Journal of Interpersonal Violence
2019, Vol. 34(6) 1261 –1286
© The Author(s) 2016
Article reuse guidelines:
The Cycle of Violence
and Future Violence
Kevin A. Wright,1 Jillian J. Turanovic,2
Eryn N. O’Neal,3 Stephanie J. Morse,1
and Evan T. Booth4
The individual and social protective factors that help break the cycle of
violence are examined. Specifically, this study investigates (a) the individual
and social protective factors that reduce violent offending among previously
victimized children, and (b) whether certain protective factors are more
or less important depending on the type and frequency of childhood
victimization experienced. Data on young adults from Wave III of the
National Longitudinal Study of Adolescent to Adult Health are used (N =
13,116). Negative binomial regression models are estimated to examine
the protective factors that promote resiliency to violent offending among
individuals who reported being physically and sexually victimized as children.
Results indicate that a number of individual and social protective factors
reduce violent offending in young adulthood. With a few exceptions,
these factors are specific to the type, frequency, and comorbidity of abuse
experienced. The results suggest a number of promising approaches to
break the cycle of violence among previously victimized children. Future
1Arizona State University, Phoenix, AZ, USA
2Florida State University, Tallahassee, FL, USA
3Sam Houston State University, Huntsville, TX, USA
4University of Denver, CO, USA
Kevin A. Wright, School of Criminology and Criminal Justice, Arizona State University, 411 N.
Central Ave. Ste. 600, Phoenix, AZ 85004, USA.
651090JIVXXX10.1177/0886260516651090Journal of Interpersonal ViolenceWright et al.
1262 Journal of Interpersonal Violence 34(6)
research should move beyond explaining the cycle of violence to examine
how the cycle may be broken.
treatment/intervention, child abuse, physical abuse, sexual abuse,
intergenerational transmission of trauma
Over the last 25 years, the cycle of violence hypothesis (Widom, 1989a;
1989b) has provided a solid foundation for studying the long-term crimino-
genic effects of childhood victimization. Research within this tradition has
established two major components to the thesis. First, victims of child abuse
are at an increased risk of perpetrating violence in adolescence and young
adulthood (Currie & Tekin, 2011; Maxfield & Widom, 1996; Watts &
McNulty, 2013), and second, not all victims of child abuse go on to perpetrate
future violence (DuMont, Widom, & Czaja, 2007; Jaffee, Caspi, Moffitt,
Polo-Tomas, & Taylor, 2007). Comparatively speaking, childhood victims
that engage in violence have received far more scholarly attention than those
who do not. Indeed, the majority of research in this area focuses on identify-
ing risk factors that increase the likelihood of violence among abused chil-
dren—factors such as biased and deficient social-information-processing
patterns (Dodge, Bates, & Pettit, 1990), endorsement of antisocial attitudes
and associations with deviant peers (Herrenkohl, Huang, Tajima, & Whitney,
2003), and genetic susceptibility to maltreatment-induced changes to neu-
rotransmitter systems (Caspi et al., 2002).
Much less understood, however, are the protective factors that explain
why some abused children can avoid completing the cycle of violence
(McGloin & Widom, 2001). And existing research suggests that most abused
children do avoid completing this cycle. Even in Widom’s widely cited 1989
Science study, only 29% of her abused and neglected sample had an adult
criminal record, and only 11% of that sample had a violent criminal record as
an adult. The relative inattention to the protective factors exhibited by these
resilient individuals is unfortunate given that they may provide valuable
information regarding how the cycle may be broken for those less resilient.
The purpose of the current study is to identify the protective factors that
increase resilience to violent offending among young adults who were previ-
ously abused as children.1 We focus specifically on a subset of young adults
from the National Longitudinal Study of Adolescent to Adult Health (Add
Wright et al. 1263
Health) who were abused prior to the sixth grade to examine two broad
Research Question 1: What factors are associated with reduced violence
among young adults who identify as being previously abused as
In answering this, we focus on determining the individual and social protec-
tive factors that are distinctive to the “negative cases” of the cycle of vio-
lence—those individuals who might be expected to engage in adult violence
given their childhood victimization yet do not.
Research Question 2: Are certain protective factors more or less impor-
tant depending on the type and frequency of abuse experienced?
