You are on page 1of 18

Psychological Assessment

A Comparison of Self-Report Measures of Psychopathy


Among Nonforensic Samples Using Item Response
Theory Analyses
Siny Tsang, Randall T. Salekin, C. Adam Coffey, and Jennifer Cox
Online First Publication, April 13, 2017. http://dx.doi.org/10.1037/pas0000481

CITATION
Tsang, S., Salekin, R. T., Coffey, C. A., & Cox, J. (2017, April 13). A Comparison of Self-Report
Measures of Psychopathy Among Nonforensic Samples Using Item Response Theory Analyses.
Psychological Assessment. Advance online publication. http://dx.doi.org/10.1037/pas0000481
Psychological Assessment © 2017 American Psychological Association
2017, Vol. 1, No. 2, 000 1040-3590/17/$12.00 http://dx.doi.org/10.1037/pas0000481

A Comparison of Self-Report Measures of Psychopathy Among


Nonforensic Samples Using Item Response Theory Analyses

Siny Tsang Randall T. Salekin, C. Adam Coffey, and


Columbia University Jennifer Cox
University of Alabama

This study investigated how well components of the psychopathy trait are measured among college
students with the Levenson Self-Report Psychopathy Scale (LSRP), the Personality Assessment
Inventory–Antisocial Features Scale (PAI ANT), the Psychopathic Personality Inventory–Short Form
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

(PPI-SF), and the Self-Report Psychopathy Scale-II (SRP-II). Using Samejima (1969)’s graded response
This document is copyrighted by the American Psychological Association or one of its allied publishers.

model (GRM), the subscales were found to vary in their ability to measure the corresponding latent traits.
The LSRP primary psychopathy factor is more precise in measuring the latent trait than the secondary
psychopathy factor. The PAI ANT items show coherent psychometric properties, whereas the PPI-SF
factors differ in their precision to measure the corresponding traits. The SRP-II factors are effective in
discriminating among individuals with varying levels of the latent traits. Results suggest that multiple
self-report measures should be used to tap the multidimensional psychopathy construct. However, there
are concerns with respect to using negatively worded items to assess certain aspects of psychopathy.

Public Significance Statement


This study examines the item properties of 4 self-report measures of psychopathy among college
students, with results showing that the factors in the instruments differ in their ability to measure the
respective latent traits. Findings suggest that multiple measures should be used to tap the multidi-
mensional concept of psychopathy, and raise concerns regarding the use of negatively worded items
to assess certain aspects of psychopathy.

Keywords: psychopathy, item response theory, self-report psychopathy measure

Supplemental materials: http://dx.doi.org/10.1037/pas0000481.supp

Conceptualized as a personality disorder with distinctive Neumann, Harrison, & Vincent, 2005), which reincorporated an-
interpersonal-affective and behavioral characteristics, the majority tisocial content. In addition to the controversies of the number of
of research on the latent trait of psychopathy was conducted dimensions underlying the construct of psychopathy (Neumann,
among forensic samples using the Psychopathy Checklist (PCL; Kosson, & Salekin, 2007), concerns have been raised regarding
Hare, 1980) and its revision, the Psychopathy Checklist-Revised whether antisocial behavior should be included as a core feature of
(PCL-R; Hare, 1991, 2003). Earlier factor analytic studies of the the latent psychopathy trait (Hare & Neumann, 2010; Skeem &
PCL and the PCL-R among adult samples suggested the latent Cooke, 2010).
psychopathy trait is underpinned by a two-factor structure, reflect- In response to the burgeoning interest in the study of the
ing interpersonal/affective traits and antisocial behavioral charac- psychopathy trait among nonforensic populations (e.g., commu-
teristics (e.g., Hare, 1980, 1991; Harpur, Hare, & Hakstian, 1989). nity, college), a number of self-report psychopathy measures have
However, subsequent research has suggested a three-factor model
been developed, including the Levenson Self-Report Psychopathy
that excluded the antisocial behavior items (Cooke & Michie,
Scale (LSRP; Levenson, Kiehl, & Fitzpatrick, 1995), the antisocial
2001), and a four-factor model (Hare, 2003; Vitacco, Roger,
features subscale in the Personality Assessment Inventory (PAI
ANT; Morey, 2007), the Psychopathic Personality Inventory–
Short Form (PPI-SF; Lilienfeld, 1990), and the Self-Report
Psychopathy-II (SRP-II; Hare, Harpur, & Hemphill, 1989). As
Siny Tsang, Department of Epidemiology, Columbia University; Ran- self-report measures of psychopathy become more widely avail-
dall T. Salekin, C. Adam Coffey, and Jennifer Cox, Department of Psy- able, they are increasingly used in research studies. In fact, among
chology, University of Alabama.
the psychopathy studies published between 2000 and February
This study was partially supported by the research training grant
5-T32-MH 13043 from the National Institute of Mental Health.
2014, more than half of the studies did not include the PCL-R
Correspondence concerning this article should be addressed to Siny and/or its variants in their abstracts or keywords, suggesting that
Tsang, Department of Epidemiology, Columbia University, New York, NY self-report psychopathy measures may have been used instead of
10032. E-mail: siny.tsang@columbia.edu the PCL-R and its variants (Kelsey, Rogers, & Robinson, 2014). A
1
2 TSANG, SALEKIN, COFFEY, AND COX

brief review of the four self-report psychopathy measures is pro- PCL-R and PPI scores, though the correlations were mainly driven
vided below. by the antisocial/lifestyle factor (Edens, Hart, Johnson, Johnson, &
Olver, 2000; Edens, Poythress, & Watkins, 2001). In addition, the
ANT subscales were found to be associated with other self-report
Self-Report Measures of Psychopathy
psychopathy measures among community samples (Benning, Pat-
The 26-item LSRP was developed to measure psychopathy in rick, Salekin, & Leistico, 2005; Salekin, Trobst, & Krioukova,
noninstitutionalized samples; two factors were generated from 2001).
initial factor analysis, representing primary and secondary psy- Lilienfeld (1990) developed the Psychopathic Personality In-
chopathy (Levenson et al., 1995). Primary psychopaths are posited ventory (PPI) to measure central psychopathy characteristics in
to be manipulative, callous, and extremely selfish, whereas sec- nonclinical samples. The 187 self-report items assess eight factor-
ondary psychopaths tend to be impulsive and engage in antisocial analytically developed domains of psychopathy: (a) blame exter-
behavior (Karpman, 1948). However, later studies were unable to nalization, (b) social potency, (c) Machiavellian egocentricity, (d)
replicate the original two-factor structure of the LSRP without fearlessness, (e) coldheartedness, (f) impulsive nonconformity, (g)
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

modification (Brinkley, Schmitt, Smith, & Newman, 2001; Ly- stress immunity, and (h) carefree nonplanfulness. Selecting the
This document is copyrighted by the American Psychological Association or one of its allied publishers.

nam, Whiteside, & Jones, 1999). Recently, a three-factor structure seven items that had the highest factor loadings on each of the
proposed to measure egocentricity, callousness, and antisocial eight dimensions, an abbreviated version of the PPI, the 56-item
features, using 19 of the 26 LSRP items, was found to fit well PPI-SF, was constructed (see Lilienfeld, 1990). Subsequent studies
among samples of incarcerated inmates (Brinkley, Diamond, showed that the eight PPI-SF domains can be combined into two
Magaletta, & Heigel, 2008; Sellbom, 2011), college students higher order dimensions paralleling the two-factor structure of the
(Brinkley et al., 2008; Salekin, Chen, Sellbom, Lester, & Mac- PCL-R, however, the PPI-SF two-factor models derived have not
Dougall, 2014; Sellbom, 2011), and community participants always been consistent (e.g., Benning, Patrick, Hicks, Blonigen, &
(Somma, Fossati, Patrick, Maffei, & Borroni, 2014). Despite Krueger, 2003; Lilienfeld, 1990; Smith, Edens, & Vaughn, 2011;
promising model fit indices, the convergent and discriminant va- Wilson, Frick, & Clements, 1999). Although the PPI-SF was found
lidity of the three-factor model remains questionable, whereas the to be highly correlated with the original version with adequate to
convergent and discriminant validity of the two-factor model holds high internal consistencies (Lilienfeld & Hess, 2001), concerns
some value (Salekin et al., 2014). have been raised regarding the item selection process for the
To the best of our knowledge, only two published studies have PPI-SF. For example, the underlying latent trait may be underrep-
examined the psychometric properties of the LSRP with an item resented by the abbreviated scales (Smith, McCarthy, & Anderson,
response theory (IRT) framework. Gummelt, Anestis, and Carbo- 2000), and rarely endorsed items that are highly discriminating at
nell (2012) found that the LSRP items vary in their ability to either ends of the latent trait continuum may be excluded (Tonnaer,
distinguish among undergraduate students with varying levels of Cima, Sijtsma, Uzieblo, & Lilienfeld, 2012).
psychopathy, with more than half of the items demonstrating Adapted from the first version of the Self-Report Psychopathy
differential item functioning (DIF) between genders. Using a Bra- Scale (SRP; Hare et al., 1989), the 60-item SRP-II was revised to
zilian community sample, Hauck-Filho and Teixeira (2014) eval- parallel the two factors in the PCL-R (Hare, 1991), with items
uated the extent to which items in the primary and secondary designated to measure interpersonal/affective features (nine items),
psychopathy subscales fit the Unidimensional Rating Scale model social deviance characteristics (13 items), and both domains (nine
(RSM; Andrich, 1978). Results showed support for the two-factor items). However, there is limited research examining the factor
model in this sample, with minimal DIF between genders. How- structure of the SRP-II in community populations, and among
ever, the authors did not further explore whether certain items are those, results have been mixed. Although some studies identified a
better or worse at discriminating respondents with varying levels two-factor structure of the SRP-II, these models failed to mirror
of the latent trait. the structure of the PCL-R from which the SRP-II was derived
The Personality Assessment Inventory (PAI; Morey, 1991, (e.g., Benning et al., 2005; Williams & Paulhus, 2004). After
2007) is a multiscale self-report instrument designed to measure an making substantial revisions to the SRP-II, a four-factor solution
array of psychopathological constructs. Among the 11 PAI clinical that parallels the four-factor model of the PCL-R was proposed
scales, the Antisocial Features Scale (ANT) consists of items that (Williams, Paulhus, & Hare, 2007). More recently, Lester, Salekin,
assess features of antisocial personality, psychopathy, and crimi- and Sellbom (2013) proposed a rationally derived four-factor
nality. Developed using Cleckley (1941) and Hare’s (1980, 1991) model from exploratory and confirmatory factor analyses. This
conceptualizations of the disorder, the PAI ANT emphasizes the 36-item four-factor SRP-II maps well onto the PCL-R’s four-
personological features of the disorder (Morey, 2003), making the factor structure, assessing features of interpersonal, disinhibition/
scale a favorable choice when assessing psychopathic characteris- impulsivity, fearlessness, and coldheartedness. Although a third
tics in a nonoffending population. As with all the PAI subscales, edition of the SRP has been developed (Paulhus, Neumann, &
the ANT items were written at a relatively low reading level Hare, in press), this revision focused on adding items to assess
(fourth grade), making it appealing for use in a wide variety of antisocial behavior (Williams, Nathanson, & Paulhus, 2003),
settings (Morey, 1996), especially when assessing psychopathy which is less relevant in nonforensic samples. The SRP-II has been
traits among community samples. Although there is limited re- found to remain useful in tapping the Cleckley psychopathic
search evaluating the ANT as a measure of psychopathy, the scale personality among nonforensic samples (Lester, Salekin, & Sell-
was found to have adequate psychometric properties (Morey, bom, 2013).
2007; Siefert, Sinclair, Kehl-Fie, & Blais, 2009). Among offend- Collectively, the majority of the research on self-report psy-
ers, ANT was reported to be positively associated with the total chopathy measures has focused on evaluating how well the self-
SELF-REPORT PSYCHOPATHY IRT 3

report instruments map onto the factor structures of the PCL-R, Method
and investigating whether correlates of psychopathic traits are
similar among forensic and nonforensic populations. Thorough Participants
examination of how individual items function within these self-
report measures of psychopathy has been lacking. With the excep- The current study is a secondary data analysis of data collected
tion of one study investigating the item properties of the LSRP from 1,257 (378 men, 869 women) undergraduate students en-
(Hauck-Filho & Teixeira, 2014), we are unaware of any published rolled in a southeastern university in the United States (see Lester
IRT studies that examined the psychometric properties of the PAI et al., 2013), with ages between 17- and 51-years-old (M ⫽ 19.3,
ANT, PPI-SF, or SRP-II. Thus, whether self-report items assessing SD ⫽ 2.3). Most participants self-identified as Caucasians
psychopathy among community populations are able to assess the (81.54%), with 10.90% African Americans, and 5.01% of other
different components of the psychopathy trait remains unknown. racial/ethnic groups. Participants received course credit in ex-
As these psychopathy measures rely on individuals’ self-reports, change for their participation in the initial study. The study was
rather than trained clinicians’ assessments, it is critical to investi- approved by the Institutional Review Board at the university (IRB
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

gate if some items are more sensitive to varying levels of the latent #02-JCH-037).
This document is copyrighted by the American Psychological Association or one of its allied publishers.

