University of Southern California Dissertations and Theses

Developmental etiology of reactive and proactive aggression, electrodermal reactivity and their relationships: a twin study

(USC Thesis Other)

Developmental etiology of reactive and proactive aggression, electrodermal reactivity and their relationships: a twin study

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Developmental Etiology of Reactive and Proactive Aggression, Electrodermal
Reactivity and their Relationships: A Twin Study

Devika Dhamija

A dissertation presented to

The Department of Psychology
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
University of Southern California
Los Angeles, California
December 2016

Table of Contents
Chapter 1: General Introduction ...........................................................................................1
Chapter 2: Developmental Etiology of Reactive and Proactive Aggression ......................9
Introduction ............................................................................................................................10
Methods ..................................................................................................................................13
Results ....................................................................................................................................18
Discussion...............................................................................................................................23
References ..............................................................................................................................28
Tables and Figures ..................................................................................................................34
Chapter 3: Genetic and Environmental Influences in the Development of
Electrodermal Reactivity ....................................................................................................54
Introduction ............................................................................................................................55
Methods ..................................................................................................................................58
Results ....................................................................................................................................63
Discussion...............................................................................................................................67
References ..............................................................................................................................72
Tables and Figures ..................................................................................................................76
Chapter 4: Longitudinal Relationship between Reactive, Proactive Aggression and
Electrodermal Reactivity: A Twin Study ............................................................................84
Introduction ............................................................................................................................85
Methods ..................................................................................................................................87
Results ....................................................................................................................................91
Discussion...............................................................................................................................94
References ..............................................................................................................................97
Tables and Figures ................................................................................................................100

Chapter 5: Discussion and Conclusions ............................................................................108

Acknowledgements
I am extremely thankful to my advisor, Dr. Laura Baker, for her unconditional support
and guidance through the course of my graduate studies at USC. Her mentorship and scientific
acumen have been inspiring, to say the least. I would also like to thank Dr. Catherine Tuvblad
for her unparalleled patience and feedback. She has been a mentor, friend and colleague and has
encouraged me to be always persistent and strive for the best. She has led me to pursue exciting
research and provided me with excellent scientific training.
I would like to give my heartfelt thanks to Dr. Michael Dawson, who has inspired me
with his passionate attitude towards research and been a mentor since my first year at USC. His
teaching philosophy and detail oriented approach to research have been a guiding force
throughout.
I am thankful to my dissertation committee, Dr. Morteza Dehghani and Dr. Yaling Yang
for their immense support and patience. I am grateful to Dr. John McArdle for his invaluable
insights, training and mentorship.
A big thanks to members of the USC Twin Project- Gemma Quick, Karina Gomez, Theo
Botwick and all research and graduate assistants- who have helped in data collection and
management throughout these years and this dissertation would not have been possible without
them. A special thanks to Dr. Pan Wang, who has been extremely responsive and provided her
technical expertise. I would like to thank NIMH for funding RFAB (risk factors for antisocial
behaviors), a longitudinal twin study from which the data for current study were drawn (R01
MH58354).

I am forever indebted to my friends and fellow graduate students at the USC Twin Lab-
Leslie Bernsten and Nicholas Jackson- for their unconditional support, motivating words, for
their incomparable camaraderie, deep insights into subject matters and long conversations about
science and life. This experience would have been incomplete without them.
I am grateful for my family- my father Mr. Sunil Dhamija, my mother Mrs. Savita
Dhamija and my brother Hemang Dhamija- who have been instrumental in my ability to pursue
my graduate studies. Their deep faith and words of encouragement have always propelled me
forward and taught me to be fearless in pursuing my dreams.
Lastly, I would like to thank my friends Nazneen Aibara, Meera George Palackan, Ankur
Saxena, Bhushan Dharmadhikari and Akshar Tandon, who have stood by my side through this
process. I am lucky to have found friends who have believed in me and pushed me to achieve
greater heights.

Abstract
Heterogeneity in the construct of human aggression has motivated researchers to identify
sub forms of aggressive behaviors. Two separable functional forms of aggressive behaviors have
been identified and which appear to be distinguishable from childhood to adulthood: reactive and
proactive aggression. A significant body of literature distinguishes these two forms based on
comorbidity with other psychopathologies, personality correlates and later predicted outcomes.
There is, however, limited research on how reactive and proactive aggression develop over time,
the genetic underpinnings of these changes and the genetic overlap with autonomic functioning.
The goal of this study is to examine longitudinal development of reactive and proactive
aggression, the genetic and environmental underpinnings of their development and their
relationship to electrodermal reactivity. The data for this study were drawn from a longitudinal
study of risk factors of antisocial behavior in a community sample of over 750 pairs of twins.
There have been five waves of measurement so far, spanning a period of ten years from late
childhood (age 9-10 years) to early adulthood (age 19-20 years). Latent growth curve analysis
revealed significant changes over time in reactive and proactive aggression. Significant sex
differences emerged in the development of proactive aggression, specifically while examining
the endorsement of these behaviors. Genetic results indicated that both genetic stability and
innovation were important across development for both reactive and proactive aggression. The
behaviors were modestly to moderately heritable from childhood, with new genetic variance
emerging at ages 11-13 years, 14-15 years and 16-18 years. Heritability estimates for proactive
aggression tend to be lower in females than in males. For electrodermal reactivity, genetic factors
explained 46-90% of the variance in scores, suggesting moderate to high heritability. Males and
females showed mean differences but no difference in the developmental patterns, at the

phenotypic or genetic level. Significant genetic correlation between proactive aggression and
orienting response was found for males (rg = -0.43), suggesting shared genetic etiology.
1

Chapter 1: General Introduction
Aggressive behavior is defined as the intent to hurt, harm or injure another person (Coie
& Dodge, 1998). This is a broad definition which encompasses a wide range of behaviors that
vary in intensity and intention. Researchers have thus worked to develop useful ways of
distinguishing between specific forms of aggression. One distinction, based on the functional
utility of these behaviors, is reactive and proactive aggression (Dodge & Coie, 1987). Whereas
reactive aggression is defined as a defensive response to a threat or provocation, proactive
aggression is defined as a goal-oriented, instrumental form of aggression with positive outcome
as a perceived reward. Historically, reactive and proactive aggression definitions emerged from
two competing theories of aggressive behaviors. Reactive aggression has its theoretical roots in
the frustration aggression model (Berkowitz, 1978), which considered aggression as a hostile
reaction to a perceived threat or provocation. Proactive aggression, on the other hand, is a
descendant of the social learning theory (Bandura, 1973), according to which aggression is a
learned behavior, acquired through social conditioning and environmental influences.
A longstanding debate has existed over the utility of this distinction. Whereas some argue
against this distinction (Bushman & Anderson, 2001), others have supported the idea (J. A.
Hubbard, McAuliffe, Morrow, & Romano, 2010; Kempes, Matthys, De Vries, & Van Engeland,
2005; Merk, Orobio de Castro, Koops, & Matthys, 2005; Polman, Orobio de Castro, Koops, van
Boxtel, & Merk, 2007; Poulin & Boivin, 2000; Frank Vitaro, Brendgen, & Barker, 2006).
Empirical support has also been provided through studies which support a two-factor structure,
with distinct, but correlated, latent phenotypes of reactive and proactive aggression (Brown,
2

Atkins, Osborne, & Milnamow, 1996; Poulin & Boivin, 2000; Raine et al., 2006; Tuvblad,
Dhamija, Berntsen, Raine, & Liu, 2016).
In addition to support from factor analytic studies, the distinction between these two
forms of aggression have been strengthened by empirical research demonstrating the differential
correlation, both concurrent and longitudinal, with a host of personality and behavior problems.
Reactive aggression tends to be more robustly correlated with psychosocial adjustment including
internalizing problems and emotional dysregulation (Card & Little, 2006) as well as impulsivity,
social anxiety hostility poor interpersonal skills (Fossati et al., 2009; Raine, et al., 2006; Seah &
Ang, 2008). Proactive aggression, on the other hand, is characterized by initiation of fights,
delinquency, hyperactivity and bullying (Fossati et al., 2009; Raine et al., 2006; Seah & Ang,
2008). Longitudinal associations of reactive and proactive aggression also differ. Whereas
proactive aggression is related to a psychopathic personality traits by adolescence and predicts
conduct problems (Raine, et al., 2006), reactive aggression predicts negative emotionality (Fite,
Raine, Stouthamer-Loeber, Loeber, & Pardini, 2009). Distinguishing between these two sub
forms is thus useful for two reasons: (1) it can explain the observed heterogeneity in aggressive
behaviors and (2) it can, more importantly, shed light on possible differences in later outcomes
associated with these behaviors, having important implications for intervention.
Development of aggressive and antisocial behaviors, at large, has been widely studied
across the lifespan. Reviews on the subject delineate the patterns of change over time and stress
on the importance of developmental factors including maturation, social changes- in the study of
aggression and violence (Liu, Lewis, & Evans, 2013; Loeber & Hay, 1997). Several notable
patterns of change have been observed in aggressive behaviors from childhood to adolescence to
adulthood. For example, qualitatively, violent aggression tends to increase from childhood to
3

adolescence whereas the prevalence of physical fights (backyard fights, fights among children)
decreases. A special emphasis in developmental studies of aggression has been given to
adolescence as it is a crucial period with respect to brain maturation and changes in the social
setting of individuals. Numerous studies which have looked at changes in antisocial behavior,
aggression, and violence and have found that these behaviors tend to peak during the late teens.
This peak in antisocial behaviors is so prominent that those exhibiting “adolescent limited”
antisocial behavior are differentiated from persistently antisocial individuals (Moffitt, 1993).
Heterogeneity in the development of aggressive behaviors has also been studied. For example, it
has been shown that aggressive and non-aggressive antisocial behaviors follow distinct
developmental patterns, peak at different times during childhood and adolescence, and are
related to different outcomes (Burt, 2012). Although there are changes in reactive and proactive
aggression over time (Barker, Tremblay, Nagin, Vitaro, & Lacourse, 2006; Fite, Colder,
Lochman, & Wells, 2008) there remain unanswered questions regarding the physiological and
social underpinnings of these behaviors as well as sex differences.
Twin studies have a long history of interest and association with the study of violence,
aggression and antisocial behaviors. Meta analytic studies of genetic and environmental
influences on aggression (Rhee & Waldman, 2002) emphasize that heterogeneity in the concept
and operationalization of these behaviors can impact the empirical results obtained. Molecular
genetic studies have also shaped our understanding of the underlying biological bases of
aggressive behaviors. For example, mutation for the gene encoding monoamine oxidase (MAO),
which is responsible for the breakdown of certain neurotransmitters (e.g., serotonin) have been
implicated in the biology of aggression (Brunner, Nelen, Breakefield, Ropers, & van Oost,
1993). Important GXE interactions have also been implicated in the development of violent and
4

aggressive behavior (Caspi et al., 2002). Taken together, a wealth of research has tried to
disentangle the genetic basis of aggression and violence, given its high cost to society.
Psychophysiological research studies have provided useful insights into the biological basis of
these behaviors. Several studies using autonomic measures, brain imaging and EEG have
delineated possible mechanisms pertaining to aggressive behaviors. Fewer studies, however,
have specifically focused on reactive and proactive aggression and their biological bases. To
understand the longitudinal associations of behavior with psychophysiological correlates, at both
phenotypic and genetic levels, it is important to first establish their development and genetic
basis. Such longitudinal examinations are very rare in the psychophysiological literature in
general.
The psychophysiological measure of interest in this study is the orienting response, a
measure of attention, arousal and regulation (for reviews, see Boucsein, 2012; Hugdahl, 1995).
The orienting response has been studied in association with antisocial behavior, psychopathy
criminality and externalizing behaviors, at large (Isen et al., 2010; Isen, Iacono, Malone, &
McGue, 2012; Raine, Venables, & Williams, 1995) and has shown some promise as a potential
biomarker for externalizing behaviors ( Isen, et al., 2012), exhibiting significant genetic overlap.
The current study aims to examine the developmental changes in reactive and proactive
aggression, their genetic and environmental etiologies as well as their psychophysiological
associations over time. A specific measure of electrodermal reactivity and attention-the skin
conductance orienting response-is used to examine these questions. Specifically, the following
questions will be addressed:
5

a) What are the developmental trajectories of reactive and proactive aggression from
late childhood (age 9-10) through emerging adulthood (age 19-20)?
b) What are the relative genetic and environmental influences on reactive and
proactive aggression across developmenty? How do the influences of genetic and
environmental factors change from late childhood to emerging adulthood?
c) What is the developmental trajectory of electrodermal reactivity, as measured by
orienting response, and what are the etiological underpinnings of these
developmental changes?
d) How are reactive and proactive aggression related to electrodermal reactivity?
e) To what extent do genetic and/or environmental factors account for overlap
between electrodermal reactivity and sub forms of aggression?
Questions a-b are examined in the first study of this dissertation (Chapter 2). Question c
is examined in Chapter 3, specifically exploring the nature of development of the orienting
response and its genetic and environmental basis. Chapter 4 examines questions d-e, regarding
the concurrent and longitudinal relationships between reactive, proactive aggression and the
orienting response, as well as their genetic overlap.

6

References
Barker, E. D., Tremblay, R. E., Nagin, D. S., Vitaro, F., & Lacourse, E. (2006). Development of
male proactive and reactive physical aggression during adolescence. Journal of child
psychology and psychiatry, and allied disciplines, 47(8), 783-790.
Boucsein, W. (2012). Electrodermal activity: Springer Science & Business Media.
Brown, K., Atkins, M. S., Osborne, M. L., & Milnamow, M. (1996). A revised teacher rating
scale for reactive and proactive aggression. Journal of abnormal child psychology, 24(4),
473-480.
Brunner, H. G., Nelen, M., Breakefield, X. O., Ropers, H. H., & van Oost, B. A. (1993).
Abnormal behavior associated with a point mutation in the structural gene for
monoamine oxidase A. Science, 262(5133), 578-580.
Card, N. A., & Little, T. D. (2006). Proactive and reactive aggression in childhood and
adolescence: A meta-analysis of differential relations with psychosocial adjustment.
International Journal of Behavioral Development, 30(5), 466-480.
Caspi, A., McClay, J., Moffitt, T. E., Mill, J., Martin, J., Craig, I. W., et al. (2002). Role of
genotype in the cycle of violence in maltreated children. Science, 297(5582), 851-854.
Fite, P. J., Colder, C. R., Lochman, J. E., & Wells, K. C. (2008). Developmental trajectories of
proactive and reactive aggression from fifth to ninth grade. Journal of clinical child and
adolescent psychology : the official journal for the Society of Clinical Child and
7

Adolescent Psychology, American Psychological Association, Division 53, 37(2), 412-
421.
Fite, P. J., Raine, A., Stouthamer-Loeber, M., Loeber, R., & Pardini, D. A. (2009). Reactive and
proactive aggression in adolescent males: Examining differential outcomes 10 years later
in early adulthood. Criminal Justice and Behavior.
Fossati, A., Raine, A., Borroni, S., Bizzozero, A., Volpi, E., Santalucia, I., et al. (2009). A cross-
cultural study of the psychometric properties of the Reactive-Proactive Aggression
Questionnaire among Italian nonclinical adolescents. Psychological assessment, 21(1),
131-135.
Hugdahl, K. (1995). Psychophysiology: The mind-body perspective: Harvard University Press.
Isen, J., Raine, A., Baker, L., Dawson, M., Bezdjian, S., & Lozano, D. I. (2010). Sex-specific
association between psychopathic traits and electrodermal reactivity in children. Journal
of Abnormal Psychology, 119(1), 216-225.
Isen, J. D., Iacono, W. G., Malone, S. M., & McGue, M. (2012). Examining electrodermal
hyporeactivity as a marker of externalizing psychopathology: a twin study.
Psychophysiology, 49(8), 1039-1048.
Liu, J., Lewis, G., & Evans, L. (2013). Understanding aggressive behaviour across the lifespan.
[Review]. Journal of psychiatric and mental health nursing, 20(2), 156-168.
Loeber, R., & Hay, D. (1997). Key issues in the development of aggression and violence from
childhood to early adulthood. Annual review of psychology, 48, 371-410.
8

Poulin, F., & Boivin, M. (2000). Reactive and proactive aggression: evidence of a two-factor
model. Psychological assessment, 12(2), 115.
Raine, A., Dodge, K., Loeber, R., Gatzke-Kopp, L., Lynam, D., Reynolds, C., et al. (2006). The
Reactive-Proactive Aggression Questionnaire: Differential Correlates of Reactive and
Proactive Aggression in Adolescent Boys. Aggressive behavior, 32(2), 159-171.
Raine, A., Venables, P. H., & Williams, M. (1995). High autonomic arousal and electrodermal
orienting at age 15 years as protective factors against criminal behavior at age 29 years.
The American journal of psychiatry, 152(11), 1595-1600.
Rhee, S. H., & Waldman, I. D. (2002). Genetic and environmental influences on antisocial
behavior: a meta-analysis of twin and adoption studies. Psychological bulletin, 128(3),
490-529.
Seah, S. L., & Ang, R. P. (2008). Differential correlates of reactive and proactive aggression in
Asian adolescents: relations to narcissism, anxiety, schizotypal traits, and peer relations.
Aggressive behavior, 34(5), 553-562.
Tuvblad, C., Dhamija, D., Berntsen, L., Raine, A., & Liu, J. (2016). Cross-Cultural Validation of
the Reactive-Proactive Aggression Questionnaire (RPQ) Using Four Large Samples from
the US, Hong Kong, and China. Journal of psychopathology and behavioral assessment,
38(1), 48-55.