To answer this question, we examine whether protective factors differ accord-
ing to the type, frequency, and comorbidity of childhood abuse. Our work
comes as part of a more general victimization literature that seeks to explain
variations in behavioral responses among victims (e.g., Boxer & Sloan-
Power, 2013; Grych, Hamby, & Banyard, 2015; Turanovic & Pratt, 2015).
Ultimately, then, our broader objective is to refocus cycle of violence research
on determining the modifiable protective factors that may break the cycle.
Child Abuse and the Cycle of Violence
Research on child abuse has come a long way since Kempe, Silverman,
Steele, Droegemueller, and Silver (1962) wrote of physical abuse in the form
of “parental assault” as part of battered-child syndrome. This body of work
has advanced our understanding of child abuse in significant ways, and yet
reviews of the literature identify a number of important conceptual and meth-
odological issues to address for any study that examines the consequences of
childhood victimization (see especially Malvaso, Delfabbro, & Day, 2018;
Thornberry et al., 2012). Four issues in particular merit special attention for
research that seeks to identify protective factors that may break the cycle of
violence. Two of these are conceptual and relate to how abuse and resilience
are defined, and two are methodological and concern the nature of the sam-
ples most often used in existing research.
First, existing works often overlook the distinctions between type, fre-
quency, and comorbidity of child abuse (Malvaso et al., 2018). The original
cycle of violence research documented that physical abuse was most impor-
tant for the prediction of violence in adulthood (Widom, 1989a; see also
1264 Journal of Interpersonal Violence 34(6)
Maxfield & Widom, 1996). Yet the type of abuse experienced by children
may produce different behavioral outcomes, with sexual abuse assuming a
critical importance alongside physical abuse (e.g., Currie & Tekin, 2011; see
also Noll, 2005; Yun, Ball, & Lim, 2011). The frequency with which abuse
occurs is also likely to impact whether a previously abused child engages in
violent behavior as an adult. For example, Heller, Larrieu, D’Imperio, and
Boris (1999) identify the number of instances of abuse as a potentially con-
founding factor that is consistently unaddressed in studies of maltreated
children (see also DuMont et al., 2007; McGloin & Widom, 2001). In addi-
tion, the comorbidity of types of abuse is likely to affect the relationship
between childhood victimization and future adult violence. This becomes
particularly important when acknowledging resilience research that suggests
personal resources (e.g., above-average intelligence) may no longer be suf-
ficient as a protective factor when the child experiences multiple forms of
stress (Jaffee et al., 2007).
Second, existing works often examine static protective factors that cannot
be addressed or altered through intervention programs. In particular, demo-
graphic characteristics are often offered as protective factors, which lead to
perplexing assertions that “being White” or “being older” serve as a buffer
against the deleterious effects of child abuse. Instead, if modifiable protective
factors such as success in school or mentorship from a caring adult can be
identified, then interventions can be tailored toward promoting these factors
within previously abused individuals (Cicchetti, 2004; Haskett, Nears, Ward,
& McPherson, 2006).
Third, existing works often use agency-based samples of abused chil-
dren that may not be representative of the larger abused population of
childhood victims. Differences between victimized children who were
referred to agencies (such as child protective services) and those who were
not may include the severity of abuse and levels of familial or extrafamil-
ial support systems (Heller et al., 1999). Furthermore, data based on
agency samples tend to underreport actual childhood victimization
instances, which could result in conservative estimates of the relationships
between abuse, resilience, and future violence (Topitzes, Mersky, Dezen,
& Reynolds, 2013). Due in part to these concerns, Thornberry and col-
leagues (2012) recommend that cycle of violence studies use a sample that
is representative of a general population, as selected using probability
sampling techniques, with a satisfactory participation rate. A large sample
of this nature would also allow for separate examinations of the protective
factors of different subtypes (e.g., physical abuse only, physical and sexual
abuse; Haskett et al., 2006).
Wright et al. 1265
Finally, existing works within the cycle of violence and resilience litera-
tures often focus on outcomes in childhood or adolescence, and less is known
regarding the protective factors that promote resilience into early adulthood
(McGloin & Widom, 2001; Topitzes et al., 2013; Topitzes, Mersky, &
Reynolds, 2012). Studies have documented that previously abused children
may appear resilient in adolescence, but not in early adulthood (DuMont et
al., 2007); youth who appear resilient in adolescence may not have truly bro-
ken the cycle. We believe that the essence of the cycle of violence is that
adults use violence toward children who then use violence toward others
when they are adults.