trait than others, and whether certain items are more frequently
endorsed than others. Measures
Levenson Self-Report Psychopathy Scale (LSRP). The
LSRP (Levenson et al., 1995) is a 26-item instrument modeled
The Current Study
after the PCL-R items; each item was scored on a 4-point ordinal
In the present study, we utilized data from a large undergraduate scale (1 ⫽ disagree strongly, 2 ⫽ disagree somewhat, 3 ⫽ agree
student sample in which participants completed four self-report somewhat, 4 ⫽ agree strongly). Initial factor analysis showed that
measures of psychopathy to explore how well the underlying trait the LSRP consists of 16 items measuring the manipulative and
of psychopathy is measured by each self-report instrument. Using uncaring attitude of primary psychopathy (PP), and 10 items
an item response theory (IRT) approach, we attempted to answer assessing impulsivity and antisocial lifestyle of secondary psy-
two main questions: chopathy (SP).1
Personality Assessment Inventory-Antisocial Features Scale
1. Assuming the subscales within each instrument are mea- (PAI ANT). The 24-item PAI ANT Scale (Morey, 2007) can be
suring components of the underlying psychopathy trait, divided into three factors of eight items each, assessing antisocial
are the different domains of psychopathy measured behavior (A), egocentricity (E), and stimulus seeking (S). Each
equally well by the different subscales within an instru- item was rated on a 4-point scale (0 ⫽ false/not at all true, 1 ⫽
ment? somewhat true, 2 ⫽ mainly true, 3 ⫽ very true).
Psychopathic Personality Inventory-Short Form (PPI-SF).
2. As the four self-report instruments were developed to The 56-item PPI-SF consists of items that have the highest factor
assess characteristics indicative of the latent psychopathy loadings on the original PPI (Lilienfeld, 1990). The PPI-SF is
trait, are the latent constructs measured with different comprised of eight 7-item factors: blame externalization (BE),
self-report instruments similar to one another? social potency (SP), Machiavellian (ME), fearlessness (F), cold-
heartedness (C), impulsive nonconformity (IN), stress immunity
To answer the first question, we examined (a) whether the items
(SI), and carefree nonplanfulness (CN). Each item was scored on
in different subscales (within the same instrument) are similarly
a 4-point scale (1 ⫽ false, 2 ⫽ mostly false, 3 ⫽ mostly true, 4 ⫽
discriminating; and (b) the range of the latent trait in which the
true).
items were positively endorsed. An optimal instrument for assess-
Self-Report Psychopathy Scale-II (SRP-II). The SRP-II
ing the features indicative of the underlying psychopathy trait
(Hare et al., 1989) includes a total of 60 items, in which each item
should mostly consist of items that are discriminating, as well as
was scored on a 7-point scale (1 ⫽ disagree strongly, 2 ⫽ disagree
items that are endorsed in a wide range on the latent trait contin-
moderately, 3 ⫽ disagree slightly, 4 ⫽ neutral, 5 ⫽ agree slightly,
uum. We aim to investigate whether certain subscales in the
6 ⫽ agree moderately, 7 ⫽ agree strongly). A rationally derived
self-report psychopathy instruments are better at assessing features
36-item four-factor model (Lester et al., 2013) was adopted in this
associated with the psychopathy trait, and whether the use of
study. The four factors are interpersonal (IP ⫽ 16 items), disinhi-
subscales can tap a broad range of psychopathic tendencies among
bition/impulsivity (DI ⫽ nine items), fearlessness (F ⫽ five items),
nonforensic samples.
and coldheartedness (C ⫽ six items).
The second question is addressed by examining the extent to
which the subscales (within and between different instruments) are
IRT Model
associated with one another. For respondents who completed all
four self-report measures, we explored such relations using (a) the Considering that ordinal response categories are used in all four
total scores on each subscale computed by summing the item self-report measures of psychopathy, Samejima’s (1969) graded
scores in each corresponding subscale, and (b) the underlying
latent trait estimated from the item response theory analysis for 1
Due to technical error, one item measuring secondary psychopathy
each subscale. We expected moderate to strong correlations be- (Item 17: “I find myself in the same kinds of trouble, time after time”) was
tween subscales from different instruments that are assessing sim- not included in this dataset. However, our findings with respect to the
ilar domain. LSRP would likely not be different if the missing item had been included.
4 TSANG, SALEKIN, COFFEY, AND COX

response model (GRM) was used. In the GRM, each item i with underlying assumption that the residuals of the estimated IRT
mi ⫹ 1 response categories is described by a slope parameter (␣i) parameters follow a normal distribution.
and j ⫽ 1 . . . mi category threshold parameters (␤ij). The estimated The differences in measurement precision across measures are
parameters can be represented graphically as category response explored through inspection of the TIFs and TICs. We examined
curves (CRCs), where the number of CRCs corresponds to the the amount of information provided by different measures, as well
number of response categories for each item i. The ␣i parameter as the ranges of the latent trait in which various instruments are
can be considered as a discrimination parameter; it reflects how better at assessing the respective latent traits.
quickly the expected item score varies as a function of the latent
trait level (Embretson & Reise, 2000). An item with a larger ␣ has Results
more narrow and peaked CRCs, whereas the CRCs of an item with
a smaller ␣ are flatter. The category thresholds (␤ij parameters)
Descriptive Statistics
indicate the underlying trait levels needed to respond above the
corresponding thresholds with 50% probability, which are re- Descriptive statistics of the self-report measures are presented in
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

flected by the intersection points between two consecutive CRCs Table 1. For each of the self-report psychopathy measures in this
This document is copyrighted by the American Psychological Association or one of its allied publishers.

for item i, at which the probability of the next response category study, a total score can be computed, with higher scores represent-
(mi) becomes equally likely as the previous category (mi-1). ing higher levels of psychopathy traits. Because these self-report
The probability that a respondent’s score (x) on item i is at or measures of psychopathy were initially developed with varying
above a category threshold (j ⫽ 1. . . mi), conditional on the latent factor structures to tap different components in the psychopathy
trait level (␪), is computed as construct, factor scores can also be generated by summing the
items in each factor, reflecting strengths of endorsement in differ-
exp关␣i(␪ ⫺ ␤ij)兴
*
Pix(␪) ⫽ , ent domains of the psychopathy measures. This information can
1 ⫹ exp[␣i(␪ ⫺ ␤ij)] also be located in Table 1.
where x ⫽ j ⫽ 1 . . ., mi. The probabilities of a respondent scoring
in each response category conditional on the latent trait level are Unidimensionality
estimated separately as
It should be noted that evaluating the factor structures of the
*
Pix(␪) ⫽ Pix
*
(␪) ⫺ Pi(x⫹1)
*
(␪). self-report measures was not the primary goal of the present study.
Although the self-report measures are all assumed to assess the
As the LSRP, PAI ANT, and PPI-SF are scored on a 4-point underlying psychopathy construct, all four measures have been
ordinal scale, three ␤ij parameters are estimated for each item in proposed to consist of correlated factor structures. Confirmatory
these measures. Six ␤ij parameters are estimated for each of the factor analyses (CFAs) were used to examine whether the unidi-
SRP-II items, corresponding to the seven response categories. All mensional models are better fit than the theoretically driven factor
GRM analyses were performed using Mplus version 7.4 (Muthén models. For the LSRP, PAI ANT, and PPI-SF, the original factor
& Muthén, 1998 –2012). structures were tested. The factor structure of the SRP-II was
In IRT, the precision of measurement at different levels of the latent previously examined with the sample used in the current study
trait can be estimated with information functions (Embretson, 1996). (Lester et al., 2013); neither Hare et al.’s (1989) original two-factor
Conceptually, item information function can be computed by taking model nor Benning, Patrick, Salekin, and Leistico (2005); 23-item
the inverse of the square of the standard error of estimate (Baker, two-factor model was found to achieve adequate fit in this sample.
1992; Nunnally & Bernstein, 1994). The sum of the corresponding Thus, Lester et al.’s (2013) rationally derived 36-item four-factor
item information functions in a scale constitutes a test information model was used to test against a unidimensional SRP-II model.
function (TIF), which reflects the collective measurement precision Across all four self-report measures, the factor models showed
provided by the items that make up the scale (Bolt, Hare, Vitale, & better fit than the unidimensional model; model fit indices are
Newman, 2004). The TIFs can also be graphically presented as test presented in supplementary material Appendix 1, Table 1.
information curves (TICs), where the amount of information provided We further examined the unidimensionality of each factor
by a scale is plotted against the latent trait continuum. Across the with the self-report measures using two criteria: (a) principal
latent trait continuum, locations in which the scale is precise in component analyses (PCAs) with polychoric correlations, and
measuring the latent construct are reflected by high TICs, whereas the (b) the fit of a one-factor model to the polychoric correlation
TICs are low in locations where the scale is not very precise. A TIC matrix. The first three eigenvalues (␭1, ␭2, and ␭3) from the
with a narrow peak suggests that the scale is precise, or able to PCAs are shown in Table 2. Using the UNIDI index: (␭1 ⫺
distinguish respondents above or below a certain latent trait level, but ␭2)/(␭2 ⫺ ␭3) ⬎ 5 (Martínez Arias, 1995, p. 297), all but three
not as precise at the extremes. On the other hand, a TIC that is smooth of the factors (LSRP’s SP, ANT-E, and PPI-SF’s CN) showed
and flat indicates that the scale is able to measure respondents across good support of unidimensionality. One-factor models were fit
a wide range of latent trait level with similar precision. using the diagonally weighted least squares estimator in the
“lavaan” package (Rosseel, 2012). Goodness of fit was assessed
using comparative fit index (Bentler, 1990), Tucker-Lewis in-
Comparisons Across Measures
dex (Tucker & Lewis, 1973), and root-mean-square error of
Mann–Whitney U tests and Kruskal-Wallis H tests are used to approximation index (Steiger, 1990). With the exception of
compare the estimated IRT parameters (␣is and ␤ijs) across the three factors (LSRP’s SP, ANT-E, and PPI-SF’s CN), all factors
self-report measures. Nonparametric tests are used as there is no achieved good fit on at least two of the fit indices. Subsequent
SELF-REPORT PSYCHOPATHY IRT 5

Table 1
Descriptive Statistics of Four Self-Report Measures of Psychopathy and Their Subscales

Scale Measure Items Mean (SD) Min Max

LSRP Primary psychopathy 16 29.44 (8.07) 16 60


Secondary psychopathyⴱ 9 19.77 (4.46) 9 35
Total 25 49.20 (10.74) 25 89
PAI ANT Antisocial behavior 8 7.53 (5.22) 0 24
Egocentricity 8 5.49 (3.80) 0 24
Stimulus seeking 8 8.18 (4.72) 0 24
Total 24 21.23 (11.33) 0 71
PPI-SF Blame externalization 7 14.60 (4.48) 7 28
Social potency 7 20.15 (4.18) 7 28
Machiavellian 7 14.94 (3.64) 7 27
Fearlessness 7 15.64 (5.04) 7 28
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Coldheartedness 7 13.84 (3.19) 7 27


This document is copyrighted by the American Psychological Association or one of its allied publishers.