9

Chapter 2: Developmental Etiology of Reactive and Proactive Aggression
Abstract
The current study examined the development of reactive and proactive aggression and its
genetic basis in a twin sample from ages 9-20 years old. Sex differences in developmental
changes were also examined. Mean levels of aggressive behaviors peaked during adolescence
(age 14-15 years). Latent growth curve analyses of reactive aggression scores suggested a
quadratic trajectory of change and significant individual differences in both initial levels as well
as change across time (slope factors). For proactive aggression, a two part model for the semi-
continuous data revealed that the increase observed in mean levels of proactive aggression stems
from an increased engagement in these behaviors across time. The probability of endorsement
followed a quadratic curve; consistent with the observed means, but changes in the continuous
part of the data showed a small but significant decline over time. Significant effect of sex was
found for both parts. Genetic analyses revealed that variance in both reactive and proactive
aggression could be explained by additive genetic and non-shared environmental influences.
Shared environmental influences were not significant in the multivariate analyses. Common
genetic factors and time-specific unique genetic factors were both important, suggesting that
genetic factors lend to both stability as well as change in aggression scores across development.
Genetic factors explained 28-50% of variance in reactive aggression scores and 14-55% in
proactive aggression scores. Genetic factors explained a greater variance in proactive scores for
males than for females.
10

Introduction
Reactive and proactive aggression are distinct forms of aggressive behavior, based on
their functional utility (Dodge & Coie, 1987). Whereas reactive aggression is a defensive
behavior in response to an immediate or perceived threat, proactive aggression is a goal-oriented,
instrumental form of aggression with positive outcome as a perceived reward. Although some
researchers have argued against this distinction (Bushman & Anderson, 2001), a significant body
of research has supported the idea of these two functionally distinct forms (Hubbard, McAuliffe,
Morrow, & Romano, 2010; Kempes, Matthys, De Vries, & Van Engeland, 2005; Merk, Orobio
de Castro, Koops, & Matthys, 2005; Polman, Orobio de Castro, Koops, van Boxtel, & Merk,
2007; Poulin & Boivin, 2000; Vitaro, Barker, Boivin, Brendgen, & Tremblay, 2006).
Not only are reactive and proactive aggression distinguished based on the motive of the
aggressor, but have been differentially correlated with a host of other personality and behavior
problems. Whereas proactive aggression is correlated with narcissism, delinquency,
hyperactivity, and bullying, reactive aggression is correlated with anxiety, poor interpersonal
skills and impulsivity (Fossati et al., 2009; Raine et al., 2006; Seah & Ang, 2008). Reactive and
proactive aggressive behaviors also differentially predict later negative outcomes. Whereas
proactive aggression predicts later conduct problems, delinquency and oppositional defiance
disorder (ODD), psychopathic traits (Raine, et al., 2006), antisocial problems (Vitaro, Gendreau,
Tremblay, & Oligny, 1998); reactive aggression predicts negative emotionality (Fite, Raine,
Stouthamer-Loeber, Loeber, & Pardini, 2009).
A wealth of research exists on aggression across the lifespan (Liu, Lewis, & Evans, 2013;
Loeber & Hay, 1997) and has revealed that the nature and form of aggressive behaviors change
considerably across time. For example, violent aggression tends to increase from childhood to
adolescence but the prevalence of physical fights decreases (Liu, et al., 2013). Only a few studies
11

have specifically examined proactive and reactive aggression longitudinally. Studies on the
stability of these constructs have found mixed results. Whereas one study found the constructs to
be unstable over time (Murray-Close & Ostrov, 2009), others have found that the constructs are
in fact stable (Fite, Colder, Lochman, & Wells, 2008; McAuliffe, Hubbard, Rubin, Morrow, &
Dearing, 2006). Longitudinal studies of reactive and proactive aggression during adolescence (5
th

to 9
th
grade) have shown a peak in 6
th
grade in the trajectory of mean levels, and a decline
thereafter (Fite, et al., 2008). However, the developmental trajectories did not appear to differ
for the two forms of aggression in adolescent males (age 13-17) (Barker, Tremblay, Nagin,
Vitaro, & Lacourse, 2006). The pattern and magnitude of change in reactive and proactive
aggression, however, has not been extensively explored. A developmental psychopathology
perspective (which in itself emphasizes maturation, adaptation, stability, continuity and change)
is imperative for clinical as well as research applications (Sroufe & Rutter, 1984). The proposed
study aims to examine developmental trajectories of both reactive and proactive aggression to
assess their differential development from childhood to early adulthood.
Longitudinal twin studies also offer the opportunity to examine development from an
etiological perspective. The classical twin design allows the variance of an observed trait to be
decomposed into three parts: additive genetic (A), shared environmental (C) and non-shared
environmental (E) factors. These analyses rely on two assumptions: (1) that monozygotic (MZ;
identical) twin pairs share 100% of their genes, while dizygotic (DZ; fraternal) twins share, on
average, 50% of their genes and (2) that MZ and DZ twins reared together are influenced by their
common family environment (C) to an equal degree. Shared environmental factors (C) refer to
non-genetic influences that contribute to similarity within pairs of twins. Non-shared
environmental factors (E) refer to experiences that make twins within a pair dissimilar.
12

Heritability (h
2
) is the proportion of total variance explained by genetic factors: h
2
= A/(A+C+E)
(Neale & Cardon, 1992).
The genetic underpinnings of aggressive behavior, broadly construed, have been well
studied. Meta-analyses and systematic literature reviews concur that that genes account for as
much as 50-70 percent of the variance in aggressive behaviors (Miles & Carey, 1997; Rhee &
Waldman, 2002; Tuvblad & Baker, 2011) and that there is little or no role of shared
environmental influences. Age and sex tend to be important moderators with higher heritability
estimates for males compared to females, and for adults compared to children and adolescents.
Some research has previously explored the topic of genetic and environmental
underpinnings of reactive and proactive aggression. Three studies (with two independent
samples) determined that both reactive and proactive aggression have moderate heritability
(Baker, Raine, Liu, & Jacobson, 2008; Brendgen, Vitaro, Boivin, Dionne, & Perusse, 2006;
Tuvblad, Raine, Zheng, & Baker, 2009). In one of these studies, unique environmental factors
explained the remainder variance (Brendgen, et al., 2006) but the other two studies found a role
of both unique and shared environmental factors (Baker, et al., 2008; Tuvblad, et al., 2009).
Moreover, two of these studies (Brendgen, et al., 2006; Tuvblad, et al., 2009) did not find any
sex differences in the genetic and environmental variation in reactive and proactive aggression,
which is contrary to the findings for overall aggression (Miles & Carey, 1997; Rhee & Waldman,
2002) whereby significant sex differences are found in the heritablity estimates for males and
females. Despite these efforts of parsing out the genetic basis of reactive and proactive
aggression, much remains to be explained. The role of genetic and environmental factors has not
been determined across childhood, adolescence through early adulthood. Sex differences in the
development and etiological bases of these behaviors also need to be studied, given prior
13

evidence suggesting that males and females differ in the expression of these behaviors (Connor,
Steingard, Anderson, & Melloni, 2003).
The present study aims to bridge these gaps in our understanding of the development of
proactive and reactive aggression from childhood to early adulthood by examining a) the
changes accompanying these behaviors from childhood to early adulthood at both mean levels
and individual variation in these changes; b) the role of genetic and environmental influences on
these behaviors at different periods of development as well as longitudinally and c) sex
differences at both phenotypic and genetic levels.
Methods
Participants
Participants were drawn from the Risk Factors for Antisocial Behavior (RFAB) twin study
at the University of Southern California (USC). RFAB is a longitudinal study of the interplay of
genetic, environmental, social and biological factors in the development of antisocial and aggressive
behaviors from childhood to emerging adulthood. A community sample of twin families was
recruited. The ethnic constitution of this sample is: 44% Hispanic, 25% Caucasian, 16% African
American, 3% Asian, and 12% mixed or other, which is representative of the diversity in the greater
Los Angeles area.
On the first assessment (Wave 1) the participants were 9–10 years old ( mean age = 9.59,
SD = 0.58); during Wave 2, the participants were 11–13 years old ( mean age = 11.79, SD =
0.92); during Wave 3, the participants were 14-15 years old ( mean age = 14.82, SD = 0.83) on
the third assessment (Wave 3) ; during Wave 4, the participants were 16-18 years old ( mean age
= 17.22, SD = 1.23), and during Wave 5 they were 19-23 (mean age=19.55, SD=1.11) years old.
A total sample of 1,564 subjects (781 twin pairs) - 169 monozygotic (MZ) male, 171 MZ female,
14

121 dizygotic (DZ) male, 120 DZ female, and 200 DZ opposite-sex twin pairs- constitutes the
sample. Due to attrition of participants, during Wave 3, a second round of enrollment was carried
out. Thus, there were families who started participation during Wave 1 and those who started
participation only at Wave 3.
Zygosity was determined through DNA microsatellite analysis (7 concordant and zero
discordant markers for MZ; one or more discordant markers for DZ) for 87% of the same-sex
twin pairs. For the remaining same-sex twin pairs, zygosity was established by questionnaire
items about the twins’ physical similarity and the frequency with which people confuse them. The
questionnaire was used only when DNA samples were insufficient for one or both twins in a pair.
When both questionnaire and DNA results were available, there was a 90% agreement between
the two (Baker et al., 2013).
Attrition
Longitudinal studies suffer a great deal from attrition. To examine whether certain
demographic variables as well as aggression levels were predictive of future dropout, attrition
analyses were conducted. Logistic regression analyses were conducted to predict likelihood of
never returning (complete dropout) to the study. For participants starting at Wave 1, gender
[OR=0.34], reactive aggression at Wave 1[OR=0.68], proactive aggression at Wave 1[OR= 2.63],
did not predict dropout from the study. However, Caucasians were less likely to dropout [OR=
0.50]. For those participants initiating at Wave 3, all variables, including baseline reactive and
proactive aggression, gender and race were not significant in predicting dropout.
Procedure:
On all five waves of data collection, families had the option to participate in 4-7 hours of
assessment during a laboratory visit. During the first wave of assessment, all participating
15

families visited the laboratory. During Waves 2 and 3, an additional option to participate through
mail surveys was included and during Waves 4 and 5, there was an additional option of
participating through online surveys. For the twins, an average laboratory assessment consisted
of two parts: cognitive and psychophysiological. During the cognitive portion, they completed
several neuropsychological tests and clinical interviews. The psychophysiological portion
consisted of multiple different tasks at each waves. Caregivers only participated through
interviews, completed either in person or via mail or online surveys.
Measures:
Reactive Proactive Questionnaire: Twins and their caregivers completed the Reactive-Proactive
Aggression Questionnaire (RPQ). The RPQ is a 23-item questionnaire with 12 items measuring
proactive aggression and 11 items measuring reactive aggression (Raine, 2006). Items are
presented in Appendix A. The response options are on a three-point scale (0 = never, 1 =
sometimes, 2 = often). The RPQ is designed for ages 8 and above and thus was appropriate for the
present twin sample (ages 9- 21). The measure has been validated for a two-factor
(reactive/proactive) fit in multiple samples (Baş & Yurdabakan, 2012; Fossati, et al., 2009; Fung,
Raine, & Gao, 2009; Tuvblad, Dhamija, Berntsen, Raine, & Liu, 2016) and this two factor
structure has also been confirmed for the present sample (Baker, et al., 2008; Tuvblad, et al.,
2009). For the present study, only data from the twin self-reports was used. The item mean for
each subscale was taken, with a possible range of 0-2.
Statistics:
Descriptive statistics were computed for each form of aggression for males and females
separately and mean differences were examined within each wave. Next, to examine
development at a phenotypic level and any sex differences in these changes, latent growth curve
16

models were fit. A pictorial representation of a growth curve with intercept, linear slope and
quadratic slope is presented in Figure 2.2. Goodness of fit of these models was assessed by a
likelihood-ratio χ2-test, which is the difference between -2 log likelihood (-2LL) of the full
model from that of the restricted model. This difference is distributed as a χ
2
. The degrees of
freedom (df) for this test are equal to the difference between the number of estimated parameters
in the full model and that in the restricted model. Model fit was also determined by comparing
Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The AIC
represents the balance between model fit and the number of parameters; with lower values of
AIC indicating the most suitable model (Akaike, 1987). Similarly, increasingly negative values
of the BIC correspond to increasingly parsimonious and better fitting models (Raftery, 1995).
The root mean square error of approximation (RMSEA) and comparative fit index (CFI) were
also used to judge model fit. RMSEA values of 0.05 indicate good fit and 0.08 an adequate fit
and CFI values range from 0 to 1.00 with higher values suggesting better fit.
For proactive aggression, a two part model was fit to assess changes in both proportion of
sample endorsing any proactive items as well as the changes in the continuous portion of the data
when the scores were more than zero. This approach has been utilized for data with a high
preponderance of zeros, commonly occurring in behaviors which are sparsely endorsed (Brown,
Catalano, Fleming, Haggerty, & Abbott, 2005; Olsen & Schafer, 2001). In this approach, the
data are divided into two parts- the first is a binary indicator of whether or not any endorsement
has been made and the second part is a continuous variable with non-zero values containing the
mean value of proactive aggression items from the questionnaire, when the value is not zero or
missing. The two parts are modelled as parallel but correlated processes. First, simpler models
17

for each part were fit and then combined together. These latent growth curve models were fit in
Mplus 6.12 (Muthén & Muthén, 1998-2011).
Genetic Models
Phenotypic twin correlations were computed across the five zygosity groups to gain a
first insight into the underlying sources of variation. Subsequently, genetic models were fit using
the structural equation modeling program Mplus (Muthén & Muthén, 1998-2007) .Due to the
high skewness in proactive scores, these scores were ranked and normalized using Blom’s
transformation in SPSS. Univariate genetic models were first fit to estimate the relative
contribution of A, C and E to the variance of reactive and proactive aggression at each wave. Sex
differences in these variance sources were also examined by first letting the variance components
differ across the sexes and then adding parameter constraints to equate them. Next, a
multivariate Cholesky decomposition was fit to examine the sources of variation across all five
waves. A Cholesky decomposition separates the variance of observed variables into an equal
number of A, C, E factors as the waves of observation. In a five wave model, this allows for five
A, five C and five E factors to be estimated. The first factor influences data from all five waves,
the second factor influences all but the first wave and so on and the last factor only influences the
fifth (final) wave. This allows for the examination of common and shared A, C , E variance
sources across different waves, giving insight into the stability and innovation in these factors
across development while simultaneously estimating the proportion of variance explained by
A,C and E at each wave. Next, an independent pathway model was fit. In this model, genetic
and environmental influences can either be shared or be specific to aggression measures at each
wave. The extent to which common or shared factors influence measures at each time point can
thus be determined, in addition to the time-specific factors which come into play across
18

development. In addition to the Cholesky decomposition, independent pathway models with one
and two common factors were fit to assess whether one or two common genetic and
environmental factors could explain the variance in aggressive measures across development. In
a one factor independent pathway model, one of each A, C and E factors are common across all
waves of measurement. Additional sources of variation include A, C and E components which
are unique to measures at each time point. Similarly, in a two factor independent pathway, two
common A, C and E factors load onto scores at each wave in addition to unique A, C and E
components. Goodness of fit of these genetic models was assessed by χ2-test, AIC, BIC and also
examining CFI and RMSEA.
Results
Descriptive Statistics
Descriptive statistics for reactive and proactive aggression are presented in Table 2.1.
There were significant sex differences in mean reactive aggression scores at Wave 1 (t
(1205)
=3.20,
p=.001), Wave 4 (t
(969)
=2.72, p=.007) and Wave 5 (t
(1011)
=2.21, p=.03), with higher mean levels
for males in caparison to females. No significant sex differences emerged during Wave 2
(t
(822)
=1.68, p=.09) and Wave 3 (t
(1172)
=.66, p=.51). For proactive aggression, significant sex
differences emerged at Wave 1 (t
(1205)
=4.16, p<.001), Wave 2 (t
(822)
=2.20, p=.03), Wave 4
(t
(969)
=3.33, p=.001) and Wave 5 (t
(1011)
=3.89, p<.001) but not during Wave 3 (t
(1172)
=1.59,
p=.11), with consistently higher mean values for males. Cross wave correlations for reactive and
proactive aggression are presented in Table 2.2a and 2.2b respectively.
Two 5 (Wave) × 2 (Sex) mixed design ANOVAs, for reactive and proactive aggression,
were conducted to examine changes in aggression scores over time and any interaction of these
changes with participant’s biological sex. For reactive aggression, the main effect of Wave was
19

significant [F(4, 1664) = 43.20, p<.001]. Pairwise comparisons suggested a significant increase
in aggression scores from Wave 2 to Wave 3 and a significant decrease thereafter with Wave 5
scores being lowest in comparison to all other waves. The interaction between Wave and Sex
was not found to be significant [F(4, 1664) = 0.68, p=.62], indicating that mean level changes
were consistent for both males and females. The same pattern was observed for proactive
aggression, with a significant main effect of Wave [F(4, 1664) = 25.81, p<.001] and no
significant interaction between Wave and Sex [F(4, 1664) = 1.40, p=.23]. Pairwise comparisons
suggested a significant increase in aggression scores from Wave 2 to Wave 3 and a significant
decrease after Wave 4. Mean levels of reactive and proactive aggression scores across the five
waves are plotted in Figure 2.1.
Latent Growth Curve Analyses:
For reactive aggression scores, first a model with only an intercept was fit. The addition of
a linear slope (∆χ
2
=228.84, ∆df=6, p<0.01) and subsequently a quadratic slope (∆χ
2
=232.79,
∆df=6, p<0.01) fit the data significantly better, as indexed by a decrease in Chi- square as well as
a decrease in AIC, BIC and RMSEA values and an increase in the CFI. Further models were
used to model invariance across males and females. First, invariance of fixed effects (intercept
and slope) was tested. Subsequently, invariance in the random effects- variance of intercept and
slopes was tested. The model fitting results are presented in Table 2.4a. The best fitting model
based on Chi square change and BIC was a model where the intercepts and unique error variance
components across males are females were different but the slope parameters, variance in
intercept and slope parameters and covariance of intercept and slope parameters could be
equated across the sexes (Model 3e in Table 2.4a). The intercept, linear and quadratic slope all
had a significant mean, suggesting significant change over time. Moreover, the variances of
20

intercept and slope parameters were significant, suggesting heterogeneity in these parameters
across individuals. A plot of the estimated sample means is presented in Figure 2.3.
For proactive aggression, the two part growth model was approached in steps. First, the
binary portion of the data was examined to assess changes in the proportion of samples
endorsing any of the items on the proactive aggression scale (i.e., a score of zero vs. non zero).
For this part of the model, the addition of a linear (∆χ
2
=45.38, ∆df=3, p<0.01) and a quadratic
slope (∆χ
2
=212.06, ∆df=4, p<0.01) both significantly reduced the model misfit, as evidenced by
a decrease in the log likelihood function. ∆χ
2
was calculated as twice the difference in the log
likelihood value of each model. The estimated probability of endorsement by wave is presented
in Figure 2.4 and the proportions across males and females are presented in Figure 2.5. The
overall proportion of endorsing items followed a quadratic curve, peaking at Wave 3for both
males and females. The second part was the model with log transformed values for the
continuous portion of the observations. Once again, the addition of a linear slope significantly
improved model fit (∆χ
2
=34.84, ∆df=3, p<0.01). The addition of the quadratic slope (∆χ
2
=5.84,
∆df=4, p=0.21) did not fit the data significantly better than a linear slope and also increased the
BIC values, suggesting that a linear slope for the continuous portion of the data was a good fit.
The growth trajectory for the continuous part is presented in Figure 2.6. Next, these two models
were combined and a conditional model with all latent variables regressed on Sex was obtained.
The parameter estimates for that model are presented in Table 2.5b
Genetic Analyses
Twin correlations for reactive and proactive aggression for all five waves are presented in
Table 2.3a and 2.3b respectively. The MZ correlations were higher than DZ correlations at all
five waves, suggesting genetic influences for reactive and proactive aggression. However, the
21

MZ correlations were not always twice as large as the DZ correlations and in some cases were
very similar to the DZ correlations, suggesting shared environmental influences.
Univariate genetic analyses for reactive and proactive aggression are presented in Tables
2.6a and 2.6b respectively. At Wave1, 36% of the variance in reactive aggression scores in males
was explained by additive genetic effects (A) and 64% by unique environment (E) effects.
Whereas in females, 36% of the variance was explained by shared environmental effects (C) and
64% by unique environmental effects. At Waves 2 and 3, the parameters could be equated across
sex and 49% of the variance in reactive aggression scores was explained by A and 51% by C at
Wave2 and 39% by A and 61% by E at Wave 3. At Wave4, 24% of variance in male reactive
scores was explained by A and 76% by E and for females 44% and 56% was explained by A and
E, respectively. Wave 5 univariate analyses revealed that parameters could be equated across sex
and 41% of the variance in reactive scores was explained by A and 59% by E.
For proactive aggression scores at Wave1, the variance contributions differed across
sexes. Whereas 49% of variance in male proactive scores was explained by A, only 14% was
explained by A in females. The remainder variance of 50% in males and 86% in females was
explained by E. At Wave 2, 39% of variance in male proactive scores was explained by C and
61% by E. In females, all three components were significant and explained 41% (A), 13% (C)
and 61% (E) of the variance in proactive scores. In males, additive genetic effects explained
35%, 40% and 42% of variance at Wave 3, Wave 4 and Wave 5 respectively and unique
environmental effects (E) explained 65%, 60% and 42% of variance. Wave 3 estimates for
females were the same, but at Wave 4 and Wave 5, A explained 32% and 18% of variance in
females, whereas 67% and 81% of variance was explained by E.
22

