Page 1
http://yvj.sagepub.com/
Justice
Youth Violence and Juvenile
http://yvj.sagepub.com/content/7/3/223
The online version of this article can be found at:
DOI: 10.1177/1541204009333830
2009 7: 223 originally published online 11 May 2009 Youth Violence and Juvenile Justice
Kevin M. Beaver, Brian B. Boutwell, J.C. Barnes and Jonathon A. Cooper
Longitudinal Sample of Twins
The Biosocial Underpinnings to Adolescent Victimization : Results From a
Published by:
http://www.sagepublications.com
On behalf of:
Academy of Criminal Justice Sciences
can be found at:
Youth Violence and Juvenile Justice
Additional services and information for
http://yvj.sagepub.com/cgi/alerts
Email Alerts:
http://yvj.sagepub.com/subscriptions
Subscriptions:
http://www.sagepub.com/journalsReprints.nav
Reprints:
http://www.sagepub.com/journalsPermissions.nav
Permissions:
http://yvj.sagepub.com/content/7/3/223.refs.html
Citations:
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.comDownloaded from
Page 2
The Biosocial Underpinnings to
Adolescent Victimization
Results From a Longitudinal Sample of Twins
Kevin M. Beaver
Brian B. Boutwell
J. C. Barnes
Florida State University
Jonathon A. Cooper
Arizona State University
Behavioral genetic research has consistently revealed that antisocial behaviors are due par-
tially to genetic factors and partially to environmental factors. Even in light of these findings,
researchers have failed to examine the genetic and environmental contributors to adolescent
victimization. The current study addressed this gap in the literature by analyzing a sample of
twin pairs drawn from the National Longitudinal Study of Adolescent Health (Add Health).
The results of the statistical models revealed that genetic factors explained about 40% to 45%
of the variance in adolescent victimization, with the remaining variance attributable to the
nonshared environment. Moreover, additional analyses revealed that 64% of the variance in
repeat victimization was due to genetic factors. The implications that these findings have for
victimization researchers are discussed.
Keywords: delinquency; victimization; adolescence; genetics; environment; twins
O
results of these studies have been remarkably consistent in showing that all phenotypes are
at least partially influenced by genetic factors and some phenotypes are almost completely
genetic in origin (Wright, Tibbetts, & Daigle, 2008). Equally important is that this same
line of research has also revealed that the environment is implicated in the development of
most behaviors and traits. As a direct result, most scientists now recognize that human
ver the past decades, there has been an explosion of empirical-based research examin-
ing the genetic contributors to virtually every imaginable human characteristic. The
Youth Violence and
Juvenile Justice
Volume 7 Number 3
July 2009 223-238
© 2009 SAgE Publications
10.1177/1541204009333830
http://yvj.sagepub.com
hosted at
http://online.sagepub.com
223
Authors’ Note: This research uses data from Add Health, a program project designed by J. Richard Udry, Peter
S. Bearman, and Kathleen Mullan Harris, and funded by a grant P01-HD31921 from the Eunice Kennedy
Shriver National Institute of Child Health and Human Development, with cooperative funding from 17 other
agencies. Special acknowledgment is due to Ronald R. Rindfuss and Barbara Entwisle for assistance in the
original design. Persons interested in obtaining data files from Add Health should contact Add Health, Carolina
Population Center, 123 W. Franklin Street, Chapel Hill, NC 27516-2524 (addhealth@unc.edu). No direct sup-
port was received from grant P01-HD31921 for this analysis. Please address all correspondence to Kevin M.
Beaver, College of Criminology and Criminal Justice, Florida State University, 634 West Call Street,
Tallahassee, FL 32306-1127; e-mail: kbeaver@fsu.edu.
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.comDownloaded from
Page 3
development is contingent on the mutual interdependence of environmental and genetic
factors (Rutter, 2006). Historically, however, criminologists have been relatively slow to
accept the likelihood that antisocial behaviors are scripted, in part, by genetic factors and
instead have narrowly studied only environmental factors (Rowe, 2002). Recently, there
has been a slight change in this tradition as a new perspective in criminology—known as
biosocial criminology—has begun to map out the complex ways in which genetic and
environmental factors work independently and interactively to produce antisocial behav-
iors (Beaver, 2009; Walsh, 2002; Walsh & Beaver, 2009).
given that the biosocial perspective is still relatively new, there remains a host of
unanswered questions surrounding the genetic and environmental basis to antisocial
phenotypes. One of the more salient gaps in the literature is the extent to which adolescent
victimization is affected by genetic factors. To our knowledge, no study has ever explored
this issue; so it is not yet possible to determine with any degree of accuracy the effect that
genetic factors play in the etiology of adolescent victimization. The current study addresses
this omission in the biosocial criminological research by analyzing a sample of twin pairs
drawn from the National Longitudinal Study of Adolescent Health (Add Health) to estimate
the genetic and environmental influences on adolescent victimization.