Impulsive nonconformity 7 13.80 (3.69) 7 27


Stress immunity 7 18.15 (4.28) 7 28
Carefree nonplanfulness 7 12.89 (3.12) 7 24
Total 56 124.04 (14.88) 84 187
SRP-II Interpersonal 16 62.57 (15.72) 20 109
Disinhibition/Impulsivity 9 22.38 (9.51) 9 61
Fearlessness 5 19.82 (6.54) 5 35
Coldheartedness 6 14.55 (6.37) 6 36
Total 36 119.31 (28.52) 45 233
Note. LSRP ⫽ Levenson Self-Report Psychopathy Scale; PAI ANT ⫽ Personality Assessment Inventory—
Antisocial Features Scale; PPI-SF ⫽ Psychopathic Personality Inventory-Short Form; SRP-II ⫽ Self- Report
Psychopathy Scale-II.

Item 17: “I find myself in the same kinds of trouble, time after time” is missing from the dataset.

IRT analyses were conducted separately for each of the factors results for ANT-E and PPI-SF’s CN are included in the sup-
in the corresponding instrument. Adopting a conservative ap- plementary material (Appendix 1, Tables 2, and 3) for refer-
proach, we refrained from conducting IRT analyses on the ences only. However, IRT results for LSRP’s SP are presented
subscales that did not meet the unidimensionality criteria. IRT for comparison with LSRP’s PP.

Table 2
Results From PCA and CFA Each of the Four Self-Report Measures of Psychopathy

Scale ␭1 ␭2 ␭3 UNIDI CFI TLI RMSEA WRMR

LSRP
Primary psychopathy 5.05 1.42 .98 8.16 .97 .96 .07 2.07
Secondary psychopathy 2.36 1.33 1.09 4.11 .83 .77 .13 2.91
PAI ANT
Antisocial behavior 3.14 1.08 .85 8.88 .96 .90 .10 3.15
Egocentricity 2.53 1.28 .95 3.90 .91 .88 .13 2.82
Stimulus seeking 3.32 1.10 .84 8.45 .99 .99 .06 1.49
PPI-SF
Blame externalization 3.41 .78 .75 76.78 .99 .99 .08 1.61
Social potency 3.15 .93 .75 12.36 .97 .96 .12 2.38
Machiavellian 2.54 .97 .93 36.17 .97 .96 .07 1.50
Fearlessness 2.95 .96 .74 9.06 .99 .99 .05 1.20
Coldheartedness 2.03 1.06 .96 9.78 .93 .90 .07 1.50
Impulsive nonconformity 2.33 1.06 .97 13.78 .94 .91 .08 1.75
Stress immunity 3.15 .83 .79 68.00 .99 .99 .05 1.13
Carefree nonplanfulness 2.26 1.25 .95 3.46 .88 .81 .14 2.92
SRP-II
Interpersonal 4.92 1.45 1.10 9.73 .96 .95 .08 2.06
Disinhibition/Impulsivity 3.30 .98 .96 139.75 .99 .99 .05 1.08
Fearlessness 2.21 .86 .84 64.91 .99 .99 .06 .75
Coldheartedness 2.74 .86 .75 17.30 1.00 .99 .05 .82
Note. ␭1, ␭2, ␭3 ⫽ first three eigenvalues of the polychoric correlation matrix from principal component analyses (PCAs); UNIDI ⫽ UNIDI index, where
UNIDI index ⬎ 5 shows good support of unidimensionality (Martínez Arias, 1995, p. 297); CFI ⫽ comparative fit index; TLI ⫽ Tucker-Lewis Index;
RMSEA ⫽ root-mean-square error of approximation; WRMR ⫽ weighted root-mean-square residual; LSRP ⫽ Levenson Self-Report Psychopathy Scale;
PAI ANT ⫽ Personality Assessment Inventory—Antisocial Features Scale; PPI-SF ⫽ Psychopathic Personality Inventory-Short Form; SRP-II ⫽ Self-
Report Psychopathy Scale-II.
6 TSANG, SALEKIN, COFFEY, AND COX

Estimated Item Parameters with different levels of latent trait traits, with similar frequencies at
which the items are endorsed.
LSRP. Item parameters for the LSRP were estimated sepa-
PPI-SF. The estimated item parameters of the eight factors
rately for the PP and SP factors (see Table 3). Items assessing
in the PPI-SF are presented in Table 5. On average, the esti-
primary psychopathy had, on average, higher ␣ parameters, than
mated ␣ parameters vary across the seven factors, H (6) ⫽
those assessing secondary psychopathy (U ⫽ 120, p ⫽ .007). The
15.05, p ⫽ .02. The ␣ parameters were higher for items loading
higher ␣ parameters in the PP factor suggest that primary psy-
on BE and SP than those on ME, C, and IN; items in the SI and
chopathy features are more discriminating than characteristics
F factors had higher ␣ parameters than those in the C factor.
associated with secondary psychopathy. That is, items assessing
There were also differences in the estimated ␤ parameters
primary psychopathy features are better at differentiating between
individuals with varying levels of psychopathic traits than items across the eight factors, H (6) ⫽ 28.47, p ⬍ .001; H (6) ⫽
measuring secondary psychopathy features. Items in the PP factor 29.86, p ⬍ .001; H (6) ⫽ 23.64, p ⬍ .001, for ␤1, ␤2, and ␤3,
had, on average, higher ␤ parameters than those in the SP factor respectively. Items in the SP and SI factors had, on average,
(U ⫽ 118, p ⫽ .008; U ⫽ 119, p ⫽ .007; U ⫽ 133, p ⬍ .001, for lower ␤1 parameters, indicating lower category thresholds, than
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

␤1, ␤2, and ␤3, respectively). With higher category thresholds, items in the other factors. On the other hand, items in the C and
This document is copyrighted by the American Psychological Association or one of its allied publishers.

items assessing primary psychopathy are rarely endorsed, espe- ME factors had, on average, higher ␤3 parameters, than those in
cially among individuals with lower levels of the latent trait. On the other factors. Taken together, results suggest that social
the other hand, items tapping secondary psychopathy are fre- potency and stress immunity features are frequently endorsed,
quently endorsed even among respondents with low levels of the whereas characteristics indicative of coldheartedness and Ma-
latent trait. chiavellian are rarely endorsed among these participants.
PAI ANT. Two GRMs were performed to estimate the item SRP-II. The estimated item parameters of the four factors in
parameters for the ANT-A and ANT-S factors of the PAI (see the SRP-II are presented in Table 6. As the SRP-II items consist of
Table 4). There were no differences in the averaged item param- seven response categories, six ␤ parameters were estimated to
eters between different factors, H (1) ⫽ 0.04, p ⫽ .83; H (1) ⫽ indicate the threshold for each transition between response cate-
0.89, p ⫽ .34; H (1) ⫽ 0.54, p ⫽ .46; H (1) ⫽ 1.10, p ⫽ .29, for gories. There were no differences between items from different
␣, ␤1, ␤2, and ␤3, respectively. Results suggest that the PAI ANT factors with respect to the ␣ parameters, H (3) ⫽ 4.00, p ⫽ .26,
items in the two factors are similar at differentiating individuals suggesting items loaded on different factors are similar in discrim-

Table 3
LSRP Estimated Parameters From GRM

Factor Item ␣ ␤1 ␤2 ␤3

Primary psychopathy 1 1.14 (.07) ⫺1.24 (.08) .36 (.07) 2.53 (.11)
2 2.74 (.17) .46 (.12) 2.99 (.17) 5.53 (.27)
3 2.62 (.16) .02 (.11) 2.48 (.15) 5.14 (.25)
4 1.71 (.11) .23 (.09) 1.88 (.11) 3.75 (.16)
5 1.16 (.08) ⫺.94 (.08) .34 (.07) 2.74 (.12)
6 1.16 (.08) ⫺.26 (.07) 1.43 (.08) 3.33 (.14)
7 1.25 (.08) ⫺.20 (.07) 1.58 (.09) 3.39 (.14)
8 1.12 (.07) ⫺1.24 (.08) .18 (.07) 2.32 (.10)
9 1.68 (.10) ⫺.47 (.09) 1.55 (.10) 4.14 (.18)
10 .87 (.07) ⫺.17 (.07) 1.47 (.08) 3.07 (.13)
11 1.12 (.08) ⫺.38 (.07) .96 (.08) 2.66 (.11)
12 1.05 (.08) ⫺.68 (.07) 2.38 (.11) 3.50 (.15)
13 1.56 (.11) 1.39 (.10) 2.85 (.14) 4.22 (.20)
14 1.12 (.09) 1.11 (.08) 2.37 (.11) 3.57 (.16)
15 1.03 (.08) ⫺.23 (.07) 1.33 (.08) 3.18 (.13)
16 1.06 (.08) .23 (.07) 1.59 (.09) 3.37 (.14)
AIC ⫽ 41,826; BIC ⫽ 42,155 (Chi-square test not available)

18 .72 (.08) ⫺1.93 (.09) ⫺.52 (.07) 1.17 (.07)


Secondary psychopathy 19 .56 (.08) ⫺.76 (.07) 1.25 (.07) 2.86 (.12)
20 .88 (.10) ⫺1.29 (.08) .20 (.07) 1.99 (.10)
21 1.39 (.15) ⫺1.84 (.12) .44 (.08) 2.68 (.16)
22 .91 (.08) ⫺.93 (.07) .57 (.07) 2.26 (.10)
23 .63 (.08) ⫺1.10 (.07) 1.02 (.07) 2.65 (.11)
24 1.01 (.14) ⫺.79 (.08) .62 (.07) 2.17 (.12)
25 1.17 (.15) ⫺.78 (.08) .65 (.08) 2.39 (.14)
26 .73 (.08) .24 (.06) 1.26 (.08) 2.53 (.11)
AIC ⫽ 28,151; BIC ⫽ 28,346; ␹2 (26,1624) ⫽ 24,361, p ⫽ 1.00
Note. ␣ ⫽ estimated slope parameters; ␤1, ␤2, and ␤3 ⫽ estimated category threshold parameters. Standard
errors of the estimated parameters are presented in parentheses.
SELF-REPORT PSYCHOPATHY IRT 7

Table 4
PAI ANT Estimated Parameters From GRM

Factor Item ␣ ␤1 ␤2 ␤3

Antisocial behavior 1 1.25 (.10) .23 (.07) 2.12 (.11) 4.04 (.19)
2 1.65 (.12) .41 (.09) 2.16 (.12) 3.07 (.15)
3 1.83 (.12) ⫺2.70 (.14) .12 (.09) 1.19 (.10)
4 1.29 (.09) ⫺1.27 (.09) 1.10 (.08) 2.14 (.10)
5 1.21 (.09) ⫺.49 (.07) 1.73 (.09) 2.94 (.13)
6 1.22 (.11) 1.60 (.10) 1.87 (.11) 2.19 (.12)
7 1.33 (.10) .57 (.08) 1.20 (.09) 1.93 (.10)
8 1.42 (.10) ⫺.03 (.08) .78 (.08) 1.79 (.10)
AIC ⫽ 20,637; BIC ⫽ 10,801; ␹2 (65,373) ⫽ 12,468, p ⫽ 1.00

Stimulus seeking 17 3.07 (.18) ⫺.27 (.12) 3.24 (.19) 5.13 (.26)
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

18 5.96 (.70) ⫺.73 (.23) 4.96 (.57) 8.48 (.91)


This document is copyrighted by the American Psychological Association or one of its allied publishers.