To examine the genetic and environmental influences on aggressive behaviors across
development, a series of multivariate models were fit. The results are presented in Table 2.7. A
Cholesky decomposition helps in parsing the variance of observed variables in a manner which
helps elucidate the different genetic and environmental factors coming into play at different
times.
For reactive aggression, the parameter estimates for males and females could be equated
(∆χ
2
=53.81, ∆df=45, p=0.17) and shared environmental influences could be dropped, without
any loss of fit (∆χ
2
=4.66, ∆df=15, p=0.99). The standardized parameter estimates are presented
in Figure 2.7. The first genetic factor loaded significantly on Waves 1-4, indicating common
genetic influences across childhood and adolescence. The second genetic factor shows additional
genetic variance (i.e., independent of the first factor influences) emerged during Wave 2 and had
significant loadings on all subsequent waves. Similarly, additional new genetic variance was
evident in both Waves 3 and 4 but not Wave 5. Genetic correlations using these estimates are
presented in Table 7a for reactive aggression. To further examine whether one or two common
A, C, E components could explain the shared variance between all five waves, one-factor and
two-factor independent pathway models were fit. The best fitting model was a one-factor
independent pathway where the parameters could be equated for males and females and all
shared environmental influences could be dropped. Standardized parameter estimates for this
model are presented in Figure 2.9. Both common and unique factors seemed important across
the five waves. Moreover, loadings from common unique environment factor were also
significant. Squaring and summing these standardized estimates for A and E gives the proportion
of variance at each Wave explained by these factors. The common genetic factor explained 15%,
37% 35%, 19% and 16% of variance at Wave 1, 2, 3, 4, and 5 respectively. In total, genetic
23

influences explained 36%, 50%, 37%, 28% and 34% of variance in reactive aggression from
Waves 1-5.
For proactive aggression, the parameter estimates for males and females could not be
equated for the Cholesky decomposition without a significant loss of fit (∆χ
2
=102.71, ∆df=45,
p<0.001). However, shared environmental influences could be dropped (∆χ
2
=22.02, ∆df=30,
p=0.85). The standardized parameter estimates for males and females are presented in Figure
2.8a and Figure 2.8b respectively. Although the first two genetic factors loaded significantly on
Waves1-3, genetic influences emerging at ages 14-15 were most influential for Wave 4 in males.
A similar pattern was observed in females, suggesting ‘new’ important genetic variance coming
into play for proactive aggression during the third wave of measurement. Genetic correlations
based on these models are presented in Table 2.8b. Further model fitting for one factor and two
factor independent pathway models revealed that the best fitting model was a one factor
independent pathway model. Parameter estimates were different across males and females but
shared environmental influences could be dropped. Standardized parameter estimates for this
model are presented in Figure 2.10a and Figure 2.10b for males and females respectively. Both
common and unique factors seemed important across the five waves. In males, the common
genetic factor explained 9%, 14%, 38%, 29% and 14% of variance from Waves 1-5 and in
females, the common genetic factor only explained 7%, 17%, 29%, 9% and 11% of variance in
females. The total variance explained by genetic factors in males ranged from 37-50% and these
estimates ranged from 14-55% in females.
Discussion
Development of overall aggressive behaviors has been well documented at both the
phenotypic and genetic levels. Development of the different forms of reactive and proactive
24

aggression, however, has not been examined in depth. The present study aimed to delineate
developmental changes in reactive aggression and proactive aggression at mean levels and
individual differences in change over time.
For reactive aggression, a quadratic trajectory of change was observed from childhood to
emerging adulthood, with levels peaking during Wave 3 when the twins were 14-15 years old.
Individual differences in both initial levels (intercept) and change (slope) were significant. This
pattern of development was similar both males and females and gender differences were only
found in the mean levels of aggression and the unique error variance.
A preponderance of zeros in the proactive aggression scores called for a different
approach to the analyses. A two-part model was used to examine changes in both probabilities of
endorsement and the mean scores (if not zero) of proactive aggression. Results suggest that the
observed increase in mean levels of proactive aggression stems from an increased engagement in
these behaviors across time. When a growth model with only the non-zero scores is fit, a linear
decrease in aggression scores is observed. The probability of endorsement followed a quadratic
curve, consistent with the observed means, but the changes in the continuous part of the data
showed a small but significant decline over time. Sex differences were significant for both the
endorsement (non-zero) as well as the continuous part, suggesting that females have a lower
probability of endorsing items on the proactive aggression questions and when the continuous
scores are observed, they have a lower intercept than males.
The phenotypic results for reactive and proactive aggression are consistent with previous
findings showing increase in aggressive behavior from childhood to adolescence ( see review-
Loeber & Hay, 1997) and the wide realm of antisocial behaviors, which show a dramatic
25

increase during adolescence (Moffitt, 1993). These behaviors are thought to accompany the
plethora of changes in biological and social factors during adolescence. However, there also exist
significant individual differences in the levels and changes accompanying both reactive and
proactive aggression, suggesting that not every adolescent will display the same level of
aggressive behaviors.
Genetic analyses within each wave revealed that variance in reactive aggression scores
could be attributed primarily to genetic and non-shared environmental influences. There was an
indication of shared environmental influences in females only during Wave 1. Male and females
parameters could be equated for Waves 2, 3 and 5 and not during Waves 1 and 4. Heritability
estimates ranged between 0.24 and 0.49. For proactive aggression, univariate analyses at each
Wave revealed that genetic and non-shared environmental factors were largely influencing the
variation in proactive aggression except at Wave 2, when the influence of shared environment
was found for both males and females. Heritability of proactive aggression scores was lower in
females for all occasions except Wave 2.
Multivariate longitudinal analyses suggested that covariation among the five waves could
again be largely attributed to one common genetic factor and one common non-shared
environmental factor. Age specific genetic and non-shared environmental factors were also
important. For reactive aggression scores, the common genetic factor explained 15-37% of the
variance in aggression scores which was 28-50% of the total genetic variance or heritability. For
proactive aggression scores in males, 9-38% of variance was explained by the common additive
genetic factor and in females, these estimates ranged from 7-30%. Overall, the heritability
estimates for proactive aggression scores were higher in males than in females. This result had
important implications for the studies examining sex differences in the etiology of aggressive
26

behaviors. As evident, even though sex differences are not found to be important for overall
aggression scores (Rhee & Waldman, 2002; Tuvblad & Baker, 2011), examining different forms
of aggression can yield different results. Other factors that may impact these results is
differences in informant (Brendgen, et al., 2006; Tuvblad, et al., 2009). For example, Brendgen
(2006) did not find any significant sex differences teacher reported reactive and proactive
aggression scores. However, in the same sample as used in the present study, assessed at Wave 1,
significant sex differences in self-reported reactive and proactive aggression emerged but no sex
differences were found for teacher and caregiver reports.
Cholesky decomposition results are important in evaluating the longitudinal structure of
genetic stability and innovation. For reactive aggression, significant genetic innovation came into
play during Waves 2, 3, and 4. Genetic correlation among different Waves ranged from 0.22
(between Waves 1 and 5) to 0.81 (between Waves 3 and 4). For proactive aggression scores,
both genetic stability and innovation were significant. Even though the parameters could not be
equated across males and females, the pattern of loadings was similar. Genetic factors present at
Wave 1 and those emerging during Wave 3 seemed to be most important. The highest genetic
correlation in males was between Waves 3 and 4 (rg = 0.81) but in females the highest genetic
correlation was between Waves 4 and 5 (rg = 0.89).
No influence of shared environment was found in the multivariate analyses. This, in part,
could result from a limitation of the model fitting methodology. In longitudinal genetic models,
shared environmental influences (or additive genetic influences) are dropped at large and the
resulting misfit is assessed. It is possible that shared environmental influences are not captured
through this methodology, due to low power for distinguishing these effects from genetic
influences and assortative mating.
27

In summary, reactive and proactive aggression change significantly from childhood to
adolescence to emerging adulthood. These changes are similar in males and females for reactive,
but not for proactive aggression. Genetic and non-shared environmental influences mostly
account for the variation, stability and innovation in aggression scores over time.

28

References
Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52(3), 317-332.
Baker, L. A., Raine, A., Liu, J., & Jacobson, K. C. (2008). Differential genetic and
environmental influences on reactive and proactive aggression in children. Journal of
abnormal child psychology, 36(8), 1265-1278.
Baker, L. A., Tuvblad, C., Wang, P., Gomez, K., Bezdjian, S., Niv, S., et al. (2013). The
Southern California Twin Register at the University of Southern California: III. Twin
research and human genetics : the official journal of the International Society for Twin
Studies, 16(1), 336-343.
Barker, E. D., Tremblay, R. E., Nagin, D. S., Vitaro, F., & Lacourse, E. (2006). Development of
male proactive and reactive physical aggression during adolescence. Journal of child
psychology and psychiatry, and allied disciplines, 47(8), 783-790.
Baş, A. U., & Yurdabakan, İ. (2012). Factor structure of the Reactive-Proactive Aggression
Questionnaire in Turkish children and gender, grade-level, and socioeconomic status
differences in reactive and proactive aggression. Journal of Psychoeducational
Assessment, 0734282911428892.
Brendgen, M., Vitaro, F., Boivin, M., Dionne, G., & Perusse, D. (2006). Examining genetic and
environmental effects on reactive versus proactive aggression. Developmental
psychology, 42(6), 1299-1312.
29

Brown, E. C., Catalano, R. F., Fleming, C. B., Haggerty, K. P., & Abbott, R. D. (2005).
Adolescent substance use outcomes in the Raising Healthy Children project: a two-part
latent growth curve analysis. Journal of consulting and clinical psychology, 73(4), 699.
Bushman, B. J., & Anderson, C. A. (2001). Is it time to pull the plug on the hostile versus
instrumental aggression dichotomy? Psychological review, 108(1), 273-279.
Connor, D. F., Steingard, R. J., Anderson, J. J., & Melloni, R. H., Jr. (2003). Gender differences
in reactive and proactive aggression. Child psychiatry and human development, 33(4),
279-294.
Dodge, K. A., & Coie, J. D. (1987). Social-information-processing factors in reactive and
proactive aggression in children's peer groups. Journal of personality and social
psychology, 53(6), 1146-1158.
Fite, P. J., Colder, C. R., Lochman, J. E., & Wells, K. C. (2008). Developmental trajectories of
proactive and reactive aggression from fifth to ninth grade. Journal of clinical child and
adolescent psychology : the official journal for the Society of Clinical Child and
Adolescent Psychology, American Psychological Association, Division 53, 37(2), 412-
421.
Fite, P. J., Raine, A., Stouthamer-Loeber, M., Loeber, R., & Pardini, D. A. (2009). Reactive and
proactive aggression in adolescent males: Examining differential outcomes 10 years later
in early adulthood. Criminal Justice and Behavior.
Fossati, A., Raine, A., Borroni, S., Bizzozero, A., Volpi, E., Santalucia, I., et al. (2009). A cross-
cultural study of the psychometric properties of the Reactive-Proactive Aggression
30

Questionnaire among Italian nonclinical adolescents. Psychological assessment, 21(1),
131-135.
Fung, A. L., Raine, A., & Gao, Y. (2009). Cross-cultural generalizability of the Reactive-
Proactive Aggression Questionnaire (RPQ). Journal of personality assessment, 91(5),
473-479.
Hubbard, J. A., McAuliffe, M. D., Morrow, M. T., & Romano, L. J. (2010). Reactive and
proactive aggression in childhood and adolescence: precursors, outcomes, processes,
experiences, and measurement. Journal of personality, 78(1), 95-118.
Kempes, M., Matthys, W., De Vries, H., & Van Engeland, H. (2005). Reactive and proactive
aggression in children A review of theory, findings and the relevance for child and
adolescent psychiatry. European child & adolescent psychiatry, 14(1), 11-19.
Liu, J., Lewis, G., & Evans, L. (2013). Understanding aggressive behaviour across the lifespan.
Journal of psychiatric and mental health nursing, 20(2), 156-168.
Loeber, R., & Hay, D. (1997). Key issues in the development of aggression and violence from
childhood to early adulthood. Annual review of psychology, 48, 371-410.
McAuliffe, M. D., Hubbard, J. A., Rubin, R. M., Morrow, M. T., & Dearing, K. F. (2006).
Reactive and proactive aggression: Stability of constructs and relations to correlates. The
Journal of Genetic Psychology, 167(4), 365-382.
Merk, W., Orobio de Castro, B., Koops, W., & Matthys, W. (2005). The distinction between
reactive and proactive aggression: Utility for theory, diagnosis and treatment? European
Journal of Developmental Psychology, 2(2), 197-220.
31

Miles, D. R., & Carey, G. (1997). Genetic and environmental architecture of human aggression.
Journal of personality and social psychology, 72(1), 207-217.
Moffitt, T. E. (1993). Adolescence-limited and life-course-persistent antisocial behavior: a
developmental taxonomy. Psychological review, 100(4), 674-701.
Murray-Close, D., & Ostrov, J. M. (2009). A longitudinal study of forms and functions of
aggressive behavior in early childhood. Child development, 80(3), 828-842.
Muthén, L. K., & Muthén, B. O. (1998-2011). Mplus User's Guide. Sixth Edition. Los Angeles,
CA: Muthén & Muthén.
Neale, M., & Cardon, L. (1992). Methodology for genetic studies of twins and families: Springer
Science & Business Media.
Olsen, M. K., & Schafer, J. L. (2001). A two-part random-effects model for semicontinuous
longitudinal data. Journal of the American Statistical Association, 96(454), 730-745.
Polman, H., Orobio de Castro, B., Koops, W., van Boxtel, H. W., & Merk, W. W. (2007). A
meta-analysis of the distinction between reactive and proactive aggression in children and
adolescents. Journal of abnormal child psychology, 35(4), 522-535.
Poulin, F., & Boivin, M. (2000). Reactive and proactive aggression: evidence of a two-factor
model. Psychological assessment, 12(2), 115.
Raftery, A. E. (1995). Bayesian model selection in social research. Sociological methodology,
111-163.
32

Raine, A., Dodge, K., Loeber, R., Gatzke-Kopp, L., Lynam, D., Reynolds, C., et al. (2006). The
Reactive-Proactive Aggression Questionnaire: Differential Correlates of Reactive and
Proactive Aggression in Adolescent Boys. Aggressive behavior, 32(2), 159-171.
Rhee, S. H., & Waldman, I. D. (2002). Genetic and environmental influences on antisocial
behavior: a meta-analysis of twin and adoption studies. Psychological bulletin, 128(3),
490-529.
Seah, S. L., & Ang, R. P. (2008). Differential correlates of reactive and proactive aggression in
Asian adolescents: relations to narcissism, anxiety, schizotypal traits, and peer relations.
Aggressive behavior, 34(5), 553-562.
Sroufe, L. A., & Rutter, M. (1984). The domain of developmental psychopathology. Child
development, 55(1), 17-29.
Tuvblad, C., & Baker, L. A. (2011). Human aggression across the lifespan: genetic propensities
and environmental moderators. Advances in genetics, 75, 171-214.
Tuvblad, C., Dhamija, D., Berntsen, L., Raine, A., & Liu, J. (2016). Cross-Cultural Validation of
the Reactive-Proactive Aggression Questionnaire (RPQ) Using Four Large Samples from
the US, Hong Kong, and China. Journal of psychopathology and behavioral assessment,
38(1), 48-55.
Tuvblad, C., Raine, A., Zheng, M., & Baker, L. A. (2009). Genetic and environmental stability
differs in reactive and proactive aggression. Aggressive behavior, 35(6), 437-452.
33

Vitaro, F., Barker, E. D., Boivin, M., Brendgen, M., & Tremblay, R. E. (2006). Do early difficult
temperament and harsh parenting differentially predict reactive and proactive aggression?
Journal of abnormal child psychology, 34(5), 685-695.
Vitaro, F., Gendreau, P. L., Tremblay, R. E., & Oligny, P. (1998). Reactive and proactive
aggression differentially predict later conduct problems. Journal of child psychology and
psychiatry, and allied disciplines, 39(3), 377-385.

34

Tables and Figures

Table 2.1: Descriptive Statistics for Aggression Scores Waves 1-5
Males Females
Reactive N Mean SD N Mean SD
Wave 1 590 .67 0.34 617 0.61 0.32
Wave 2 396 .65 0.34 428 0.61 0.33
Wave 3 577 .68 0.32 597 0.66 0.31
Wave 4 467 .63 0.32 504 0.58 0.30
Wave5 465 .48 0.32 548 0.44 0.31
Proactive
Wave 1 590 0.09 0.17 617 0.06 0.12
Wave 2 396 0.08 0.15 428 0.06 0.11
Wave 3 577 0.11 0.15 597 0.10 0.14
Wave 4 467 0.11 0.15 504 0.08 0.11
Wave 5 465 0.07 0.14 548 0.04 0.08

Table 2.2a: Correlations Across Waves for Reactive Aggression
W1 (617) W2 (428) W3 (597) W4 (504) W5 (548)
W1 (590) 0.27** 0.25** 0.23** 0.21**
W2 (396) 0.31** 0.41** 0.32** 0.30**
W3 (577) 0.32** 0.41** 0.54** 0.50**
W4 (467) 0.19** 0.30** 0.55** 0.56**
W5 (465) 0.26** 0.22** 0.43** 0.58**
Note: Males are below the diagonal and females are above. ** indicates p<0.01
Number of participants in the parentheses.

Table 2.2b: Correlations Across waves for Proactive Aggression
W1 (617) W2 (428) W3 (597) W4 (504) W5 (548)
W1 (590) 0.17** 0.18** 0.11** 0.01
W2 (396) 0.16** 0.21** 0.19** 0.07
W3 (577) 0.24** 0.37** 0.36** 0.27**
W4 (467) 0.14** 0.25** 0.38** 0.26**
W5 (465) 0.21** 0.33** 0.34** 0.53**
Note: Males are below the diagonal and females are above. ** indicates p<0.01.
Number of participants in the parentheses.

35

Table 2.3a: Twin Correlations Across Zygosity Groups for Reactive Aggression
MZM DZM MZF DZF DZOS
W1 0.39** 0.22* 0.35** 0.32** 0.05
W2 0.41** 0.36** 0.55** 0.39** 0.15
W3 0.31** 0.17 0.46** 0.24** 0.19*
W4 0.27** 0.05 0.45** 0.26* 0.22*
W5 0.39** 0.22 0.49** 0.30** 0.08
Note: MZM=Monozygotic Males, DZM= Dizygotic Males, MZF=Monozygotic
Females, DZF=Dizygotic Females, DZOS= Dizygotic Opposite-Sex

Table 2.3b: Twin Correlations Across Zygosity Groups for Proactive Aggression
MZM DZM MZF DZF DZOS
W1 0.52** 0.19 0.33** 0.08 0.11
W2 0.40** 0.40** 0.45** 0.30** 0.34**
W3 0.30** 0.23* 0.33** 0.10 0.08
W4 0.38** 0.08 0.27** 0.19 0.10
W5 0.25** 0.05 0.23* -0.01 0.08
Note: MZM=Monozygotic Males, DZM= Dizygotic Males, MZF=Monozygotic
Females, DZF=Dizygotic Females, DZOS= Dizygotic Opposite-Sex
36

Table 2.4a: Phenotypic Latent Growth Curve Model- Fit Statistics for Reactive Aggression
compare
LL df with ∆χ
2
∆df p CFI AIC BIC RMSEA
Reactive
1 No Growth- intercept only -1279.44 0.40 2570.88 2583.92 0.14
2 Linear slope -1165.02 28 1 228.84 6 <.01 0.65 2354.03 2380.12 0.12
3 Quadratic slope -1048.62 20 2 232.79 8 <.01 0.89 2137.25 2180.74 0.08
3a Quadratic slope equate β
0
-1054.52 21 3 11.79 1 <.01 0.88 2147.03 2188.34 0.08
3b Quadratic growth equate β
1
-1049.86 21 3 2.47 1 0.12 0.89 2137.72 2179.03 0.08
3c Quadratic growth equate
β
1
, β
2

-1049.94 22 3b 0.16 1 0.69 0.89 2135.87 2175.01 0.08
3d Quadratic growth equate
β
1
, β
2,
σ
0
2
σ
1
2
σ
2
2

-1050.92 25 3c 1.96 3 0.58 0.89 2131.83 2164.45 0.07
3e Quadratic growth equate
β
1
, β
2,
σ
0
2
σ
1
2
σ
2
2
, σ
01
2
σ
02
2
σ
12
2

-1052.98 28 3d 4.13 3 0.25 0.89 2129.97 2156.06 0.07
3f Quadratic growth equate
β
1
, β
2,
σ
0
2
σ
1
2
σ
2
2
, σ
01
2
σ
02
2
σ
12
2
and σ
u
2

-1055.60 29 3e 5.24 1 0.02 0.89 2133.20 2157.12 0.07

Note: β
0
= intercept, β
1=
linear slope, β
2=
quadratic slope
,
σ
0
2
=variance of intercept, σ
1
2
=variance of linear slope,

σ
2
2
=variance of quadratic slop,
σ
01
2
=covariance of intercept and linear slope, σ
02
2
=covariance of intercept and quadratic slope,

σ
12
2
=covariance of linear and quadratic slopes and
σ
u
2
=residual variance

Table 2.4b: Phenotypic Latent Growth Curve Model Parameter Estimates for Reactive Aggression
Fixed Effect Random Effect
Intercept
β
0

G1-Slope
β
1

G2- Slope
β
2

Intercept
σ
0
2

U1-Slope
σ
1
2

U2-Slope
σ
2
2

C01-Covar
σ
01
2

C01-
Covar σ
02
2

C01-
Covar
σ
12
2

U[t]-
Unique
σ
u
2

Males 0.65** 0.08** -0.03** 0.05** 0.03** 0.001** -0.02** 0.002 -0.004** 0.06**
Females 0.60** 0.08** -0.03** 0.05** 0.03** 0.001** -0.02** 0.002 -0.004** 0.05**

37

Table 2.5a: Phenotypic Latent Growth Curve Mode-Fit Statistics for Proactive Aggression
Proactive compare
LL df with ∆χ
2
∆df p AIC BIC
Part 1- Binary
1 No Growth- intercept only -3499.76 7003.51 7007.86
2 Linear slope -3477.07 26 1 45.38 3 <0.01 6964.14 6975.01
3 Quadratic slope -3371.04 22 2 212.06 4 <0.01 6760.09 6779.65
Part Two- continuous
1 Intercept only -2269.54 4553.09 4566.41
2 Linear slope -2252.12 10 1 34.84 3 <0.01 4524.23 4543.27
3 Quadratic slope -2249.20 6 2 5.84 4 0.21 4526.39 4553.04

Table 2.5b: Phenotypic Latent Growth Curve Model Parameter Estimates for Proactive Aggression
Proactive Wave1 status Linear Growth Quadratic growth

Variable Estimate SE Estimate SE Estimate SE

Part1: Endorsement vs non endorsement
Growth Factor intercept -0.66** 0.10 0.93** 0.11 -0.22** 0.03
Gender -0.44** 0.14 0.37* 0.16 -0.09* 0.04

Part two: continuous

Growth Factor intercept -1.89** 0.04 -0.04** 0.01
Gender -0.16** 0.05 -0.01 0.02

38

Table 2.6a: Univariate Genetic Model Fitting Results for Reactive aggression Scores for Waves 1-5
Reactive Parameter Estimates
Model χ
2
df p CFI LL AIC BIC RMSEA A C E
Wave 1
Full ACE 16.23 17 0.51 1.00 -330.01 676.03 685.86 0.00
Equate variances 24.99 20 0.20 0.91 -334.40 678.79 684.93 0.05
ACE m≠f, Drop C in
males, A in females
16.66 19 0.61 1.00 -320.23 672.46 679.83 0.00 0.36
(0.00)
0.00
(0.36)
0.64
(0.64)

Wave 2
Full ACE 20.74 17 0.24 0.95 -226.23 468.46 475.26 0.05
ACE m=f 24.04 20 0.24 0.94 -227.88 465.76 470.01 0.05
ACE m=f, Drop C 24.20 21 0.28 0.95 -227.96 463.92 467.32 0.04 0.49 - 0.51

Wave 3
Full ACE 12.89 17 0.74 1.00 -287.62 591.25 600.92 0.00
ACE m=f 17.00 20 0.65 1.00 -289.68 465.76 470.01 0.00
ACE m=f, Drop C 17.05 21 0.71 1.00 -289.70 587.41 592.24 0.00 0.39 - 0.61

Wave 4
Full ACE 15.84 17 0.54 1.00 -223.59 463.17 471.62 0.00
ACE m=f 24.43 20 0.22 0.89 -227.88 465.77 471.05 0.05
ACE m≠f, Drop C 16.68 19 0.61 1.00 -224.01 460.01 466.35 0.00 0.24
(0.44)
0.76
(0.56)

Wave 5
Full ACE 5.56 17 0.99 1.00 -251.88 519.76 528.89 0.00
ACE m=f 10.90 20 0.95 1.00 -254.55 519.11 524.81 0.00
ACE m=f, Drop C 10.90 21 0.96 1.00 -254.55 517.11 521.67 0.00 0.41 0.59

39

Table 2.6b: Univariate Genetic Modeling Fitting Results Proactive Aggression Scores for Waves 1-5
Proactive Parameter Estimates
Model χ
2
df p CFI LL AIC BIC RMSEA A C E
Wave 1
Full ACE 11.89 17 0.81 1.00 -1378.57 2773.13 2782.96 0.00
ACE m=f 38.85 20 0.01 0.55 -1392.05 2794.09 2800.23 0.09
ACE m≠f, Drop C 12.77 19 0.85 1.00 -1379.00 2770.00 2777.38 0.00 0.49
(0.14)
0.50
(0.86)
Wave 2
Full ACE 22.00 17 0.15 0.91 -936.39 1888.79 1895.58 0.07
ACE m=f 35.61 20 0.02 0.75 -942.69 1895.38 1899.63 0.09
ACE m≠f, Drop A m 23.07 18 0.19 0.92 -936.42 1886.85 1892.80 0.06 -
(0.41)
0.39
(0.13)
0.61
(0.46)
Wave 3
Full ACE 15.03 17 0.59 1.00 -1477.69 2971.38 2981.05 0.00
ACE m=f 21.12 20 0.39 0.97 -1480.74 2971.47 2977.51 0.02
ACE m=f, Drop C 21.49 21 0.43 0.99 -1480.92 2969.84 2987.38 0.01 0.35 - 0.65

Wave 4
Full ACE 11.72 17 0.82 1.00 -1192.52 2401.04 2409.49 0.00
ACE m=f 23.40 20 0.27 0.90 -1198.35 2406.71 2411.99 0.04
ACE m≠f, Drop C m, f 12.61 19 0.85 1.00 -1192.96 2397.91 2404.25 0.00 0.40
(0.32)
- 0.60
(0.67)
Wave 5
Full ACE 19.21 17 0.32 0.92 -1142.36 2300.72 2309.85 0.03
Equate variances 41.55 20 0.00 0.22 -1153.53 2317.06 2322.77 0.10
ACE m≠f, Drop C 19.21 19 0.44 0.99 -1142.36 2296.72 2322.61 0.01 0.42
(0.18)
0.58
(0.81)

40

Table 2.7: Multivariate Genetic Model Fitting Results
Model

χ
2
df CFI comparison ∆χ
2
/∆df p LL AIC BIC
RMSEA
Reactive Aggression
1 Cholesky m≠f 256.29 225 0.97 -862.64 1925.28 2390.83 0.03
1a Cholesky m=f 310.10 270 0.97 1 53.81/45 0.17 -889.55 1889.10 1970.50 0.03
1b Cholesky m=f Drop C 314.76 285 0.97 1a 4.66/15 0.99 -891.88 1863.76 1922.96 0.03
1c Cholesky m=f Drop A 337.10 285 0.95 1a 27/15 0.03 -903.05 1886.09 1945.29 0.03
2 1 fac IP 282.64 255 0.97 1 26.35/30 0.66 -875.82 1891.64 1995.23 0.03
2a 1 fac IP m=f 320.95 285 0.97 2 38.31/30 0.14 -894.97 1869.95 1929.15 0.03
2b 1 fac IP m=f Drop C 334.67 295 0.97 2a 13.72/10 0.19 -901.83 1863.66 1908.06 0.03
2c 1 fac IP m=f Drop A 367.66 295 0.94 2a 46.71/10 <.001 -918.33 1896.66 1941.06 0.04
3 2 factor IP m≠f 259.42 231 0.98 1 3.13/6 0.79 -864.21 1916.42 2055.54 0.03
3a 2 factor IP m=f 313.83 273 0.96 3 54.41/42 0.09 -891.41 1886.82 1963.78 0.03
3b 2 factor IP m=f Drop C 317.82 287 0.97 3a 3.99/14 0.99 -893.41 1862.82 1919.06 0.03
3c 2 factor IP m=f Drop A 339.06 287 0.95 3a 25.23/14 0.03 -904.03 1884.06 1940.30 0.03

Proactive Aggression
1 Cholesky m≠f 303.51 225 0.88 -5913.58 12027.16 12175.29 0.05
1a Cholesky m=f 406.21 270 0.79 1 102.7/45 <.001 -5964.93 12039.87 12121.33 0.06
1b Cholesky m≠f Drop C 325.53 255 0.89 1 22.02/30 0.85 -5924.59 11989.19 12092.87 0.04
1c Cholesky m≠f Drop A 345.37 255 0.86 1 41.86/30 0.07 -5934.51 12009.03 12112.71 0.05
2 1 fac IP m≠f 322.96 255 0.90 1 19.45/30 0.93 -5923.31 11986.61 12090.30 0.04
2a 1 fac IP m=f 413.29 285 0.80 2 90.33/30 <.001 -5968.48 12016.95 12076.20 0.05
2b 1 fac IP m≠f Drop C 349.39 275 0.89 2 26.43/20 0.15 -5936.52 11973.05 12047.11 0.04
2c 1 fac IP m≠f Drop A 368.24 275 0.86 2 45.28/20 0.001 -5945.95 11991.90 12224.73 0.05
3 2 factor IP m≠f 306.94 231 0.89 1 3.43/6 0.75 -5915.30 12018.60 12157.84 0.05
3a 2 factor IP m=f 394.70 273 0.80 3 87.76/42 <.001 -5965.22 12034.43 12111.46 0.06
3b 2 factor IP m≠f Drop C 327.40 259 0.90 3 20.46/28 0.85 -5925.53 11983.05 12080.81 0.04
3c 2 factor IP m≠f Drop A 340.14 259 0.88 3 33.2/48 0.95 -5931.90 11995.79 12093.56 0.05

Note:1 fac IP= One Factor Independent Pathway, 2 fac IP= 2 factor Independent Pathway
41

Table 2.8a: Genetic Correlations (Rg) Among Five Waves for Reactive Aggression
Wave 1 Wave 2 Wave 3 Wave 4 Wave 5

Wave 1 -
Wave 2 0.51 -
Wave 3 0.46 0.73 -
Wave 4 0.34 0.58 0.81 -
Wave 5 0.22 0.57 0.70 0.70 -

Table 2.8b: Genetic Correlations (Rg) Among Five Waves for Proactive Aggression
Wave 1 Wave 2 Wave 3 Wave 4 Wave 5

Wave 1 - 0.60 0.67 0.02 0.43
Wave 2 0.41 - 0.42 0.22 0.37
Wave 3 0.41 0.68 - 0.62 0.79
Wave 4 0.07 0.22 0.82 - 0.89
Wave 5 0.23 0.34 0.62 0.72 -

Note: males below diagonal, females above diagonal

42

Figure 2.1: Mean Scores for Reactive and Proactive Aggression

.30
.40
.50
.60
.70
.80
1 2 3 4 5
Mean +/- S.E.
Wave of Meaurement
Reactive Aggression
Males
Females
.00
.02
.04
.06
.08
.10
.12
.14
1 2 3 4 5
Mean +/- S.E.
Wave of Meaurement
Proactive Aggression
Males
Females
43

Figure 2.2: Latent Growth Curve Model with Linear and Quadratic Slopes
W1
W2 W5
Go
G1
1
1
1
1 1
1 2
3
4
σ
0
2
σ
1
2
σ01
2
β1
β0
0
W3
σ
u
2
W4
σ
u
2
σ
u
2
σ
u
2
σ
u
2
σ
2
2 G2
4
9 16
1
0
β1
σ02
2
σ12
2

Note:W1- W5= Waves 1-5

44

Figure 2.3: Estimated Means for Reactive Aggression from Latent Growth Curve Model

45

Figure 2.4: Estimated Probability of Endorsement for Proactive Aggression from the Growth Model

46

Figure 2.5: Proportions of Endorsement across Five Waves for Proactive Aggression in Males and Females

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1 2 3 4 5
Proportion of non-zero score respondents
Wave of Measurement
Males
Females
47

Figure 2.6: Estimated Log Mean for Proactive Aggression--Continuous Part of Two- Part Growth Model

48

Figure 2.7: Cholesky Decomposition Parameter Estimates for Reactive Aggression

Reactive
9-10 yrs
Reactive
11-13 yrs
Reactive
14-15 yrs
Reactive
16-18 yrs
Reactive
19-23 yrs
A1 A4 A5
E5
E4 E3 E2
A2
A3
E1
0.59*
0.36*
0.29*
0.20*
0.14
0.81*
0.09*
0.13*
0.09
0.13*
0.60*
0.36*
0.27*
0.33*
0.71*
0.12*
0.12*
0.02
0.76*
0.30*
0.21*
0.34*
0.15
0.42*
0.33*
0.27*
0.74*
0.31*
0.67*
0.41

49

Figure 2.8a: Cholesky Decomposition Parameter Estimates for Proactive Aggression in Males

Proactive
9-10 yrs
Proactive
11-13 yrs
Proactive
14-15 yrs
Proactive
16-18 yrs
Proactive
19-23 yrs
A1 A4 A5
E5 E4
E3 E2
A2
A3
E1
0.69*
0.26*
0.25*
0.04
0.15
0.72*
0.05
0.09
0.13
0.05
0.58*
0.34*
0.14
0.17
0.77*
0.04
0.19*
0.14
0.80*
0.04
0.02
0.16
0.41
0.43*
0.63*
0.34*
0.71*
0.22*
0.72*
0.28

50

Figure 2.8b: Cholesky Decomposition Parameter Estimates for Proactive Aggression in Females

Proactive
9-10 yrs
Proactive
11-13 yrs
Proactive
14-15 yrs
Proactive
16-18 yrs
Proactive
19-23 yrs
A1 A4 A5
E5
E4 E3 E2
A2
A3
E1
0.35*
0.44*
0.41*
0.01
0.20
0.94*
0.01
0.01
0.11
0.01
0.58*
0.01
0.15
0.06
0.68*
0.10
0.23*
0.10
0.78*
0.22*
0.07
0.30
0.27
0.45*
0.47*
0.31*
0.75*
0.31*
0.82*
0.01

51

Figure 2.9: Independent Pathway Model Parameter Estimates for Reactive Aggression
Reactive
9-10 yrs
Reactive
11-13 yrs
Reactive
14-15 yrs
Reactive
16-18 yrs
Reactive
19-23 yrs
Ac
0.39*
Au
0.46*
0.61*
0.59*
0.43*
0.40*
0.41*
Ec
0.08
0.11*
0.71*
0.54*
Eu
0.80*
Eu
0.70*
Au
0.36*
Eu
0.16
Eu
0.46*
Au
0.42*
Eu
0.61*
Au
0.68* 0.31*
Au

Note: Ac= Common genetic factor; Au= Unique additive genetic factor, Eu= Unique non-shared
environmental factor

52

Figure 2.10a: Independent Pathway Model for Proactive Aggression in Males
Proactive
9-10 yrs
Proactive
11-13 yrs
Proactive
14-15 yrs
Proactive
16-18 yrs
Proactive
19-23 yrs
Ac
0.30*
Au
0.64*
0.37*
0.62*
0.38*
0.54*
Ec
0.17 0.56*
Eu
0.70*
Eu
0.75*
Au
0.52*
Eu
0.78*
Eu
0.34*
Au
0.47*
Eu
0.57*
Au
0.07
Au
0.59*
0.05
0.50*
0.05
Note: Ac= Common genetic factor; Au= Unique additive genetic factor, Eu= Unique non-shared
environmental factor

53

Figure 2.10b: Independent Pathway Model for Proactive Aggression in Females
Proactive
9-10 yrs
Proactive
11-13 yrs
Proactive
14-15 yrs
Proactive
16-18 yrs
Proactive
19-23 yrs
Ac
0.26*
0.42*
0.54*
0.33*
0.30*
Ec
0.77*
0.31*
Eu
0.93*
Eu
0.65*
Au
Eu
0.74*
Eu
0.51*
Au
0.33*
Eu
0.83*
Au
0.28*
Au
0.22
0.05
0.30*
0.61*
0.16*
Au
0.27*
Note:
Ac= Common genetic factor; Au= Unique additive genetic factor, Eu= Unique non-shared environmental factor

54

Chapter 3: Genetic and Environmental Influences in the Development of Electrodermal
Reactivity
Abstract
Development of electrodermal reactivity, as measured by the skin conductance orienting
response (SCOR), was examined in a longitudinal twin sample from childhood to emerging
adulthood. SCOR was measured at five different time points between the ages 9-20years old.
Latent growth curve analyses revealed a non-linear change in SCOR, with peak average SCOR
magnitude during ages 14-15. Although mean level sex differences emerged in SCOR during
childhood and adulthood, with males exhibiting higher OR magnitude, the overall pattern of
change was quite similar for males and females. Within each wave of analysis, SCOR had
moderate to high heritability of 0.50 or higher. Shared environmental factors did not contribute
to the variance in SCOR and additive genetic and non-shared environmental factors accounted
for individual differences. There were no differences in males and females in the heritability
estimates for SCOR. Longitudinal genetic analyses revealed important stable genetic factors as
well as genetic innovation during adolescence, with genetic correlations ranging from 0.40 to
0.90. A single genetic factor accounted for the covariation among five waves while wave-
specific genetic factors remained significant. In conclusion, genetic factors play an important
role in the stability and change in SCOR magnitude over time.

55

Introduction
Electrodermal activity refers to the measurement of changes in the electrical properties of
the skin. Activity of eccrine sweat glands causes variation in these properties. When a small
electrical current is passed through the skin (usually the palmar surface), the conductance of this
electrical current depends on sweat gland activity. Greater activity allows for an easier passage
of current, increasing conductance. Activity of the eccrine sweat gland is directly under the
influence of the sympathetic nervous system, responsible for fight-or-flight responses. The
central influences on electrodermal activity consist of limbic as well as prefrontal structures,
making electrodermal activity a useful measure of emotional components and attention. The
sweat glands on palmar and plantar surfaces are especially known to be related to emotional
rather than thermoregulatory functions (for review, see (Boucsein, 2012; Hugdahl, 1995).
Both tonic (i.e., baseline) and phasic (i.e., event- or stimulus-related) measures of
electrodermal activity are used to index many physiological and psychological constructs
including attention, arousal, and regulation. Whereas electrodermal activity refers to overall
changes in the electrical properties of the skin, electrodermal reactivity refers to the changes in
response to a stimulus or situation. The skin conductance orienting response (SCOR) is a phasic
electrodermal response elicited by non-signal novel stimuli. The orienting response magnitude
(the size of the response averaged across trials) has been studied in relation to different
psychopathologies including schizophrenia (for review -see Dawson & Schell, 2002) and
ADHD. Schizophrenia patients tend to be either non-responders or hyper-responsive in an
orienting paradigm but hyperactivity has been associated with poorer outcomes in schizophrenic
patients. Similarly, ADHD is associated with decreased reactivity (or responsivity) to orienting
stimuli (Rosenthal & Allen, 1978). SCOR It has also been studied in the context of electrodermal
56

(hypo) reactivity as related to externalizing behaviors (Herpertz et al., 2001; Isen et al., 2010;
Isen, Iacono, Malone, & McGue, 2012; A. Raine, Venables, & Williams, 1995).
Given the importance of orienting response as a biomarker for pathologies and
dysregulated affective and attention processes, it is imperative to understand developmental
patterns and the genetic underpinnings of these developmental processes. The current study aims
to examine development of electrodermal orienting responses from childhood to early adulthood
at both a phenotypic and genetic level. Moreover, sex differences in the development of the
orienting response are also examined.
Development of Electrodermal Activity
Limbic and cortical structures affect electrodermal activity and thus changes in brain
maturation would be expected to influence both tonic and phasic electrodermal activity. A few
studies have investigated the relationship between age and electrodermal activity. A cross-
sectional study of 640 subjects revealed a significant relationship between age and both non-
specific skin conductance responses (NS-SCR) as well as reactivity to loud non-signaled stimuli
(Venables & Mitchell, 1996). Another study found significant changes in initial orienting
response in children from ages 3-8 (Gao, Raine, Dawson, Venables, & Mednick, 2007), peaking
at age six and leveling off thereafter, suggesting developmental changes in these measures map
onto changes in brain maturation and neural networks. Similarly, El-Sheikh (2007) found a
significant relationship between age and skin conductance baseline levels, but not for reactivity,
with younger children displaying higher baseline levels. Most other studies focus on changes
from adulthood to older ages (as reviewed in Boucsein, 2012) to investigate the changes related
to aging and study the effect of changes in the properties of the skin to EDA. Little is known
57

about the development of EDA from childhood to early adulthood. Considering that this period
of time is marked by significant changes in brain structure and function, one would expect this to
be mapped onto changes in autonomic reactivity. Orienting response is thus expected to change
across development in meaningful ways. To the author’s knowledge, no longitudinal study has
examined the development of SCOR from childhood to emerging adulthood during the critical
period of adolescence.
Genetic and Environmental Underpinnings of Electrodermal Reactivity
Measures of skin conductance level as well as reactivity tend to be heritable, with an
average heritability estimate of 25-50% (Crider et al., 2004; Isen, et al., 2012; Tuvblad et al.,
2012; Vaidyanathan et al., 2014). For both the amplitude and frequency of the skin conductance
orienting response, about 45-55% of variance (Isen, et al., 2012; Tuvblad, et al., 2012;Isen, et al.,
2012) can be accounted for by genetic factors, with the remainder of variance explained by
unique environmental factors. No effect of shared environmental factors was observed. Data
from longitudinal studies is limited (Tuvblad, et al., 2012) and a more comprehensive
understanding of genetic and environmental influences on the orienting response across
development is necessary to understand to what extent these risk factors play a role in shaping
physiological reactivity and how it maps onto developmental changes. In a longitudinal study
with the first three waves of SCOR in the current sample, Tuvblad et al. (2012) found moderate
to high heritability and that a single common genetic factor accounted for the stability in SCOR
from childhood to adolescence. In this paper, we extend these findings from childhood to
emerging adulthood.

58

Sex Differences in Orienting Response
Sex differences in electrodermal activity have only been studies sparsely and changes
across development are not clear. As reviewed by Bouscien (2012), some studies have shown
that females exhibit greater skin conductance levels (tonic activity) but males exhibit greater
reactivity to stimuli. However, there are significant effects of menstrual cycle and other
hormonal influences. Conversely, other studies have failed to find significant gender differences
in orienting responses (Raine, Venables, Mednick, & Mellingen, 2002). It is also possible that
sex differences are influences by the age of measurement and there exists a more nuanced
relationship between the development of electrodermal reactivity and sex.
The present study conducts a holistic examination of the development of electrodermal
reactivity, as measured by the orienting response (SCOR). Three main questions were examined:
(a) What are the phenotypic changes accompanying SCOR from childhood to emerging
adulthood? (b) To what extent do genetic and environmental factors influence SCOR and do
these influences change across time? (c) Are changes at the phenotypic and genetic level
different for males and females?
Methods
Participants
The data for the current study were collected as a part of a longitudinal twin study, the
Risk Factors for Antisocial Behavior (RFAB) twin study at the University of Southern California.
RFAB is a longitudinal study of the interplay of genetic, environmental, social and biological
factors in the development of antisocial and aggressive behaviors from childhood to emerging
adulthood. A community sample of twin families was recruited. The ethnic constitution of this
59

sample is: 44% Hispanic, 25% Caucasian, 16% African American, 3% Asian, and 12% mixed or
other, which is representative of the diversity in the greater Los Angeles area (Baker et al., 2013).
The sample contains a total of 1,564 subjects (781 twin pairs) - 169 monozygotic (MZ)
male, 171 MZ female, 121 dizygotic (DZ) male, 120 DZ female, and 200 DZ opposite-sex twin
pairs. During the first assessment (Wave 1) the participants were 9–10 years old ( mean age =
9.59, SD = 0.58); during Wave 2, the participants were 11–13 years old ( mean age = 11.79, SD =
0.92); during Wave 3, the participants were 14-15 years old ( mean age = 14.82, SD = 0.83) on
the third assessment (Wave 3) ; during Wave 4, the participants were 16-18 years old ( mean age
= 17.22, SD = 1.23), and during Wave 5 they were 19-23 (mean age=19.55, SD=1.11) years old.
Zygosity was determined through DNA microsatellite analysis (7 concordant and zero
discordant markers for MZ; one or more discordant markers for DZ) for 87% of the same-sex
twin pairs. For the remaining same-sex twin pairs, zygosity was established by questionnaire
items about the twins’ physical similarity and the frequency with which people confuse them. The
questionnaire was used only when DNA samples were insufficient for one or both twins in a pair.
When both questionnaire and DNA results were available, there was a 90% agreement between
the two (Baker et al., 2006, 2007).
Procedure:
At each of the five waves of data collection, families had the option to participate in 4-7
hours of assessment during a laboratory visit. During the first wave of assessment, all
participating families visited the laboratory. During Waves 2 and 3, an additional option to
participate through mail surveys was included and during Waves 4 and 5, there was an additional
option of participating through online surveys. For the twins, an average laboratory assessment
60

consisted of two parts: cognitive and psychophysiological. The psychophysiological portion
consisted of multiple different tasks at each waves.
Measures:
Electrodermal Activity: Electrodermal activity was recorded from the surface of the distal
phalanges of the non-dominant hand using Ag-Ag Cl electrodes. The cavities of these electrodes
were filled with either a water soluble lubricant (Waves 1-4) or a gel by Biopac consisting of
0.5% saline in a neutral base (Wave 5) and the electrodes were fastened with waterproof tape.
This was done to maintain a continuous conducting medium.
During Waves 1-4, electrodermal activity was recorded with equipment and software from
the James Long Company (1999; Caroga Lake, New York). A 31 channel Isolated Bioelectric
Amplifier was used with sampling rate of 512 Hz. All measurements except non-specific skin
conductance responses were scored through the James Long Company software. During Wave 5,
a bio amplifier from Biopac equipment (BIOPAC Systems) was used and online recordings were
made using the Acqknowledge software. A digitized signal was sent out with a sampling rate of
1000 Hz.
Orienting Task: Subjects were first briefed about the procedures that would be followed
during psychophysiological testing. Following attachment of electrodes, there was a three minute
rest period used for collection of baseline measures. The rest period was followed by an orienting
task during which the subjects were presented with four different auditory stimuli. The stimuli
consisted of: a) 1000 Hz tone (75 dB) presented twice , b) a consonant vowel sound (“da”, 65-
95dB) presented three times, c) four novel sounds (cuckoo clock, bird chirping, and rooster
crowing, 70-105 dB) and d) the sound of a baby crying (90-95 dB), presented three times. The
61

inter-stimulus intervals ranged from 18 to 28 seconds. The stimuli consisted of complex
waveforms that never exceeded a sound pressure level of 105 dB.
To measure the orienting response, 1-4.5 seconds after each stimulus onset and that
exhibited a minimum increase of 0.02 μS/second met the criteria for an orienting response. The
magnitude of this response was calculated as the difference between the peak value of skin
conductance and the skin conductance baseline level just before the onset of the response. If no
response was detected, then the magnitude for that particular trial assumed a value of zero rather
than being omitted.
Statistics:
Descriptive statistics were computed for males and females separately and mean
differences were examined within each wave. For phenotypic longitudinal analyses, latent
growth curve models were fit, using the SEM software Mplus (Muthén & Muthén, 1998-2011),
to examine developmental changes in mean SCOR magnitude and associated sex differences in
the development of SCOR magnitude. To incorporate the non-linear changes in SCOR
magnitude, a growth model with latent bases or a shape factor growth model was fit (McArdle,
2004; Meredith & Tisak, 1990). In this parametrization of the growth model, the time scores are
not constrained to increase in any linear or polynomial function. Instead, two bases are fixed
(usually as 0 and 1) and the rest are estimated. For the present study, the first basis was fixed to
be zero, so that the mean intercept provided the initial values at Wave 1. The last time score (or
basis) was fixed to be 1, making the interpretation of the slope as the mean change between
Waves 1 and 5. The estimated time scores then represent the proportion of change that occurs at
each time point relative to the overall change from Waves 1 through 5. To examine sex
62

differences, a sequential process of invariance testing was adopted. Details of this sequential
model fitting process are provided in the results section.
Goodness of fit of all growth models was assessed by a likelihood-ratio by computing the
difference between two times the log likelihood (-2*LL) of the full model from that of the
restricted model. This difference is distributed as a χ
2
. The degrees of freedom (df) for this test
are equal to the difference between the number of estimated parameters in the full model and that
in the restricted model. Model fit was also determined by comparing Akaike Information
Criterion (AIC) and Bayesian Information Criterion (BIC). The AIC represents the balance
between model fit and the number of parameters; with lower values of AIC indicating the most
suitable model (Akaike, 1987) .Similarly, increasingly negative values of the BIC correspond to
increasingly parsimonious and better fitting models (Raftery, 1995). The root mean square error
of approximation (RMSEA) and comparative fit index (CFI) were also used to judge model fit.
RMSEA values of 0.05 indicate good fit and 0.08 an adequate fit and CFI values range from 0 to
1.00 with higher values suggesting better fit.
Phenotypic twin correlations were computed across the five zygosity groups to gain a
first insight into the underlying sources of variation. All genetic models were fit using the
structural equation modeling program Mplus (Muthén & Muthén, 1998-2011). Univariate
genetic models were first fit to estimate the relative contribution of A, C and E to the variance of
SCOR magnitude at each wave. Sex differences in these variance sources were also examined.
Next, a multivariate Cholesky decomposition was fit to examine the sources of variation across
all five waves (Neale & McArdle, 2000). A Cholesky decomposition separates the variance of
observed variables into an equal number of A, C, and E factors as the waves of observation. In a
five wave model, this allows for five A, five C and five E factors to be estimated. The first factor
63

influences data from all five waves, the second factor influences all but the first wave and so on
and the last factor only influences the fifth (final) wave. This allows for the examination of
common and shared A, C , E variance sources across different waves, giving insight into the
stability and innovation in these factors across development while simultaneously estimating the
proportion of variance explained by A,C and E at each wave. Next, a more restrictive
independent pathway model was fit. This model allows for one shared or common A, C, E
factors across all waves as well as wave-specific factors for each wave. This model specifically
answers the questions about a single factor explaining the covariation among all five waves.
Next, a two factor independent pathway model was fit to examine whether more than one A, C,
E shared factor was required to account for the covariation among the five waves. In a two factor
independent pathway, two common A, C and E factors load onto scores at each wave in addition
to unique A, C and E components. As with phenotypic growth models, the goodness of fit was
assessed using Chi square difference tests, AIC, BIC, CFI and RMSEA.
Results
Descriptive Statistics
Descriptive statistics of SCOR magnitude for five waves are presented in Table 3.1.
There were significant sex differences in mean reactive aggression scores at Wave 1 (t
(1143)
=2.53,
p=0.01), Wave 4 (t
(315)
=2.86, p=0.005) and Wave 5 (t
(484)
=2.52, p=0.012). No significant sex
differences emerged during Wave 2 (t
(148)
=1.84, p=.07) and Wave 3 (t
(636)
=1.60, p=.11).
A 5 (Wave) × 2 (Sex) mixed design ANOVA was conducted to examine changes in OR
scores over time and any interaction of these changes with participant’s biological sex. The main
effect of Wave was significant (F(4, 48) = 3.39, p<0.05). Pairwise comparisons suggested a
significant increase in OR scores from Wave 2 to Wave 3 and a significant decrease to Wave 4,
64

with Wave 3 having a higher mean SCOR magnitude in comparison with Waves 2, 4 and 5. The
interaction between Wave and Sex was not found to be significant [F(4, 48) = 0.57, p=.69],
indicating that these mean level changes were consistent for both males and females. A plot of
the mean scores is presented in Figure 3.1.
Latent Growth Curve Analyses:
A repeated measures ANOVA only utilizes part of the data due to listwise deletion (i.e., a
case must have data at all five waves to be included). Moreover, it does not fully capture the
nuanced changes with time in an observed variable. For these reasons, latent growth curve
models were fit to closely examine the changes in orienting magnitudes over time. First, a model
with only an intercept was fit and subsequently a slope with free estimated time scores for Waves
2-4 was added, which encompassed non-linear changes over time. The addition of a slope factor
improved model fit (∆χ
2
= 134.52, ∆df = 8, p < 0.001), also reflected by a drastic increase in CFI
and decrease in AIC, BIC and RMSEA values. Next, a series of models were used to test for
invariance across males and females in both the fixed and random components. These models are
presented in Table 3.4. First, invariance of fixed effects (intercept and slope) were tested.
Subsequently, invariance in the random effects- variance of intercept and slopes were tested. To
account for dependence in the data from twin pairs, family was used as a clustering variable to
obtain corrected standard errors. The best fitting model based on Chi square, AIC and BIC was a
model where the intercepts and intercept variance across males and females were different but all
other parameters including latent bases, slope mean, variance in slope parameters and covariance
of intercept and slope parameters could be equated. The slope had a significant mean, suggesting
significant change over time. The variances of intercept parameters were significant, signifying
heterogeneity in these parameters across individuals. However, the variance of the slope was not
65

significant, suggesting fairly consistent trajectories across individuals. These results are
tabulated in Table 3.4b. The mean slope is negative, indicating a decrease in SCOR magnitude
from Wave 1 to 5. The time score for Wave 3 is negative, indicating an increase in SCOR
magnitude, given the overall negative mean of slope.
Genetic Analyses
Twin correlations SCOR magnitude for all five waves are presented in Table 3.2. Data
from Waves 1-3 have already been presented previously by Tuvlad et al. (2012). The MZ
correlations were higher than DZ correlations at all five waves, suggesting genetic influences in
both males and females. However, the MZ correlations were not always twice as large as the DZ
correlations for females, suggesting shared environmental influences might be at play. Univariate
genetic models are presented in Table 3.3.
Univariate genetic are presented in Table 3.5 (results from Waves 1-3 presented in
Tuvblad et al., 2012). At Waves 1 through 4, the parameter estimates could be equated across sex
without any significant loss of fit (as indicated by non-significant χ
2
tests. At Wave 5, even
though equating the parameter estimates for males and females did not result in a significant
increase in misfit as indicated by ∆χ
2
= 7.42, ∆df = 3, p = 0.06, the AIC value was slightly higher
and thus the full model was assessed. For all Waves, shared environment influences could be
dropped without a significant loss of fit and an increase in the parsimony of the model (as
indicated by lower AIC and BIC values). The AE sub model was the best fit to the data for all
five Waves, indicating no significant influence of shared environmental influences. Variance
estimates suggested moderate to high heritability of the mean SCOR magnitude for all five
Waves. The heritability of estimates ranged from 0.49-0.80 for males and females.
66

Multivariate genetic models were fit to examine the influence of genetic and
environmental factors from childhood to early adulthood. The results are presented in Table 3.6.
For the Cholesky decomposition, the model with equated parameter estimates for males and
females was more parsimonious (lower AIC, BIC) and did not result in a significant loss of fit
(∆χ
2
= 40.92, ∆df = 45, p = 0.65). Shared environmental influences could also be dropped
without significant loss of fit (∆χ
2
= 6.46, ∆df = 15, p = 0.97). Figure 3.2 presents standardized
parameter estimates for this Cholesky decomposition. The first genetic factor loaded
significantly on all five waves, suggesting stable genetic influences from an early time in
development. New genetic variance was significant at Waves 2 and 3 but genetic factors at
Waves 4 and 5 did not load significantly on scores at Waves 4 or 5. Non shared environmental
influences were time specific and only loaded significantly at each wave that they came into play
(i.e., E1 loaded significantly only at SCOR magnitude at Wave 1, E2 on Wave 2, etc). Overall,
the results suggested both genetic stability and genetic innovation. Genetic correlation computed
from this model are presented in Table 3.6b. The highest genetic correlation was observed
between Waves 3 and 4 (r
g
= 0.90) and the lowest between SCOR magnitudes at Waves 2 and 4,
5 (r
g
= 0.40).
To further examine whether one or two common A, C, E components could explain the
shared variance between all five waves, a one factor and two factor independent pathway model
were fit. The best fitting model was a one factor independent pathway model, as indicated by the
lowest AIC and BIC values. Parameter estimates could be equated for males and females (∆χ
2
=
31, ∆df = 30, p = 0.42) and shared environmental influences could be dropped further, without
any loss of fit (∆χ
2
= 8.74, ∆df = 10, p = 0.56). Moreover, common non-shared environmental
influences could be dropped and additive genetic influences remained as the only significant
67

factor explaining the covariance between Waves 1 through 5. The standardized parameter
estimates are presented in Figure 3.3. Unique genetic influences were also significant at all five
waves. Squaring and summing these standardized estimates for A and E gives the proportion of
variance at each Wave explained by these factors. The common genetic factor explained 36%,
35%, 38%, 30% and 24% of variance at Wave 1, 2, 3, 4, and 5 respectively. In totality, genetic
influences explained 54%, 80%, 49%, 46% and 56% of variance from Waves 1-5, results
consistent with the univariate analyses.
Due to a lower number of participants at Wave 2, the covariance coverage across waves
was really low; resulting in non-adequate fit statistics (CFI values of ~0.60, RMSEA~0.09/1).
The analyses were thus repeated without Wave 2, to confirm whether the same conclusions could
be drawn if Wave 2 scores were eliminated. As expected, the overall fit statistics improved, but
the best fitting models and overall conclusions remained the same. For comparison, the results
are presented in Table 3.7.
Discussion
In this longitudinal study, the developmental changes in skin conductance orienting
response (SCOR) were examined from both a phenotypic as well as a genetic standpoint.
Moreover, sex differences in mean levels, phenotypic changes and etiology of SCOR from
childhood to early adulthood were also examined. The main findings of this study are: a) there
are significant phenotypic changes observed through the course of development from childhood
to early adulthood with mean SCOR magnitudes peaking during adolescence; b) variance in
mean SCOR magnitude at each wave exhibited moderate to high heritability and c) a single
common genetic factor explains the covariance across development in SCOR magnitude from
childhood to emerging adulthood.
68

Phenotypic changes in SCOR magnitude revealed an interesting pattern. Although small
in magnitude, there are significant changes which occur over time. Highest SCOR magnitudes
were observed when the twins were 14-15 years old and the lowest SCOR magnitudes were
estimated around the ages of 16-18 years. Given the complex nature of the relationship between
SCOR, arousal, attention, the mechanisms behind these changes can only be speculated. The
neural substrates of the orienting response include the ventromedial prefrontal cortex,
hippocampus and anterior cingulate cortex (Williams et al., 2000) and the high SCOR
magnitudes observed during adolescence might reflect changes in brain maturation and
development (Pfefferbaum et al., 1994). Given the preponderance of changes during the time, it
is possible that the attentional resources allocated to external stimuli are increased. In a study
examining the effects of puberty on startle reflex, Quevedo and colleagues (2009) observed
heightened overall startle magnitudes, but not its modification, in mid/late pubertal adolescents
and speculated that this may be correlated with changes in brain areas responsible for arousal and
alertness. Another study (Silk et al., 2009) found increased pupil dilation in an affective word
identification task in mid/late pubertal children, speculating increased cognitive processing and
affective reactivity as potential mechanisms.
An examination of sex differences in SCOR magnitude revealed that males displayed
consistently higher SCOR magnitudes than females at each wave although these differences were
only significant statistically at Waves 1, 4 and 5. However, there were no differences in the
developmental patterns or magnitude of change between males and females, as indexed by equal
slope parameters in the growth model. Higher SCOR magnitudes in males have been previously
reported see for review (Boucsein, 2012)
69

Genetic analyses of SCOR revealed moderate to high heritability, ranging from 0.49-0.80
in both males and females. At all waves except Wave 5, the parameter estimates from genetic
models could be equated across sex. All variance in SCOR magnitude were mainly explained by
additive genetic factors and non-shared environmental factors. No evidence for shared
environmental factors was found, which is consistent with previous findings from this sample
(Tuvblad, et al., 2012) and other samples (Vaidyanathan, et al., 2014).
Longitudinal assessment of the etiology of SCOR suggested evidence for both genetic
stability and genetic innovation. The Cholesky decomposition parameter estimates presented in
Figure 3.2 show the extent of this stability and innovation. Additive genetic factors present at the
first wave of measurement when twins were only 9-10 years old, load significantly (albeit
moderately sized loadings) even at Wave 5. Using this decomposition, the genetic correlations
between the five waves were estimated and fell between 40-90%, once again suggesting stable
genetic factors at play with some genetic innovation across development. The highest genetic
correlation appeared between Waves 3 and 4- 0.90, indicating a high overlap in the genetic
variance during adolescence and emergence to adulthood.
The best fitting genetic model – a one factor independent pathway – suggests that
covariance among all five waves could be best explained by a single common additive genetic
factor and neither common shared or non-shared environment influences were important.
Moreover, wave specific genetic factors stayed significant at each assessment, signifying genetic
innovation across development. This common factor explained 24-38% of the variance in SCOR
magnitudes but accounted for 43-78% of the genetic variance at each wave. Consistent with prior
studies, no sex differences in the heritability of SCOR were apparent.
70

The stable genetic effects found in this study have important implications for SCOR as a
potential endophenotype. Establishing the heritability of a potential endophenotype is important
in then establishing its genetic overlap with a behavioral outcome.
There are several limitations of the present study. Firstly, the sample size at the second
wave of assessment was considerably lower than at all other waves. However, the presence of
four other waves of data collection and usage of full information maximum likelihood
procedures to account for missing data may partly overcome the problem of low power. It has
been posited that twin studies suffer from a low power to estimate shared environmental
influences. Thus, replication of these results would strengthen the literature on the contribution
of genetic and environmental influences on SCOR across development. A second limitation of
this study is the change in measurement procedures at Wave 5. The psychophysiological
recording system was replaced with updated, industry standard equipment from a different
company. Checks such as recording and measuring with a resistor were implemented to ensure
that the equipment was correct. However, equipment parameters related to measurement error
can vary with hardware and although small, such measurement errors can affect the data. Finally,
a considerable proportion of the sample at Wave 5 (roughly 20 per cent) had unusable data due
to either excessive participant movement or bad or noisy electrodes. All efforts were made to
collect and process data such that any usable data would be included. A third and important
limitation of the study is that we assumed no gene –environment correlations across
development. Any such correlation, if present, would bias the heritability estimates positively,
leading to overestimation.
In conclusion, this is one of the first studies examining the development of skin
conductance orienting response, a measure of orienting and autonomic reactivity, from childhood
71

to emerging adulthood and to examine the etiology of SCOR during this developmental period.
Results of the study indicate that SCOR magnitude changes in small but significant ways during
this period. Stable genetic influences explain a significant proportion of variance and time-
specific genetic factors also play an important role. Non-shared environmental influences tend to
be specific for the time of measurement and no influence of shared environmental factors is
apparent.

72

References
Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52(3), 317-332.
Baker, L. A., Tuvblad, C., Wang, P., Gomez, K., Bezdjian, S., Niv, S., et al. (2013). The
Southern California Twin Register at the University of Southern California: III. Twin
research and human genetics : the official journal of the International Society for Twin
Studies, 16(1), 336-343.
Boucsein, W. (2012). Electrodermal activity: Springer Science & Business Media.
Crider, A., Kremen, W. S., Xian, H., Jacobson, K. C., Waterman, B., Eisen, S. A., et al. (2004).
Stability, consistency, and heritability of electrodermal response lability in middle-aged
male twins. Psychophysiology, 41(4), 501-509.
Dawson, M. E., & Schell, A. M. (2002). What does electrodermal activity tell us about prognosis
in the schizophrenia spectrum? Schizophrenia research, 54(1-2), 87-93.
Gao, Y., Raine, A., Dawson, M. E., Venables, P. H., & Mednick, S. A. (2007). Development of
skin conductance orienting, habituation, and reorienting from ages 3 to 8 years: a
longitudinal latent growth curve analysis. Psychophysiology, 44(6), 855-863.
Herpertz, S. C., Wenning, B., Mueller, B., Qunaibi, M., Sass, H., & Herpertz-Dahlmann, B.
(2001). Psychophysiological responses in ADHD boys with and without conduct
disorder: implications for adult antisocial behavior. Journal of the American Academy of
Child and Adolescent Psychiatry, 40(10), 1222-1230.
Hugdahl, K. (1995). Psychophysiology: The mind-body perspective: Harvard University Press.
73

Isen, J., Raine, A., Baker, L., Dawson, M., Bezdjian, S., & Lozano, D. I. (2010). Sex-specific
association between psychopathic traits and electrodermal reactivity in children. Journal
of Abnormal Psychology, 119(1), 216-225.
Isen, J. D., Iacono, W. G., Malone, S. M., & McGue, M. (2012). Examining electrodermal
hyporeactivity as a marker of externalizing psychopathology: a twin study.
Psychophysiology, 49(8), 1039-1048.
McArdle, J. J. (2004). Latent growth curve analysis using structural equation modeling
techniques. Handbook of research methods in developmental science, 340-366.
Meredith, W., & Tisak, J. (1990). Latent curve analysis. Psychometrika, 55(1), 107-122.
Neale, M. C., & McArdle, J. J. (2000). Structured latent growth curves for twin data. Twin
research : the official journal of the International Society for Twin Studies, 3(3), 165-
177.
Pfefferbaum, A., Mathalon, D. H., Sullivan, E. V., Rawles, J. M., Zipursky, R. B., & Lim, K. O.
(1994). A quantitative magnetic resonance imaging study of changes in brain morphology
from infancy to late adulthood. Archives of neurology, 51(9), 874-887.
Quevedo, K. M., Benning, S. D., Gunnar, M. R., & Dahl, R. E. (2009). The onset of puberty:
effects on the psychophysiology of defensive and appetitive motivation. Development
and psychopathology, 21(1), 27-45.
Raftery, A. E. (1995). Bayesian model selection in social research. Sociological methodology,
111-163.
74

Raine, A., Venables, P. H., Mednick, S., & Mellingen, K. (2002). Increased psychophysiological
arousal and orienting at ages 3 and 11 years in persistently schizotypal adults.
Schizophrenia research, 54(1), 77-85.
Raine, A., Venables, P. H., & Williams, M. (1995). High autonomic arousal and electrodermal
orienting at age 15 years as protective factors against criminal behavior at age 29 years.
The American journal of psychiatry, 152(11), 1595-1600.
Rosenthal, R. H., & Allen, T. W. (1978). An examination of attention, arousal, and learning
dysfunctions of hyperkinetic children. Psychological bulletin, 85(4), 689.
Silk, J. S., Siegle, G. J., Whalen, D. J., Ostapenko, L. J., Ladouceur, C. D., & Dahl, R. E. (2009).
Pubertal changes in emotional information processing: pupillary, behavioral, and
subjective evidence during emotional word identification. Development and
psychopathology, 21(1), 7-26.
Tuvblad, C., Gao, Y., Isen, J., Botwick, T., Raine, A., & Baker, L. A. (2012). The heritability of
the skin conductance orienting response: a longitudinal twin study. Biological
psychology, 89(1), 47-53.
Vaidyanathan, U., Isen, J. D., Malone, S. M., Miller, M. B., McGue, M., & Iacono, W. G.
(2014). Heritability and molecular genetic basis of electrodermal activity: a genome-wide
association study. Psychophysiology, 51(12), 1259-1271.
Venables, P. H., & Mitchell, D. A. (1996). The effects of age, sex and time of testing on skin
conductance activity. Biological psychology, 43(2), 87-101.
75

Williams, L. M., Brammer, M. J., Skerrett, D., Lagopolous, J., Rennie, C., Kozek, K., et al.
(2000). The neural correlates of orienting: an integration of fMRI and skin conductance
orienting. Neuroreport, 11(13), 3011-3015.

76

Tables and Figures

Table 3.1: Descriptive Statistics for SCOR magnitude
Males Females
N Mean SD N Mean SD
Wave 1 559 0.43 0.30 586 0.39 0.28
Wave 2 77 0.35 0.28 73 0.25 0.30
Wave 3 304 0.46 0.32 334 0.42 0.30
Wave 4 141 0.34 0.25 176 0.27 0.23
Wave 5 242 0.41 0.29 244 0.35 0.27

Table 3.2: Twin correlations for SCOR magnitude across five zygosity groups
MZM DZM MZF DZF DZOS
Wave 1 0.57** 0.14 0.54** 0.35** 0.24
Wave 2 0.80** 0.12 0.88** 0.34 0.40
Wave 3 0.56** 0.30 0.49** 0.37** 0.22
Wave 4 0.62** -0.17 0.50** 0.37 0.31
Wave 5 0.51** 0.17 0.67** 0.47** 0.45**

Table 3.3: Cross-Wave correlations for mean SCOR magnitude
Wave 1 Wave 2 Wave 3 Wave 4 Wave 5
Wave 1 1
Wave 2 0.38** 1
Wave 3 0.43** 0.51** 1
Wave 4 0.37** 0.42** 0.44** 1
Wave 5 0.36** 0.45** 0.29** 0.28** 1

77

Table 3.4: Phenotypic Latent Growth curve model fit statistics for SCOR magnitude
compare
LL df with ∆χ
2
∆df p CFI AIC BIC RMSEA

1 No Growth- intercept only -423.47 30 0.48 866.94 888.01 0.07
2 Growth with latent bases -356.21 22 1 134.52 8 <.001 0.99 748.43 786.36 0.01
Invariance testing
3a Equate β
0
-358.60 23 2 4.78 1 0.03 0.98 751.20 841.03 0.02
3b Equate time scores -356.62 25 2 0.82 3 0.84 1.00 743.24 774.85 0.00
3c Equate time scores, β
1
-356.75 26 3b 0.26 1 0.61 1.00 741.50 771.01 0.00
3d Equate time scores, β
1,
σ
0
2
-358.40 27 3c 3.30 1 0.07 1.00 742.80 770.20 0.00
3e Equate time scores, β
1,
σ
1
2
-356.77 27 3c 0.40 1 0.53 1.00 739.53 766.93 0.00
3f Equate time scores, β
1,
σ
1
2
,
σ
01
2

-357.07 28 3e 0.6 1 0.44 1.00 738.14 763.43 0.00
3g Equate time scores, β
1,
σ
1
2
, σ
01
2
, σ
u
2

-357.14 29 3f 0.14 1 0.71 1.00 736.27 759.45 0.00

Note: β
0
= intercept, β
1=
slope
,
σ
0
2
=variance of intercept, σ
1
2
=variance of slope, σ
01
2
=covariance of intercept and slope and
σ
u
2
=residual variance

Table 3.4b: Phenotypic Latent Growth Curve model parameter estimates for SCOR Magnitude
Fixed Effect Random Effects
Intercept β
0
G1-Slope β
1
α
i2
, α
i3,
α
i4
Intercept σ
0
2
U1-Slope σ
1
2
C01-Covar σ
01
2
U[t]-Unique σ
u
2

Males 0.43* -0.04* 2,-0.98,
3.04
0.04* -0.001 -0.003 0.05
Females 0.38* 0.03*

78

Table 3.5: Univariate biometric results for SCOR Magnitude
Parameter Estimates
Model χ
2
df p ∆χ
2
/∆df p CFI LL AIC BIC
RMSEA
A C E
Wave1
1.Full ACE 16.60 17 0.48 1.00 -154.86 325.72 335.29 0.00
2.ACE m=f 18.47 20 0.56 1 1.87/3 0.60 1.00 -155.79 321.58 327.57 0.00
3ACE m=f Drop C 18.47 21 0.62 2 0/1 1.00 1.00 -155.79 319.58 324.37 0.00 0.53 - 0.47

Wave2
1.Full ACE 14.21 17 0.65 1.00 -3.36 22.73 16.26 0.00
2.ACE m=f 19.85 20 0.47 1 5.64/3 0.13 1.00 -6.18 22.36 18.32 0.00
3.ACE m=f Drop C 19.85 21 0.53 2 0/1 1.00 1.00 -6.18 20.36 17.13 0.00 0.80 - 0.20

Wave3
1.Full ACE 17.56 17 0.42 0.99 -134.65 285.29 290.89 0.02
2.ACE m=f 19.84 20 0.46 1 2.28/3 0.52 1.00 -135.78 281.56 285.06 0.00
3.ACE m=f Drop C 20.67 21 0.48 2 0.83/1 0.36 1.00 -136.20 280.40 283.20 0.00 0.49 - 0.51

Wave4
1.Full ACE 13.91 17 0.67 1.00 22.48 -28.96 -28.89 0.00
2.ACE m=f 14.74 20 0.79 1 0.83/3 0.84 1.00 22.07 -34.13 -34.09 0.00
3.ACE m=f Drop C 14.74 21 0.84 2 0/1 1.00 1.00 22.07 -36.13 -36.10 0.00 0.51 0.49

Wave5
1.Full ACE 10.21 17 0.89 1 1.00 -46.25 109.45 113.02 0.00
2.ACE m=f 17.63 20 0.61 2 7.42/3 0.06 1.00 -49.96 109.92 112.74 0.00
3.ACE m≠f Drop C 12.63 19 0.86 1 2.40/1 0.12 1.00 -47.46 106.92 110.31 0.00 0.49
(0.70)
0.51
(0.30)

79

Table 3.6: Longitudinal multivariate model fitting results for SCOR Magnitude

Compare

#
LL χ
2
df with ∆χ
2
/∆df p CFI AIC BIC
RMSEA

1 Cholesky m≠f -197.55 533.09 225 0.60 595.09 739.30 0.10
1a Cholesky m=f -217.60 574.01 270 1 40.92/45 0.65 0.61 545.19 624.50 0.09
1b Cholesky m=f Drop C -220.82 580.47 285 1a 6.46/15 0.97 0.62 521.64 579.33 0.08
1c Cholesky m=f Drop A -238.52 615.85 285 1a 41.84/15 <0.01 0.57 557.03 614.71 0.09

2 1 fac IP -205.89 550.61 255 1 17.52/30 0.97 0.62 551.79 652.73 0.09
2a 1 fac IP m=f -221.39 581.61 285 2 31/30 0.42 0.62 522.78 580.47 0.08
2b 1 fac IP m=f Drop C -225.76 590.35 295 2a 8.74/10 0.56 0.62 511.53 554.79 0.08
2c 1 fac IP m=f Drop A -244.43 627.69 295 2a 46.08/10 <0.01 0.57 548.87 592.13 0.09

3 2 factor IP m≠f -199.63 538.08 231 1 4.99/6 0.55 0.60 587.26 722.81 0.09
3a 2 factor IP m=f -217.90 574.62 273 3 36.54/42 0.71 0.61 539.80 614.79 0.09
3b 2 factor IP m=f Drop C -221.75 582.32 287 3a 7.7/14 0.90 0.62 519.50 574.29 0.08
3c 2 factor IP m=f Drop A -240.26 619.35 287 3a 44.73/14 <0.01 0.57 556.53 611.32 0.09

Note: 1 fac IP= one factor Independent Pathway, 2 fac IP= 2 factor independent pathway

Table 3.6b: Genetic correlations across five waves for SCOR Magnitude
Wave 1 Wave 2 Wave 3 Wave 4 Wave 5

Wave 1 -
Wave 2 0.50 -
Wave 3 0.67 0.62 -
Wave 4 0.67 0.40 0.90 -
Wave 5 0.60 0.40 0.47 0.66 -

80

Table 3.7: Longitudinal multivariate model fitting results for SCOR Magnitude without Wave 2

Compare

#
LL χ
2
df with ∆χ
2
/∆df p CFI AIC BIC
RMSEA

1 Cholesky m≠f -216.34 224.08 152 0.85 568.68 739.30 0.07
1a Cholesky m=f -228.18 247.75 182 1 23.67/30 0.78 0.86 532.35 587.15 0.05
1b Cholesky m=f Drop C -231.36 254.12 192 1a 6.37/10 0.78 0.87 518.72 559.10 0.05
1c Cholesky m=f Drop A -245.19 281.78 192 1a 34.03/10 <0.01 0.81 546.39 586.76 0.06

2 1 fac IP -219.07 229.54 164 1 5.46/12 0.94 0.86 550.14 630.89 0.05
2a 1 fac IP m=f -229.55 250.50 188 2 20.96/24 0.64 0.87 523.10 569.25 0.05
2b 1 fac IP m=f Drop C -233.57 258.53 196 2a 8.03/8 0.43 0.87 515.13 549.74 0.05
2c 1 fac IP m=f Drop A -247.87 287.15 196 2a 36.65/8 <0.01 0.80 543.75 578.36 0.06

3 2 factor IP m≠f -216.46 224.31 146 1 0.23/6 0.99 0.83 580.92 687.63 0.06
3a 2 factor IP m=f -228.18 247.75 179 3 23.44/33 0.89 0.85 538.35 597.47 0.05
3b 2 factor IP m=f Drop C -231.36 244.12 190 3a 3.63/11 0.98 0.86 522.72 565.98 0.05
3c 2 factor IP m=f Drop A -245.19 281.78 190 3a 34.03/11 <0.01 0.80 550.39 593.65 0.06

Note: 1 fac IP= one factor Independent Pathway, 2 fac IP= 2 factor independent pathway

81

Figure 3.1: Mean SCOR Scores

.0000
.1000
.2000
.3000
.4000
.5000
.6000
1 2 3 4 5
Mean +/- S.E.
Wave of Measurement
SCOR Magnitude
Males
Females
82

Figure 3.2: Cholesky Decomposition Parameter Estimates for SCOR Magnitude

SCOR
9-10 yrs
SCOR
11-13 yrs
SCOR
14-15 yrs
SCOR
16-18 yrs
SCOR
19-23 yrs
A1 A4 A5
E5
E4 E3 E2
A2
A3
E1
0.73*
0.44*
0.47*
0.46*
0.45*
0.68*
0.00
0.07
0.08
0.01
0.77*
0.24
0.05
0.09
0.47*
0.08
0.16
0.20
0.70*
0.05
0.06
0.25
0.44
0.47*
0.44*
0.04
0.70*
0.05
0.63*
0.40

83

Figure 3.3: One factor independent pathway for SCOR magnitude

SCOR
11-13 yrs
SCOR
14-15 yrs
SCOR
16-18 yrs
SCOR
19-23 yrs
Ac
0.60*
Au
0.42*
0.59*
0.62*
0.49*
0.55*
Eu
0.68*
Eu
0.45*
Au
0.67*
Eu
0.33*
Eu
0.73*
Au
0.57*
Eu
0.66*
Au
0.71* 0.40*
Au
SCOR
9-10 yrs

Note: Ac= Common genetic factor; Au= Unique additive genetic factor, Eu= Unique non-shared environmental factor

84

Chapter 4: Longitudinal Relationship between Reactive, Proactive Aggression and
Electrodermal Reactivity: A Twin Study
Abstract
It has been previously theorized that reactive and proactive aggression may have distinct
underlying physiological substrates. Whereas reactive aggression is theorized to be associated
with a hyperactive autonomic response system, proactive aggression may be characterized by a
hypoactive autonomic response system. In addition, the genetic and environmental etiologies of
these associations are unknown. In the present study, longitudinal association between reactive
and proactive aggression with skin conductance orienting response (SCOR) were examined in a
twin sample between the ages of 9-20 years old. While several significant concurrent and
longitudinal correlations emerged between aggressive behaviors and SCOR, no lucid
longitudinal pattern of associations could be discerned. Within-wave analyses did not reveal any
genetic overlap between SCOR and aggression measures. However, when the stable components
of these measures computed by averaging across waves were computed, a significant negative
association emerged between SCOR and proactive aggression in males. The genetic overlap
between SCOR and proactive aggression was significant in males, shedding light on the
underlying physiology.

85

Introduction
Autonomic functioning in aggressive and antisocial individuals has been studied
extensively, used to indicate deficits in attention, emotional arousal and regulation. Low
autonomic arousal and reactivity is consistently related to, and sometimes predictive of,
antisocial and aggressive behavior (Raine, 2002; Scarpa & Raine, 1997, 2004; van Goozen,
Fairchild, Snoek, & Harold, 2007). Neurobiological models of antisocial and aggressive behavior
posit that these results can be explained on the basis of multiple theoretical constructs including
fearlessness, low arousal, low orienting and sensation-seeking in aggressive and antisocial
individuals (Raine, 2002; Scarpa & Raine, 1997). Given that reactive and proactive aggressive
behaviors are differentially related to measures of personality, later conduct problems and
emotional responding, they are expected to have physiologically distinct profiles.
Electrodermal activity is directly related to sympathetic function and has been used as an
index of arousal and emotional responding. Skin conductance orienting response, specifically, is
a phasic electrodermal response to non-negative novel stimuli. Studies have found that smaller
orienting responses (Herpertz et al., 2003; Herpertz et al., 2001; Siddle, Nicol, & Foggitt, 1973);
lower frequency (Herpertz, et al., 2003; Herpertz, et al., 2001; Isen, Iacono, Malone, & McGue,
2012) and faster habituation of the orienting response (Gilbert, Gilbert, Johnson, & McColloch,
1991) are related to aggressive, antisocial and conduct problem behaviors. However, these
results from different studies do not always converge. Whereas some find only the frequency of
the orienting response to be related with antisocial problems (Isen, et al., 2012), others find that
only a smaller amplitude and not faster habituation (Siddle, et al., 1973) is associated with
aggressive behaviors. A meta-analysis on the psychophysiology of aggression and antisocial
behaviors found that differences in electrodermal arousal and reactivity differed across the
86

specific construct (psychopathy, aggressive behaviors, conduct problems) being measured
(Lorber, 2004), suggesting that sub forms of externalizing psychopathology can have varied
psychophysiological profiles.
Findings addressing the relationship between skin conductance, at large, and reactive and
proactive aggression are also mixed. One study found that greater electrodermal reactivity to an
aversive stimulus was predictive of reactive, but not proactive aggression (Hubbard et al., 2002).
Lower skin conductance level has shown to be associated with reactive aggression and higher
skin conductance level was associated with proactive aggression (Scarpa, Haden, & Tanaka,
2010). Another study, using the current sample of twins, found poor fear conditioning was
associated with persistent proactive, but not reactive aggression (Gao, Tuvblad, Schell, Baker, &
Raine, 2015). Scarpa and Raine suggest that these results can be explained by an under-active
and over-active autonomic system in proactive and reactive aggression, respectively (Scarpa &
Raine, 1997). However, more research needs to be done to assert and clarify the under and over
reactive autonomic system in reactive and proactive aggression to better understand their
biological bases.
The biological basis of these phenotypes can also be assessed through the lens of
behavior genetics. The heritability of both reactive and proactive aggression in the current
sample have been established. In the second chapter of this dissertation, it was demonstrated that
both reactive and proactive aggression have moderate heritability across development from
childhood to early adulthood. In the third chapter, it was established that the skin conductance
orienting response has a moderate to high heritability in both males and females, with a high
stability in genetic factors across development from childhood to early adulthood. Having
87

established that both these measures are heritable, this paper aims to examine the genetic overlap
in these constructs.
Sex differences in the relationship between autonomic responding and antisocial
behaviors, at large, have also been demonstrated. For example, in a study with the current sample
at ages 9-10, Isen and colleagues (2010) found sex differences in the relationship between
electrodermal hyporeactivity and psychopathy but only for boys and not for girls. Using the same
twin sample, a previous study (Niv, 2013) found a relationship between male antisocial behavior
and arousal, as measured by EEG. In the second chapter of this dissertation, we established that
there were significant quantitative sex differences in the etiology of proactive aggression, which
also affirm the plausibility of sex differences in the genetic overlap between aggression and
SCOR.
The main questions addressed in this study are: (a) Does electrodermal reactivity as
measured by SCOR related to reactive or proactive aggression at phenotypic level and do these
relationships differ for proactive and reactive aggression? (b) Do SCOR and aggression share
etiologies, and to what extent to the same genetic and environmental factors explain the
covariance between them? (c) Do these effects, if present, differ for males and females?
Methods
Participants
Data for the current study were drawn from a longitudinal twin study of Risk Factors for
Antisocial Behavior (RFAB) at the University of Southern California (USC). RFAB is a
longitudinal study of the interplay of genetic, environmental, social and biological factors in the
development of antisocial and aggressive behaviors from childhood to emerging adulthood.
88

Participants in this community sample of twin families were recruited from the greater Los Angeles
area and the ethnic constitution of this sample is representative of the diversity of this area: 44%
Hispanic, 25% Caucasian, 16% African American, 3% Asian, and 12% mixed or other.
The total sample consists of 781 twin pairs. On the first assessment (Wave 1) the
participants were 9–10 years old (mean age = 9.59, SD = 0.58; during Wave 2, the participants
were 11–13 years old (mean age = 11.79, SD = 0.92); during Wave 3, the participants were 14-15
years old (mean age = 14.82, SD = 0.83) on the third assessment (Wave 3) ; during Wave 4, the
participants were 16-18 years old ( mean age = 17.22, SD = 1.23), and during Wave 5 they were
19-23 (mean age=19.55, SD=1.11) years old.
Zygosity was determined through DNA microsatellite analysis (7 concordant and zero
discordant markers for MZ; one or more discordant markers for DZ) for 87% of the same-sex
twin pairs. For the remaining same-sex twin pairs, zygosity was established by questionnaire
items about the twins’ physical similarity and the frequency with which people confuse them. The
questionnaire was used only when DNA samples were insufficient for one or both twins in a pair.
When both questionnaire and DNA results were available, there was a 90% agreement between
the two (Baker et al., 2006, 2007, 2013).
Procedures
On all five waves of data collection, families had the option to participate in 4-7 hours of
assessment during a laboratory visit. During the first wave of assessment, all participating
families visited the laboratory. During Waves 2 and 3, an additional option to participate through
mail surveys was included and during Waves 4 and 5, there was an additional option of
participating through online surveys. For the twins, an average laboratory assessment consisted
89

of two parts: cognitive and psychophysiological. During the cognitive portion, they completed
several neuropsychological tests and clinical interviews. The psychophysiological portion
consisted of multiple different tasks at each waves. Caregivers only participated through
interviews, completed either in person or via mail or online surveys.
Measures:
Reactive Proactive Questionnaire: As delineated in Chapter 2, twins as well as their caregiver
completed the Reactive-Proactive Aggression Questionnaire (RPQ), a 23-item questionnaire with
12 items measuring proactive aggression and 11 items measuring reactive aggression (Raine,
2006). For the present study, only data from the twin subjects (not their caregivers) was used. The
mean score for each subscale was taken.
Electrodermal Activity: The procedures for measuring electrodermal activity have been
described in Chapter 3 of this dissertation. Briefly, electrodermal activity was recorded from the
surface of the distal phalanges of the non-dominant hand using Ag-Ag Cl electrodes, filled with
either a water soluble lubricant or a 0.5% saline gel. During Waves 1-4, electrodermal activity
was recorded with equipment and software from the James Long Company (1999; Caroga Lake,
New York). A 31 channel Isolated Bioelectric Amplifier was used with sampling rate of 512 Hz.
During Wave 5, a bio amplifier from Biopac equipment (BIOPAC Systems) was used and online
recordings were made using the Acqknowledge software. A digitized signal was sent out with a
sampling rate of 1000 Hz.
Orienting Task: Only a brief description of the task is included here, as details have been
provided in Chapter 3. Following a 3 minute rest period, an orienting task was presented during
which the subjects received four different auditory stimuli. The stimuli consisted of: a) 1000 Hz
90

tone (75 dB) presented twice , b) a consonant vowel sound (“da”, 65-95dB) presented three
times, c) four novel sounds (cuckoo clock, bird chirping, and rooster crowing, 70-105 dB) and d)
the sound of a baby crying (90-95 dB), presented three times. The orienting response was
measured during a window of 1-4.5 s after stimulus onset and the magnitude of this response was
calculated as the difference between the peak value of skin conductance and the skin
conductance baseline level just before the onset of the response. If no response was detected,
then the value for that particular trial assumed a value of zero rather than being omitted and a
mean SCOR magnitude was computed, which was the average amplitude (including zeroes)
across all trials.
Statistical analyses:
Descriptive statistics of aggression and SCOR magnitude were computed. To gain a first
insight into the association of SCOR magnitude and aggression, within and cross-wave
correlations between aggression measures and SCOR were computed- separately for males and
females.
To assess the genetic overlap between SCOR and aggression, genetic models were fit
using the structural equation modeling program Mplus (Muthén & Muthén, 1998-2007). Due to
the high skewness in proactive scores, these scores were ranked and normalized using Blom’s
transformation in SPSS. Bivariate Cholesky decompositions were used to examine the sources of
covariation among SCOR and aggression at each wave. In a bivariate Cholesky decomposition,
the variance and covariance of observed variables is proportioned into two A, C, E factors. The
first set of biometric A, C and E factors load onto both variables, giving the estimation for a) the
heritability of the first phenotype and b) the overlap between the two through a loading onto the
second phenotype. The second set of A,C and E factors only load on to the second phenotype,
91

representing the unique contribution of these influences. A re-parametrization of the Cholesky is
a correlated factors model, which assesses the correlation between the A factors (genetic
correlation or rg), C factors (shared environmental correlation or rc). Goodness of fit of these
genetic models was assessed by χ2-test, AIC, BIC and also examining CFI and RMSEA.
A second approach was used to examine the association between the stable components
of SCOR and aggression across Waves. Given that the genetic correlations among the five waves
could be explained by a single common genetic factor for both SCOR and aggression scores, an
averaged score for all waves was computed for SCOR, reactive and proactive aggression scores.
Data were normalized within each wave to eliminate the mean differences observed across time
in each of these measures, to alienate the stable components. This was done separately for boys
and girls, since mean differences between sexes were observed. These normalized scores were
then averaged across all five waves. These average scores were computed for participants who
had at least three valid data points (for each measure separately). This resulted in 372
participants with a valid average score on mean SCOR across all five waves and 1125
participants with valid scores on each of the aggression measures. Correlations between these
averaged scores were then computed.
Results
Within wave and cross-wave correlations for SCOR magnitude and reactive and
proactive aggression are presented in Table 4.1 and Table 4.2 respectively. Although a general
trend of negative correlations appeared, and more so for proactive aggression in males, no
conclusive pattern could be discerned. Cholesky decompositions for a within wave analyses did
not reveal any significant genetic overlap for aggression scores. The model fitting results are
92

presented in Tables 4.3 for SCOR and reactive and Table 4.4 for SCOR and proactive
aggression.
The second approach used was to examine the association between the stable components
of aggression. Correlations between SCOR and aggression measures are presented in Table 4.5.
A significant phenotypic correlation was observed between proactive aggression and SCOR
averaged across waves (r= -0.26, p<0.001) for males but not for females (r= -0.13, p=0.06). The
correlations between reactive aggression and SCOR averaged across waves were not significant
for either males (r= -0.11, p=0.11) or for females (r= -0.12, p=0.09). Bivariate Cholesky
decompositions between these averaged scores were fit to assess whether there was any
significant genetic overlap between aggression and SCOR. The results for these models are
presented in Table 4. 6.
For reactive aggression and SCOR measures, equating parameters resulted in a
significant drop in fit (∆χ
2
=21.34, ∆df=9, p=0.01). Thus, further analyses were conducted
without equating parameters across males and females. Dropping shared environmental
influences did not result in a significant drop in fit (∆χ
2
=11.50, ∆df=6, p=0.07) and the AIC and
BIC values also lowered, suggesting a more parsimonious model. Although dropping the shared
genetic path between SCOR and aggression in males did not result in a significant loss in fit
(∆χ
2
=2.81, ∆df=3, p=0.09), the AIC for the model was slightly higher. However, the shared
genetic path could be dropped in females, without a significant loss of fit (∆χ
2
=0.53, ∆df=1,
p=0.47) and lower AIC and BIC values. This resultant model is presented in Figure 4.1.
Interestingly, the paths from E1 to SCOR are significant and in opposing directions for males
and females.
93

Parameter estimates for Cholesky decomposition for proactive aggression and SCOR
could not be equated across males and females due to a significant loss in fit (∆χ
2
=23.96, ∆df=9,
p=0.004). However, shared environmental influences could be dropped without a significant
increase in misfit (∆χ
2
=4.78, ∆df=6, p=0.57) and led to a more parsimonious model, as indicated
by lower AIC and BIC values. The fit could be further improved by dropping the path from E1 to
SCOR, as presented in Figure 4.2. Dropping the shared genetic paths responsible for the
covariance between SCOR and proactive aggression in males led to a significant drop in fit
(∆χ
2
=6.61, ∆df=1, p=0.01). For females, this didn’t significantly increase the misfit (∆χ
2
=3.87,
∆df=1, p=0.05) but did lead to higher AIC and BIC values. The best fitting model is presented in
bold in Table 4.6.
Using estimated from these models, the heritability of these averaged scores as well as
the genetic correlations and environmental correlations can be calculated. The heritability
estimate for reactive aggression in males was calculated as 0.67
2
=0.45 and for SCOR, it could be
calculated as the sum of all squared genetic paths (0.20
2
+0.90
2
=0.85). The genetic correlation
(rg) for males could be calculated as the product of the paths emerging from A1 divided by the
square root of the product of the heritability for SCOR and proactive aggression, i.e.,
rg=
0.67∗(−0.20)
√0.85∗0.45

This gave a genetic correlation of 0.27 for males. The non-shared environmental
correlation could be similarly calculated and for males this was estimated to be 0.36. For
females, the heritability estimates were 0.53 and 0.64 for reactive aggression and SCOR,
respectively and the environmental correlation was -0.24.
94

For proactive aggression, the heritability estimates were 0.58 for males and 0.31 for
females. SCOR heritability estimates were 0.59 and 0.60 for males and females, respectively.
The genetic correlation for proactive aggression and SCOR was significant in males (rg= -0.40)
but not in females 0.44.
Prior research has demonstrated ethnic differences in electrodermal activity, specifically
lower skin conductance levels in African-American individuals (Boucsein, 2012). To ensure that
the results obtained were not due to ethnic differences, genetic analyses were also conducted on
residuals of SCOR and proactive aggression aggregate scores from regressions with ethnicity.
These results are presented in Table 4.7. Parameter estimates for the model were similar to those
obtained without controlling for ethnicity. The genetic correlation between SCOR and proactive
aggression was still significant in males (rg= -0.36), confirming that these results were not due to
masked effect of ethnic differences in either SCORor proactive aggression.
Discussion
In this longitudinal study, concurrent and longitudinal associations between conductance
orienting response (SCOR) and reactive and proactive aggression were examined from both a
phenotypic as well as a genetic standpoint. Sex differences in these associations were also
examined. Both within-wave and longitudinal relationships were examined to assess whether
SCOR characteristics were associated with risks for two different types of aggression. In a
second approach, aggregated scores across waves were examined to assess whether the stable
components of these measures exhibited significant associations. The main finding of this study
was the significant negative association between aggregated orienting responses and proactive
95

aggression in males. A significant genetic correlation of -0.40 was found between these
aggregated measures.
Within-wave phenotypic examination of SCOR and reactive and proactive aggression
revealed several negative correlations. Some of these correlations were significant, and majority
of these significant associations were for proactive aggression in males. No conclusive
longitudinal pattern was obvious from these correlations but nonetheless, the negative
correlations associated with proactive aggression were in line with our initial hypothesis. An
interesting point to be noted form the correlations was that none of the significant correlations
(very few) between reactive aggression and SCOR were in the positive direction. We had
hypothesized that if an over-aroused and over-active autonomic system underlies reactive
aggression, there should be positive associations with SCOR. However, all significant
correlations between SCOR and reactive aggression were negative.
The significant correlation, both at a phenotypic and genetic level, between proactive
aggression and SCOR magnitude in males has two prominent implications. Firstly, the
significant negative phenotypic correlation suggests that individuals who consistently score high
on proactive aggression display reduced orienting magnitudes. Secondly, this association is also
observed at the genetic level and there is significant genetic overlap between SCOR and
proactive aggression- suggesting shared etiology. In the current sample of twins, Gao et al.
(2015) examined the relationship of reactive and proactive aggression and fear conditioning and
concluded that those who were persistently high on proactive aggression exhibited poorer fear
conditioning. The results of the current study allude to a similar effect of consistently higher
proactive individuals exhibiting lower SCOR magnitudes. The exact nature of this effect can be
interpreted in two different ways. The first interpretation can be an underlying hyporeactive
96

autonomic system. However, a second explanation can also be reduced prefrontal activity and
attentional deficits as related to orienting (Raine, 2002). The third interesting finding of this
study is that the relationship between proactive aggression and SCOR only appeared for males. It
is possible that we failed to find this association because of a lower effect size in females and
less power. However, given that the aggregated sample had a larger number of females as
compared to males, it seems like a decreased SCOR magnitude may not represent a biological
risk factor in females and the effect is actually specific to males. Given the dearth of literature
specifically looking at these constructs and the associated sex differences, one can not wholly
disentangle these effects.
There are several limitations of the present study. Firstly, the sample size obtained by
averaging across all waves limited the use of all participants. Only 372 participants had scores on
both SCOR and aggression for at least three waves of assessment. This may bias the results and
decrease the generalizability to a larger population. A second limitation, which has been
previously discussed in Chapter 3 of this dissertation, is the change in psychophysiological
recording equipment at Wave 5. To reiterate, measures were taken to ensure that the readings
were accurate. However systematic error variance due to this change can not be ruled out.
In conclusion, this was one of the first studies examining the association of SCOR with
reactive and proactive aggression from a behavioral genetic standpoint. Phenotypic longitudinal
effects were inconclusive with no concrete pattern of association across waves. When the data
from all waves were aggregated after normalizing, a significant negative association was found
between proactive aggression and SCOR in males and this relationship was, in part, mediated
through genetic factors. No significant association between SCOR and reactive aggression were
found after data were aggregated across the five waves.
97

References
Boucsein, W. (2012). Electrodermal activity: Springer Science & Business Media.
Gao, Y., Tuvblad, C., Schell, A., Baker, L., & Raine, A. (2015). Skin conductance fear
conditioning impairments and aggression: a longitudinal study. Psychophysiology, 52(2),
288-295.
Gilbert, D. G., Gilbert, B. O., Johnson, S., & McColloch, M. A. (1991). Electrocortical and
electrodermal activity differences between aggressive adolescents and controls. Journal
of Social Behavior & Personality.
Herpertz, S. C., Mueller, B., Wenning, B., Qunaibi, M., Lichterfeld, C., & Herpertz-Dahlmann,
B. (2003). Autonomic responses in boys with externalizing disorders. Journal of neural
transmission, 110(10), 1181-1195.
Herpertz, S. C., Wenning, B., Mueller, B., Qunaibi, M., Sass, H., & Herpertz-Dahlmann, B.
(2001). Psychophysiological responses in ADHD boys with and without conduct
disorder: implications for adult antisocial behavior. Journal of the American Academy of
Child and Adolescent Psychiatry, 40(10), 1222-1230.
Hubbard, J. A., Smithmyer, C. M., Ramsden, S. R., Parker, E. H., Flanagan, K. D., Dearing, K.
F., et al. (2002). Observational, physiological, and self–report measures of children’s
anger: Relations to reactive versus proactive aggression. Child development, 73(4), 1101-
1118.
98

Isen, J. D., Iacono, W. G., Malone, S. M., & McGue, M. (2012). Examining electrodermal
hyporeactivity as a marker of externalizing psychopathology: a twin study.
Psychophysiology, 49(8), 1039-1048.
Lorber, M. F. (2004). Psychophysiology of aggression, psychopathy, and conduct problems: a
meta-analysis. Psychological bulletin, 130(4), 531-552.
Niv, S. (2013). Patterns of EEG spectral power in 9-10 year old twins and their relationships
with aggressive and nonaggressive antisocial behavior in childhood and adolescence:
University of Southern California.
Raine, A. (2002). Annotation: the role of prefrontal deficits, low autonomic arousal, and early
health factors in the development of antisocial and aggressive behavior in children.
Journal of child psychology and psychiatry, and allied disciplines, 43(4), 417-434.
Raine, A. (2002). Biosocial studies of antisocial and violent behavior in children and adults: A
review. Journal of abnormal child psychology,30(4), 311-326.
Scarpa, A., Haden, S. C., & Tanaka, A. (2010). Being hot-tempered: autonomic, emotional, and
behavioral distinctions between childhood reactive and proactive aggression. Biological
psychology, 84(3), 488-496.
Scarpa, A., & Raine, A. (1997). Psychophysiology of anger and violent behavior. The
Psychiatric clinics of North America, 20(2), 375-394.
Scarpa, A., & Raine, A. (2004). The psychophysiology of child misconduct. Pediatric annals,
33(5), 296-304.
99

Siddle, D. A., Nicol, A. R., & Foggitt, R. H. (1973). Habituation and over-extinction of the GSR
component of the orienting response in anti-social adolescents. The British journal of
social and clinical psychology, 12(3), 303-308.
van Goozen, S. H., Fairchild, G., Snoek, H., & Harold, G. T. (2007). The evidence for a
neurobiological model of childhood antisocial behavior. Psychological bulletin, 133(1),
149-182.
100

Tables and Figures

Table 4.1: Correlations Across Waves for Mean OR Magnitude and Reactive Aggression
SCOR
Magnitude 
Wave 1 Wave 2 Wave 3 Wave 4 Wave 5
Reactive M F M F M F M F M F
Wave 1 -0.04 -0.03 -0.13 0.05 -0.10 -0.03 -0.13 0.03 -0.02 -0.02
Wave 2 -0.17* 0.04 0.001 0.06 -0.21** 0.04 -0.06 -0.16 -0.12 0.17*
Wave 3 0.04 -0.06 -0.21 -0.14 -0.05 -0.14** -0.08 -0.08 0.02 -0.08
Wave 4 0.07 -0.01 0.01 -0.001 -0.02 0.05 -0.03 -0.04 0.004 -0.12
Wave 5 -0.04 -0.02 -0.20 -0.13 -0.08 -0.01 -0.03 -0.03 0.04 -0.21**
Note: ** indicates p<0.01, * indicates p<0.05 and (m) indicates p<0.1

Table 4.2: Correlations Across Waves for Mean OR Magnitude and Proactive Aggression
SCOR
Magnitude 
Wave 1 Wave 2 Wave 3 Wave 4 Wave 5
Proactive M F M F M F M F M F
Wave 1 -0.01 -0.05 -0.14 -0.08 -0.12 0.01 -0.20* 0.01 -0.02 -0.01
Wave 2 -0.10* 0.05 -0.05 -0.12 -0.23** 0.09 -0.24* -0.07 -0.11 0.09
Wave 3 -0.01 -0.13** -0.28* -0.12 -0.12* -0.10 -0.23* -0.18* -0.02 -0.12
Wave 4 -0.03 0.01 -0.25 -0.13 -0.05 0.06 -0.17* -0.18* 0.04 0.02
Wave 5 0.02 0.05 -0.15 -0.12 -0.13 0.04 -0.14 0.19* 0.04 -0.15*
Note: ** indicates p<0.01, * indicates p<0.05