Behavioral Genetics
Behavioral genetics is a field of study interested in studying both the genetic and
environmental contributors to all types of phenotypes, including crime, delinquency, and
other behaviors of interest to criminologists (Plomin, DeFries, McClearn, & Mcguffin,
2008). Most behavioral genetic studies decompose the variance in a phenotype (e.g.,
criminal involvement) into three different components: a heritability component, a shared
environmental component, and a nonshared environmental component. The heritability
component indexes the proportion of phenotypic variance in the population that is
attributable to genetic factors. A heritability of .30, for example, would indicate that .30 (or
30%) of the variance in the phenotype is due to genetic factors. The variance that is not
accounted for by heritability is then parceled into the shared environmental component and
the nonshared environmental component.
given that the criminological research rarely separates the shared environment and the
nonshared environment, the distinction between these two types of environments requires
additional clarification (Beaver, 2008). Shared environments are those environments that
are the same between siblings from the same household. For example, socioeconomic
status is typically considered to be a shared environment because all siblings experience the
same socioeconomic status. Shared environmental factors work to make siblings more
similar to each other in terms of behaviors, personality traits, and any other measurable
phenotype. To see why this is the case, consider the shared environmental factor of being
reared in poverty. Poverty is a shared environment because if a family is living in poverty,
then all siblings in that family would also be living in poverty. And, if poverty has an effect
on human development, then it should affect all siblings equally, thereby making them
similar to each other.
224 Youth Violence and Juvenile Justice
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.com Downloaded from
Page 4
Nonshared environments, in contrast, are any environments that are not the same
between siblings. Peer groups, for example, are often considered nonshared environments
because siblings frequently have very distinct friendship networks. Where shared
environments make siblings similar to each other, nonshared environments make siblings
dissimilar from each other (Plomin & Daniels, 1987). Logically, this makes sense because
if an environment is influential and if that environment is experienced by one sibling but
not another, then the two siblings should be differentially affected by the environment. For
instance, all else equal, a sibling who is embedded in an antisocial peer group should be
more delinquent than their sibling who is embedded in a prosocial peer group. For this
reason, nonshared environments hold the potential to explain why siblings often mature
into very different people (Dunn & Plomin, 1990).
All phenotypes, including antisocial phenotypes, are due to some combination of genetic
factors (i.e., heritability), shared environmental factors, and nonshared environmental
factors. One of the main research questions that behavioral geneticists are interested in
answering is the proportion of phenotypic variance explained by each of these three
variance components. To do so, behavioral geneticists typically analyze samples of kinship
pairs, especially twin pairs. By using twin pairs, it is possible to estimate genetic and
environmental effects on a phenotype by comparing the similarity of dizygotic (DZ) twins
to the similarity of monozygotic (MZ) twins. DZ twins share, on average, 50% of their
DNA, whereas MZ twins share 100% of their DNA. Both DZ and MZ twins, however,
share the same environments. As a result, the only reason that MZ twins should be more
similar to each other than DZ twins is because they share more genetic material. If MZ
twins are no more similar to each other than DZ twins, then genetic factors are unlikely to
be important to the phenotype being studied. By using this logic, behavioral geneticists are
able to gain relatively accurate estimates of the proportion of phenotypic variance due to
genetic factors, shared environmental factors, and nonshared environmental factors (Plomin
et al., 2008).
Mounds on mounds of behavioral genetic research have been conducted to examine the
genetic and environmental underpinnings to virtually every imaginable phenotype. Although
the precise estimates vary from study to study, from sample to sample, and from phenotype
to phenotype, the overwhelming majority of research has revealed that genetic factors
explain approximately 40% to 60% of phenotypic variance, shared environmental factors
explain approximately 0% to 10% of phenotypic variance, and nonshared environmental
factors explain approximately 40% to 60% of phenotypic variance. The same pattern of
results holds for antisocial phenotypes as well. For example, three meta-analyses have been
conducted to examine the heritability of antisocial behaviors (Mason & Frick, 1994; Miles
& Carey, 1997; Rhee & Waldman, 2002). The results of these meta-analyses that span
hundreds of kinship pairs have been remarkably consistent in showing that about 50% of
the variance is attributable to genetic factors, with about 0% to 20% of the variance due to
the shared environment, and the remaining variance due to the nonshared environment
(Moffitt, 2005). Taken together, these findings cast serious doubt on criminological theories
that ignore or downplay the significance of genetic factors in the etiology of antisocial
behaviors (Walsh, 2002).
Beaver et al. / Biosocial Underpinnings To Adolescent Victimization 225
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.com Downloaded from
Page 5
226 Youth Violence and Juvenile Justice
Genetics Influences on Adolescent Victimization
The results culled from the behavioral genetic research thus provide strong and
compelling evidence that antisocial phenotypes are partially the result of genetic factors.
Still, much remains unknown about the genetic origins to many different antisocial
phenotypes. Of particular interest is the extent to which genetic factors play a role in
adolescent victimization. Although no research has ever examined this issue, there are at
least three converging lines of literature to suggest that adolescent victimization should be
influenced by genetic and environmental factors (Beaver et al., 2007).
First, adolescents who engage in acts of delinquency—especially in acts of violent
delinquency—are much more likely to be victimized when compared to youths who are
nondelinquents (Esbensen & Huizinga, 1991; Lauritsen, Sampson, & Laub, 1991; Wright
& Decker, 1994, 1997). This comorbidity necessarily raises the possibility that the same
factors that promote delinquency may also promote victimization. Since genetic factors are
known to affect delinquent involvement (Beaver, 2009; Moffitt, 2005), it stands to reason
that genetic factors may also affect the odds of being victimized. Only one study has
examined this possibility, however. Using a sample of twin pairs, Hines and Saudino
(2004) examined the genetic and environmental effects on intimate partner aggression. The
results of their model-fitting analyses revealed that genetic factors explained 25% of the
variance in the receipt of psychological aggression by an intimate partner and 15% of the
variance in the receipt of physical aggression by an intimate partner. Moreover, their
research also indicated that the genetic factors that influenced the use of aggression also
were implicated in the receipt of aggression—that is, there was a shared genetic pathway
to the use and receipt of aggression between intimate partners. Although only suggestive,
the findings from this study point to the likelihood that genetic factors may also affect
adolescent victimization.
Second, crime victims are not always innocent bystanders who are targeted at random.
Criminologists have long recognized that many victims are active participants in constructing
their victimization experiences (Wolfgang, 1957). To illustrate, homicides often occur only
after a heated exchange between two people, which ultimately culminates in one person
murdering the other. In this case, the murder victim was not targeted at random; instead they
were actively involved in the transactional process that eventually led to their murder
(Luckenbill, 1977; Wolfgang, 1957). Criminologists, however, have been slow to investigate
how, why, and under what conditions people actively construct their victimization experiences.
This omission in the criminological literature can easily be filled by borrowing the concept
of an evocative gene × environment correlation from the behavioral genetic literature.
Evocative gene × environment correlations refer to the effects that genetic factors have
in the elicitation of certain responses from the environment (Walsh, 2002). An adolescent
with a genetic predisposition for a bad temper, for example, is likely to evoke negative
reactions from their parents, their teachers, and their peers. The reason for these negative
environmental reactions is due, in part, to genetic tendencies. It is possible that a similar
process is at work with victimization, where people with certain genetic factors may be
more or less likely to evoke harsh negative reactions from their environment. Some of these
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.com Downloaded from
Page 6
Beaver et al. / Biosocial Underpinnings To Adolescent Victimization 227
negative reactions might include reactions that turn violent. If this is the case, then genetic
factors might be partially responsible for why certain youths are victimized.
A third reason to suspect that adolescent victimization may be due to genetic factors is
because recent research has revealed that individual personality traits are implicated in the
odds of being victimized. For example, Schreck and his colleagues have conducted a
number of studies revealing that persons with relatively low levels of self-control are at risk
for being victimized (Schreck, 1999; Schreck, Wright, & Miller, 2002; Stewart, Elifson, &
Sterk, 2004). This is a particularly important finding because research has revealed that all
personality traits, including self-control, are influenced by genetic factors (Wright &
Beaver, 2005; Wright, Beaver, DeLisi, & Vaughn, 2008). If this is the case, then genetic
factors may affect victimization by affecting levels of self-control.
The above discussion provided three different reasons for why it is likely that adolescent
victimization is due partially to genetic factors. But it is also important to note that genetic
factors may be able to shed some light on the phenomenon of repeat victimization.
Victimization researchers have consistently found that persons who are victimized once are
at risk for being victimized again in the future (Lauritsen & Quinet, 1995). Still, there
remains much confusion and debate over the causes of repeat victimization (Ellingworth,
Farrell, & Pease, 1995; Osborn, Ellingworth, Hope, & Trickett, 1996). One potential
explanation for repeat victimization is that the genetic factors that promote victimization at
one point in time are the same genetic factors that promote victimization at a later point in
time. Unfortunately, no empirical research has ever examined this possibility.
The Current Study
The purpose of the current study is threefold. First, we estimate a series of statistical
models to decompose the variance in adolescent victimization into a genetic component, a
shared environmental component, and a nonshared environmental component. Second, we
examine whether low self-control, delinquent peers, and delinquent involvement are significant
sources of nonshared variance in adolescent victimization. Third, we examine the extent to
which genetic and environmental factors are implicated in repeat victimization.
Method
Data
The current study employs data drawn from the National Longitudinal Study of
Adolescent Health (Add Health; Udry, 2003). The Add Health is a longitudinal and
nationally representative sample of American youths enrolled in grades 7 through 12
during the 1994-1995 school year. Three waves of data have been collected thus far. The
first round of interviews was conducted in 1994 when approximately 90,000 youths
completed self-report surveys at school (i.e., the Wave 1 in-school interview). Adolescents
were asked questions about their family life, their social life, and their involvement in
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.com Downloaded from
Page 7
228 Youth Violence and Juvenile Justice
different activities. To gain more detailed information about some of the respondents,
20,745 adolescents, along with 17,700 of their primary caregivers (usually their mother),
were reinterviewed at their home (i.e., the Wave 1 in-home interview). A wide range of
questions were asked during the Wave 1 in-home interviews, including questions concerning
delinquent involvement, drug and alcohol use, romantic relationships, and personal
victimization (Harris et al., 2003).
Approximately 1.5 years after the Wave 1 in-home interviews were administered, the
Wave 2 in-home interviews were completed. As most of the respondents were still
adolescents, most of the questions asked at Wave 1 were asked once again at Wave 2. For
example, youths were asked about their delinquent behaviors, their personality traits, and
their social relationships. Altogether, 14,738 respondents participated in the Wave 2
in-home interviews. Then, during 2001-2002, the third wave of data was collected. Since
the respondents were young adults at Wave 3, many of the questions that were asked
previously were removed from the survey instruments and replaced with newer, more age-
appropriate questions. Respondents, for instance, were asked about their marital status and
their employment and educational histories. In total, 15,197 young adults were reinterviewed
at Wave 3 (Harris et al., 2003).
Nested within the Add Health data are pairs of siblings, including pairs of twins. During
Wave 1 interviews, youths were asked whether they had a twin or a sibling living with
them. If their sibling was between the ages of 11 and 20 years, then they were added to the
sample. Altogether, 3,139 sibling pairs were included in the sample at Wave 1 and,
importantly, the demographic characteristics of the sibling sample do not appear to deviate
from the demographic characteristics of the nationally representative sample (Jacobson &
Rowe, 1999). For reasons to be discussed shortly, the analyses were restricted to twins only
(dizygotic same-sex twin pairs: n = 247; monozygotic twin pairs: n = 289).
Measures
Victimization. A series of questions pertaining to the frequency with which adolescents
were victimized were included in the Add Health data. Specifically, during Wave 1 inter-
views, respondents were asked how many times in the past 12 months they had had a knife
or gun pulled on them, had been shot, had been cut or stabbed, and had been jumped.
Adolescents were also asked how many times they had seen someone shot or stabbed in the
past year. This latter question indexes exposure to violence and also captures variation in
the odds of being victimized in the future. All five of the items were coded such that 0 =
never, 1 = once, and 2 = more than once. Responses to each of the questions were summed
together to create the Wave 1 adolescent victimization scale (α = .66). The same questions
were asked during Wave 2 interviews, and thus the Wave 2 adolescent victimization scale
is a duplicate of the Wave 1 scale (α = .64). These scales are similar to the ones used by
previous researchers analyzing adolescent victimization in the Add Health data (Beaver
et al., 2007; Haynie & Piquero, 2006). Table 1 contains the descriptive statistics for the
victimization scales and all of the other variables/scales used in the analyses.
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011 yvj.sagepub.comDownloaded from
Page 8
Beaver et al. / Biosocial Underpinnings To Adolescent Victimization 229
Low self-control. A rich line of empirical research has revealed a relatively strong and
consistent association between levels of self-control and criminal and delinquent involve-
ment (Pratt & Cullen, 2000). There is also emerging evidence to indicate that levels of
self-control are also associated with the odds of being victimized (Schreck, 1999; Schreck
et al., 2002; Stewart et al., 2004). To take this finding into account, we developed a
Wave 1 low self-control scale and a Wave 2 low self-control scale. During Wave 1 inter-
views, respondents were asked to indicate their ability to keep their mind focused, to con-
centrate, and to get along with their teachers among others. A total of five items were used
to construct the Wave 1 low self-control scale (α = .65). The same items were used to con-
struct the Wave 2 low self-control scale (α = .62). Higher values on both scales indicated
lower levels of self-control. These low self-control scales have been employed by prior
researchers analyzing the Add Health data (Beaver, 2008; Boutwell & Beaver, 2008;
Perrone, Sullivan, Pratt, & Margaryan, 2004).
Delinquent peers. Research has revealed that exposure to delinquent peers is a risk factor
for adolescent victimization (Haynie & Piquero, 2006; Schreck et al., 2004; Schreck &
Fisher, 2004). As a result, we created a Wave 1 delinquent peers scale and a Wave 2 delin-
quent peers scale, both of which have been used previously (Beaver, 2008; Beaver &
Wright, 2005; Bellair, Roscigno, & McNulty, 2003). During Wave 1 interviews, respondents
were asked to indicate how many of their three best friends smoked at least one cigarette
each day, smoked pot more than once per month, and drank alcohol at least once per month.
Responses to these questions were summed together to form the Wave 1 delinquent peers
scale (α = .76). The Wave 2 delinquent peers scale was created from the same three items
(α = .77). For both scales, higher values indicated greater contact with delinquent peers.
Delinquent involvement. Involvement in crime and delinquency is associated with
increased odds of being victimized (Esbensen & Huizinga, 1991; Lauritsen et al., 1991).
