/07/2007

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Antisocial Process Screening Device,child maltreatment,juvenile psychopathy,multitrait-multimethod,OAI-PMH Harvest,psychometric

Language English

Advisor Trickett, Penelope K. (committee chair), Manis, Franklin R. (committee member), Mennen, Ferol E. (committee member), Schwartz, David (committee member)

Creator Email mfung@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m536

Unique identifier UC168888

Identifier etd-Fung-20070619 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-507033 (legacy record id),usctheses-m536 (legacy record id)

Legacy Identifier etd-Fung-20070619.pdf

Dmrecord 507033

Document Type Dissertation

Rights Fung, Michelle Tien-Yee

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu