Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Biometric models of psychopathic traits in adolescence: a comparison of item-level and sum-score approaches
(USC Thesis Other)
Biometric models of psychopathic traits in adolescence: a comparison of item-level and sum-score approaches
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Running head: BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 1
Biometric Models of Psychopathic Traits in Adolescence:
A Comparison of Item-Level and Sum-Score Approaches
Shannon K. Potts
University of Southern California
Author Note
This thesis submitted for fulfillment of the degree of Master of Arts (PSYCHOLOGY) at
the USC Graduate School in August 2018.
Shannon K. Potts, Department of Psychology, Dana and David Dornsife College of
Letters, Arts and Sciences, University of Southern California.
Correspondence concerning this manuscript should be addressed to Shannon K. Potts,
SGM 501, 3620 South McClintock Ave., Los Angeles, CA 90089-1061. Email:
shannon.potts@usc.edu
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 2
Table of Contents
Abstract .......................................................................................................................................... 4
Biometric Models of Psychopathic Traits in Adolescence: A Comparison of Item Level and
Sum Score Approaches .................................................................................................................. 5
Behavioral Genetics and Twin Study Rationale ................................................................ 6
Bayesian Markov Chain Monte Carlo Estimated Item-Response Models ........................ 8
Adolescent Psychopathy and the Manipulative/Deceitful Psychopathy Factor .............. 12
Method ......................................................................................................................................... 14
Participants ....................................................................................................................... 14
Materials .......................................................................................................................... 15
Child Psychopathy Scale ...................................................................................... 15
Failure to accept responsibility ............................................................... 16
Lack of guilt ............................................................................................. 16
Manipulative ............................................................................................ 16
Parasitic lifestyle ..................................................................................... 16
Glibness ................................................................................................... 16
Untruthfulness .......................................................................................... 17
Statistical Software and Analyses .................................................................................... 17
Results .......................................................................................................................................... 19
Genetic Models ................................................................................................................ 20
MLE Univariate Models ...................................................................................... 21
Bayesian MCMC Estimation Univariate Models ................................................ 21
MLE Common Pathway Models ......................................................................... 22
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 3
Bayesian MCMC Estimation Item-Response Models ......................................... 23
Discussion .................................................................................................................................... 25
Future Directions ............................................................................................................. 27
Conclusion ....................................................................................................................... 29
References .................................................................................................................................... 30
Tables ........................................................................................................................................... 37
Table 1 ............................................................................................................................. 37
Table 2 ............................................................................................................................. 39
Table 3 ............................................................................................................................. 40
Table 4 ............................................................................................................................. 41
Table 5 ............................................................................................................................. 42
Figures .......................................................................................................................................... 43
Figure 1 ............................................................................................................................ 43
Figure 2 ............................................................................................................................ 44
Figure 3 ............................................................................................................................ 45
Figure 4 ............................................................................................................................ 46
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 4
Abstract
Traditional behavioral genetics models use maximum likelihood estimation (MLE) and sum
scores to decompose the variance of a construct into its genetic and environmental components.
This can lead to inflated error variance and biased heritability estimates, since sum scores
eliminate the item-level intricacies in the data and MLE performs unreliably in cases of non-
normally distributed or discrete data. Twin study data (N = 864) for the manipulative/deceitful
psychopathy factor in the USC Risk Factors for Antisocial Behavior (RFAB) exhibit these
common statistical artifacts, and therefore may not be appropriate for use with MLE sum-score
models. To address these concerns, Bayesian Markov chain Monte Carlo (MCMC) estimated
genetic item-response models were fit to the manipulative/deceitful psychopathy factor’s six
subscales, and resulting heritability estimates and fit indices were compared across MLE
univariate, Bayesian MCMC univariate, and MLE common pathway models. For all subscales,
the Bayesian MCMC item-response and MLE common pathway models performed best, yielding
higher heritability estimates than either univariate model. This held especially true for the item-
response models fit to those subscales with severe skew and kurtosis, and those whose items had
the least consistent factor loadings. If the items’ factor loadings were relatively uniform and
there were no egregious departures from normality in the distribution of the observed data, the
MLE common pathway model was a potential substitute for the more complex item-response
model, producing similarly less-biased heritability estimates. Future simulation studies
comparing MLE and Bayesian MCMC estimation for behavioral genetic models could greatly
improve the field’s understanding of these methods’ performance when applied to item-level
data.
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 5
Biometric Models of Psychopathic Traits in Adolescence: A Comparison of Item Level and Sum
Score Approaches
The current research is chiefly a comparison of quantitative genetic models for
decomposing the genetic and environmental variance components, as applied to a well-studied
instrument. Methodological background and justifications are presented first, as well as
descriptions of the models to be compared. Then, the psychological concept of psychopathy and
the manipulative/deceitful psychopathy factor—to which these methods are applied—are briefly
introduced. Details regarding the relevant twin study and associated observed data are presented
after the overview of psychopathy, followed by a description of measurement details.
Behavioral Genetics and Twin Study Rationale
Behavioral genetics is principally concerned with determining the extent to which
genetic and environmental influences contribute to the development of psychological constructs
(Martin & Eaves, 1977; Martin, Eaves, Kearsey, & Davies, 1978; Nivard, Dolan, Middeldorp, &
Boomsma, 2017). In twin studies, these constructs are assessed in monozygotic (MZ; i.e.,
identical) and dizygotic (DZ; i.e., fraternal) twin pairs, and are often quantified for each
individual twin as a single sum score (Novick, 1966; Stevens, 1946). Then the total variance of
the sum scores is decomposed into its genetic and environmental components under the
assumption that MZ twins share 100% of their genetic material and 100% of their shared
environment (e.g., home, family, community), while DZ twins share on average 50% of their
genetic material and 100% of their shared environment (Martin & Eaves, 1977; Nivard et al.,
2017).
The univariate ACE model (see Figure 1) is one such model for decomposing the
variance of a single observed construct (phenotype) into its additive genetic (A), shared (or
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 6
common) environmental (C), and non-shared environmental (E) components, which yields
expected twin correlations of rMZ = A + C and rDZ = ½A + C. As such, the components of
variance can be roughly estimated as: A = 2 (rMZ – rDZ), C = rMZ – A, and E = 1 – rMZ (Martin &
Eaves, 1977; Martin et al., 1978; Nivard et al., 2017; Sullivan & Eaves, 2002). The total
phenotypic variance is constrained to unity (i.e., 1 = A + C + E), meaning that each of the
model’s variance components can be expressed as accounting for a proportion of the variance in
the phenotype (MacGregor, Snieder, Schork, & Spector, 2000). For instance, A (i.e., the
heritability h
2
of the given construct) is interpreted as the proportion of variation due to genetic
influence.
Reliance on sum scores in quantitative genetic modeling can introduce bias into the
parameter estimates and inflate the fitted model’s error variance (Liao et al., 2014). Sum scores
are a product of Classical Test Theory, which operates under the assumption than an individual’s
observed score is equal to their true score plus some random error component (i.e., measurement
error), where the random error is independent of the true score and has the distribution N(0, σ
2
;
Novick, 1966). When an observed score is a sum score, or multiple item responses from an
instrument added together or averaged into a single value, as is common in psychological
research, the measurement errors compound and bias the results (Paulhus, 1991). If the variance
of a sum score is decomposed into genetic and environmental components, the measurement
error variance is incorporated into the non-shared environmental variance (E) such that the
estimated heritability of the trait (A) is attenuated (Liao et al., 2004; van den Berg, Glas, &
Boomsma, 2007). As the reliability of the instrument falls, the error inherent in the sum score
increases and the heritability estimate declines, resulting in misleadingly low estimates of twin
correlations and of the genetic influence on a given construct.
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 7
Additionally, genetic and environmental etiology are typically estimated using maximum
likelihood methods in a structural equation modeling framework. For maximum likelihood
estimation (MLE) to arrive at unbiased estimates of the model parameters, the data need to be
relatively free of artifacts such as skewness, excess kurtosis, missing data, and floor- and ceiling-
effects (Enders, 2001; Sullivan & Eaves, 2002). Furthermore, the resulting variance components
(A, C, and E) are only interpretable if the underlying data are interval or continuous in nature
(Neale & Maes, 2004; van den Berg, 2007). Data with an approximately normal distribution and
lack of missingness, measured on an interval or continuous construct, rarely occur in real-world
psychological research, leaving fitted genetic models vulnerable to unreliable—and in extreme
cases, meaningless—parameter estimates (Enders, 2001; Sullivan & Eaves, 2002).
The common pathway ACE model takes a step in the right direction by using item-level
data instead of sum score data to determine the genetic and environmental etiology, which
eliminates sum score bias. Common pathway models analyze multivariate data to estimate the
heritability of a single latent factor (e.g., a psychological construct) given that latent factor’s
underlying observed phenotypes (e.g., a collection of items measuring that construct; see Figure
2). Common genetic (AC) and environmental (CC and EC) components influence each of the
observed phenotypes (i.e., items in this case) equally through the latent factor, while specific
genetic (AS) and environmental (CS and ES) components individually account for the remainder
of each observed phenotypic (item) variance (Neale & Maes, 2004).