101

Table 4.3: Bivariate Cholesky Results: Reactive Aggression and SCOR Magnitude

Compare

Model χ
2
df CFI
with
∆χ
2
/∆df P LL AIC BIC RMSEA
Wave1
1.1 Cholesky m≠f 58.70 48 0.94 -483.15 1010.29 1037.36 0.04
1.2. Cholesky m=f 69.06 57 0.93 1.1 10.36/9 0.32 -488.33 1002.67 1018.65 0.04
1.3. Cholesky m=f Drop C 70.49 60 0.94 1.2 1.43/3 0.70 -489.04 998.09 1010.39 0.04

Wave2
2.1 Cholesky m≠f 60.58 48 0.90 -224.13 492.26 510.96 0.06
2.2 Cholesky m=f 78.02 57 0.84 2.1 17.44/9 0.04 -232.85 491.69 502.75 0.07
3. Cholesky m≠f Drop C 67.66 54 0.90 2.1 7.08/6 0.31 -227.67 487.33 500.94 0.06

Wave3
3.1 Cholesky m≠f 45.60 48 1.00 -417.69 879.39 905.98 0.00
3.2 Cholesky m=f 54.18 57 1.00 3.1 8.58/9 0.48 -421.98 869.97 885.68 0.00
3.3 Cholesky m=f Drop C 55.95 60 1.00 3.2 1.77/3 0.62 -422.87 865.74 877.83 0.00

Wave4
4.1 Cholesky m≠f 56.53 48 0.88 -204.12 452.24 475.48 0.04
4.2 Cholesky m=f 66.73 57 0.87 4.1 10.20/9 0.33 -209.22 444.45 458.18 0.04
4.3 Cholesky m=f Drop C 67.10 60 0.90 4.2 0.37/3 0.95 -209.41 438.82 449.39 0.03

Wave5
5.1 Cholesky m≠f 31.14 48 1.00 -291.19 626.39 651.68 0.00
5.2 Cholesky m=f 53.67 57 1.00 5.1 22.53/9 .007 -302.46 630.91 645.86 0.00
5.3 Cholesky m≠f Drop C 36.09 54 1.00 5.1 4.95/6 0.55 -293.67 619.34 637.74 0.00

102

Table 4.4: Bivariate Cholesky Results: Proactive Aggression and SCOR Magnitude

Compare

Model χ
2
df
with
∆χ
2
/∆df P CFI LL AIC BIC RMSEA
Wave1
1.1 Cholesky m≠f 57.00 48 0.95 -1532.03 3108.07 3135.14 0.04
1.2 Cholesky m=f 86.84 57 1.1 29.84/9 <.001 0.82 -1546.96 3119.91 3177.18 0.07
1.3 Cholesky m≠f Drop C 59.03 54 1.1 2.03/6 0.92 0.97 -1533.05 3098.09 3117.78 0.03

Wave2
2.1 Cholesky 50.86 48 0.97 -938.74 1921.49 1940.19 0.03
2.2 Cholesky m=f 70.67 57 2.1 19.81/9 0.02 0.87 -948.65 1923.30 1934.35 0.05
2.3 Cholesky m≠f Drop C 55.57 54 2.1 4.71/6 0.58 0.99 -941.10 1914.20 1927.81 0.02

Wave3
3.1 Cholesky m≠f 54.17 48 0.94 -1609.29 3262.58 3289.17 0.03
3.2 Cholesky m=f 63.76 57 3.1 9.59/9 0.38 0.93 -1614.08 3254.16 3269.88 0.03
3.3 Cholesky m=f Drop C 64.99 60 3.2 1.23/3 0.75 0.95 -1614.70 3249.39 3261.48 0.03

Wave4
4.1 Cholesky m≠f 39.67 48 1.00 -1168.26 2380.52 2403.76 0.00
4.2 Cholesky m=f 55.42 57 4.1 15.75/9 0.07 1.00 -1176.13 2378.26 2391.99 0.00
4.3 Cholesky m=f Drop C 55.97 60 4.2 0.55/3 0.91 1.00 -1176.40 2372.82 2383.38 0.00

Wave5
5.1 Cholesky m≠f 52.47 48 0.95 -1195.38 2434.76 2460.06 0.03
5.2 Cholesky m=f 83.42 57 5.1 30.95/9 <.001 0.68 -1210.85 2447.70 2462.65 0.06
5.3 Cholesky m≠f Drop C 55.64 54 5.1 3.17/6 0.79 0.98 -1196.97 2425.93 2444.33 0.02

103

Table 4.5: Correlations -Aggregated SCOR and Aggression
SCOR (averaged across waves)
Aggression
(averaged across waves)
M (172) F (200)
Reactive Aggression -0.11 -0.12
Proactive aggression -0.26** -0.13

Note: M=males, F=females, number of participants in parentheses
104

Table 4.6: Bivariate Cholesky Results for Proactive, Reactive Aggression and SCOR – Aggregates Scores

Compare

Model χ
2
df
with
∆χ
2
/
∆df
p CFI LL AIC BIC RMSEA
Reactive Aggression
1 Cholesky m≠f 39.02 48 1.00 -1550.15 3144.30 3170.59 0.00
2 Cholesky m=f 60.36 57 1 21.34/9 0.01 1.00 -1560.82 3147.64 3163.18 0.03
3 Cholesky m≠f Drop C 50.52 54 1 11.50/6 0.07 1.00 -1555.90 3143.81 3162.93 0.03
4 Cholesky m≠f Drop C,
A
12
in males
53.33 55 3 2.81/1 0.09 1.00 -1557.31 3144.61 3162.54 0.00
5 Cholesky m≠f Drop
C, A
12
in females
51.05 55 3 0.53/1 0.47 1.00 -1556.17 3142.33 3160.26 0.00
Proactive Aggression
1 Cholesky 63.57 48 0.90 -1392.75 2829.50 2855.79 0.05
2 Cholesky m=f 87.53 57 1 23.96/9 .004 0.81 -948.65 2835.46 2850.99 0.07
3 Cholesky m≠f Drop C 68.35 54 1 4.78/6 0.57 0.91 -1395.14 2822.28 2841.41 0.05
4 Cholesky m≠f Drop
C, E
12

71.31 56 3 2.96/2 0.23 0.90 -1396.62 2821.25 2837.98 0.05
5 Cholesky m≠f Drop C,
E
12
, A
12
in females
75.17 57 4 3.87/1 0.05 0.89 -1398.55 2823.11 2838.65 0.05
6 Cholesky m≠f Drop C,
E
12
, A
12
in males
77.92 57 4 6.61/1 0.01 0.87 -1399.93 2825.86 2841.40 0.06

105

Table 4.7: Bivariate Cholesky Results for Proactive Aggression and SCOR –Controlling for Ethnicity

Compare

Model χ
2
df
with
∆χ
2
/
∆df
p CFI LL AIC BIC RMSEA
Proactive Aggression
1 Cholesky 64.75 48 0.89 -1382.52 2809.04 2835.33 0.06
2 Cholesky m=f 91.02 57 1 26.27/9 0.001 0.78 -1395.65 2817.31 2832.85 0.07
3 Cholesky m≠f Drop C 69.81 54 1 5.06/6 0.57 0.54 -1385.05 2802.09 2821.41 0.05
4 Cholesky m≠f Drop
C, E
12

72.51 56 3 2.70/2 0.26 0.89 -1386.40 2800.79 2817.52 0.05
5 Cholesky m≠f Drop C,
E
12
, A
12
in females
75.42 57 4 2.91/1 0.09 0.88 -1387.85 2801.70 2817.24 0.05
6 Cholesky m≠f Drop C,
E
12
, A
12
in males
79.12 57 4 6.61/1 0.01 0.85 -1389.70 2805.41 2820.94 0.06

106

Figure 4.1: Bivariate Model for Reactive Aggression and SCOR

Reactive
Agg
SCOR
A1
E2
A2
E1
0.67*
(0.73*)
0.74*
(0.69*)
0.90*
(0.80*)
-0.20
(0.00)
0.38*
(0.57*)
0.13*
(-.16*)

Note: Parameters for females presented in parentheses; *=p<0.05

107
Figure 4.2: Bivariate Model for Proactive Aggression and SCOR

Proactive
Agg
SCOR
A1
E2
A2
E1
0.76*
(0.56*)
0.83*
(0.83*)
0.77*
(0.78*)
-0.33*
(-0.25)
0.40*
(0.58*)

Note: Parameters for females presented in parentheses; *=p<0.05

108
Chapter 5: Discussion and Conclusions
Heterogeneity in aggressive behaviors has led researchers to operationalize myriad
manifestations of related yet distinct constructs. One such distinction has been based on the
functional utility of aggression- reactive and proactive aggression. This distinction has been
supported both by theorists supporting the utility of this distinction as well as from empirical
research demonstrating their differential correlates and predictors. The field of developmental
psychopathology addresses important questions pertinent to the development of both normative
and deviant behaviors. Limited research exists on the developmental course of reactive and
proactive aggression, the genetic and environmental underpinnings of these changes and the
genetic overlap with autonomic functioning over time. The goal of this study was to specifically
address these gaps in the literature. The results can be summarized as follows:
 Reactive and proactive aggression peak during adolescence and wane down during
the transition to adulthood
 Changes in reactive aggression from childhood to emerging adulthood are similar in
males and females. However, sex differences appear in the developmental course
of proactive aggressive behaviors over time, specifically in the binary presence or
absence of these behaviors. Females are less likely to endorse proactive
aggression items but those with non-zero scores at the initial time points develop
similarly to males.
 Genetic and environmental influences on the development of aggression show both
stability and innovation. Additive genetic and unique environmental influences
explain the variance in aggressive behaviors, with little influence of shared
environment.
109
 Whereas genetic and unique environmental influences on reactive aggression are
similar in males and females, they differ for proactive aggression. Heritability of
proactive aggression tends to be higher in males.
 The covariance in reactive aggression across the five waves of measurement could be
explained by a single genetic and a single genetic and a single unique
environmental factor. Wave-specific genetic and environmental influences
remained significant.
 Similarly, the covariance in proactive aggression scores could also be reduced down
to a single latent genetic factor and a single latent unique environmental factor.
 Genetic factors explained 46-90% variance in electrodermal reactivity, suggesting
moderate to high heritability.
 Significant mean differences in electrodermal reactivity emerged between males and
females but the longitudinal pattern of change was similar.
 Significant genetic overlap was found (r
g
= -0.43) between proactive aggression and
orienting response for males, suggesting shared etiology.
These results have several important implications. Firstly, important differences in
longitudinal development of reactive and proactive aggression- both at the phenotypic and
genetic level emerged. Although both peak during adolescence, the nature of these changes is
heterogeneous. For proactive aggression, sex differences, albeit small, come into play more than
just at the initial level. Proactive and not reactive aggression, in males, was associated with
SCOR, suggesting separate etiologies. Taken together, these results lend further support for the
separation of these constructs. The utility in studying variable manifestations of aggression lies
not just in providing a theoretical framework, but also to understand the antecedents and
110
underlying physiological underpinnings which can have important implications for treatment and
policy considerations.
A second implication of these results is especially pertinent to our understanding of
developmental psychopathology. These results don’t lend themselves simply to tell us how
aggressive behaviors change across development, but also shape our understanding about
important implications regarding different stages at which careful attention is warranted. A
central goal of examining the developmental course of antisocial behaviors at large is to devise
informed interventions and policies which can be used to protect those at highest risk of
developing later antisocial problems. Extant literature on adolescent development has already
focused on the significant biological and psycho-social changes and suggests it may be an
important period from a developmental psychopathology perspective (Spear, 2009). Policy
considerations, especially geared towards high-risk youth could learn deeply from these results.
The Center for Disease Control (2015) statistics estimate youth violence to be the third leading
cause of death for people 15-24 years old and delinquent peer association, drug use and prior
history of violence, amongst others, are recognized as risk factors. Early intervention and
recognition of high risk behaviors could help identify those at highest risk.
Here, it is also important to emphasize the significance of individual differences in
development of aggressive behaviors. Not all individuals experiencing a similar upbringing or
the same genetic risk will develop in the same exact manner. In the present study, significant
individual differences appear in the phenotypic growth models for both reactive and proactive
aggression. Careful examination of proactive aggression at both the categorical as well as
continuous level further suggested that although the endorsement rates for proactive aggression
rise during adolescence, those with initial non-zero levels actually decline in their levels. Thus,
111
there are significant individual differences in how these behaviors develop through adolescence,
suggesting that not every individual follows a single set trajectory. Literature on early predictors
of adult antisocial behavior suggests that studying the individual differences playing a role in
development of these behaviors is extremely important for identifying individuals who are at
high risk. The genetic and environmental underpinning of aggression, as assessed in this study,
also stress on the emphasis of these individual differences. Behavioral genetic studies do not
solely aim at quantifying the heritability of phenotypes, but also lend us the opportunity to look
at what forms of environmental factors (family or individual) are responsible for variation in
these behaviors. It is important to note that not only are non- shared environmental factors
important in explaining the variability in aggression across development but, for reactive
aggression scores, these also lend to the observed stability from childhood to merging adulthood.
Endeavors to examine which environmental factors are most important in maintaining the
stability of these behaviors would benefit the field from identifying targets for potential
interventions at an earlier age. Given that roughly fifty percent or more variation in aggressive
behaviors was explained by non-shared environmental influences emphasizes the importance of
assessing the role of these factors across development.
Longitudinal genetic analyses conducted in this study also reveal important genetic
innovation during the transition to adolescence, furthering our understanding of a
developmentally dynamic genome. The transition stages to adolescence and adulthood are
important from a psycho social perspective and many biological (hormonal) and social
(adjustment, growth, changes in social structure) factors are at play during this time. For both
aggressive behaviors and SCOR, we observe important genetic and environmental influences
coming into play during this period while genetic factors from the childhood also stay
112
significant, again emphasizing on the importance of both early recognition of those at risk and
the provision of protective environments during this period of adjustment and changes. The
results regarding genetic innovation should be observed with the caveat that gene by
environment interactions were not explored in this study, but could have an important
contribution in the development of aggressive behaviors. Genetic factors do place individuals at
a risk for aggressive behaviors but this does not mean that these influences can not be modulated
through the environment. For example, in a sample of Chinese adolescents, Zhang (2015)
reported significant interaction between maternal parenting and polymorphisms in two candidate
genes associated with aggressive behaviors -Monoamine Oxidase and well as catechol-O-
methyltransferase- with reactive aggression but not with proactive aggression. Further
identification and examination of environmental factors which either protect or place individuals
at risk for later aggressive behaviors through interactions with genetic risks would be an
important next step in examining the etiological development of these behaviors.
A third important implication of this study comes from the results on sex differences
during development. Mean differences observed in aggressive behaviors inform us that at large,
males have higher scores on reactive and proactive aggression. This can be seen as an extension
of findings of greater mean level antisocial behaviors in males, at large. However, it is important
to consider that other forms of aggressive behaviors such as relational aggression are more
prevalent in females than in males. Developmentally, it is interesting to note that reactive
aggression follows similar trajectories in males and females but proactive aggression develops
differentially. During adolescence, females tend to display similar levels of proactive aggressive
behaviors but sex differences emerge again during the transition to adulthood, suggesting that
both males and females may be at a higher risk for developing antisocial tendencies during
113
adolescence. Sex differences at the genetic level suggested higher heritability of proactive
aggression in males in comparison to females, emphasizing on the role of environmental risk
factors. The implications of these results are especially important for policy and programs in
place for prevention of youth violence and aggressive behaviors. Differential strategies and
programs for males and females may be needed. Further investigation into what differential risk
factors come into play for males and females would be an important next step.
Sex differences also emerge in the bivariate genetic analyses of SCOR and aggression.
The significant genetic overlap between orienting and male proactive aggression suggests that
males who score consistently higher on proactive aggression may display lower electrodermal
reactivity/ orienting as a potential biomarker. This is also consistent with findings from the
current sample demonstrating that those who score persistently higher on proactive aggression
display reduced fear conditioning (Gao et al., 2015). However, the exact nature of these results is
hard to disentangle. Whether it is resource and attention allocation or under arousal and
fearlessness, can only be determined through precise experimental investigations of these effects.
In conclusion, these studies provide a detailed account of development of reactive and
proactive aggression, electrodermal reactivity and their relationships. The longitudinal genetic
structure of these constructs is delineated and suggests that genetic and non-shared
environmental factors are instrumental in the stability and changes in these constructs. Sex
differences in these longitudinal structures as well as the relationship between aggressive
measures and orienting suggest different developmental etiologies in males and females and have
important implications for both research and policy.

Abstract (if available)

Abstract Heterogeneity in the construct of human aggression has motivated researchers to identify sub forms of aggressive behaviors. Two separable functional forms of aggressive behaviors have been identified and which appear to be distinguishable from childhood to adulthood: reactive and proactive aggression. A significant body of literature distinguishes these two forms based on comorbidity with other psychopathologies, personality correlates and later predicted outcomes. There is, however, limited research on how reactive and proactive aggression develop over time, the genetic underpinnings of these changes and the genetic overlap with autonomic functioning. The goal of this study is to examine longitudinal development of reactive and proactive aggression, the genetic and environmental underpinnings of their development and their relationship to electrodermal reactivity. The data for this study were drawn from a longitudinal study of risk factors of antisocial behavior in a community sample of over 750 pairs of twins. There have been five waves of measurement so far, spanning a period of ten years from late childhood (age 9-10 years) to early adulthood (age 19-20 years). Latent growth curve analysis revealed significant changes over time in reactive and proactive aggression. Significant sex differences emerged in the development of proactive aggression, specifically while examining the endorsement of these behaviors. Genetic results indicated that both genetic stability and innovation were important across development for both reactive and proactive aggression. The behaviors were modestly to moderately heritable from childhood, with new genetic variance emerging at ages 11-13 years, 14-15 years and 16-18 years. Heritability estimates for proactive aggression tend to be lower in females than in males. For electrodermal reactivity, genetic factors explained 46-90% of the variance in scores, suggesting moderate to high heritability. Males and females showed mean differences but no difference in the developmental patterns, at the phenotypic or genetic level. Significant genetic correlation between proactive aggression and orienting response was found for males (rg = -0.43), suggesting shared genetic etiology.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

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

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

PDF

Heritability of schizotypal traits during adolescence

PDF

Multivariate biometric modeling among multiple traits across different raters in child externalizing behavior studies

PDF

Different genetic and environmental structures for the overlap of three antisocial behavior factors with alcohol initiation

PDF

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

PDF

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

PDF

Biometric models of psychopathic traits in adolescence: a comparison of item-level and sum-score approaches

PDF

Sources of stability and change in the trajectory of openness to experience across the lifespan

PDF

How parental psychological control is related to adolescent anger and aggression

PDF

Emotion regulation as a mechanism linking parents’ marital aggression to adolescent behavioral problems: a longitudinal analysis

PDF

Autonomic and behavioral responses of children with autism to auditory stimulation

PDF

You always say that: physical aggression perpetration, linguistic behavior, and conflict intensity in young adult dating couples

PDF

Spouse aggression, depression, and physical health: a multivariate longitudinal study of midlife couples

PDF

Stigma-based peer aggression and social status in middle adolescence: the unique implications of weight-related aggression

PDF

Family aggression exposure and community violence exposure associated with brain volume in late adolescence: a comparison of automated versus manual segmentation

PDF

Nice by nature? A twin study of the development and physiology of prosocial personality

PDF

Emotional reactivity and memory biases in older and younger adults

PDF

Psychophysiological reactivity to stress, psychopathic traits, and developmental trajectories of antisocial behavior: a prospective, longitudinal study

PDF

Social dynamics of intragroup aggression and conflict resolution in wild chimpanzees (Pan troglodytes) at Kanyawara, Kibale National Park, Uganda

Asset Metadata

Creator Dhamija, Devika (author)

Core Title Developmental etiology of reactive and proactive aggression, electrodermal reactivity and their relationships: a twin study

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Psychology

Publication Date 10/07/2016

Defense Date 08/18/2016

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

Tag aggression,electrodermal reactivity,genetic,heritability,OAI-PMH Harvest,proactive,reactive,skin conductance

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Baker, Laura (committee chair), Dawson, Michael (committee member), Dehnghani, Morteza (committee member), Yang, Yaling (committee member)

Creator Email ddhamija@usc.edu,devika87@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-312982

Unique identifier UC11214639

Identifier etd-DhamijaDev-4868.pdf (filename),usctheses-c40-312982 (legacy record id)

Legacy Identifier etd-DhamijaDev-4868.pdf

Dmrecord 312982

Document Type Dissertation

Format application/pdf (imt)

Rights Dhamija, Devika

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