As a result, we follow the lead of prior researchers examining the correlates to victimiza-
tion in the Add Health data (Beaver et al., 2007; Haynie & Piquero, 2006) and include a
Wave 1 delinquency scale and a Wave 2 delinquency scale. During Wave 1 interviews,
respondents were asked to indicate the number of times in the past year they had engaged
Table 1
Descriptive Statistics for the Add Health Study Variables
Variable M SD Min-Max
Victimization at Wave 1
Victimization at Wave 2
Low self-control at Wave 1
Low self-control at Wave 2
Delinquent peers at Wave 1
Delinquent peers at Wave 2
Delinquent involvement at Wave 1
Delinquent involvement at Wave 2
0.87
0.30
6.31
5.86
2.55
2.87
4.21
2.48
1.52
0.86
3.11
2.92
2.71
2.81
5.28
3.52
0-12
0-8
1-19
1-19
0-9
0-9
0-45
0-22
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.com Downloaded from
Page 9
230 Youth Violence and Juvenile Justice
in 15 delinquent acts, including selling drugs, damaging property, and stealing something
worth more than 50 dollars. Responses to these items were then summed together to create
the Wave 1 delinquent involvement scale (α = .84). Similarly, during Wave 2 interviews,
respondents were asked to indicate how many times in the past year they had engaged in
14 acts of delinquency. The responses were summed together to form the Wave 2 delin-
quent involvement scale (α = .77). Both of these scales have been used by prior researchers
(Beaver, 2008; Beaver & Wright, 2005), and for both scales, higher scores corresponded to
more delinquent involvement.
Plan of Analysis
To examine the genetic and environmental effects on adolescent victimization, DeFries-
Fulker (DF) analysis is employed (DeFries & Fulker, 1985). DF analysis is a regression-
based statistical technique that can be applied to samples of kinship pairs (e.g., twins) to
estimate the proportion of variance in an outcome measure (e.g., victimization) that is
accounted for by genetic factors, shared environmental factors, and nonshared environmen-
tal factors. The baseline DF formula takes the following form:
K1 = b0 + b1K2 + b2R + b3(R * K2) + e (1)
where K1 is the score on the outcome measure for one twin, K2 is the score on the same
outcome measure for the cotwin, R captures genetic similarity (R = 1.0 for MZ twins, and
R = .5 for DZ twins), and R * K2 is an interaction term created by multiplying R and K2. In
Equation 1, b0 is the constant, b1 is the proportion of variance in K1 that is explained by the
shared environment, b2 is not typically reported in DF analysis, b3 is the proportion of vari-
ance in K1 that is explained by genetic factors (i.e., heritability), and e is the proportion of
variance that is due to the nonshared environment and error.
Recently Rodgers and Kohler (2005) have made a number of modifications to the DF
equation presented above. This new DF formula takes the following form:
K1 = b0 + b1(K2 – Km) + b2(R * (K2 – Km)) + e (2)
where K1 is still the score on the outcome measure for one twin, K2 is still the value on the
same outcome measure for the cotwin, and R still captures genetic similarity. The modified
DF equation includes an additional term, Km, that was not included in Equation 1. Km is the
mean value for K2. In Equation 2, then, K2 is being mean centered, whereas in Equation 1,
K2 was not mean centered. The main effect of R has also been removed from Equation 2.
The interpretation of the slopes, however, is unchanged: b1 is the proportion of variance in
K1 that is explained by the shared environment, and b2 is the proportion of variance in K1
explained by genetic factors. We use equation 2 to estimate the genetic, shared environmen-
tal, and nonshared environmental effects on the Wave 1 adolescent victimization scale and
the Wave 2 adolescent victimization scale.
It is also possible to estimate the effects of specific nonshared environments when using
DF analysis (Rodgers, Rowe, & Li, 1994) by modifying Equation 2 slightly. This newly
modified DF equation takes the following form:
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.comDownloaded from
Page 10
Beaver et al. / Biosocial Underpinnings To Adolescent Victimization 231
K1 = b0 + b1(K2 – Km) + b2(R * (K2 – Km)) + b4(ENVDIF) + e (3)
The only difference between Equations 2 and 3 is that Equation 3 introduces an additional
term, ENVDIF. ENVDIF is a variable that measures the difference between twins on a
specific measure. To illustrate, one twin’s score on the Wave 1 low self-control scale could
be subtracted from their cotwin’s score on the Wave 1 low self-control scale. The new vari-
able, ENVDIF, would capture the difference between twins on the low self-control scale.
ENVDIF is a specific nonshared environment that estimates whether differences between
twins on the independent variables can explain differences between twins on the outcome
measure (e.g., victimization). We estimate a series of DF models using Equation 3 to deter-
mine whether differences in low self-control, differences in delinquent peers, and differ-
ences in delinquent involvement predict differences in adolescent victimization.
The equations presented thus far can, for the most part, be thought of as univariate equa-
tions because genetic and environmental effects are being estimated for a single variable
(e.g., victimization). Sometimes, however, researchers are interested in examining more
than one variable. For example, suppose a researcher was interested in the association
between two measures. Suppose further that this researcher found a significant correlation
between these two measures, perhaps r = .40. Now they might be concerned with determin-
ing the percentage of that correlation that is due to genetic factors, shared environmental
factors, and nonshared environmental factors. This type of question requires bivariate or
multivariate genetic modeling techniques, not univariate ones. Typically, bivariate genetic
modeling is carried out in structural equation modeling programs, where Cholesky decom-
position models or correlated factors models are estimated (Neale & Cardon, 1992).
However, Rodgers, Kohler, Kyvik, and Christensen (2001) showed that the DF model
could be transformed into a cross-variable DF model to estimate bivariate analyses.