The child abuse and resilience literatures address the above issues to vary-
ing degrees—with available reviews suggesting that the bulk of this literature
falls short on conceptual and methodological rigor (Haskett et al., 2006;
Heller et al., 1999; Thornberry et al., 2012). Nevertheless, this work suggests
that studies that examine the factors that can break the cycle of violence
should (a) differentiate between the type, frequency, and comorbidity of child
abuse, (b) focus on dynamic protective factors that can be modified, (c) use a
large and diverse sample, and (d) use self-reported outcomes of violence that
are measured in adulthood.
A key disclaimer of the original cycle of violence hypothesis research is
that future adult violence is far from inevitable. Widom (1989a, p. 169)
suggests, “It is important to understand the potential protective factors that
intervene in the child’s development and to compare the development of
those who succumb and those who are ‘resilient’ and do not.” We take this
advice and begin with a sample of young adults who have reported being
abused as children. According to the cycle of violence, we would expect
most of these individuals to “succumb” and be at an increased risk of engag-
ing in violence. So what sets those who perpetrate violence apart from those
who do not? Based on the existing child abuse and resilience literatures, our
current study has two objectives. First, we identify the protective factors
that reduce violent offending in early adulthood among individuals who
were abused as children. Second, we examine whether certain protective
factors are more or less important depending on the type and frequency of
child abuse experienced. By addressing the above objectives, we identify
the protective factors that contribute to resilience in young adulthood, and
we also hope to encourage future work on how victims of child abuse are
able to break the cycle of violence.
1266 Journal of Interpersonal Violence 34(6)
We use data from Add Health, which is an ongoing, nationally representative
study of adolescent health and well-being (Harris, 2009). A sample of 80 high
schools and 52 feeder middle schools and junior high schools was selected
through a disproportionately stratified, school-based, clustered sampling
design. The sample was representative of U.S. schools with respect to region
of the country, urbanicity, school type, school size, and ethnicity (Harris,
2011). At Wave I in 1994 to 1995, in-school surveys were administered to
more than 90,000 students enrolled in grades 7 to 12, from which a random
subsample of 20,745 adolescents was selected to participate in the Wave I,
in-home component of the study. Wave III follow-up interviews with the
Wave I sample were conducted 7 years later during 2001 to 2002. The aver-
age age of participants was 15 years at Wave I (ranging from 11 to 21 years)
and 22 years at Wave III (ranging from 18 to 28 years). Of the original Wave
I respondents, approximately 15,000 participated in the Wave III in-home
interview (N = 14,322 with valid sample weights).2 We focus primarily on the
Wave III survey as it contains information on childhood physical and sexual
abuse that was not captured in previous waves.
As is common with large-scale survey data, information was missing on
some of our key variables due to item nonresponse (10.7% of Wave III
respondents had item-missing data). To address the potential bias produced
by missing data, multiple imputation was used (Allison, 2000). This involved
a procedure in which 10 imputed data sets were generated by a missingness
equation that included all variables in the present study, and which adjusted
estimates according to the clustered surveying of respondents in schools
(using the mi suite in Stata 13). The results from 10 imputed data sets were
combined using pooled parameter estimates to account for the possible
underestimation of standard errors observed in single imputation procedures.3
Cases with missing information on the dependent variable (i.e., violent
offending), and those without information on childhood victimization were
excluded (n = 1,206). As a result, 91.6% of Wave III respondents were
retained (N = 13,116).