19 2.07 (.12) ⫺.84 (.10) 1.92 (.12) 3.42 (.16)


20 .73 (.07) .37 (.07) 1.80 (.09) 2.97 (.13)
21 .70 (.08) 1.16 (.07) 2.20 (.10) 3.01 (.13)
22 .82 (.07) ⫺2.00 (.09) ⫺.21 (.07) .94 (.07)
23 1.44 (.09) ⫺.53 (.08) 1.70 (.10) 3.31 (.14)
24 1.15 (.08) ⫺2.37 (.11) ⫺1.04 (.08) .98 (.08)
AIC ⫽ 20,665; BIC ⫽ 20,830; ␹2 (65,379) ⫽ 10,410, p ⫽ 1.00
Note. ␣ ⫽ estimated slope parameters; ␤1, ␤2, and ␤3 ⫽ estimated category threshold parameters. Standard
errors of the estimated parameters are presented in parentheses. Results for the egocentricity subscale is not
available as the subscale did not meet the unidimensional assumption for IRT analyses.

inating across individuals with varying trait levels. With respect to measurement precision decreases outside the range. Overall, the
the ␤ parameters, there were substantial differences between items TIFs of the PP factor were higher than those of the SP factor
loaded on different factors, H (3) ⫽ 22.84, p ⬍ .001; H (3) ⫽ across the latent trait continuum, meaning that the latent pri-
24.98, p ⬍ .001; H (3) ⫽ 25.75, p ⬍ .001; H (3) ⫽ 25.38, p ⬍ mary psychopathy trait is measured with greater precision than
.001; H (3) ⫽ 22.77, p ⬍ .001; H (3) ⫽ 16.41, p ⬍ .001, for ␤1, characteristics indicative of the latent secondary psychopathy
␤2, ␤3, ␤4, ␤5, and ␤6, respectively. Items in the C and DI factors construct.
had, on average, higher ␤ parameters than those in the IP and F. PAI ANT. As shown in Figure 2a, the TIC of the ANT-A is
These findings suggest that coldheartedness and disinhibition/ relatively symmetric around ␪ ⫽ 0.5 to 1.5, suggesting that the
impulsivity features of psychopathy are less frequently endorsed latent antisocial behavior trait is estimated with some precision
than interpersonal and fearlessness features among participants in around moderate to moderately high levels of the latent trait. A
this study. spike in test information was observed in the TIC for the
ANT-S in the range of ␪ ⫽ ⫺1 and 2 (Figure 2b), indicating that
Test Information relative precise estimation of the stimulus seeking trait may also
be limited to individuals with moderate to moderate high levels
The TIFs of each factor in the four self-report measures of
of stimulus seeking. Of note, the maximum values of TIFs were
psychopathy across the latent trait continuum are shown in Table
7, and the corresponding TICs are presented in Figures 1 to 4. The much higher in the ANT-S than those in the ANT-A, suggesting
TICs reflect the amount of information provided by the subscales that the assessment of stimulus seeking is relatively more
across the whole range of the latent trait continuum. The latent trait precise than antisocial behavior.
is estimated with relative precision at the peak of the TIC, where PPI-SF. The TICs for the BE and F factors had the highest
a maximum amount of information is provided by the subscale. A peaks (Figures 3a and 3d), suggesting that the estimation of
TIC with a narrow peak indicates that the scale is relatively precise these two latent traits are relatively precise, especially within
around the corresponding latent trait levels, but not as precise at the range of ␪ ⫽ ⫺1.5 to 2.5 and ␪ ⫽ ⫺1 to 1.5, respectively.
the extremes. On the other hand, a TIC that is smooth and flat There are relatively flat TIC peaks for the SP, ME, and SI
indicates that the scale is able to measure respondents across a factors (␪ ⫽ ⫺2 to 1, ␪ ⫽ ⫺1 to 2, and ␪ ⫽ ⫺2 to 1.5 in Figures
wide range of latent trait level with similar, albeit less, precision. 3b, 3c, and 3g, respectively), meaning that the corresponding
LSRP. The TIFs of the PP factor varied across the latent latent traits are measured with similar precision within these
trait continuum, as reflected in the relatively steep TIC in ranges, but decrease rapidly beyond them. In comparison, the
Figure 1a. Such variation indicates that the primary psychopa- TICs of the C and IN factors were relatively flat (Figures 3e and
thy trait is measured with unequal precision at varying latent 3f), indicating that these respective latent traits are measured
trait levels. In comparison, the TIC of the SP factor was with similar precision. However, the relatively low levels of
relatively flat over the range ␪ ⫽ ⫺1 to 1.5 (Figure 1b), TIFs indicate that the C and IN factors yield small amount of
suggesting that the latent trait of secondary psychopathy is information, suggesting that these two corresponding traits are
measured with similar precision within this range, whereas estimated with limited precision.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Table 5
PPI-SF Estimated Parameters From GRM

Factor Item ␣ ␤1 ␤2 ␤3 Factor Item ␣ ␤1 ␤2 ␤3

Blame externalization 1 2.63 (.17) ⫺1.60 (.13) 2.27 (.15) 4.49 (.23) Coldheartedness 1 1.09 (.11) ⫺.31 (.07) 1.90 (.10) 3.92 (.19)
2 2.73 (.17) ⫺1.99 (.14) 2.12 (.15) 4.47 (.23) 2 1.25 (.12) .54 (.08) 2.69 (.13) 3.93 (.19)
3 1.90 (.12) ⫺.55 (.09) 2.13 (.12) 3.69 (.17) 3 1.20 (.11) ⫺.67 (.08) 1.90 (.10) 3.77 (.18)
4 1.11 (.08) ⫺1.18 (.08) 1.20 (.08) 2.49 (.11) 4 1.12 (.11) .49 (.07) 2.19 (.11) 3.66 (.17)
5 1.44 (.09) ⫺.47 (.08) 1.25 (.09) 2.49 (.11) 5 .45 (.07) ⫺.92 (.07) .92 (.07) 2.35 (.10)
6 1.58 (.09) ⫺1.62 (.10) 1.32 (.09) 3.06 (.13) 6 1.03 (.10) ⫺.92 (.08) 1.26 (.08) 3.16 (.14)
7 1.32 (.08) ⫺2.20 (.10) .53 (.08) 2.21 (.10) 7 .52 (.08) ⫺2.27 (.10) ⫺1.00 (.07) .48 (.06)
AIC ⫽ 19,278; BIC ⫽ 19,422; ␹2 (16,286) ⫽ 8,144, p ⫽ 1.00 AIC ⫽ 19,543; BIC ⫽ 19687; ␹2 (16,304) ⫽ 7,838, p ⫽ 1.00

Social potency 1 1.59 (.11) ⫺5.21 (.27) ⫺3.00 (.14) .62 (.08) Impulsive 1 1.04 (.10) ⫺.07 (.07) 1.98 (.10) 3.49 (.16)
nonconformity
2 1.70 (.11) ⫺3.51 (.16) ⫺.99 (.09) 1.66 (.10) 2 1.19 (.10) ⫺1.11 (.08) 1.32 (.09) 2.95 (.14)
3 1.64 (.11) ⫺3.23 (.15) ⫺.70 (.09) 1.85 (.11) 3 1.16 (.09) ⫺1.50 (.09) .79 (.08) 2.87 (.13)
4 1.79 (.11) ⫺2.68 (.13) ⫺.77 (.09) 1.49 (.10) 4 1.36 (.11) ⫺1.52 (.10) .76 (.08) 2.79 (.13)
5 1.90 (.12) ⫺3.19 (.15) ⫺1.26 (.10) 1.18 (.10) 5 1.26 (.12) .59 (.08) 1.89 (.11) 3.29 (.16)
6 1.27 (.09) ⫺3.34 (.14) ⫺1.05 (.08) 1.46 (.09) 6 1.13 (.11) .47 (.07) 1.52 (.09) 2.45 (.12)
7 1.10 (.08) ⫺2.55 (.11) ⫺.85 (.08) 1.43 (.08) 7 .37 (.07) ⫺1.24 (.07) .57 (.06) 2.00 (.09)
AIC ⫽ 19,734; BIC ⫽ 19,878; ␹2 (16,291) ⫽ 7,339, p ⫽ 1.00 AIC ⫽ 20,169; BIC ⫽ 20,313; ␹2 (16,291) ⫽ 10,264, p ⫽ 1.00

Machiavellian 1 1.94 (.14) ⫺2.08 (.13) 1.19 (.10) 3.67 (.19) Stress immunity 1 1.89 (.11) ⫺1.97 (.11) ⫺.10 (.09) 2.66 (.13)
2 1.84 (.13) ⫺1.30 (.10) 1.54 (.11) 3.70 (.18) 2 2.33 (.14) ⫺3.44 (.17) ⫺.78 (.11) 2.65 (.15)
3 .88 (.08) ⫺1.51 (.08) .72 (.07) 2.83 (.12) 3 2.16 (.13) ⫺4.22 (.20) ⫺1.36 (.11) 2.10 (.13)
4 .68 (.07) ⫺2.08 (.09) ⫺.17 (.06) 2.03 (.09) 4 1.26 (.08) ⫺2.27 (.10) ⫺.38 (.07) 1.85 (.09)
5 .99 (.09) .11 (.07) 1.89 (.09) 3.37 (.15) 5 1.47 (.09) ⫺3.41 (.15) ⫺1.40 (.09) 1.42 (.09)
TSANG, SALEKIN, COFFEY, AND COX

6 1.08 (.08) ⫺1.31 (.08) 1.21 (.08) 2.70 (.12) 6 1.07 (.07) ⫺2.02 (.09) ⫺.17 (.07) 1.84 (.09)
7 1.23 (.09) ⫺1.57 (.09) .97 (.08) 3.23 (.14) 7 1.02 (.07) ⫺.96 (.07) .85 (.07) 2.88 (.12)
AIC ⫽ 20,255; BIC ⫽ 20,399; ␹2 (16,281) ⫽ 7,937, p ⫽ 1.00 AIC ⫽ 20,422; BIC ⫽ 20,566; ␹2 (16,286) ⫽ 8,088, p ⫽ 1.00

Fearlessness 1 .93 (.07) ⫺.82 (.07) .14 (.07) 1.45 (.08)


2 2.06 (.14) ⫺.65 (.10) .82 (.10) 2.74 (.15)
3 1.52 (.10) ⫺.93 (.09) .23 (.08) 1.76 (.10)
4 1.21 (.09) ⫺.01 (.07) 1.16 (.08) 2.57 (.12)
5 1.47 (.10) ⫺1.33 (.09) ⫺.01 (.08) 1.60 (.09)
6 1.90 (.13) ⫺.61 (.09) .61 (.09) 1.82 (.11)
7 1.09 (.08) ⫺1.01 (.08) .73 (.07) 2.57 (.11)
AIC ⫽ 21,691; BIC ⫽ 21,835; ␹2 (16,330) ⫽ 14,900, p ⫽ 1.00
Note. ␣ ⫽ estimated slope parameters; ␤1, ␤2, and ␤3 ⫽ estimated category threshold parameters. Standard errors of the estimated parameters are presented in parentheses. Results for the carefree
nonplanfulness subscale is not available as the subscale did not meet the unidimensional assumption for IRT analyses.
SELF-REPORT PSYCHOPATHY IRT 9

Table 6
SRP-II Estimated Parameters From GRM

Factor Item ␣ ␤1 ␤2 ␤3 ␤4 ␤5 ␤6

Interpersonal 1 1.16 (.07) ⫺3.10 (.13) ⫺1.93 (.09) ⫺1.41 (.08) ⫺.44 (.07) .99 (.08) 2.35 (.10)
2 1.29 (.08) ⫺1.06 (.08) ⫺.11 (.07) .44 (.07) 1.37 (.08) 2.25 (.10) 3.28 (.14)
3 2.10 (.11) ⫺3.19 (.15) ⫺1.96 (.11) R1.16 (.10) .04 (.09) 2.02 (.12) 3.75 (.16)
4 1.38 (.08) ⫺3.02 (.13) ⫺1.79 (.10) ⫺.87 (.08) .05 (.08) 1.61 (.10) 3.07 (.13)
5 1.86 (.10) ⫺3.08 (.14) ⫺1.68 (.10) ⫺.97 (.09) .15 (.09) 1.81 (.10) 3.34 (.14)
6 .68 (.06) ⫺2.98 (.13) ⫺2.09 (.09) ⫺1.40 (.08) ⫺.33 (.06) .85 (.07) 2.10 (.09)
7 .94 (.07) ⫺2.56 (.11) ⫺1.45 (.08) ⫺.64 (.07) .20 (.66) 1.68 (.08) 2.89 (.12)
8 1.36 (.08) ⫺2.84 (.12) ⫺1.80 (.09) ⫺.86 (.08) .22 (.08) 1.52 (.09) 2.86 (.12)
9 .72 (.07) ⫺2.61 (.11) ⫺1.66 (.08) ⫺1.02 (.07) ⫺.03 (.06) .89 (.07) 2.04 (.09)
10 .83 (.07) ⫺2.32 (.10) ⫺1.46 (.08) ⫺.77 (.07) .17 (.07) 1.30 (.08) 2.63 (.11)
11 1.11 (.07) ⫺1.49 (.09) ⫺.45 (.07) .18 (.07) 1.09 (.08) 2.00 (.10) 3.08 (.13)
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

12 .69 (.06) ⫺1.69 (.08) ⫺.88 (.07) ⫺.32 (.06) .77 (.07) 1.41 (.08) 2.19 (.10)
This document is copyrighted by the American Psychological Association or one of its allied publishers.