Specifically, the cross-variable DF model takes the following form:
K1wave2 = b0 + b1(K2wave1 – Km) + b2(R * (K2wave1 – Km)) + e (4)
where K1wave2 is the score on the Wave 2 outcome measure (e.g., the Wave 2 victimization
scale) for one twin, K2wave1 is the score on the Wave 1 measure (e.g., the Wave 1 victimiza-
tion scale) for the cotwin, and R is still the coefficient of genetic relatedness.1 Note that in
Equation 4, K2wave1 is being mean centered (i.e., the mean for Wave 1). The interpretation
of the coefficients in Equation 4 varies from those of the previous equations. In the cross-
variable DF model presented in Equation 4, b1 is the proportion of the correlation between
the Wave 1 and Wave 2 measures (e.g., the Wave 1 and Wave 2 victimization scales) that
is due to shared environmental effects, while b2 is the proportion of the correlation between
the Wave 1 and Wave 2 measures (e.g., the Wave 1 and Wave 2 victimization scales) that
is due to common genetic factors. The variance in the correlation that is unexplained by b1
and b2 is due to nonshared environmental factors plus error.
With DF analysis, it is allowable to double enter twins, where each twin acts once as the
dependent variable and once as the independent variable. When using double-entry, the
standard errors must be corrected to take into account the nonindependence in observations.
We opted to randomly select one twin from each twin pair to be used as the dependent
variable, and their cotwin was thus used as the independent variable. In this way, the twin
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.com Downloaded from
Page 11
232 Youth Violence and Juvenile Justice
pair is the unit of analysis and all of the observations are independent of each other. The
robustness of the models was examined by reestimating the models using double-entry, and
the substantive results remained the same. All of the DF models were estimated using the
structural equation modeling program, AMOS (Beaver, DeLisi, Vaughn, Wright, &
Boutwell, 2008). AMOS uses a full-information maximum likelihood (FIML) imputation
algorithm to estimate missing values, which allows all of the models to be calculated using
the full sample size of N = 536 twin pairs.
Results
The analysis for this study began by estimating a DF model using the Wave 1
victimization scale as the outcome measure. To facilitate interpretation of the coefficients,
the coefficients for b1 and b2 (presented in Equation 2) have been labeled shared environment
and heritability, respectively. Table 2 displays the results of these models. As can be seen
in Model 1, the heritability of the Wave 1 victimization scale was statistically significant
with a value of .45. This finding can be interpreted to mean that about 45% of the variance
in the Wave 1 victimization scale was due to genetic factors. Also of importance is that the
shared environmental effect was nonsignficant (b = –.05, p > .05), meaning that 0% of the
variance in the Wave 1 victimization scale was explained by shared environmental factors.2
The remaining 55% of variance that was unexplained by heritability was due to the
nonshared environment and error.
Since a significant amount of variance in the Wave 1 victimization scale was explained
by the nonshared environment, the next step in the analysis was to introduce the measures
that tapped sources of nonshared variance. Recall that the nonshared variables were created
by subtracting one twin’s score from their cotwin’s score. These differences scores were
then introduced into the models (estimated by using Equation 3) to determine whether they
were predictive of the Wave 1 victimization scale. All of the nonshared variables in Table
2 were measured at Wave 1, and thus the effects were contemporaneous.
Table 2
DF Analysis of Adolescent Victimization at Wave 1 (N = 536 Twin Pairs)
Model 1 Model 2 Model 3 Model 4 Model 5
b SE b SE b SE b SE b SE
DF analysis components
Heritability
Shared environment
Sources of nonshared variance
Low self-control
Delinquent peers
Delinquent involvement
.45*
–.05
.13
.10
.44*
–.04
.12
.10
.43*
–.02
.12
.10
.47*
.02
.12
.09
.45*
.04
.12
.09
.03*
.01
.10*
.02
.10*
.01
.01
.04*
.09*
.01
.02
.01
NOTE: Unstandardized coefficients presented.
*p ≤ .05, two-tailed tests.
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.com Downloaded from
Page 12
Beaver et al. / Biosocial Underpinnings To Adolescent Victimization 233
The effects of the nonshared variables on victimization are presented in Models 2
through 5. Model 2 shows that low self-control was a positive and statistically significant
predictor of victimization. Not surprisingly, this finding can be interpreted to mean that the
twin (from each twin pair) who scored higher on the low self-control scale also scored
higher on the victimization scale. Similarly, in Model 3, the effect of delinquent peers was
also positive and statistically significant, meaning that the twin (from each twin pair) who
reported a greater number of delinquent peers also reported more incidents of victimization.
Model 4 shows the same effect for delinquent involvement, where the twin (from each twin
pair) who was more involved in delinquency also was victimized more frequently. Last, in
Model 5, all three of the nonshared variables were entered into the equation simultaneously.
In this model, delinquent peers and delinquent involvement retained a significant association
with victimization, but the effect of low self-control dropped from statistical significance.