However, like univariate ACE models, common pathway models are often implemented
using MLE, which may prove unreliable when data are distributed neither continuously nor
normally. Incorporating item-level data into the common pathway model can exacerbate these
incompatibilities, since measurement items frequently employ nominal or ordinal response
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 8
systems. It is only when the latent factor connecting the items is assumed to be continuous and
the researchers do not interpret AS, CS, and ES, that the common pathway model may cautiously
be considered an acceptable alternative to the sum-score univariate ACE model (Lynch &
Walsh). Although the use of item-level data in the common pathway ACE model means that the
sum-score bias is eliminated and the inflated error components are reduced, researchers with
idiosyncrasies in their item-level data may be best suited looking elsewhere for an appropriate
variance decomposition model (Neale & Maes, 2004; van den Berg et al., 2007).
Bayesian Markov Chain Monte Carlo Estimated Item-Response Models
Sum scores and MLE are suitable for genetic modeling in cases where the instrument of
measurement has demonstrated high reliability and construct validity, and where the data have
sufficiently high between-group variation, a relatively normal distribution, and no missing cases
(Enders, 2001; Liao et al., 2004; Martin, Eaves, Kearsey, & Davies, 1978; van den Berg et al.,
2007). However, as described above, these requirements are seldom met by real-world
psychological data (Reynolds & Suzuki, 2012). The item-response theory (IRT) measurement
model provides a way to extend traditional variance decomposition models and could offer a
practical solution to these issues (Eaves et al., 2005; van den Berg, 2006).
IRT models the probability of a certain response to each of an instrument’s items based
on an underlying latent construct X, with the assumption that X can be estimated by any item(s)
with a known item response function (Embretson & Reise, 2000). This paper will consider and
give a brief overview of the two-parameter logistic (2PL) IRT model, which is used for items
with binary responses (1 = endorsed and 0 = not endorsed). For a more complete description of
IRT and other IRT models, please refer to Embretson and Reise (2000) or Lord, Novick, and
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 9
Birnbaum (1968). Under the 2PL model, the probability of an individual n endorsing an item i
can be written as
P (Y
in
) =
1
1 + e
-z
where z = ai (Xn – bi), Xn is individual n’s latent construct score, ai is the discrimination of item i,
and bi is the difficulty of item i. In an IRT model, discrimination represents the relatedness of the
given item to the underlying latent construct, just as a factor loading relates an observed
phenotype to its underlying common factor (Reise, Ainsworth, & Haviland, 2005). Item
discrimination can also be thought of as the “slope” of the item characteristic curve (Embretson
& Reise, 2000). The difficulty parameter is the latent construct score X at which there is a .5
probability of an individual endorsing the given item (Embretson & Reise, 2000; Thissen, Cai, &
Bock, 2010). Therefore, difficulty may also be conceptualized as the point on the item
characteristic curve where the probabilities of endorsing (1) and not endorsing (0) an item are
equal.
The 2PL model can be easily modified to include genetic and environmental variance
components based on a sample of MZ and DZ twin pairs (Eaves et al., 2005). To do so, the
above formula is changed to
P (Y
i jk
) =
1
1 + e
-z
where Yijk is the response to item i for twin j in family k, and z = ai (Xjk – bi) where Xjk is the
latent construct score X of twin j in family k, ai is the discrimination of item i, and bi is the
difficulty of item i. Under this new model, it is assumed that there is a single underlying
construct variable X with identical factor loadings for MZ and DZ twins as well as for twins 1
and twins 2 within the families. Additive genetic (A), shared environmental (C), and nonshared
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 10
environmental (E) influences for the MZ and DZ twins determine the covariances of the latent
construct’s assumed bivariate normal distribution (Eaves et al., 2005). The item parameters are
scaled in this model so that the latent construct’s variance is 1 and can be decomposed into
interpretable estimates of the A, C, and E components (Eaves et al., 2005).
According to Eaves et al. (2005, p. 766), “Part of the problem of applying IRT to twin
data stems from the fact that although it is relatively easy to write the likelihood for a ‘twin IRT’
model…it is far harder to maximize the likelihood with respect to the parameters of a genetic
IRT model because of the numerical problem of integrating the likelihood of individual pairs
over the infinity of possible latent trait values in multiple dimensions.” In other words, due to the
complexity of these multivariate genetic models and IRT’s basic assumption of the underlying
latent construct’s distribution, MLE is computationally intractable (Eaves et al., 2005; van den
Berg, Beem, & Boomsma, 2006). And even if structural equation modeling using MLE of the
model parameters were practical, the observed binary item responses and frequency of missing
data in self-report survey responses would raise the same issues regarding accuracy and
interpretability of the variance components as with the univariate and common pathway models
(Enders, 2001; Sullivan & Eaves, 2002; van den Berg et al., 2007).
A viable alternative to MLE in this situation is Bayesian inference via Markov chain
Monte Carlo (MCMC) estimation (Curtis, 2010; Thomas, 2014; van den Berg, Beem, &
Boomsma, 2006). Frequentist methods of inference approach statistics from the standpoint of
determining P(Evidence | Hypothesis). In contrast, Bayesian inference aims to determine
P(Hypothesis | Evidence). In terms of model fit and parameter estimation, Bayesian inference
seeks the posterior probability, where the posterior density function for the model parameters η
and data Y, is P(η | Y). P(η | Y) is proportional to the product of the prior distribution P(η)
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 11
specified a priori by the researcher and the likelihood function P(Y | η). For a detailed description
of the principles of Bayesian inference, refer to Bayesian Data Analysis by Gelman, Carlin,
Stern, and Rubin (2004).
Because the model parameters’ joint posterior density function is too complex to sample
from directly, MCMC algorithms (specifically, Gibbs sampling) instead sample the model
parameters from the conditional distributions over a sequence of iterations. In MCMC
algorithms, each Markov chain is composed of sequential, correlated iterations. After a series of
burn-in iterations, the conditional distributions converge on the joint posterior density function,
and it can be considered that the MCMC algorithm is sampling model parameters from this
desired distribution (Casella & George, 1992; van den Berg et al., 2006). Once additional
samples have been taken, the mean or median of the iterations’ parameter estimates is used as the
model parameter’s point estimate (Casella & George, 1992; van den Berg et al., 2006).
Estimating heritability within this Bayesian MCMC item-response framework results in
flexible estimation of the genetic and environmental variance components, allows for skewness,
categorical observations, and other departures from normality in the data, eliminates sum-score
bias, and tolerates missingness (Curtis, 2010; Eaves et al., 2005; Liao et al., 2004; van den Berg
et al., 2006; van den Berg et al., 2007). Because the fitted models are intended for use with
dichotomous item-level data, Bayesian MCMC estimated item-response ACE models are more
flexible than MLE univariate or common pathway ACE models, and grant the researchers
freedom in the data they choose for variance decomposition. Thus, heritability estimates from
these methods should better approximate the true heritability of the construct of interest than
those from the sum score models.
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 12
Theoretical superiority of the item-response ACE model has been demonstrated by a brief
simulation study of MLE sum score biases (van den Berg et al., 2007) and two applications of
different Bayesian item-response ACE models to well-validated, approximately normally-
distributed, dichotomous (Eaves et al., 2005) and polytomous data (van den Berg et al., 2006)
with consistently high factor loadings on the latent construct. However, these variance
decomposition methods have not yet been tested on skewed data with suboptimal factor loadings,
even though these types of data would potentially benefit most from variance decomposition
using item-response models as opposed to sum score models.
The current study aims to fill this gap in the literature by fitting both MLE and Bayesian
MCMC estimated genetic models (i.e., univariate, common pathway, and item-response) to a
more realistic data set (i.e., manipulative/deceitful factor of adolescent psychopathy) which falls
short of meeting the assumptions for MLE genetic models. It is hypothesized that the Bayesian
MCMC genetic item-response models will produce the highest heritability estimates and
correspondingly the lowest non-shared environmental and error variance components for these
data. Due to their incorporation of item-level data as opposed to sum scores, MLE common
pathway models are hypothesized to produce the next-best estimates of heritability. Both item-
response and common pathway models are expected to produce superior parameter estimates
compared to the MLE and Bayesian MCMC univariate ACE models.
Adolescent Psychopathy and the Manipulative/Deceitful Psychopathy Factor
For decades psychologists and behavior geneticists have measured psychopathic
tendencies in children and adolescents in an attempt to accurately predict which individuals are
most at-risk of maintaining this potentially dangerous behavior into adulthood (Bezdjian, Raine,
Baker, & Lynam, 2011; Fink, Tant, Tremba, & Kiehl, 2012; Lynam, 1966; Lynam, Caspi,
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 13
Moffitt, Loeber, & Stouthamer-Loeber, 2007; Lynam et al., 2009). Psychopathy may be briefly
described as a set of longitudinally stable personality traits including but not limited to
remorselessness, egocentricity, manipulativeness, ingenuine emotion, impulsivity, unreliability,
and antisocial behavior (Lynam et al., 2009; Tuvblad, Bezdjian, Raine, & Baker, 2014; Wang,
Baker, Gao, Raine, & Lozano, 2012). Adolescent psychopathy has been found to be strongly
predictive of adult psychopathy (Lynam et al., 2007; Lynam et al., 2009).
Although researchers agree that it is vital to identify and attempt to mitigate psychopathy
early in one’s life, there is disagreement on how best to conceptualize this combination of traits
in adolescence versus adulthood (Frick, O’Brien, Wootton, & McBurnett, 1994; Lynam, 1997;
Bezdjian, Raine, et al., 2011). Frick et al. (1994) proposed a two-factor structure of a callous
unemotional factor and impulsivity/conduct problems factor for adolescent psychopathy based
on Hare’s (1990) adult psychopathy factors of “selfish, callous, and remorseless use of others”
and “chronically unstable, antisocial, and socially deviant lifestyle”. However, these factors were
found to have poor performance when applied to alternate samples (Bezdjian, Raine, et al., 2011;
Fink et al., 2012). After failing to confirm Frick et al.’s (1994) two-factor structure, Bezdjian,
Raine, et al. (2011) found evidence for a callous/disinhibited factor and a manipulative/deceitful
factor in a sample of 9- to 10-year-old twins.
Bezdjian, Raine, et al.’s (2011) manipulative/deceitful factor is composed of six
constructs (i.e., sum-score scales) from the Child Psychopathy Scale: failure to accept
responsibility, lack of guilt, manipulativeness, parasitic lifestyle, glibness, and untruthfulness
(Lynam, 1997). In adolescent samples, the constructs in this factor have been found to have
moderate factor loadings, and the factor itself was found to have a moderate-to-high heritability
of around .46 to .64 (Bezdjian, Raine, et al., 2011; Tuvblad et al., 2014). Like overall
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 14
psychopathy, scores for the manipulative/deceitful factor are strongly positively skewed, as items
for this factor are rarely endorsed. The manipulative/deceitful psychopathy factor will be used to
fit genetic variance decomposition models in the current study.
Method
Participants
Data were drawn from the University of Southern California (USC) longitudinal twin
study of Risk Factors for Antisocial Behavior (RFAB). The USC RFAB sample is
demographically diverse and includes over 600 monozygotic (MZ) and dizygotic (DZ) twin pairs
from the Greater Los Angeles Area, which covers five counties and an estimated population of
18.69 million (U.S. Census Bureau, 2017). Twins and their families were recruited through local
school enrollment records, advertisements, and twin clubs. Data collection is ongoing, and began
when participants were 9-10 years old (Wave 1) and continued with follow-up waves spaced
approximately two years apart through ages 19-20 (Wave 5). At each wave, twins completed a
battery of psychosocial and psychophysiological measures while primary caregivers completed a
battery of psychosocial measures regarding family environment and behaviors of each twin.
During school-age years, teachers completed mail-in surveys regarding the twins’ behavior. For
details regarding USC RFAB’s recruitment and data collection processes, see Baker, Barton, et
al. (2006) and Baker, Tuvblad, et al. (2013).
The current research uses data from Wave 3 of the USC RFAB, originally composed of
604 participating twin pairs. However, incomplete surveys and families lacking data from one or
both twins were excluded from analyses, leaving a final sample of 432 twin pairs (864
individuals) with a mean age of 13.9 years. 175 (40.4%) of these were MZ pairs (87 male MZ
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 15
and 88 female MZ) and 257 (59.3%) were DZ pairs (74 male DZ, 78 female DZ, and 105 cross-
sex DZ). In total, 437 (50.6%) of participants were female and 427 (49.4%) were male.
In addition to having a relatively equal gender split, USC RFAB’s Wave 3 sample was
approximately representative of the Greater Los Angeles Area’s racial diversity (U.S. Census
Bureau, 2016). 32.9% (142) of the twin pairs were Hispanic, 30.8% (133) were Caucasian,
11.8% (51) were African-American, 3.9% (17) were Asian, and 19.7% (85) were of mixed race.
A further 0.9% (4) of families chose not to report their race.
Materials
Child Psychopathy Scale. Lynam (1997) developed the Child Psychopathy Scale (CPS)
as a translation of Hare’s (1991) well-validated adult Psychopathy Checklist. The CPS was
administered to participating twins as a self-report measure composed of 58 items broken into 14
subscales (Lynam, 1997). Of these, only the six subscales (23 items) that make up Bezdjian,
Raine, et al.’s (2011) manipulative/deceitful psychopathy factor were used in the current
analyses: failure to accept responsibility (three items), lack of guilt (three items), manipulative
(three items), parasitic lifestyle (four items), glibness (five items), and untruthfulness (five
items). The questionnaire asks participants about their “way of doing things” with “yes”, “no”,
and “don’t know” response options for each item. Items were scored as endorsed (1) or not
endorsed (0). The manipulative/deceitful factor’s 8 reverse-coded items were scored as endorsed
(0) or not endorsed (1). To calculate sum scores according to Lynam’s (1997) original
formulation, each participant’s responses for all items in a given subscale were averaged. To
calculate Bezdjian, Raine, et al.’s (2011) manipulative/deceitful psychopathy factor, all six
subscales’ sum scores were averaged. Thus, each subscale’s sum scores and the
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 16
manipulative/deceitful factor were distributed from zero to one about the overall subscale mean
(for subscale items, means, and standard deviations, see Table 1).
Failure to accept responsibility. The failure to accept responsibility subscale is
composed of three items measuring the degree to which the participant attempts to lie in order to
deny responsibility for their behavior, or to shift blame for their own actions onto others.
Cronbach’s α = .19.
Lack of guilt. Lack of guilt also contains three items but asks participants about feelings
of remorse for their “wrong” actions. For example, “Do you wish you could take back many
things that you have done wrong?” All items on the lack of guilt subscale are reverse-coded so
that higher scores indicate lower intensity of guilt. Cronbach’s α = .66.
Manipulative. The manipulative subscale is composed of three items, none of which are
reverse-coded. Higher manipulative scores suggest a participant who is liable to act
charismatically around others for their own, selfish purposes. Cronbach’s α = .61.
Parasitic lifestyle. Four items (two reverse-coded) are in the parasitic lifestyle subscale.
The parasitic lifestyle subscale measures the extent to which a participant takes from others
without giving back in return. For instance, one item reads, “Do you try to see how much you
can get away with?” Cronbach’s α = .35.
Glibness. Glibness is defined by Merriam-Webster’s Collegiate Dictionary as, “fluency
in speaking or writing often to the point of being insincere or deceitful,” (Glib, n.d.).
Accordingly, this subscale measures the degree to which a participant is talkative, attention-
seeking, and insincere. The glibness subscale is composed of five items total, with only one
reverse-coded item (i.e., “Are you shy?”). Cronbach’s α = .43.
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 17
Untruthfulness. Finally, untruthfulness simply measures tendency toward dishonesty.
Adolescents who score highly on the untruthfulness subscale indicate that they cannot be trusted,
and tend to lie frequently and convincingly. Two items on the untruthfulness subscale are
reverse-coded. Cronbach’s α = .14. These subscales were formulated according to Lynam’s
(1997) research and as a translation of Hare’s Psychopathy Checklist, in spite of its subscales
yielding some rather low internal consistencies.
Statistical Software and Analyses
Prior to analyses, data management and file formatting were completed using IBM’s
SPSS (Version 22.0) and Microsoft Excel (Version 1710). All maximum likelihood analyses
were carried out using the ‘OpenMx’ (Version 2.7.18), ‘psych’ (Version 1.7.8), and ‘lavaan’
(Version 0.5-23.1097) packages for the R programming language (Version 3.4.1; Neale et al.,
2016; Revelle, 2017; Rosseel, 2012). Bayesian analyses were carried out using OpenBUGS
(Version 3.2.3; Thomas, 2014). OpenBUGS utilizes a Markov Chain Monte Carlo (MCMC)
algorithm commonly used in Bayesian inference known as Gibbs Sampling (Thomas, 2014).
Like all MCMC estimation, Gibbs Sampling requires a sequence of burn-in iterations but
employs randomness in its algorithms, making it a flexible option for parameter estimation
(Congdon, 2007; Curtis, 2010; Thomas, 2014). For each model, parameters were estimated on
four chains of 9000 iterations following a 1000-iteration-per-chain burn-in period (i.e., 36,000
samples total distributed over four Markov chains).
Four categorically different models were fit to each of the six CPS subscales as well as to
the overall manipulative/deceitful psychopathy factor: (1) MLE univariate genetic models (e.g.,
Figure 1), (2) Bayesian MCMC univariate genetic models, (3) MLE common pathway genetic
models (e.g., Figure 2), and (4) Bayesian MCMC item-response genetic models. Binary CPS
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 18
item responses for each subscale were used in the common pathway models and item-response
models. Both MLE and Bayesian MCMC univariate models used subscale sum scores. For the
manipulative/deceitful factor models, an average of all subscales’ sum scores was calculated and
used in fitting the univariate models. The manipulative/deceitful factor’s common pathway and
item-response models were fit using dichotomized (by median split so 0 = low and 1 = high)
versions of the six subscales’ sum scores as item-level data, as opposed to the 23 total items in
the factor. This was done to hold model size (i.e., number of estimated parameters) consistent
across analyses and so that number of estimated parameters would not exceed total number of
twin pairs for any model (Jackson, 2009).