13 1.19 (.07) ⫺1.72 (.09) ⫺.51 (.07) .28 (.07) 1.17 (.08) 2.07 (.10) 3.30 (.14)
14 1.31 (.08) ⫺1.49 (.09) ⫺.59 (.08) .03 (.07) .89 (.08) 2.20 (.10) 3.39 (.14)
15 .84 (.07) ⫺1.93 (.09) ⫺.90 (.07) ⫺.04 (.07) .64 (.07) 1.26 (.08) 2.43 (.11)
16 1.75 (.09) ⫺1.95 (.11) ⫺.55 (.09) .49 (.09) 1.58 (.10) 2.97 (.13) 4.43 (.19)
AIC ⫽ 68,839; BIC ⫽ 69,412 (Chi-square test unavailable)

Disinhibition/Impulsivity 17 .79 (.07) ⫺1.07 (.07) .05 (.06) .72 (.07) 1.67 (.08) 2.75 (.12) 4.00 (.19)
18 .67 (.06) ⫺1.76 (.08) ⫺.38 (.06) .79 (.07) 1.39 (.08) 2.32 (.10) 3.26 (.13)
19 1.99 (.12) .15 (.09) 1.40 (.11) 2.06 (.12) 2.62 (.13) 3.34 (.15) 4.16 (.18)
20 1.19 (.08) ⫺.53 (.07) .37 (.07) .84 (.08) 1.44 (.09) 2.53 (.11) 3.43 (.14)
21 1.39 (.10) .83 (.09) 1.36 (.09) 1.68 (.10) 2.06 (.11) 2.67 (.12) 3.28 (.14)
22 4.57 (.44) 1.75 (.24) 3.37 (.34) 4.48 (.41) 5.81 (.50) 7.29 (.59) 8.39 (.67)
23 1.61 (.10) ⫺.53 (.08) .67 (.09) 1.38 (.10) 2.26 (.11) 3.30 (.14) 4.36 (.19)
24 1.56 (.10) .41 (.09) 1.22 (.09) 1.72 (.10) 2.58 (.12) 3.58 (.15) 4.49 (.21)
25 1.44 (.10) .82 (.09) 1.41 (.10) 1.89 (.11) 2.58 (.12) 3.31 (.15) 4.00 (.18)
AIC ⫽ 31,527; BIC ⫽ 31,849 (Chi-square test unavailable)

Fearlessness 26 .76 (.07) ⫺1.85 (.09) ⫺1.32 (.08) ⫺.99 (.07) ⫺.45 (.07) .39 (.06) 1.57 (.08)
27 3.10 (.30) ⫺2.28 (.20) ⫺.91 (.14) ⫺.00 (.12) 1.07 (.15) 2.69 (.23) 4.57 (.35)
28 1.35 (.09) ⫺2.96 (.13) ⫺1.97 (010) ⫺1.16 (.08) ⫺.22 (.08) 1.16 (.09) 2.67 (.12)
29 1.05 (.08) ⫺2.02 (.10) ⫺1.14 (.08) ⫺.51 (.07) .20 (.07) 1.02 (.08) 2.15 (.10)
30 1.47 (.10) ⫺1.24 (.09) ⫺.17 (.08) .45 (.08) 1.24 (.09) 2.25 (.11) 3.40 (.15)
AIC ⫽ 22,243; BIC ⫽ 22,422; ␹ (16725) ⫽ 14556, p ⫽ 1.00
2

Coldheartedness 31 2.42 (.16) ⫺.76 (.11) 1.40 (.12) 2.76 (.16) 3.90 (.20) 4.74 (.24) 5.57 (.28)
32 1.90 (.12) ⫺.51 (.09) 1.74 (.11) 2.70 (.14) 3.60 (.17) 4.39 (.21) 5.21 (.25)
33 .97 (.07) ⫺1.21 (.08) .08 (.07) .85 (.07) 1.67 (.09) 2.58 (.11) 3.67 (.17)
34 1.35 (.09) ⫺1.16 (.09) .21 (.08) 1.39 (.09) 2.25 (.11) 2.94 (.13) 3.82 (.17)
35 1.90 (.12) ⫺1.19 (.10) .69 (.09) 1.85 (.11) 2.96 (.15) 3.73 (.17) 4.78 (.22)
36 1.35 (.10) .73 (.08) 1.59 (.10) 2.03 (.11) 3.03 (.14) 3.72 (.17) 4.39 (.21)
AIC ⫽ 21,131; BIC ⫽ 21,346; ␹2 (117356) ⫽ 11387, p ⫽ 1.00
Note. ␣ ⫽ estimated slope parameters; ␤1, ␤2, ␤3, ␤4, ␤5, and ␤6 ⫽ estimated category threshold parameters. Standard errors of the estimated parameters
are presented in parentheses.

SRP-II. The TIC of the DI factor showed a tall, narrow peak ing data. For these participants, two estimates of the latent trait
between ␪ ⫽ 0 and 2 (Figure 4b), indicating that this latent trait is underlying each subscale were available: the total score (Ts) and
measured very precisely within this range of the disinhibition/ the estimate from the GRM (␪s). We used these estimates to
impulsivity trait continuum. Beyond this narrow range, measure- examine the extent to which the factors are associated with one
ment precision of the disinhibition/impulsivity trait decreases another. In Table 8, the correlation coefficients among Ts (rT) are
markedly. As for the IP, F, and C factors, the TICs are relatively presented in the lower diagonal, and those among ␪s (r␪) are
flat over ␪ ⫽ ⫺1.5 to 2, ␪ ⫽ ⫺1 to 1.5, and ␪ ⫽ ⫺0.5 to 2.5, displayed in the upper diagonal.
respectively (Figures 4a, 4c, and 4d). Results suggest that these The two factors in the LSRP were moderately correlated with
respective latent traits are measured with similar precision within each other (rT ⫽ .43; r␪ ⫽ .42). The correlations among the two
the flat TIC ranges, and decrease as the TIC slopes decrease PAI ANT factors were moderate (rT ⫽ .56; r␪ ⫽ .59). Among the
toward both ends of the corresponding latent trait continuums. seven PPI-SF factors, the associations varied (rs ⫽ ⫺.19 to .43),
suggesting that some of the factors may not be tapping the same
Comparison Among Measures underlying construct of psychopathy as the others. Most of the
correlations among the SRP-SF factors were moderate to strong
Of the 1,257 participants in this study, 1,065 respondents com- (rs ⫽ .32 to .61), with the exception of the weak association
pleted all four self-report measures of psychopathy with no miss- between F and C.
10 TSANG, SALEKIN, COFFEY, AND COX

Table 7
Information Functions Across the Latent Trait Continuum

Information at different levels of latent trait (␪)

Measure ⫺3.0 ⫺2.5 ⫺2.0 ⫺1.5 ⫺1.0 ⫺.5 0 .5 1.0 1.5 2.0 2.5 3.0

LSRP
Primary psychopath 1.71 2.17 2.86 3.89 5.47 7.88 10.32 11.03 11.36 11.16 11.06 9.26 6.91
Secondary psychopath 1.88 2.20 2.55 2.88 3.09 3.21 3.27 3.30 3.28 3.25 3.17 2.95 2.63
PAI ANT
Antisocial behavior 1.44 1.85 2.47 3.14 3.76 4.45 5.16 5.56 5.53 5.15 4.50 3.68 2.91
Stimulus seeking 1.54 1.73 2.01 2.51 3.65 9.14 13.86 10.03 14.57 14.90 6.22 3.38 2.32
PPI⫺SF
Blame externalization 1.52 1.94 2.70 4.17 6.44 7.41 6.72 7.47 8.30 8.29 7.02 4.56 2.83
Social potency 3.58 4.60 5.51 5.86 5.87 5.88 5.80 5.78 5.39 4.31 3.08 2.16 1.61
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Machiavellian 1.72 2.11 2.72 3.48 4.03 4.07 4.07 4.29 4.39 4.38 4.29 3.77 3.05
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Fearlessness 1.42 1.74 2.29 3.14 4.27 5.32 5.79 5.83 5.60 4.92 3.82 2.82 2.14
Coldheartedness 1.40 1.57 1.80 2.09 2.40 2.66 2.84 2.94 3.00 3.04 3.06 3.05 2.97
Impulsive 1.50 1.78 2.16 2.55 2.88 3.13 3.34 3.50 3.58 3.60 3.54 3.35 2.98
nonconformity
Stress immunity 2.61 3.77 5.14 6.02 6.32 6.42 6.03 5.86 6.16 5.44 3.92 2.72 2.02
SRP-II
Interpersonal 4.12 5.44 7.02 8.33 8.99 9.20 9.19 9.09 9.07 8.95 8.44 7.27 5.81
Disinhibition/Impulsivity 1.37 1.54 1.84 2.35 3.19 4.57 7.82 11.95 12.57 12.52 10.61 6.30 4.61
Fearlessness 1.80 2.10 2.51 3.30 4.83 5.77 5.91 5.81 5.69 5.37 3.97 2.79 2.17
Coldheartedness 1.30 1.54 1.99 2.82 4.14 5.49 6.04 6.32 6.57 6.68 6.64 6.08 4.80
Note. LSRP ⫽ Levenson Self-Report Psychopathy Scale; PAI ANT: Personality Assessment Inventory—Antisocial Features Scale; PPI-SF ⫽ Psycho-
pathic Personality Inventory-Short Form; SRP-II ⫽ Self- Report Psychopathy Scale-II.

There are some noteworthy associations between the four self- dents may mostly be impulsive actions. The F and C factors in the
report measures. For example, LSRP PP is strongly associated with PPI-SF are moderately associated with the corresponding SRP-II
PPI-SF ME, SRP-II IP, SRP-II DI, and SRP-II C (rs ⫽ .54 to .60), factors (rs ⫽ .33 to .58), indicating that there are some differences in
suggesting that these features may be indicative of primary psychop- the latent traits measured by the PPI-SF and SRP-II items.
athy. Participants who endorsed positively on ANT-S also had high Inspection of Table 8 reveals that the directions of the associ-
levels of F and DI on the SRP-II (rs ⫽ .40 to .62). The moderate to ations among factors are the same between the Ts and the ␪s.
strong correlations between ANT-A, PPI-SF IN, and SRP-II DI (rs ⫽ However, there are differences in the correlation magnitudes be-
.36 to .62) suggest that antisocial behavior endorsed by these respon- tween some factors. Fisher’s z tests were used to compare the two
12

12
10

10
8

8
Information

Information
6

6
4

4
2

2
0

−4 −2 0 2 4 −4 −2 0 2 4

(a) Primary Psychopathy (b) Secondary Psychopathy


Figure 1. Test information curves for LSRP subscales. See the online article for the color version of this figure.
SELF-REPORT PSYCHOPATHY IRT 11
15

15
10

10
Information

Information
5

5
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

0
−4 −2 0 2 4 −4 −2 0 2 4

(a) Antisocial behavior ( b ) Stimulus seeking


Figure 2. Test information curves for PAI ANT subscales. See the online article for the color version of this
figure.