The results thus far indicate that about 45% of the variance in adolescent victimization is
due to genetic factors, and the remaining 55% of variance is attributable to the nonshared
environment. Remember that these estimates were garnered by using data drawn from Wave
1 interviews. To examine the robustness of these findings, we replicated the same models
using Wave 2 measures. Table 3 contains the results of these models, which are almost
identical to the ones reported with the Wave 1 measures. Model 1 shows, for example, that the
heritability of adolescent victimization was .42, with the remaining variance accounted for by
the nonshared environment. Once again the shared environment had no effect on victimization.
The remaining models in Table 3 examined whether the three sources of nonshared variance—
low self-control, delinquent peers, and delinquent involvement—were associated with
adolescent victimization. As can be seen in Models 2 through 4, the twin who had lower levels
of self-control, who had more delinquent peers, and who was more involved in delinquency
was the same twin who also scored higher on the Wave 2 adolescent victimization scale. When
all three nonshared sources of variance were entered into the equation simultaneously, the
effect of low self-control dissipated from statistical significance, but the effects of delinquent
peers and delinquent involvement continued to remain statistically significant.
Table 3
DF Analysis of Adolescent Victimization at Wave 2 (N = 536 Twin Pairs)
Model 1 Model 2 Model 3 Model 4 Model 5
b SE b SE b SE b SE b SE
DF analysis components
Heritability
Shared environment
Sources of nonshared variance
Low self-control
Delinquent peers
Delinquent involvement
.42*
–.03
.14
.12
.43*
–.03
.14
.12
.39*
.01
.14
.12
.43*
.01
.13
.12
.41*
.03
.13
.12
.03*
.01
.05*
.01
.06*
.01
.01
.03*
.05*
.01
.01
.01
NOTE: Unstandardized coefficients presented.
*p ≤ .05, two-tailed tests.
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011 yvj.sagepub.comDownloaded from
Page 13
234 Youth Violence and Juvenile Justice
In the last set of models, we examined repeat victimization by estimating the stability
between victimization at Wave 1 and victimization at Wave 2. We then calculated a cross-
variable DF analysis (see Equation 4 presented above) to estimate the percentage of the
correlation that was due to genetics, shared environmental factors, and nonshared
environmental factors. Table 4 contains the results for both of these analyses. As can be seen
in Model 1, there was a strong degree of wave-to-wave stability in victimization, indicating
the tendency for some respondents to consistently score high (low) on both victimization
scales (β = .52). Model 2 displays the results of the cross-variable DF analysis, which shows
that 64% of the wave-to-wave stability in victimization was due to genetic factors, 0% was
due to the shared environment, and 36% was due to the nonshared environment.
Discussion
Rates of adolescent victimization are exceedingly high, with anywhere between 30%
and 50% of all youths being victimized each year (Christiansen & Evans, 2005; Esbensen
& Huizinga, 1991; Menard, 2002; Snyder & Sickmund, 1999). As a result, researchers have
devoted a considerable amount of time to identifying the various factors that are linked to
the odds of being victimized. For the most part, studies have focused on how environmental
factors, such as daily life routines, may relate to victimization (Felson, 2002, 2006).
Although this line of research has provided a considerable amount of information about the
correlates to victimization, much still remains unknown about the various factors that may
promote youthful victimization. In particular, researchers have been slow to examine the
influence that genetic factors play in the genesis of victimization experiences (Beaver et al.,
2007). The current study addressed this gap in the literature and estimated the genetic and
environmental effects on adolescent victimization.
Analysis of twin pairs drawn from the Add Health revealed four broad findings. First,
the results of the DF models revealed that genetic factors explain somewhere between 40%
Table 4
Cross-Variable DF Analysis of Adolescent Victimization at Waves 1 and 2
(N = 536 Twin Pairs)
Model 1 Model 2
b SE
β b SE
Stability estimate
Victimization at Wave 1
DF analysis components
Heritability
Shared environment
.29* .02 .52 — —
—
—
—
—
—
—
.64*
–.17
.18
.15
NOTE: Standardized and unstandardized coefficients presented in Model 1; Unstandardized coefficients pre-
sented in Model 2.
*p ≤ .05, two-tailed tests.
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.com Downloaded from
Page 14
Beaver et al. / Biosocial Underpinnings To Adolescent Victimization 235
and 45% of the variance in adolescent victimization. These estimates are in line with a
wealth of behavioral genetic research indicating that about 50% of the variance in antisocial
phenotypes is due to genetic factors (Mason & Frick, 1994; Miles & Carey, 1997; Moffitt,
2005; Rhee & Waldman, 2002). Hopefully, these findings challenge victimization researchers
to take seriously the possibility that adolescent victimization is driven, in part, by genetic
factors (Beaver et al., 2007).
Second, the shared environment explained none of the variance in the adolescent
victimization scales. At first glance, this finding may seem somewhat counterintuitive—
after all, shared environments are often thought to be a dominant force in shaping an
adolescent (Harris, 1998). It should be noted, however, that behavioral genetic research has
consistently found that very little—if any—phenotypic variance is explained by the shared
environment, a finding that is especially true for antisocial phenotypes (Harris, 1998;
Rowe, 1994; Wright & Beaver, 2005; Wright, Beaver, DeLisi, and Vaughn, 2008). To the
extent that these findings are replicated, victimization researchers should cautiously move
away from searching for the shared environmental factors that affect youthful
victimization.