Prior distributions for the Bayesian univariate and item-response models were
purposefully vague and minimally informed, so as to facilitate a comparison of the fitted
frequentist and Bayesian models without introducing the further discrepancy of informative
priors in the Bayesian models. In other words, reasonable priors were informed by the
characteristics of genetic and environmental variance components and IRT 2PL model item
parameters, but were not informed by previous findings regarding the heritability of the
manipulative/deceitful psychopathy factor or the CPS subscales themselves. The univariate
models’ parameters were assigned the following prior distributions: subscale means μ ~ N(0.5,
0.1), families φ ~ N(μ, σ
2
Genetic), twins ~ N(φ, σ
2
Environment), rMZ ~ U(0,1), and rDZ ~ U(0,1). The
item-response models required additional priors: item responses Y ~ B(1, .5), item difficulty a ~
U(−1, 3), item discrimination b ~ U(−1, 6), and latent construct θ ~ N(0, 1)—one of IRT’s
foundational assumptions (Reise et al., 2005). This method of choosing priors that are “vague
within a realistic range for the data set under consideration,” is the method recommended by
Lambert, Sutton, Burton, Abrams, and Jones (2005, p. 2424) based on the findings of their
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 19
comprehensive simulation study regarding the impact of vague priors in Bayesian MCMC
estimation, and is further supported by a comparative meta-analysis by Smith, Spiegelhalter, and
Thomas (1995).
Results
For each item in the manipulative/deceitful psychopathy factor, the means and standard
deviations of male and female responses were calculated and are displayed in Table 1. Gender
differences in response means were statistically significant for only two of these 23 items. “Are
you talkative?” from the glibness subscale was more frequently endorsed by females [t(865) = –
4.79, p < .0005], while “Will you usually tell a lie if you think you can get away with it?” from
the untruthfulness subscale was more frequently endorsed by males [t(865) = 3.10, p = .002]. No
statistically significant gender differences were found for any of the six subscales’ sum scores, or
for the overall manipulative/deceitful sum score (consistent with Larsson, Andershed, &
Lichtenstein, 2006). Thus, all further analyses were conducted using the full twin sample (i.e.,
male same-sex twin pairs, female same-sex twin pairs, and cross-sex twin pairs), and the models
fit to these data did not control for gender either in means or variance components.
Skewness of the items is also presented in Table 1, and ranged from −0.71 (“Are you
talkative?” from the glibness subscale) to 3.86 (reverse-scored “Do you usually return what you
borrow?” from the parasitic lifestyle subscale). Nineteen of the 23 items were positively skewed.
The four negatively skewed items were: “Do people usually believe you when you tell a lie?”
(untruthfulness subscale), “Are you shy?” (reverse-scored; glibness subscale), “Do you think you
get blamed for things you did not do?” (failure to accept responsibility subscale), and “Are you
talkative?” (reverse-scored; glibness subscale). In other words, more twins endorsed the
previously-listed four items than not, indicating that most participants: are usually believed when
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 20
they lie to others, are shy, get blamed for things they did not do, and are talkative. Skewness of
the sum scores of the six subscales ranged from 0.21 (failure to accept responsibility) to 1.43
(lack of guilt). Further, subscale sum-score distributions exhibited excess kurtosis values ranging
from −0.60 (untruthfulness) to 1.19 (parasitic lifestyle; see Table 1).
To examine the fit of the 23 items to their respective CPS subscales, a six-factor
confirmatory factor analysis (CFA) was conducted using the ‘lavaan’ package (Version 0.5-
23.1097) in R (Rosseel, 2012). Table 1 contains factor loadings for this CFA, while Table 2
reports the associated factor correlations. The six-factor model was a significant improvement
over a one-factor model, with a comparative fit index (CFI) of .77 and root mean squared error
of approximation (RMSEA) of .06 [χ
2
(215) = 521.986, p < .001]. Findings were consistent with
Bezdjian, Raine, et al.’s (2011) CFA results, in that the items loaded erratically onto their
designated subscales. Of these, the manipulative subscale’s three items had the most consistent
factor loadings, ranging from .43 to .58, while the lack of guilt subscale’s three items were most
disparate, ranging from .03 to .79 (see Table 1). Again, this is consistent with Bezdjian, Raine, et
al.’s (2011) poor factor loadings for the lack of guilt subscale. Inter-factor correlations between
these six subscales in the manipulative/deceitful psychopathy factor were moderate to high
(Table 2).
Genetic Models
Table 3 reports the tetrachoric twin correlations for the MZ and DZ twin pairs, as
obtained using the R package, ‘psych’ (Version 1.7.8; Revelle, 2017). The twin correlations are
higher for the MZ pairs than the DZ pairs for all but one item (“When you get in trouble, can you
talk your way out of it?” in the fail to accept responsibility subscale), signifying a genetic
influence (A) on the item responses.
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 21
MLE univariate models. Both univariate ACE and AE models were fit to each of the
CPS subscales’ sum-score data using MLE carried out through the ‘OpenMx’ (Version 2.7.18)
package in R (Neale et al., 2016). The subscale sum scores used in these models were calculated
by averaging item responses (0 = No, 1 = Yes; if reverse-scored, 1 = Yes, 0 = No) for each
subscale and for the overall manipulative/deceitful psychopathy factor. ACE and AE models’
goodness-of-fit were compared using the –2 Log Likelihood (–2LL) and Akaike’s Information
Criterion (AIC). Lower values of both these fit indices are indicative of more parsimonious,
better-fitting models (Bozdogan, 1987). For every subscale, the common environmental variance
component (C) of the ACE model was found to be nonsignificant at the α = .05 level, leaving the
more parsimonious AE model as the best fit for these data. This is consistent with Bezdjian,
Raine, et al.’s (2011) findings regarding the non-significance of C for the manipulative/deceitful
factor. Given the non-significance of C, and for the sake of consistency when comparing the
final models, AE models were fit to the data for the remaining analyses. Fit indices for the six
subscales’ and the total manipulative/deceitful factor’s univariate AE models are displayed in
Table 5.
Heritability estimates for these models ranged from h
2
= .16 for the parasitic lifestyle
subscale to h
2
= .35 for the untruthfulness subscale and h
2
= .46 for the manipulative/deceitful
psychopathy factor. See Table 4 for all heritability estimates and accompanying standard errors.
Additionally, see Table 5 for MLE univariate AE models’ −2LL and AIC fit indices.
Bayesian MCMC estimation univariate models. Univariate and item-response
Bayesian models were fit to the data using the Bayesian inference program OpenBUGS (Version
3.2.3; Thomas, 2014). Model parameters were estimated from 9000 samples on four Markov
chains (36000 samples total per model) after 1000 burn-in iterations per chain. The subscale sum
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 22
scores used in the Bayesian univariate models were the same used in the MLE univariate models
and were calculated by averaging item responses for each subscale and for the overall
manipulative/deceitful psychopathy factor. To make the Bayesian models as comparable to the
MLE models as possible, uninformative (and identical across all subscales) priors were used for
parameter estimation, and OpenBUGS generated randomized initial values for the MCMC
chains.
The Bayesian univariate AE model fit was assessed using D = 2 * rDZ – rMZ. Negative or
near-zero positive D values suggest that the data lack a shared environmental component, C, and
are therefore better modeled using an AE model as opposed to a full ACE model (Nivard, Dolan,
Middledorp, & Boomsma, 2017). D values for each subscale’s fitted AE model were drawn from
the medians of the four Markov chains’ estimated D values, and are reported in Table 5. Unlike
the MLE models, the Bayesian MCMC univariate models suggest the possible presence of a
shared environmental component for all the subscales’ sum scores.
The heritability estimates for these univariate AE models were also drawn from the
medians of the four Markov chains’ estimated heritability parameters. The Bayesian univariate
AE models produced heritability estimates with substantially greater variances than the models
obtained through MLE (see Table 4). However, the Bayesian MCMC heritability estimates were
not consistently higher or lower than the MLE heritability estimates. The manipulative
subscale’s sum score obtained the lowest heritability of these fitted models (h
2
= .16), while lack
of guilt obtained the highest (h
2
= .39; Table 4). The manipulative/deceitful psychopathy factor’s
sum score had a heritability of .38 with a standard error of .02.
MLE common pathway models. After fitting univariate AE models to the subscales’
sum scores, MLE common pathway models and Bayesian MCMC estimated item-response
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 23
models were fit to the individual items of each subscale. Item scores were binary: 0 = no and 1 =
yes, or for reverse-coded items, 0 = yes and 1 = no. For the manipulative/deceitful factor, the six
associated subscales were converted into binary variables and treated as “item” data (see the
Methods section for details). Items were incorporated into each common pathway model as
observed variables being influenced by genetic and environmental effects through a single latent
phenotype (i.e., the items’ associated subscale). ACE and AE common pathway models were fit
to the data using MLE in the ‘OpenMx’ (Version 2.7.18) package in R (Neale et al., 2016).
Like the MLE univariate sum-score models, shared environmental effects, C, were not
statistically significant in any of the subscales’ common pathway ACE models. Therefore, AE
models were fit to the data. Fit indices and heritability estimates for these common pathway AE
models are reported in Tables 5 and 4, respectively. Heritability estimates for all subscales were
greater than those found using either univariate method, with the exceptions of parasitic lifestyle
(h
2
= .12) and untruthfulness (h
2
= .24). Parasitic lifestyle had the lowest common pathway AE
model heritability estimate, while failure to accept responsibility (h
2
= .61) and the
manipulative/deceitful psychopathy factor (h
2
= .68) had the highest heritability estimates.
Bayesian MCMC estimation item-response models. Bayesian single-parameter item-
response models and genetic variance decomposition models were fit to the item data
simultaneously (i.e., single-step) using the Bayesian inference software OpenBUGS (Version
3.2.3; Thomas, 2014). Model parameters were estimated from 9000 samples on four Markov
chains (36000 samples total per model) after 1000 burn-in iterations per chain. Like the
univariate Bayesian models, uninformative priors were used for estimating the model
parameters, and OpenBUGS randomly generated initial values for the MCMC chains. AE model
fit and appropriateness of including the shared environmental component, C, were assessed using
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 24
the D parameter. Item data was identical to that used in the common pathway models, including
the binary overall subscale scores for the manipulative/deceitful psychopathy factor.
Unlike the Bayesian univariate models, D was near-zero or negative for all AE item-
response models. The highest D value was .26 for both the manipulative and untruthfulness
subscales. Glibness had the lowest estimated D value of −.41, which—while not indicative of
shared environmental influence on this subscale—may suggest the presence of non-additive
genetic effects.
For all subscales the item-response AE model resulted in greater heritability estimates
than those from either the MLE univariate or Bayesian MCMC univariate models. Furthermore,
the item-response models produced higher heritability estimates than the common pathway
models in all but two cases: the failure to accept responsibility subscale (common pathway h
2
=
.61; item-response h
2
= .45) and the manipulative/deceitful factor (common h
2
= .68; item-
response h
2
= .54). Although the parasitic lifestyle subscale obtained the lowest heritability
estimates amongst all the fitted models for the MLE univariate and common pathway models (h
2
= .16 and h
2
= .12, respectively), the Bayesian MCMC item-response AE model estimated
parasitic lifestyle heritability to be .72. For a full report of all model heritability estimates, see
Table 4. Figure 3 is a graphical representation of the estimated heritabilities for all subscales and
associated fitted models.
Figure 4 displays the differences between the item-response and univariate heritability
estimates as well as the item-response and common pathway heritability estimates. Figure 4’s top
panel displays the skew and kurtosis of each subscale’s sum-score distribution. The three items
with the greatest combined skew and kurtosis (lack of guilt, parasitic lifestyle, and
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 25
untruthfulness) exhibited the greatest increase in estimated heritability when switching from the
univariate or common pathway models to the item-response models.
Discussion
The results support the original hypotheses that Bayesian MCMC estimated item-
response models would result in higher heritability estimates and lower corresponding error
variance than the simpler competing models, except in the cases of the failure to accept
responsibility subscale and the manipulative/deceitful psychopathy factor itself, where the MLE
common pathway models produced higher heritability estimates than the item-response models.
Common pathway and item-response heritability estimates were most discrepant when
factor loadings for the items on a particular subscale were not uniform (i.e., when items had
inconsistent factor loadings on their given subscale). This can be seen in the lack of guilt,
parasitic lifestyle, and untruthfulness heritability estimates (Table 1 & Figure 4). These subscales
were all highly skewed and had high excess kurtosis (Figure 4). The superiority of the fitted
Bayesian MCMC item-response models’ parameter estimates over the MLE univariate and
common pathway models’ parameter estimates for these subscales may be attributed to several
factors. First, MLE methods are ill equipped at estimating model parameters from non-normally
distributed data (Enders, 2001; Sullivan & Eaves, 2002). And while Bayesian MCMC estimation
is not a robust method, a brief simulation study by van den Berg et al. (2007) found that for non-
normal or discrete data, Bayesian MCMC estimated item-response models produced more
accurate heritability estimates than MLE sum scores models. Second, the common pathway
model assumes that all of a latent construct’s observed phenotypes (i.e., items) have identical
factor loadings on the latent construct. When this assumption does not hold, the model’s
resulting estimates may prove biased, reducing the heritability estimate (A) and adding error to
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 26
the nonshared environmental variance component (E). This pitfall is avoided in the item-
response model because in addition to using the item data itself, the individual items each have
their own item characteristic curves and conditional distributions from which to draw parameter
estimates (Casella & George, 1992; Embretson & Reise, 2000; van den Berg et al., 2006).
The parasitic lifestyle subscale is perhaps the best example of non-normally distributed
data in this study’s sample. The univariate models and common pathway model all exhibited
evidence of high error components for the parasitic lifestyle subscale, which was both positively
skewed and had high excess kurtosis (Table 1). The parasitic lifestyle subscale also had several
items with negative DZ tetrachoric twin correlations (Table 2), and factor loadings ranging from
.1 to .65 for the four items. For this subscale, Bayesian MCMC estimation of an item-level model
leads to substantial gains in the heritability estimate and error reduction compared to any other
fitted model. These results suggest that for real-world psychological data, Bayesian MCMC
estimation in combination with item-level data is superior to MLE for behavior genetic variance
decomposition. Additionally, allowing the item parameters to vary (e.g., in the item-response
model) instead of holding them constant within the subscale (e.g., in the common pathway
model) is preferable for instruments lacking internal consistency.
MLE and Bayesian MCMC estimation performance can be directly compared in the two
fitted univariate models for each subscale. Neither the MLE nor Bayesian MCMC univariate
model proved superior, since the MLE model produced higher heritability estimates for three
subscales (i.e., manipulative, glibness, and untruthfulness) plus the manipulative/deceitful
psychopathy factor while Bayesian MCMC produced higher heritability estimates for the others
(i.e., failure to accept responsibility, lack of guilt, and parasitic lifestyle; Figure 3). It is
noteworthy, however, that the Bayesian MCMC heritability point estimates had higher standard
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 27
errors than the MLE estimates. In totality, neither univariate model proved sufficient in
estimating the heritabilities of these subscales, as they consistently fell short of the item-response
model’s estimates. Neither MLE nor Bayesian MCMC estimation could compensate for
increased measurement error and lack of item-level data in these univariate models.
Future Directions
The next step in evaluating the performance of these Bayesian MCMC estimated item-
response models for genetic and environmental variance decomposition is to set up a systematic
simulation study to evaluate the performance across a variety of conditions. A large-scale
simulation study could examine the effects of varying the distribution of item factor loadings,
number of items, items’ difficulty and discrimination, and items’ MZ and DZ twin correlations
on the performance of all four types of quantitative genetic models presented in this paper.
Additionally, varying skewness and kurtosis of the instrument and item responses, as well as the
number of response options per item, could inform researchers as to when departures from a
continuous, normal distribution become severe enough to warrant fitting item-response models
over the more ubiquitous and more easily fit MLE common pathway models. There is very little
information on the performance of these models so far, so it is vital that these investigations
continue so model choices can be as informed and appropriate to the given data as possible.
Incorporating robust methods into the Bayesian MCMC models could be another
important step in improving the accuracy of these models. Simulated sensitivity analyses have
found that using a heavier-tailed Student’s t-distribution prior in place of a normal distribution
prior had little effect on the resulting posterior parameter estimates (Smith et al., 1995).
However, this has not been studied in behavior genetic models, and would be a natural extension
of the current research given the plethora of statistical artifacts, discrete data, and non-normality
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 28
present in psychological research. Carefully examining the effects of different, robust priors on
behavior genetic models estimated with Bayesian MCMC may provide analysts with the
information necessary to confidently apply these item-response models to their own data.
The overall findings of this study highlight several situations under which the item-
response model demonstrated clear advantages over the competing models. However, some
technical limitations were encountered that, if remedied, could improve the applicability of these
models. Many common MCMC software have shortcomings when used for variance
decomposition with these models. OpenBUGS, a widely used Bayesian Gibbs sampling
software, is an opaque program with poor exception handling, and is unstable for large models
(Gelman, 2009; Lunn, Thomas, Best, & Spiegelhalter, 2000). Additionally, due to the lack of a
parallel implementation of the MCMC algorithm, OpenBUGS’ convergence times can be very
slow for computationally intensive models such as the multivariate variance decomposition
models used in this study (Gelman, 2009; Lunn, Thomas, Best, & Spiegelhalter, 2000).
Improving OpenBUGS’ transparency when compiling and updating models, as well as
incorporating into the program a parallel implementation for independent Markov chains would
improve the user-friendliness of OpenBUGS and give analysts more control over their fitted
models.
JAGS is a Gibbs sampler that in many instances can act as an alternative to OpenBUGS,
such as with large datasets (Depaoli, Clifton, & Cobb, 2016). However, JAGS encounters similar
trouble fitting hierarchical multivariate models as OpenBUGS, making it unsuitable for larger
implementations of the genetic item-response model presented in this paper. Stan, a Hamiltonian
Monte Carlo sampler, may be an alternative to OpenBUGS for fitting models similar to those
described in this paper (Carpenter, 2017). However, Stan cannot handle discrete parameters,
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 29
which is cause for concern when it comes to the analysis of psychological data (Carpenter,
2017).
Conclusion
Bayesian estimation with item-response models is flexible enough to handle the artifacts
inherent in psychological data, such as non-ideal response distributions and categorical data,
making it a more reliable choice than MLE common pathway models. However, provided that
factor loadings on the latent construct are relatively uniform across items and there are no
egregious departures from normality in the distribution of the observed data, MLE genetic
common pathway models are also an improvement over univariate ACE models, producing
reduced variance error and less-biased heritability estimates. Future simulation studies
comparing quantitative genetic univariate, common pathway, and item-response models using
MLE and Bayesian MCMC estimation would greatly improve the field’s understanding of these
models’ performance, and would aid researchers in making the most informed choices possible
when decomposing the genetic and environmental etiologies of their data.
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 30
References
Baker, L. A., Barton, M., Lozano, D. I., Raine, A., & Fowler, J. H. (2006). The southern
California twin register at the University of Southern California: II. Twin Research and
Human Genetics, 9(6): 933–940. doi: 10.1375/183242706779462912
Baker, L. A., Tuvblad, C., Wang, P., Gomez, K., Bezdjian, S., Niv, S., & Raine, A. (2013). The
southern California twin register at the University of Southern California: III. Twin
Research and Human Genetics, 16(1), 336–343.
Bezdjian, S., Raine, A., Baker, L. A., & Lynam, D. R. (2011). Psychopathic personality in
children: Genetic and environmental contributions. Psychological Medicine, 41, 589–
600. doi: 10.1017/S0033291710000966
Bozdogan, H. (1987). Model selection and Akaike’s Information Criterion (AIC): The general
theory and its analytical extensions. Psychometricka, 52(3), 345–370. Retrieved from
https://link.springer.com/article/10.1007/BF02294361
Carpenter, B., Gelman, A., Hoffman, M. D., Lee, D., Goodrich, B., Betancourt, M., … Riddell,
A. (2017). Stan: A probabilistic programming language. Journal of Statistical Software,
76(1). doi: 10.18637/jss.v076.i01
Casella, G. & George, E. I. (1992). Explaining the Gibbs sampler. The American Statistician,
46(3). doi: 10.2307/2685208
Congdon, P. (2007). Bayesian statistical modelling (2nd ed.). Chichester, England: John Wiley
& Sons Ltd.
Curtis, S. M. (2010). BUGS code for item response theory. Journal of Statistical Software, 36(1),
1–34.
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 31
Depaoli, S., Clifton, J. P., & Cobb, P. R. (2016). Just another Gibbs sampler (JAGS): Flexible
software for MCMC implementation. Journal of Educational and Behavioral Statistics,
41(6), 628–649. doi: 10.3102/1076998616664876
Eaves, L., Erkanli, A., Silberg, J., Angold, A., Maes, H. H., & Foley, D. (2005). Application of
Bayesian inference using Gibbs sampling to item-response theory modeling of multi-
symptom genetic data. Behavior Genetics, 35(6), 765–780.
Embretson, S. E. & Reise, S. P. (2000). Item response theory for psychologists. Mahwah, NJ: L.
Erlbaum Associates.
Enders, C. K. (2001). The impact of nonnormality on full information maximum-likelihood
estimation for structural equation models with missing data. Psychological Methods,
6(4), 352–370. doi: 10.1037/1082-989X.6.4.352
Fink, B. C., Tant, A. S., Tremba, K., & Kiehl, K. A. (2012). Assessment of psychopathic traits in
an incarcerated adolescent sample: A methodological comparison. Journal of Abnormal
Child Psychology, 40, 971–986. doi: 10.1007/s10802-012-9614-y
Frick, P. J., O’Brien, B. S., Wootton, J. M., & McBurnett, K. (1994). Psychopathy and conduct
problems in children. Journal of Abnormal Psychology, 103, 700–707.
Gelman, A. (2009). Some thoughts on the BUGS package for Bayesian analysis. Statistics in
Medicine. Retrieved from
http://www.stat.columbia.edu/~gelman/research/published/bugsnext2.pdf
Gelman, A., Carlin, J. B., Stern, H. S., & Rubin, D. B. (2004). Bayesian data analysis (2nd ed.).
Washington, DC: Chapman and Hall.
Glib. (n.d.). In Merriam-Webster’s collegiate dictionary. Retrieved from https://www.merriam-
webster.com/dictionary/glib
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 32
Hare, R. D., Harpur, T. J., Hakstian, A. R., Forth, A. E., & Hart, S. D. (1990). The revised
psychopathy checklist: Reliability and factor structure. Psychological Assessment: A
Journal of Consulting and Clinical Psychology, 2(3), 338–341.
Hare, R. D. (1991). Manual for the Hare Psychopathy Checklist—Revised. North Tonawanda,
NY: Multi-Health Systems.
Jackson, D. L. (2009). Revisiting sample size and number of parameter estimates: Some support
for the N:q hypothesis. Structural Equation Modeling: A Multidisciplinary Journal,
10(1), 128–141. doi: 10.1207/S15328007SEM1001_6
Lambert, P. C., Sutton, A. J., Burton, P. R., Abrams, K. R., & Jones, D. R. (2005). How vague is
vague? A simulation study of the impact of the use of vague prior distributions in MCMC
using WinBUGS. Statistics in Medicine, 24, 2401–2428. doi: 10.1002/sim.2112
Larsson, H., Andershed, H., Lichtenstein, P. (2006). A genetic factor explains most of the
variation in the psychopathic personality. Journal of Abnormal Psychology, 115, 221–
230.
Liao, J., Li, X., Wong, T., Wang, J. J., Khor, C. C., Tai, E. S., …, Cheng, C. (2004). Impact of
measurement error on testing genetic association with quantitative traits. PLOS ONE,
9(1). doi: 10.1371/journal.pone.0087044
Lord, F. M., Novick, M. R., & Birnbaum, A. (1968). Statistical Theories of Mental Scores.
Oxford, England: Addison-Wesley.
Lunn, D. J., Thomas, A., Best, N., & Spiegelhalter, D. (2000). WinBUGS – A Bayesian
modelling framework: Concepts, structure, and extensibility. Statistics and Computing,
10(4), 325–337. doi: 10.1023/A:1008929526011
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 33
Lynam, D. R. (1996). Early identification of chronic offenders: Who is the fledgling psychopath?
Journal of Abnormal Psychology, 106(3), 425–438.
Lynam, D. R. (1997). Pursuing the psychopath: Capturing the fledgling psychopath in a
nomological net. Journal of Abnormal Psychology, 107(4), 566–575.
Lynam, D. R., Caspi, A., Moffitt, T. E., Loeber, R. & Stouthamer-Loeber, M. (2007).
Longitudinal evidence that psychopathy scores in early adolescence predict adult
psychopathy. Journal of Abnormal Psychology, 116(1), 155–165. doi: 10.1037/0021-
843X.116.1.155
Lynam, D. R., Charnigo, R., Moffitt, T. E., Raine, A., Loeber, R., & Stouthamer-Loeber, M.
(2009). The stability of psychopathy across adolescence. Development and
Psychopathology, 21(4), 1133–1153. doi: 10.1017/S0954579409990083
Lynch, M. & Walsh, B. (1998). Genetics and analysis of quantitative traits. Sunderland, MA:
Sinauer.
MacGregor, A. J., Snieder, H., Schork, N. J., & Spector, T. D. (2000). Twins. Novel uses to
study complex traits and genetic diseases. Trends in Genetics, 16(3), 131–4. doi:
10.1016/S0168-9525(99)01946-0
Martin, N. G. & Eaves, L. J. (1977). The genetical analysis of covariance structure. Heredity,
38(1), 79–95.
Martin, N. G., Eaves, L. J., Kearsey, M. J., & Davies, P. (1978). The power of the classical twin
study. Heredity, 40(1), 97–116.
Neale, M. C., Hunter, M. D., Pritkin, J. N., Zahery, M., Brick, T. R., Kirkpatrick, R. M., …
Boker, S. M. (2016). OpenMx 2.0: Extended structural equation and statistical modeling.
Psychometrika, 81(2), 535–549. doi:10.1007/s11336-014-9435-8
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 34
Neale, M. C. & Maes, H. H. M. (2004). Methodology for Genetic Studies of Twins and Families.