correlation coefficients (rT and r␪); significant differences are by the SP factor suggest that the underlying trait of secondary
illustrated in bold in Table 8. Such differences suggest that the psychopathy is estimated with limited precision among undergrad-
associations between sum scores (rT) may not represent the rela- uate students in the current study.
tions between underlying latent traits (r␪) estimated by different One possible explanation is that behavior identified to be char-
factors. acteristic of secondary psychopathy (e.g., impulsivity and antiso-
cial behavior) is normative behavior among young adults, and thus
Discussion are heavily endorsed by normal college students. It is possible that
these behavioral characteristics are not particularly sensitive in
The current study investigated how well the latent construct of
distinguishing among individuals with varying levels of secondary
psychopathy is measured among a large undergraduate sample, in
psychopathy, especially among college students who may endorse
which the levels of psychopathy are presumably quite low. The
these behaviors regardless of their underlying psychopathic ten-
LSRP, PPI-SF, PAI ANT, and the SRP-II were initially developed
dencies. Taken together, it is questionable whether the LSRP SP
to assess the underlying psychopathy construct by tapping differ-
factor is an informative measure of secondary psychopathy
ent components of the latent psychopathy trait. Although it has
traits within an undergraduate student sample such as the cur-
been argued that these measures are best suited for nonforensic
rent one.
community samples, this is the first study to investigate and
compare how well these self-report items assess the varying di- Results showed moderate to high discrimination ability for the
mensions of the psychopathy construct among undergraduate stu- items in the PAI ANT (all ␣s ⬎ 1.21 for ANT-A; all ␣s ⬎ .70 for
dents using IRT models. ANT-S). The frequencies at which items are endorsed are rela-
Overall, items in the LSRP PP factor were moderately to highly tively similar between the two factors, suggesting that the PAI
discriminating (all ␣s ⬎ .8; Baker, 2001), suggesting that LSRP ANT items are informative in the assessment of their respective
items indicative of primary psychopathy features (e.g., emotional underlying traits within similar trait levels. However, the two PAI
detachment, selfishness, manipulativeness) are relatively good at ANT factors estimate their respective latent trait levels with dif-
distinguishing among individuals with varying levels of the latent fering degrees of precision, even within the trait range at which the
primary psychopathy trait. The PP factor showed high levels of test amount of test information is at maximum. The general level of the
information at moderate to moderately high levels of the latent TIF is the highest for ANT-S, suggesting that the latent trait is
trait, indicating that the latent primary psychopathy trait is esti- estimated with relative precision across varying levels of stimulus
mated with relative precision among individuals with moderate to seeking trait. This corresponds to the observation that ANT-S
moderately high levels of the trait. contains some of the most discriminating items measuring thrill-
On the other hand, five out of the nine LSRP SP items (Items 20, pursuing behavior (e.g., Items 17 and 18). The low general TIF
21, 22, 24, and 25) were at least moderately discriminating, sug- level for ANT-A indicates that the items may not be as precise in
gesting that some of the SP items may not be sensitive enough to assessing the construct of social deviance among individuals with
distinguish individuals with different levels of the latent secondary different levels of antisocial behavior. In fact, many of the behav-
psychopathy trait. The general low levels of information provided ior assessed in ANT-A may be normative adolescent behavior
12 8
TSANG, SALEKIN, COFFEY, AND COX

8
6

6
Information

Information

Information
4

4
2

2
0

0
−4 −2 0 2 4 −4 −2 0 2 4 −4 −2 0 2 4
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

(a) Blame externalization (b) Social potency (c) Machiavellian


This document is copyrighted by the American Psychological Association or one of its allied publishers.

8
6

6
Information

Information

Information
4

4
2

2
0

0
−4 −2 0 2 4 −4 −2 0 2 4 −4 −2 0 2 4

(d) Fearlessness (e) Coldheartedness (f) Impulsive nonconformity


8
6
Information

4
2
0

−4 −2 0 2 4

(g) Stress immunity

Figure 3. Test information curves for PPI-SF subscales. See the online article for the color version of this
figure.

(e.g., Items 3 and 4), which the majority of the college students in is less effective in doing so as it showed lower discrimination. The
this study may have endorsed in the past. items in the SP and SI factors are more frequently endorsed,
Given that the seven factors in the PPI-SF were developed to tap whereas those in the CN and C are less frequently endorsed. The
different aspects of the latent psychopathy construct, results re- TIC graphs show that the PPI-SF factors are informative at varying
vealing differences in item discrimination and category threshold levels along their corresponding latent trait continuums, meaning
levels suggest that the factors vary in their ability to assess their that the respective latent traits are estimated with different degrees
respective underlying latent traits. With higher discrimination, the of precision. Across the PPI-SF factors, the PPI-SF C factor, with
BE and SP factors are more effective in distinguishing among all reverse-scored items, had the smallest TIFs, indicating that the
respondents who differ in their respective trait levels; the C factor coldheartedness construct is estimated with relatively low accuracy.
SELF-REPORT PSYCHOPATHY IRT 13
12

12
10

10
8

8
Information

Information
6

6
4

4
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

2
This document is copyrighted by the American Psychological Association or one of its allied publishers.

0
−4 −2 0 2 4 −4 −2 0 2 4

(a) Interpersonal (b) Disinhibition/Impulsivity


12

12
10

10
8

8
Information

Information
6

6
4

4
2
2

0
0

−4 −2 0 2 4 −4 −2 0 2 4

(c) Fearlessness (d) Coldheartedness


Figure 4. Test information curves for SRP-II subscales. See the online article for the color version of this
figure.

These findings have theoretical implications. Assuming that the the PPI and PPI-SF, the PPI-R (Lilienfeld & Widows, 2005) and
latent construct of psychopathy is made up of the eight dimensions the PPI-R-40 (Eisenbarth, Lilienfeld, & Yarkoni, 2015), further
assessed by the PPI-SF factors, the current results indicate that research is needed to examine the item properties of these revised
some components (e.g., blame externalization, social potency) measures, and whether the items in different subscales vary in their
may be more effective in distinguishing individuals who differ in ability to assess the different components of psychopathy.
the latent trait levels. At best, the subscales were only moderately Items in the four SRP-II factors have similar ability to distin-
correlated with each other, providing further evidence that the guish among respondents with different levels of the respective
PPI-SF is a multidimensional measure of psychopathy. In addition, latent traits. Results provide some evidence that the rationally
extending the current study of the PPI-SF to forensic and other derived four factors for the SRP-II (Lester et al., 2013) are equally
nonforensic samples may help delineate why differential correlates good at discriminating across individuals with different levels of
between the PPI-SF subscales in offenders (Neumann, Malterer, & the respective latent traits. Items in the IP and F factors are more
Newman, 2008) and college students (Lilienfeld, 1990) were pre- frequently endorsed than those in the C and DI factors, indicating
viously reported. With the development of the revised versions of that some features (e.g., coldheartedness) of the psychopathy con-
14 TSANG, SALEKIN, COFFEY, AND COX

Table 8
Correlation Matrix of Total Scores and Estimated Latent Traits (␪) Between Subscales of Self-Report Measures of Psychopathy

LSRP PAI PPI-SF SRP-II


PP SP ANT-A ANT-S BE SP ME F C IN SI IP DI F C

LSRP
PP .42ⴱⴱⴱ .49ⴱⴱⴱ .45ⴱⴱⴱ .30ⴱⴱⴱ .03 .59ⴱⴱⴱ .21ⴱⴱⴱ .24ⴱⴱⴱ .22ⴱⴱⴱ .00 .59ⴱⴱⴱ .54ⴱⴱⴱ .37ⴱⴱⴱ .55ⴱⴱⴱ
SP .43ⴱⴱⴱ .44ⴱⴱⴱ .38ⴱⴱⴱ .37ⴱⴱⴱ ⫺.05 .42ⴱⴱⴱ .14ⴱⴱⴱ .03 .33ⴱⴱⴱ ⫺.20ⴱⴱⴱ .36ⴱⴱⴱ .47ⴱⴱⴱ .28ⴱⴱⴱ .33ⴱⴱⴱ
PAI
ANT-A .49ⴱⴱⴱ .44ⴱⴱⴱ .59ⴱⴱⴱ .28ⴱⴱⴱ .09ⴱⴱ .50ⴱⴱⴱ .31ⴱⴱⴱ .14ⴱⴱⴱ .40ⴱⴱⴱ .05 .48ⴱⴱⴱ .62ⴱⴱⴱ .45ⴱⴱⴱ .38ⴱⴱⴱ
ANT-S .46ⴱⴱⴱ .44ⴱⴱⴱ .56ⴱⴱⴱ .22ⴱⴱⴱ .20ⴱⴱⴱ .34ⴱⴱⴱ .59ⴱⴱⴱ .12ⴱⴱⴱ .47ⴱⴱⴱ .14ⴱⴱⴱ .45ⴱⴱⴱ .52ⴱⴱⴱ .71ⴱⴱⴱ .30ⴱⴱⴱ
PPI-SF
BE .31ⴱⴱⴱ .39ⴱⴱⴱ .31ⴱⴱⴱ .25ⴱⴱⴱ ⫺.03 .32ⴱⴱⴱ .06 ⫺.06ⴱ .24ⴱⴱⴱ ⫺.19ⴱⴱⴱ .26ⴱⴱⴱ .19ⴱⴱⴱ .16ⴱⴱⴱ .17ⴱⴱⴱ
SP .04 ⫺.06 .09ⴱⴱ .20ⴱⴱⴱ ⫺.05 .02 .13ⴱⴱⴱ .00 .01 .23ⴱⴱⴱ .14ⴱⴱⴱ .09ⴱⴱ .22ⴱⴱⴱ .01
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ME .56ⴱⴱⴱ .40ⴱⴱⴱ .45ⴱⴱⴱ .32ⴱⴱⴱ .35ⴱⴱⴱ ⫺.01 .11ⴱⴱⴱ .09ⴱⴱ .27ⴱⴱⴱ ⫺.12ⴱⴱⴱ .52ⴱⴱⴱ .43ⴱⴱⴱ .27ⴱⴱⴱ .37ⴱⴱⴱ
.20ⴱⴱⴱ .30ⴱⴱⴱ .58ⴱⴱⴱ .06ⴱ .13ⴱⴱⴱ .09ⴱⴱ .12ⴱⴱⴱ .42ⴱⴱⴱ .27ⴱⴱⴱ .24ⴱⴱⴱ .29ⴱⴱⴱ .58ⴱⴱⴱ .13ⴱⴱⴱ
This document is copyrighted by the American Psychological Association or one of its allied publishers.

F .15
C .21ⴱⴱⴱ .00 .11ⴱⴱⴱ .11ⴱⴱⴱ ⫺.10ⴱⴱⴱ .06ⴱ .01 .07ⴱ .09ⴱⴱ .32ⴱⴱⴱ .20ⴱⴱⴱ .22ⴱⴱⴱ .09ⴱⴱ .40ⴱⴱⴱ
IN .18ⴱⴱⴱ .30ⴱⴱⴱ .36ⴱⴱⴱ .46ⴱⴱⴱ .22ⴱⴱⴱ .02 .15ⴱⴱⴱ .43ⴱⴱⴱ .08ⴱ .05 .30ⴱⴱⴱ .40ⴱⴱⴱ .36ⴱⴱⴱ .23ⴱⴱⴱ
SI .03 ⫺.18ⴱⴱ .06ⴱ .15ⴱⴱⴱ ⫺.21ⴱⴱⴱ .24ⴱⴱⴱ ⫺.16ⴱⴱⴱ .27ⴱⴱⴱ .35ⴱⴱⴱ .13ⴱⴱⴱ .16ⴱⴱⴱ .03 .18ⴱⴱⴱ .09ⴱⴱ
SRP-II
IP .57ⴱⴱⴱ .34ⴱⴱⴱ .45ⴱⴱⴱ .44ⴱⴱⴱ .28ⴱⴱⴱ .15ⴱⴱⴱ .48ⴱⴱⴱ .25ⴱⴱⴱ .16ⴱⴱⴱ .24ⴱⴱⴱ .16ⴱⴱⴱ .41ⴱⴱⴱ .49ⴱⴱⴱ .35ⴱⴱⴱ
DI .54ⴱⴱⴱ .47ⴱⴱⴱ .61ⴱⴱⴱ .52ⴱⴱⴱ .20ⴱⴱⴱ .10ⴱⴱ .39ⴱⴱⴱ .26ⴱⴱⴱ .20ⴱⴱⴱ .37ⴱⴱⴱ .05 .38ⴱⴱⴱ .42ⴱⴱⴱ .61ⴱⴱⴱ
F .33ⴱⴱⴱ .27ⴱⴱⴱ .40ⴱⴱⴱ .69ⴱⴱⴱ .17ⴱⴱⴱ .22ⴱⴱⴱ .20ⴱⴱⴱ .58ⴱⴱⴱ .08ⴱ .33ⴱⴱⴱ .22ⴱⴱⴱ .48ⴱⴱⴱ .03ⴱⴱⴱ .19ⴱⴱⴱ
C .54ⴱⴱⴱ .32ⴱⴱⴱ .38ⴱⴱⴱ .30ⴱⴱⴱ .17ⴱⴱⴱ .04 .36ⴱⴱⴱ .13ⴱⴱⴱ .33ⴱⴱⴱ .21ⴱⴱⴱ .11ⴱⴱⴱ .32ⴱⴱⴱ .62ⴱⴱⴱ .15ⴱⴱⴱ
Note. Correlation coefficients between total scores (rT) are presented in the lower diagonal; correlation coefficients between estimated underlying latent
traits (r␪) are presented in the upper diagonal. Bolded values indicate significant differences between rT and r␪. LSRP ⫽ Levenson Self-Report Psychopathy
Scale; PP ⫽ Primary Psychopathy; SP ⫽ Secondary Psychopathy; PAI ⫽ Personality Assessment Inventory; ANT-A ⫽ Antisocial Features Scale,
Antisocial Behavior; ANT-S ⫽ Antisocial Features Scale, Stimulus Seeking; PPI ⫽ Psychopathic Personality Inventory-Short Form; BE ⫽ Blame
Externalization; SP ⫽ Social Potency; ME ⫽ Machiavellian; F ⫽ Fearlessness; C ⫽ Coldheartedness; IN ⫽ Impulsive Nonconformity; SI ⫽ Stress
Immunity; SRP-II ⫽ Self-Report Psychopathy Scale-II; IP ⫽ Interpersonal; DI ⫽ Disinhibition/Impulsivity.