During the Wave III interview, Add Health respondents were asked to ret-
rospectively report information on physical and sexual victimization that
occurred during childhood.4 The following two questions were asked: “By
Wright et al. 1267
the time you started 6th grade, how often had your parents or other adult
caregivers slapped, hit, or kicked you?” and “By the time you started 6th
grade, how often had one of your parents or other adult caregivers touched
you in a sexual way, forced you to touch him or her in a sexual way, or
forced you to have sexual relations?” Responses to each question ranged
from 0 (this has never happened) to 5 (more than 10 times), and approxi-
mately 29.9% of respondents reported experiencing at least one instance of
childhood physical or sexual victimization.5 These questions were adminis-
tered using audio computer assisted self-interview (A-CASI), which is
thought to elicit more accurate reporting of sensitive information involving
victimization and sexual encounters (Turner et al., 1998). To improve the
accuracy of lifetime event data, the Wave III interview also used an event
history calendar as a memory aid. Other incidents of physical and sexual
victimization that occurred after respondents reached the sixth grade were
not captured in the Add Health survey. Measures of childhood victimization
in the Add Health were adapted from previous surveys and have been used
frequently in the literature (see, for example, Hussey, Chang, & Kotch,
Consistent with our research objectives, the sample is split into several
groups that reflect different forms and frequencies of childhood victimiza-
tion. For physical abuse, these include respondents who experienced no
physical abuse (70.9%, n = 9,303), a low frequency of physical abuse
(meaning 1-2 times; 14.2%, n = 1,856), and a high frequency of physical
abuse (meaning 3 or more times; 14.9%, n = 1,957). Similarly, for sexual
abuse, we categorize respondents as those who experienced no sexual abuse
(95.4%, n = 12,510), a low frequency of sexual abuse (1-2 times; 2.9%, n =
379), and a high frequency of sexual abuse (3 or more times; 1.7%, n =
227). We also developed categories to examine individuals who experi-
enced both physical and sexual abuse. These include respondents who did
not experience both physical and sexual abuse (96.2%, n = 12,617), those
who reported a low frequency of both physical and sexual abuse (no more
than two instances of each form of abuse; 1.8%, n = 240), and those who
reported a high frequency of both physical and sexual abuse (experiencing
both forms of abuse, at least one of which happened 3 or more times; 2.0%,
n = 259). Individuals who experienced three or more instances of either
physical or sexual abuse scored above the 90th percentile of childhood vic-
timization in the Add Health data. Consistent with prior research on child-
hood exposure to trauma (e.g., van der Wal, de Wit, & Hirasing, 2003), this
percentile was chosen as the cutoff point for the “high frequency” categori-
zations of child abuse. Sample statistics for all groups of childhood victims
are available by contacting the authors.6
1268 Journal of Interpersonal Violence 34(6)
The dependent variable, violent offending, is a four-item variety score that
reflects whether participants committed the following types of violence dur-
ing the year prior to the Wave III interview: “hurt someone badly in a physi-
cal fight,” “used or threatened to use a weapon to get something from
someone,” “used a weapon in a fight,” “and “pulled a knife or gun on some-
one.” All forms of violence were fairly rare in the full sample (5.5%, 2.0%,
1.8%, and 1.3%, respectively), and approximately 7.7% of Wave III respon-
dents reported committing a violent offense at Wave III.7
Individual Protective Factors
To better understand heterogeneity in adult violence among victims of child
abuse, the effects of several individual and social protective factors on violent
offending are assessed. Specifically, we include four individual protective
factors commonly associated with positive life outcomes: self-control, low
depression, self-esteem, and verbal intelligence.
Self-control at Wave III is measured using nine items from the novelty-
seeking dimension of Cloninger’s (1987) Tridimensional Personality
Questionnaire (e.g., “I sometimes get so excited that I lose control of myself,”
“I like it when people can do whatever they want, without strict rules and
regulations”). These nine items are often used to measure self-control in early
adulthood (see, for example, Turanovic, Reisig, & Pratt, 2015). Each item
featured a 5-point response set, ranging from 1 (very true) to 5 (not true). The
scale exhibits a high level of internal consistency (α = .87), and is coded so
that higher scores indicate higher levels of self-control. Principal components
analysis indicated that the self-control scale was unidimensional (λ = 4.34;
factor loadings > .66).