Nivard, M., Dolan, C. V., Middeldorp, C. M., & Boomsma, D. I. (2017). Behavior genetics:
From heritability to gene finding. In J. T. Cacioppo, L. G. Tassinary, & G. G. Berntson,
(Eds.), Handbook of psychophysiology. Cambridge: Cambridge University Press.
doi:10.1017/9781107415782
Novick, M. R. (1966). The axioms and principal results of classical test theory. Journal of
Mathematical Psychology, 3(1), 1–18. doi: 10.1016/0022-2496(66)90002-2
Paulhus, D. L. (1991). Measurement and control of response bias. In J. P. Robinson, P. R. Shaver
& L. S. Wrightsman (Eds.), Measures of Personality and Social Psychological
Attitudes (pp. 17–59). San Diego, CA: Academic Press, Inc.
Polderman, T. J., Benyamin, B., de Leeuw, C. A., Sullivan, P. F., van Bochoven, A., Visscher, P.
M., Posthuma, D. (2015). Meta-analysis of the heritability of human traits based on fifty
years of twin studies. Nature Genetics, 47(7), 702–709. doi: 10.1038/ng.3285
Reise, S. P., Ainsworth, A. T., & Haviland, M. G. (2005). Item response theory: Fundamentals,
applications, and promise in psychological research. Current Directions in Psychological
Science, 14(2), 95–101. Retrieved from www.jstor.org/stable/20182996
Revelle, W. (2017). psych: Procedures for personality and psychological research [Computer
software]. Evanston, IL: Northwestern University. Retrieved from https://CRAN.R-
project.org/package=psych
Reynolds, C. R. & Suzuki, L. A. (2012). Bias in psychological assessment: An empirical review
and recommendations. In I. B. Weiner, J. R. Graham, & J. A. Naglieri (Eds.), Handbook
of psychology (Vol. 10) Assessment Psychology (2nd Ed.) (pp. 82–113). Hoboken, NJ:
John Wiley & Sons, Inc.
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 35
Rosseel, Y. (2012). lavaan: An R package for structural equation modeling. Journal of Statistical
Software, 48(2), 1–36. Retrieved from http://www.jstatsoft.org/v48/i02/
Smith, T. C., Spiegelhalter, D. J., & Thomas, A. (1995). Bayesian approaches to random-effects
meta-analysis: A comparative study. Statistics in Medicine, 14(24), 2685–2699. doi:
10.1002/sim.4780142408
Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103, 677–680.
Sullivan, P. F. & Eaves, L. J. (2002). Evaluation of analyses of univariate discrete twin data.
Behavior Genetics, 32(3), 221–227. doi: 10.1023/A:1016025229858
Thissen, D., Cai, L., & Bock, R. D. (2010). The nominal categories item response model. In M.
L. Nering & R. Ostini (Eds.), Handbook of polytomous item response theory models (pp.
43–75). Abingdon, VA: Routledge.
Thomas, A. (2014). OpenBUGS [Computer software]. OpenBUGS Foundation. Retrieved from
openbugs.net/w/OpenBUGS_3_2_3?action=AttachFile&do=get&target=OpenBUGS-
3.2.3.tar.gz
Tuvblad, C., Bezdjian, S., Raine, A., & Baker, L. A. (2014). The heritability of psychopathic
personality in 14- to 15-year-old twins: A multirater, multimeasure approach.
Psychological Assessment, 26(3), 704–716. doi: 10.1037/a0036711
U.S. Census Bureau (2016). American Community Survey 1-year estimates. Retrieved from
https://censusreporter.org/profiles/16000US0644000-los-angeles-ca/
U.S. Census Bureau, Population Division. (2017). Annual estimates of the resident population:
April 1, 2010 to July 1, 2016. Retrieved from
https://factfinder.census.gov/faces/tableservices/jsf/pages/productview.xhtml?src=bkmk
BIOMETRIC MODELS OF PSYCHOPATHIC TRAITS 36
van den Berg, S. M., Beem, L., & Boomsma, D. I. (2006). Fitting genetic models using Markov
chain Monte Carlo algorithms with BUGS. Twin Research and Human Genetics, 9(3),
334–342.
van den Berg, S. M., Glas, C. A. W., & Boomsma, D. I. (2007). Variance decomposition using
an IRT measurement model. Behavior Genetics, 37, 604–616. doi: 10.1007/s10519-007-
9156-1
Wang, P., Baker, L. A., Gao, Y., Raine, A., & Lozano, D. I. (2012). Psychopathic traits and
physiological responses to aversive stimuli in children aged 9-11 years. Journal of
Abnormal Child Psychology, 40, 759–769. doi: 10.1007/s10802-011-9606-3
ESTIMATING THE HERITABILITY OF PSYCHOPATHIC TRAITS 37
Tables
Table 1
Standardized Factor Loadings & Descriptive Statistics for Manipulative/Deceitful Psychopathy Factor Subscales & Items
Item Factor
Loadings
Male
M(SD)
Female
M(SD)
Skew Kurtosis
Failure to Accept Responsibility (Cronbach’s α = .19) .45(.29) .41(.30) 0.21 −0.47
Do you try to blame other people for things you have done? .29 .24(.43) .23(.42) 1.25 −0.45
Do you think you get blamed for things you did not do? .17 .65(.48) .60(.49) −0.51 −1.75
When you get in trouble, can you talk your way out of it? .33 .46(.50) .39(.49) 0.31 −1.91
Lack of Guilt (Cronbach’s α = .66) .18(.27) .17(.24) 1.43 1.51
Do you usually feel bad or guilty after doing something wrong?
†
.79 .13(.34) .08(.27) 2.56 4.56
Does it bother you when you do something wrong?
†
.63 .16(.37) .11(.32) 2.10 2.43
Do you wish you could take back many things that you have done?
†
.03 .24(.42) .32(.47) 1.00 −1.00
Manipulative (Cronbach’s α = .61) .24(.30) .25(.30) 0.97 −0.08
Do you try to act charming or likable in order to get your way? .58 .27(.45) .35(.48) 0.82 −1.33
Do you try to take advantage of other people? .43 .10(.30) .06(.24) 3.03 7.22
Do you try to get others to do what you want by getting on their good side? .57 .34(.47) .31(.46) 0.75 −1.44
Parasitic Lifestyle (Cronbach’s α = .35) .16(.20) .13(.18) 1.27 1.19
Do you try to see how much you can get away with? .65 .32(.47) .27(.44) 0.90 −1.19
Do you give or share things?
†
.10 .08(.27) .05(.22) 3.59 10.92
Do you usually return what you borrow?
†
.29 .05(.22) .06(.24) 3.86 12.93
Do you take a lot and not give much in return? .23 .18(.39) .13(.34) 1.89 1.56
Glibness (Cronbach’s α = .43) .34(.26) .34(.23) 0.55 0.07
Do you try to be the center of attention? .58 .18(.38) .20(.40) 1.59 0.53
Are you talkative? .26 .59(.49)* .74(.44)* −0.71 −1.51
Are you shy?
†
.19 .55(.50) .52(.50) −0.15 −1.98
Do you tell stories to make yourself look good? .41 .18(.39) .12(.33) 1.95 1.81
Do you show off to get people to pay attention to you? .54 .16(.37) .12(.33) 2.06 2.23
Untruthfulness (Cronbach’s α = .14) .34(.27) .29(.25) 0.44 −0.60
Are you open and honest?
†
.33 .20(.40) .14(.35) 1.76 1.11
Will you usually tell a lie if you think you can get away with it? .61 .53(.50)* .43(.50)* 0.09 −2.00
Can you be trusted?
†
.09 .03(.16) .02(.13) 6.62 41.86
ESTIMATING THE HERITABILITY OF PSYCHOPATHIC TRAITS 38
Do people usually believe you when you tell a lie? .55 .53(.50) .50(.50) −0.06 −2.00
Are you a good liar? .68 .37(.48) .33(.47) 0.62 −1.61
All Items (Cronbach’s α = .74) .28(.16) .26(.15) 0.69 −0.09
†
reverse-scored item (yes = 0, no = 1).
* significant difference between male and female mean scores at αcorrected = .002.
Note. Standardized factor loadings are for a six-factor CFA of the CPS subscales in Bezdjian, Raine, et al.’s (2011)
Manipulative/Deceitful Psychopathy Factor. CFI = .77; RMSEA = .06. χ2(215) = 521.986; p < .001.
ESTIMATING THE HERITABILITY OF PSYCHOPATHIC TRAITS 39
Table 2
Factor Correlations for 6-Factor Confirmatory Model of Manipulative/Deceitful Psychopathy Factor
Factor 1 2 3 4 5 6
1. Failure to Accept Responsibility –
2. Lack of Guilt .74 –
3. Manipulative .53 .21 –
4. Parasitic Lifestyle .64 .41 .90 –
5. Glibness .78 .20 .74 .65 –
6. Untruthfulness .64 .42 .55 .76 .47 –
ESTIMATING THE HERITABILITY OF PSYCHOPATHIC TRAITS 40
Table 3
Tetrachoric Twin Correlations for Manipulative/Deceitful Psychopathy Factor Items by Subscale
Item rtMZ rtDZ
Failure to Accept Responsibility
Do you try to blame other people for things you have done? .26 .20
Do you think you get blamed for things you did not do? .23 .23
When you get in trouble, can you talk your way out of it? .37 .41
Lack of Guilt
Do you usually feel bad or guilty after doing something wrong?