p ⬍ .05. ⴱⴱ p ⬍ .01. ⴱⴱⴱ p ⬍ .001.

struct may be rarely endorsed among college students. With the factor (Item 19: “I find that I am able to pursue one goal for a long
exception of the weak association between the C and F factors, the time” and Item 23: “Before I do anything, I carefully consider the
moderate to strong correlations between the SRP-II factors suggest possible consequences”) are the least discriminating items as com-
relative uniformity in factor associations. pared to the rest of the items in the factor. Our results raised
There is evidence to suggest that some components of the questions as to whether these items are working as they intended
psychopathy construct are more associated with each other. For to be. The only reverse-scored item in the PPI-SF IN factor
example, respondents who were estimated to have high levels of (“Fitting in and having things in common with other people my
primary psychopathy (in the LSRP) also had high tendencies of age has always been important to me”) is the least discriminating
psychopathic features such as egocentricity (in the PAI ANT), item in the subscale. All items in the PPI-SF C factor are reverse-
Machiavellian (in the PPI-SF), interpersonal (in the SRP-II), and scored; the absence of empathy characteristics is inferred to rep-
coldheartedness (in the SRP-II). The latent trait of stimulus seek- resent the latent trait of coldheartedness, yet they are among the
ing (in the PAI ANT) is associated with characteristics of fear- ones with the worst discrimination parameters within the PPI-SF.
lessness and disinhibition/impulsivity (both in the SRP-II), sug- The PPI-SF C factor also had the smallest TIF values among all the
gesting that individuals who are fearless and impulsive may be factors examined in this study, suggesting that the coldheartedness
more likely to pursue activities that give them excitement or trait is estimated with little precision.
“thrills.” Likewise, features of impulsive nonconformity (in the A regular-scored item measures the presence of the latent trait,
PPI-SF) and disinhibition/impulsivity (in the SRP-II) may be with higher scores indicating higher levels of the underlying trait,
strongly associated with the possibly of engaging in social deviant whereas a reverse-scored is worded negatively, with lower scores
behavior (in the PAI ANT). Two latent traits, fearlessness and representing higher levels of the latent trait. Our finding that these
coldheartedness, are both measured in the PPI-SF and the SRP-II. reverse-scored items may have limited ability in distinguishing
The corresponding factors are, at best, moderately correlated. among respondents with varying levels of the latent trait adds to
Thus, it is questionable whether the corresponding subscales are the growing body of literature reporting problems with negatively
tapping the same underlying latent constructs, and whether the worded items in personality assessment (e.g., Crego & Widiger,
estimation of the latent traits can be improved by combining the 2014; Ebesutani et al., 2012; Lindwall et al., 2012; Ray, Frick,
respective subscales. Thornton, Steinberg, & Cauffman, 2016; Rodebaugh, Woods, &
Further examination of the estimated item parameters reveals Heimberg, 2007; Stansbury, Ried, & Velozo, 2006). These results
that items’ abilities to measure the underlying latent traits, within raise two practical questions. First, is the absence of a trait (e.g.,
each factor, may differ between those that are regular- and reverse- empathy) the same as the endorsement of its supposedly opposing
scored. For instance, the negatively worded items in the LSRP SP trait (e.g., coldheartedness)? As both traits are latent constructs, it
SELF-REPORT PSYCHOPATHY IRT 15

is unclear how well the absence of one indicates the presence of reveal that, although some of the latent constructs assessed by the
the other (Crego & Widiger, 2014). Second, to what extent is the different factors are correlated with one another, none of the
different discrimination ability between regular- and reverse- associations are exceptionally strong, suggesting that the instru-
scored items a method bias? The methodological literature has ments may be measuring different but related domains of psychop-
identified many potential sources of method bias, such as acqui- athy. Thus, it appears that the construct of psychopathy may be
escence and confirmation bias (see Podsakoff, MacKenzie, Lee, & best conceptualized as a multidimensional construct. In practice,
Podsakoff, 2003 for a review). However, whether the observed clinicians and researchers may wish to use multiple measures in
difference in the estimated parameters observed between regular- order to best tap the broader concept of psychopathy. In addition
and reverse-scored items in these self-report psychopathy mea- to the total scores of the self-report psychopathy measures, indi-
sures is due to method bias or true difference in discrimination vidual factor scores should be used to better understand how the
ability is not yet known. Considering that many of the items in the different dimensions of psychopathy is associated with one an-
self-report psychopathy measures are worded negatively, studies other.
investigating whether the response bias is present between differ-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ently keyed items are much warranted to validate the psychometric


This document is copyrighted by the American Psychological Association or one of its allied publishers.

properties of these instruments. References


In addition to the above findings regarding item properties of the Andrich, D. (1978). A rating formulation for ordered response categories.
four self-reported psychopathy measures, our results showed that Psychometrika, 43, 561–573. http://dx.doi.org/10.1007/BF02293814
two of the subscales (PAI ANT-E and PPI-SF’s CN) were not Baker, F. B. (1992). Item response theory: Parameter estimation tech-
unidimensional. The lack of unidimensionality suggests that these niques. New York, NY: Marcel Dekker.
two subscales may not be assessing one underlying construct Baker, F. B. (2001). The basics of item response theory. College Park, MD:
(egocentricity and carefree nonplanfulness, respectively). Future ERIC Clearinghouse on Assessment and Evaluation.
studies should further examine the dimensionality of these sub- Benning, S. D., Patrick, C. J., Hicks, B. M., Blonigen, D. M., & Krueger,
R. F. (2003). Factor structure of the psychopathic personality inventory:
scales to determine whether the items in the subscale were assess-
Validity and implications for clinical assessment. Psychological Assess-
ing one underlying construct, and whether certain items need to be
ment, 15, 340 –350. http://dx.doi.org/10.1037/1040-3590.15.3.340
excluded for scale refinement. In the meantime, we recommend Benning, S. D., Patrick, C. J., Salekin, R. T., & Leistico, A. M. (2005).
caution when interpreting and using composite scores from these Convergent and discriminant validity of psychopathy factors assessed
two subscales in clinical and research settings. via self-report: A comparison of three instruments. Assessment, 12,
This study examined the item functioning of four self-report 270 –289. http://dx.doi.org/10.1177/1073191105277110
psychopathy measures using an IRT framework, however, several Bentler, P. M. (1990). Comparative fit indexes in structural models. Psy-
limitations should be noted. The current sample consisted of col- chological Bulletin, 107, 238 –246. http://dx.doi.org/10.1037/0033-2909
lege students, in which the average levels of psychopathy are likely .107.2.238
to be relatively low. It is unclear whether some of the more Bolt, D. M., Hare, R. D., Vitale, J. E., & Newman, J. P. (2004). A
multigroup item response theory analysis of the Psychopathy Check-
frequently endorsed features (e.g., impulsiveness, antisocial be-
list—Revised. Psychological Assessment, 16, 155–168. http://dx.doi
havior) are indicative of endorsement of certain psychopathic
.org/10.1037/1040-3590.16.2.155
components or reflect normative young adult behavior. In addition, Brinkley, C. A., Diamond, P. M., Magaletta, P. R., & Heigel, C. P. (2008).
some of the items may have different meanings for college stu- Cross-validation of Levenson’s Psychopathy Scale in a sample of federal
dents versus adults in the community and/or criminal justice- female inmates. Assessment, 15, 464 – 482. http://dx.doi.org/10.1177/
involved individuals and whether this possibility exists needs 1073191108319043
further examination. Replication of this study in offender samples Brinkley, C. A., Schmitt, W. A., Smith, S. S., & Newman, J. P. (2001).
is much needed, and future studies should consider examining the Construct validation of a self-report psychopathy scale: Does Leven-
item-level psychometric properties of these self-report measures son’s self-report psychopathy scale measure the same constructs as
using forensic and prison samples. Nonetheless, it should be noted Hare’s Psychopathy Checklist-Revised? Personality and Individual Dif-
ferences, 31, 1021–1038. http://dx.doi.org/10.1016/S0191-8869
that we identified some items with good discriminating parameters
(00)00178-1
(e.g., Item 18 in PAI ANT: “Engage in wild activities for fun;”
Cleckley, H. M. (1941). The mask of sanity an attempt to reinterpret the
Item 31 in SRP-II: “Not hurting others’ feelings is important to so-called psychopathic personality. St. Louis, MO: Mosby.
me”), which suggests that some items are able to distinguish Cooke, D. J., & Michie, C. (2001). Refining the construct of psychopathy:
respondents with different trait levels, even at the lower end of the Towards a hierarchical model. Psychological Assessment, 13, 171–188.
trait continuums. http://dx.doi.org/10.1037/1040-3590.13.2.171
In conclusion, the current study provides some insight concern- Crego, C., & Widiger, T. A. (2014). Psychopathy, DSM–5, and a caution.
ing the statements developed to assess different components of the Personality Disorders, 5, 335–347. http://dx.doi.org/10.1037/
psychopathy construct among a large college sample, in which the per0000078
average levels of psychopathy are assumed to be lower than those Ebesutani, C., Drescher, C. F., Reise, S. P., Heiden, L., Hight, T. L.,
among forensic and offending respondents. Results from the IRT Damon, J. D., & Young, J. (2012). The Loneliness Questionnaire-Short
Version: An evaluation of reverse-worded and non-reverse-worded
analyses suggest that the factors in the four self-report psychopa-
items via item response theory. Journal of Personality Assessment, 94,
thy instruments differ in their ability to measure the respective 427– 437. http://dx.doi.org/10.1080/00223891.2012.662188
latent traits. Considering that the factors were developed to assess Edens, J. F., Hart, S. D., Johnson, D. W., Johnson, J. K., & Olver, M. E.
various psychopathy dimensions, different factors within each (2000). Use of the Personality Assessment Inventory to assess psychop-
instrument may actually complement each other by providing athy in offender populations. Psychological Assessment, 12, 132–139.
information at across different latent trait ranges. Our findings http://dx.doi.org/10.1037/1040-3590.12.2.132
16 TSANG, SALEKIN, COFFEY, AND COX