Third, the nonshared environment accounted for about 55% to 60% of the variance in
adolescent victimization. Recall that with DF analysis, it is possible to introduce measured
nonshared sources of variance. Using this technique, the DF models revealed that two
nonshared sources of variance—delinquent peers and delinquent involvement—were
associated with adolescent victimization. Stated differently, the twin from each twin pair
who had more delinquent peers or who engaged in more delinquency was the same twin
who reported more victimization experiences. Interestingly, the effect that differences in
low self-control had on victimization dropped from statistical significance once delinquent
peers and delinquent involvement were included as control variables. What this can be
interpreted to mean is that the effect of low self-control on victimization was completely
mediated by delinquent peers and delinquent involvement. In light of these findings,
victimization researchers should begin to explore the ways in which various nonshared
environmental factors may differentially lead to adolescent victimization.
Fourth, the DF models revealed that genetic factors accounted for 64% of the variance
in repeat victimization. In line with the previous models, none of the variance in repeat
victimization was due to shared environmental factors, and 36% of the variance in repeat
victimization was due to nonshared environmental factors. This finding is particularly
intriguing and sheds new light on why the same people are victimized repeatedly over time.
Unfortunately, researchers have, in general, failed to take into account the possibility that
repeat victimization is due to enduring genetic propensities. The results from the current
study provide some evidence that searching for the causes of repeat victimization virtually
requires taking into account genetic factors.
With these findings in mind, it is important to point out the main limitations of the
current study. To begin with, although the Add Health data were designed to be a nationally
representative sample, the use of twin pairs obviously calls into question the generalizability
of the findings. It could be the case that the findings reported here may not extend to
samples of singletons. Although it is not possible to discount this possibility, we should
note that a recent analysis examining potential differences between twins and nontwins
drawn from the Add Health revealed no substantive differences (Beaver, 2008). Even so,
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.comDownloaded from
Page 15
236 Youth Violence and Juvenile Justice
future researchers should replicate this study as a way of determining the robustness and
generalizability of the results.
Another limitation of this study is that only adolescent victimization was examined,
leaving open the possibility that victimization in adulthood is due to a different arrangement
of genetic and environmental factors. Until this possibility is explored, we caution against
making any conclusions about the origins of adulthood victimization. In addition, the
victimization scales only included a limited number of items. Perhaps if a broader array of
victimization experiences had been measured, the heritability and environmental estimates
would change. These issues should be confronted by future researchers examining the
correlates to adolescent victimization.
In light of these limitations, we should note that to our knowledge, this is the first study
to estimate the genetic and environmental influences on adolescent victimization. The
findings reinforce the importance of employing a biosocial perspective when examining all
types of phenotypes, including victimization. At the same time, if the results of the current
study are to be believed, then victimization researchers will need to reexamine their
theories and research and begin to integrate genetic factors. Although this will no doubt be
a challenging feat, it most likely will provide a much more complete explanation to
adolescent victimization.
Notes
1. Note that prior to estimating Equation 4, the Wave 1 and Wave 2 measures must be standardized for the
values of the coefficients to be interpretable.
2. If the shared environmental effect is nonsignficant, then it is often dropped from the model, and the equa-
tion is recalculated. We followed this procedure in this model as well as the other models where the shared
environmental effect was nonsignficant. Overall, the results remained very similar, and only the equations
where the shared environmental effect (b1) was retained were presented.
References
Beaver, K. M. (2008). Nonshared environmental influences on adolescent delinquent involvement and adult
criminal behavior. Criminology, 46, 341-370.
Beaver, K. M. (2009). Biosocial criminology: A primer. Dubuque, IA: Kendall/Hunt.
Beaver, K. M., DeLisi, M., Vaughn, M. g., Wright, J. P., & Boutwell, B. B. (2008). The relationship between
self-control and language: Evidence of a shared etiological pathway. Criminology, 46, 939-970.
Beaver, K. M., & Wright, J. P. (2005). Biosocial development and delinquent involvement. Youth Violence and
Juvenile Justice, 3, 168-192.
Beaver, K. M., Wright, J. P., DeLisi, M., Daigle, L. E., Swatt, M. L., & gibson, C. L. (2007). Evidence of a
gene × environment interaction in the creation of victimization: Results from a longitudinal sample of ado-
lescents. International Journal of Offender Therapy and Comparative Criminology, 51, 620-645.
Bellair, P. E., Roscigno, V. J., & McNulty, T. L. (2003). Linking local labor market opportunity to violent ado-
lescent delinquency. Journal of Research in Crime and Delinquency, 40, 6-33.
Boutwell, B. B., & Beaver, K. M. (2008). A biosocial explanation of delinquency abstention. Criminal
Behaviour and Mental Health, 18, 59-74.
Christiansen, E. J., & Evans, W. P. (2005). Adolescent victimization: Testing models of resiliency by gender.
Journal of Early Adolescence, 25, 298-316.
at SAM HOUSTON STATE UNIV LIBRAR on January 20, 2011yvj.sagepub.com Downloaded from
Download full-text 