†
.37 .16
Does it bother you when you do something wrong?
†
.22 .18
Do you wish you could take back many things that you have done?
†
.25 .15
Manipulative
Do you try to act charming or likable in order to get your way? .23 .21
Do you try to take advantage of other people? .17 .07
Do you try to get others to do what you want by getting on their good side? .31 .29
Parasitic Lifestyle
Do you try to see how much you can get away with? .22 .04
Do you give or share things?
†
.18 .09
Do you usually return what you borrow?
†
.12 −.09
Do you take a lot and not give much in return? .07 −.12
Glibness
Do you try to be the center of attention? .24 −.01
Are you talkative? .45 .04
Are you shy?
†
.51 .17
Do you tell stories to make yourself look good? .55 .11
Do you show off to get people to pay attention to you? .31 .19
Untruthfulness
Are you open and honest?
†
.36 .22
Will you usually tell a lie if you think you can get away with it? .34 .17
Can you be trusted?
†
.27 .12
Do people usually believe you when you tell a lie? .20 .13
Are you a good liar? .42 .28
†
reverse-scored item (yes = 0, no = 1).
ESTIMATING THE HERITABILITY OF PSYCHOPATHIC TRAITS 41
Table 4
Heritability Estimates for Manipulative/Deceitful Psychopathy Factor Subscales
Child Psychopathy Scale Subscale Heritability Coefficient Mean and Standard Error Point Estimates
Univariate Bayesian Univariate Common Pathway Bayesian Item-Response
Failure to Accept Responsibility .30 (.01) .31 (.02) .61 (.01) .45 (.01)
Lack of Guilt .33 (.01) .39 (.02) .42 (.003) .63 (.01)
Manipulative .32 (.01) .16 (.01) .52 (.01) .60 (.01)
Parasitic Lifestyle .16 (.004) .28 (.01) .12 (.003) .72 (.01)
Glibness .29 (.01) .26 (.01) .54 (.01) .59 (.01)
Untruthfulness .35 (.01) .26 (.01) .24 (.01) .62 (.01)
Manipulative/Deceitful Total .46 (.01) .38 (.02) .68 (.01)
†
.54 (.01)
†
†
Manipulative/Deceitful Total common pathway and item-response models were fit using the six CPS subscales’ sum scores recoded
via median split to 0 = low, 1 = high. Thus, these values cannot be directly compared to the subscales’ common pathway and item-
response models.
ESTIMATING THE HERITABILITY OF PSYCHOPATHIC TRAITS 42
Table 5
Fit Indices for Quantitative Genetic Models of CPS Subscales and Manipulative/Deceitful Psychopathy Factor
Child Psychopathy Scale Subscale Univariate AE Bayesian Univariate Common Pathway Bayesian Item-Response
AIC −2LL D (SE) AIC −2LL D (SE)
Failure to Accept Responsibility −1414.59 307.41 .62 (.03) −1826.47 3323.53 .12 (.01)
Lack of Guilt −1658.64 63.36 .56 (.03) −3300.71 1849.29 −.05 (.01)*
Manipulative −1437.41 284.59 .78 (.02) −2944.40 2205.60 .26 (.01)
Parasitic Lifestyle −2190.54 −470.54 .69 (.03) −5416.58 1451.41 −.21 (.03)*
Glibness −1748.74 −26.74 .70 (.03) −4387.89 4198.11 −.41 (.02)*
Untruthfulness −1681.24 40.76 .69 (.03) −5498.48 3087.52 .26 (.01)
Manipulative/Deceitful Total −2543.35 −821.35 .60 (.03) −3864.94 3919.06 .06 (.01)
Note. No significant improvement (α = .05) was found in the fitted ACE models over the more parsimonious AE models in any
condition or subscale. All reported AIC, −2LL, and D values are for the fitted AE models, where D = 2rDZ − rMZ.
ESTIMATING THE HERITABILITY OF PSYCHOPATHIC TRAITS 43
Figures
Figure 1
Univariate ACE Model where T1 = phenotype sum score for twin 1 and T2 = phenotype sum score for twin 2
A
1
C
1
T
1
E
1
e c a
E
2
C
2
T
2
A
2
a c e
r
MZ
=1; r
DZ
=.5
r
= 1
ESTIMATING THE HERITABILITY OF PSYCHOPATHIC TRAITS 44
Figure 2
Common Pathway ACE Model where T1 = latent trait score for twin 1, T2 = latent trait score for twin 2, P1n = item n for twin 1, and
P2n = item n for twin 2
ESTIMATING THE HERITABILITY OF PSYCHOPATHIC TRAITS 45
Figure 3
Heritability Estimates by Fitted Model for the Manipulative/Deceitful Psychopathy Factor
Note. Subscales displayed in order from least number of items to greatest number of items. Manipulative/deceitful factor common
pathway and Bayesian item-response models were fit using the six CPS subscales’ sum scores recoded to 0 = absence, 1 = presence.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Failure to Accept
Responsibility
Lack of Guilt Manipulative Parasitic Lifestyle Glibness Untruthfulness Mani/Deceit Total
Univariate Bayes Univariate Common Pathway Bayes Item-Response
ESTIMATING THE HERITABILITY OF PSYCHOPATHIC TRAITS 46
Figure 4
Differences between Fit Models’ Heritability Estimates as compared to Subscales’ Skew
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
Failure to Accept
Responsibility
(3 Items)
Lack of Guilt
(3 Items)
Manipulative
(3 Items)
Parasitic Lifestyle
(4 Items)
Glibness
(5 Items)
Untruthfulness
(5 Items)
Manipulative/Deceitful
Factor
(6 Subscales)
Bayes IRT − ML Univariate
Bayes IRT − Common Pathway
-1
-0.5
0
0.5
1
1.5
2
Failure to Accept
Responsibility
(3 Items)
Lack of Guilt
(3 Items)
Manipulative
(3 Items)
Parasitic Lifestyle
(4 Items)
Glibness
(5 Items)
Untruthfulness
(5 Items)
Manipulative/Deceitful
Factor
(6 Subscales)
Skew
Kurtosis
Abstract (if available)
Abstract
Traditional behavioral genetics models use maximum likelihood estimation (MLE) and sum scores to decompose the variance of a construct into its genetic and environmental components. This can lead to inflated error variance and biased heritability estimates, since sum scores eliminate the item-level intricacies in the data and MLE performs unreliably in cases of non-normally distributed or discrete data. Twin study data (N = 864) for the manipulative/deceitful psychopathy factor in the USC Risk Factors for Antisocial Behavior (RFAB) exhibit these common statistical artifacts, and therefore may not be appropriate for use with MLE sum-score models. To address these concerns, Bayesian Markov chain Monte Carlo (MCMC) estimated genetic item-response models were fit to the manipulative/deceitful psychopathy factor’s six subscales, and resulting heritability estimates and fit indices were compared across MLE univariate, Bayesian MCMC univariate, and MLE common pathway models. For all subscales, the Bayesian MCMC item-response and MLE common pathway models performed best, yielding higher heritability estimates than either univariate model. This held especially true for the item-response models fit to those subscales with severe skew and kurtosis, and those whose items had the least consistent factor loadings. If the items’ factor loadings were relatively uniform and there were no egregious departures from normality in the distribution of the observed data, the MLE common pathway model was a potential substitute for the more complex item-response model, producing similarly less-biased heritability estimates. Future simulation studies comparing MLE and Bayesian MCMC estimation for behavioral genetic models could greatly improve the field’s understanding of these methods’ performance when applied to item-level data.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Multivariate biometric modeling among multiple traits across different raters in child externalizing behavior studies
PDF
Heritability of schizotypal traits during adolescence
PDF
Psychometric and biometric evaluation of psychopathic personality traits in 9 and 10 year old twins
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
A comparison of standard and alternative measurement models for dealing with skewed data with applications to longitudinal data on the child psychopathy scale
Asset Metadata
Creator
Potts, Shannon Kimberly
(author)
Core Title
Biometric models of psychopathic traits in adolescence: a comparison of item-level and sum-score approaches
School
College of Letters, Arts and Sciences
Degree
Master of Arts
Degree Program
Psychology
Publication Date
08/06/2018
Defense Date
08/03/2018
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
adolescence,behavior genetics,biometrics,item-response theory,OAI-PMH Harvest,psychopathy,twin study
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Baker, Laura (
committee chair
), John, Richard (
committee member
), Wilcox, Rand (
committee member
)
Creator Email
shannokp@usc.edu,skp48@nau.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-56994
Unique identifier
UC11671824
Identifier
etd-PottsShann-6669.pdf (filename),usctheses-c89-56994 (legacy record id)
Legacy Identifier
etd-PottsShann-6669.pdf
Dmrecord
56994
Document Type
Thesis
Format
application/pdf (imt)
Rights
Potts, Shannon Kimberly
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
behavior genetics
biometrics
item-response theory
psychopathy
twin study