Low depression is measured using nine items from the 20-item Center for
Epidemiologic Studies Depression (CES-D) scale (Radloff, 1977) that are
available in the Add Health data. Participants were asked to report how often
they experienced feelings related to depression in the past 7 days (e.g., “you
were sad” [reverse-scored], “you cried a lot” [reverse-scored], “you enjoyed
life”). Closed ended responses for each item ranged from 0 (never/rarely) to 3
(most of the time/all of the time), and were summed to create a scale where
larger values reflect lower levels of depression (range 0-27; α = .81). Previous
research has shown the 20-item CES-D to cluster into four subfactors—
somatic-retarded activity, depressed affect, positive affect, and interpersonal
relationships (Ensel, 1986)—and all four components are represented in the
nine items used here. The CES-D has been previously validated among
Wright et al. 1269
adolescents and adults (e.g., Radloff, 1991), and principal components analysis
confirmed that the scale was unidimensional (λ = 3.74; factor loadings > .44).
Self-esteem is assessed using four items from Rosenberg’s (1965) Self-
Esteem Scale: “you have many good qualities,” “you like yourself just the
way you are,” “you have a lot to be proud of,” and “you feel like you are
doing everything just about right.” Items ranged from 0 (strongly disagree) to
4 (strongly agree), and were summed so that higher scores indicate higher
levels of self-esteem (range 0-16; α = .78). Prior research has shown the
Rosenberg scale to be highly reliable (e.g., if a person completes the scale on
two occasions, the two scores tend to be similar) and unidimensional (e.g.,
Baumeister, Campbell, Krueger, & Vohs, 2003). Principal components analy-
sis confirmed that the items used here are associated with a single construct
(λ = 2.46; factor loadings > .74).
Verbal intelligence is captured using respondents’ age-normed Add Health
Picture Vocabulary Test (PVT) score. Add Health PVT scores come from a
shorter, computerized version of the Peabody Picture Vocabulary Test
(Revised) that was administered to participants at the beginning of the Wave
III interview. During this test, interviewers would read a series of words
aloud, and respondents would select pictures that best fit the words’ mean-
ings. Each word in the PVT corresponded to four simple, black-and-white
illustrations arranged in a multiple-choice format. There were 87 items in the
Add Health PVT, and raw scores were standardized by age.
Social Protective Factors
In addition to the individual factors, five forms of social protective factors
are also assessed at Wave III: marriage, job satisfaction, mentorship, religi-
osity, and educational attainment. These factors are considered protective as
they can provide victimized children with supportive coping resources to
overcome adversity (Agnew, 2006), and they can serve as important sources
of restraint that prevent child victims from engaging in crime and violence
Marriage reflects whether respondents were currently married at the time
of the Wave III interview (1 = yes, 0 = no). Nearly 17.3% of young adults
reported being married, and this proportion is consistent with estimates from
the 2000 U.S. Census for young adults between the ages of 20 and 24 (Kreider
& Simmons, 2003). Although data limitations prevent assessing the quality
of these marriages (e.g., marital attachment, connectedness to spouse, and
marital satisfaction), it is important to examine marital status in light of the
body of work indicating that married persons are less likely to engage in
crime than their unmarried counterparts (Sampson & Laub, 1993). Still,
1270 Journal of Interpersonal Violence 34(6)
because this measure cannot differentiate between people who have healthy
marriages and those who do not, the effects of marriage observed here may
be conservative (see, for example, Kuhl, Warner, & Wilczak, 2012).
Job satisfaction is captured using a single-item indicator for whether
respondents had a job that they were satisfied with (1 = yes, 0 = no).
Approximately 70.0% of young adults reported being employed at Wave III,
and 53.2% of all respondents reported having a satisfying job. While job
satisfaction is more commonly measured using different multi-item indexes
(e.g., Hackman & Oldham, 1975), such scales were not available in the data.
The use of a single global indicator of job satisfaction is consistent with prior
research using the Add Health (e.g., Siennick, 2007).
Mentorship is captured using the following survey question at Wave III:
“Other than your parents or step-parents, has an adult made an important
positive difference in your life at any time since you were 14 years old?”
(1 = yes, 0 = no). The majority of Wave III respondents indicated that an adult
had made a positive difference in their life (75.8%), and, most commonly,
these mentors came in the form of siblings and extended family members
(34.7%), teachers/guidance counselors (19.7%), and friends (17.1%).
Three dichotomous indicators of educational attainment were included.