Edens, J. F., Poythress, N. G., & Watkins, M. M. (2001). Further validation sonality Assessment, 94, 196 –204. http://dx.doi.org/10.1080/00223891
of the Psychopathic Personality Inventory among offenders: Personality .2011.645936
and behavioral correlates. Journal of Personality Disorders, 15, 403– Lynam, D. R., Whiteside, S., & Jones, S. (1999). Self-reported psychop-
415. http://dx.doi.org/10.1521/pedi.15.5.403.19202 athy: A validation study. Journal of Personality Assessment, 73, 110 –
Eisenbarth, H., Lilienfeld, S. O., & Yarkoni, T. (2015). Using a genetic 132. http://dx.doi.org/10.1207/S15327752JPA730108
algorithm to abbreviate the Psychopathic Personality Inventory-Revised Martínez Arias, R. (1995). Psicometría: Teoría de los tests psicológicos y
(PPI-R). Psychological Assessment, 27, 194 –202. http://dx.doi.org/10 educativos [Psychometrics: Educational and psychological test theory].
.1037/pas0000032 Madrid, Spain: Síntesis.
Embretson, S. E. (1996). The new rules of measurement. Psychological Morey, L. C. (1991). The Personality Assessment Inventory (PAI), Profes-
Assessment, 8, 341–349. http://dx.doi.org/10.1037/1040-3590.8.4.341 sional manual. Odessa, FL: Psychological Assessment Resources.
Embretson, S. E., & Reise, S. P. (2000). Item response theory for psychol- Morey, L. C. (1996). An interpretative guide to the Personality Assessment
ogists. Mahwah, NJ: Erlbaum Publishers. Inventory. Odessa, FL: Psychological Assessment Resources.
Gummelt, H. D., Anestis, J. C., & Carbonell, J. L. (2012). Examining the Morey, L. C. (2003). Essentials of PAI Assessment. Hoboken, NJ: Wiley.
Levenson Self-Report Psychopathy Scale using a graded response mod- Morey, L. C. (2007). The Personality Assessment Inventory (PAI), Profes-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

el. Personality and Individual Differences, 53, 1002–1006. http://dx.doi sional manual (2nd ed.). Odessa, FL: Psychological Assessment Re-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

.org/10.1016/j.paid.2012.07.014 sources.
Hare, R. D. (1980). A research scale for the assessment of psychopathy in Muthén, L., & Muthén, B. (1998 –2012). Mplus user’s guide (7th ed.). Los
criminal populations. Personality and Individual Differences, 1, 111– Angeles, CA: Author.
119. http://dx.doi.org/10.1016/0191-8869(80)90028-8 Neumann, C. S., Kosson, D. S., & Salekin, R. T. (2007). Exploratory and
Hare, R. D. (1991). The Hare Psychopathy Checklist-Revised PCL-R. confirmatory factor analysis of the psychopathy construct: Methodolog-
Toronto, Ontario, Canada: Multi-Health Systems. ical and conceptual issues. In H. Hervé & J. C. Yuille (Eds.), The
Hare, R. D. (2003). The Hare Psychopathy Checklist-Revised PCL-R (2nd psychopath: Theory, research, and practice (pp. 79 –104). Mahwah, NJ:
ed.). Toronto, Ontario, Canada: Multi-Health Systems. Erlbaum.
Hare, R. D., Harpur, T. J., & Hemphill, J. D. (1989). Scoring pamphlet for Neumann, C. S., Malterer, M. B., & Newman, J. P. (2008). Factor structure
the Self-Report Psychopathy Scale: SRP-II. Vancouver, British Colum- of the Psychopathic Personality Inventory (PPI): Findings from a large
bia, Canada: Simon Fraser University. incarcerated sample. Psychological Assessment, 20, 169 –174. http://dx
Hare, R. D., & Neumann, C. S. (2010). The role of antisociality in the .doi.org/10.1037/1040-3590.20.2.169
psychopathy construct: Comment on Skeem and Cooke (2010). Psycho- Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory (3rd ed.).
logical Assessment, 22, 446 – 454. http://dx.doi.org/10.1037/a0013635 New York, NY: McGraw-Hill.
Harpur, T. J., Hare, R. D., & Hakstian, A. R. (1989). Two-factor concep- Paulhus, D. L., Neumann, C. S., & Hare, R. D. (in press). Manual for the
tualization of psychopathy: Construct validity and assessment implica- Hare Self-Report Psychopathy scale. Toronto, Ontario, Canada: Multi-
tions. Psychological Assessment: A Journal of Consulting and Clinical Health Systems.
Psychology, 1, 6 –17. http://dx.doi.org/10.1037/1040-3590.1.1.6 Podsakoff, P. M., MacKenzie, S. B., Lee, J. Y., & Podsakoff, N. P. (2003).
Hauck-Filho, N., & Teixeira, M. A. (2014). Revisiting the psychometric Common method biases in behavioral research: A critical review of the
properties of the Levenson Self-Report Psychopathy Scale. Journal of literature and recommended remedies. Journal of Applied Psychology,
Personality Assessment, 96, 459 – 464. http://dx.doi.org/10.1080/ 88, 879 –903. http://dx.doi.org/10.1037/0021-9010.88.5.879
00223891.2013.865196 Ray, J. V., Frick, P. J., Thornton, L. C., Steinberg, L., & Cauffman, E.
Karpman, B. (1948). The myth of the psychopathic personality. The (2016). Positive and negative item wording and its influence on the
American Journal of Psychiatry, 104, 523–534. http://dx.doi.org/10 assessment of callous-unemotional traits. Psychological Assessment, 28,
.1176/ajp.104.9.523 394 – 404. http://dx.doi.org/10.1037/pas0000183
Kelsey, K. R., Rogers, R., & Robinson, E. V. (2014). Self-report measures Rodebaugh, T. L., Woods, C. M., & Heimberg, R. G. (2007). The reverse
of psychopathy: What is their role in forensic assessments? Journal of of social anxiety is not always the opposite: The reverse-scored items of
Psychopathology and Behavioral Assessment, 37, 380 –391. http://dx.doi the social interaction anxiety scale do not belong. Behavior Therapy, 38,
.org/10.1007/s10862-014-9475-5 192–206. http://dx.doi.org/10.1016/j.beth.2006.08.001
Lester, W. S., Salekin, R. T., & Sellbom, M. (2013). The SRP-II as a rich Rosseel, Y. Y. (2012). lavaan: An R Package for structural equation
source of data on the psychopathic personality. Psychological Assess- modeling. Journal of Statistical Software, 48, 1–36. http://dx.doi.org/10
ment, 25, 32– 46. http://dx.doi.org/10.1037/a0029449 .18637/jss.v048.i02
Levenson, M. R., Kiehl, K. A., & Fitzpatrick, C. M. (1995). Assessing Salekin, R. T., Chen, D. R., Sellbom, M., Lester, W. S., & MacDougall, E.
psychopathic attributes in a noninstitutionalized population. Journal of (2014). Examining the factor structure and convergent and discriminant
Personality and Social Psychology, 68, 151–158. http://dx.doi.org/10 validity of the Levenson self-report psychopathy scale: Is the two-factor
.1037/0022-3514.68.1.151 model the best fitting model? Personality Disorders, 5, 289 –304. http://
Lilienfeld, S. (1990). Development and preliminary validation of a self- dx.doi.org/10.1037/per0000073
report measure of psychopathic personality (Unpublished doctoral dis- Salekin, R. T., Trobst, K. K., & Krioukova, M. (2001). Construct validity
sertation). University of Minnesota, Minneapolis, MN. of psychopathy in a community sample: A nomological net approach.
Lilienfeld, S. O., & Hess, T. H. (2001). Psychopathic personality traits and Journal of Personality Disorders, 15, 425– 441. http://dx.doi.org/10
somatization: Sex differences and the mediating role of negative emo- .1521/pedi.15.5.425.19196
tionality. Journal of Psychopathology and Behavioral Assessment, 23, Samejima, F. (1969). Estimation of latent ability using a response pattern
11–24. http://dx.doi.org/10.1023/A:1011035306061 of graded scores. Chicago, IL: Psychometric Society.
Lilienfeld, S. O., & Widows, M. R. (2005). Psychopathic Personality Sellbom, M. (2011). Elaborating on the construct validity of the Levenson
Inventory-Revised: Professional manual. Lutz, FL: Psychological As- Self-Report Psychopathy Scale in incarcerated and non-incarcerated
sessment Resources, Inc. samples. Law and Human Behavior, 35, 440 – 451. http://dx.doi.org/10
Lindwall, M., Barkoukis, V., Grano, C., Lucidi, F., Raudsepp, L., Liuk- .1007/s10979-010-9249-x
konen, J., & Thøgersen-Ntoumani, C. (2012). Method effects: The Siefert, C. J., Sinclair, S. J., Kehl-Fie, K. A., & Blais, M. A. (2009). An
problem with negatively versus positively keyed items. Journal of Per- item-level psychometric analysis of the Personality Assessment Inven-
SELF-REPORT PSYCHOPATHY IRT 17

tory: Clinical scales in a psychiatric inpatient unit. Assessment, 16, and Behavioral Assessment, 35, 153–161. http://dx.doi.org/10.1007/
373–383. http://dx.doi.org/10.1177/1073191109333756 s10862-012-9333-2
Skeem, J. L., & Cooke, D. J. (2010). Is criminal behavior a central Tucker, L. R., & Lewis, C. (1973). A reliability coefficient for maximum
component of psychopathy? Conceptual directions for resolving the likelihood factor analysis. Psychometrika, 38, 1–10. http://dx.doi.org/10
debate. Psychological Assessment, 22, 433– 445. http://dx.doi.org/10 .1007/BF02291170
.1037/a0008512 Vitacco, M. J., Roger, R., Neumann, C. S., Harrison, K. S., & Vincent,
Smith, G. T., McCarthy, D. M., & Anderson, K. G. (2000). On the sins of G. M. (2005). A comparison of factor models on the PCL-R with
short-form development. Psychological Assessment, 12, 102–111. http:// mentally disordered offenders: The development of a four-factor model.
dx.doi.org/10.1037/1040-3590.12.1.102 Criminal Justice and Behavior, 32, 526 –545. http://dx.doi.org/10.1177/
Smith, S. T., Edens, J. F., & Vaughn, M. G. (2011). Assessing the external 0093854805278414
correlates of alternative factor models of the Psychopathic Personality Williams, K. M., Nathanson, C., & Paulhus, D. L. (2003, August). Struc-
ture and validity of the self-report psychopathy scale-III in normal
Inventory-short form across three samples. Journal of Personality As-
populations. Paper presented at the 11th Annual Convention of the
sessment, 93, 244 –256. http://dx.doi.org/10.1080/00223891.2011
American Psychological Association, Toronto, Canada.
.558876
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Williams, K. M., & Paulhus, D. L. (2004). Factor structure of the Self-


Somma, A., Fossati, A., Patrick, C., Maffei, C., & Borroni, S. (2014). The
Report Psychopathy scale (SRP-II) in non-forensic samples. Personality
This document is copyrighted by the American Psychological Association or one of its allied publishers.

three-factor structure of the Levenson Self-Report Psychopathy Scale:


and Individual Differences, 37, 765–778. http://dx.doi.org/10.1016/j
Fool’s gold or true gold? A study in a sample of Italian adult non-clinical
.paid.2003.11.004
participants. Personality and Mental Health, 8, 337–347. http://dx.doi Williams, K. M., Paulhus, D. L., & Hare, R. D. (2007). Capturing the
.org/10.1002/pmh.1267 four-factor structure of psychopathy in college students via self-report.
Stansbury, J. P., Ried, L. D., & Velozo, C. A. (2006). Unidimensionality Journal of Personality Assessment, 88, 205–219. http://dx.doi.org/10
and bandwidth in the Center for Epidemiologic Studies Depression .1080/00223890701268074
(CES-D) Scale. Journal of Personality Assessment, 86, 10 –22. http://dx Wilson, D. L., Frick, P. J., & Clements, C. B. (1999). Gender, somatiza-
.doi.org/10.1207/s15327752jpa8601_03 tion, and psychopathic traits in a college sample. Journal of Psychopa-
Steiger, J. H. (1990). Structural model evaluation and modification: An thology and Behavioral Assessment, 21, 221–235. http://dx.doi.org/10
interval estimation approach. Multivariate Behavioral Research, 25, .1023/A:1022825415137
173–180. http://dx.doi.org/10.1207/s15327906mbr2502_4
Tonnaer, F., Cima, M., Sijtsma, K., Uzieblo, K., & Lilienfeld, S. O. (2012). Received September 23, 2016
Screening for psychopathy: Validation of the Psychopathic Personality Revision received February 24, 2017
Inventory-short form with reference scores. Journal of Psychopathology Accepted February 27, 2017 䡲

You might also like