At Wave III, respondents were asked to identify the number of years of
schooling they had received, as well as the educational degrees they
received. Using this information, the following indicators of highest level
of educational attainment were created: high school graduate (35.4%),
some college (36.8%), and college graduate (18.7%), where no high school
degree serves as the reference category. College attendance included both
2-year and 4-year postsecondary institutions. Approximately 12.5% of par-
ticipants at Wave III had not received a high school diploma or graduate
Religiosity is a four-item summated scale composed of the following sur-
vey items at Wave III: “How important is your religious faith to you?” “How
important is your spiritual life to you?” “To what extent are you a spiritual
person?” and “To what extent are you a religious person?” Responses to each
item ranged from 0 (not at all/not important) to 3 (very/more important than
anything else), where higher values indicate greater religiosity (α = .88).
Principal components analysis confirmed that the items used to measure reli-
giosity were unidimensional (λ = 2.95; factor loadings > .83).
Additional Explanatory Variables
Demographic variables and several important correlates of child abuse and
violent offending are also included in the multivariate analyses. Financial
Wright et al. 1271
hardship is a three-item scale at Wave III reflecting whether respondents
or someone in their household did not have enough money in the past year
to: “pay the full amount of rent or mortgage,” “pay the full amount of a
gas, electricity, or oil bill,” or “had services turned off by the gas or elec-
tric company or the oil company wouldn’t deliver because payments were
not made.” Items were dummy-coded and summed to create an index
where higher scores reflect greater financial hardship (range 0-3). Factor
analysis of tetrachoric correlations confirmed that these items are associ-
ated with a single construct (λ= 2.08; factor loadings > .76). A single-item
indicator of childhood neglect available in the data is also included that
reflects how often, before the sixth grade, respondents’ parents or other
adult caregivers “had not taken care of your basic needs, such as keeping
you clean or providing food or clothing.” Closed ended responses to the
childhood neglect item ranged from 0 (never) to 5 (more than 10 times),
and 11.15% of respondents reported at least one instance of neglect.
Finally, variables are included for age (the respondent’s age in years at
Wave III), male (1 = male, 0 = female), and race/ethnicity (including Black,
Hispanic, and Other minority, where non-Hispanic White serves as the
The analyses proceed in two stages. First, several diagnostic tests are con-
ducted to rule out the presence of harmful levels of collinearity. Next, a series
of multivariate regression models are estimated to assess whether the indi-
vidual and social protective factors reduce violent offending within each sub-
sample of victims. As descriptive statistics indicate that the distributions for
violent offending are overdispersed within each subsample of victims (e.g.,
M = .18, variance = .31 among those experiencing low frequency physical
abuse), negative binomial regression is used (Long & Freese, 2006). The
negative binomial model is a generalized linear regression model for count
data that is appropriate to use when there is overdispersion in the dependent
variable (i.e., when the conditional variance is greater than the mean; see
Cameron & Trivedi, 2013; Hilbe, 2011).8
In addition, coefficient estimates and standard errors may be biased if fea-
tures of the Add Health sampling design are not taken into account (Chen &
Chantala, 2014). As a result, the multivariate models are estimated using the
Wave III Add Health sampling weights adjusted for subpopulation analyses,
and clustered robust standard errors that account for the school-based sam-
pling design.9 All analyses are conducted using Stata 13 (StataCorp, College
1272 Journal of Interpersonal Violence 34(6)
Before proceeding with the multivariate analyses, a series of model diagnos-
tics were examined. Bivariate correlations between independent variables in
all models did not exceed an absolute value of .40, and variance inflation
factors were under 1.6. Furthermore, the condition index values did not
exceed 28, which puts them beneath the commonly used threshold of 30
(Tabachnick & Fidell, 2012). According to this evidence, the relationships
between independent variables should not result in biased estimates or inef-
ficient standard errors due to multicollinearity.
Tables 1 to 3 display the negative binomial regression models predicting
violent offending among the different groups of childhood victims—physical
abuse victims, sexual abuse victims, and victims of both physical and sexual
abuse. Each model is estimated using all of the individual and social factors
to determine which have more “general” protective effects on violent offend-
ing across groups, and which tend to be more specific to particular types of
childhood victims. For purposes of comparison, each table also includes a
reference group of individuals who did not experience each type of childhood
Childhood Physical Abuse
Table 1 displays the effects of various protective factors on violent offending
according to the amount of physical abuse experienced. As can be seen, sev-
eral key variables are negatively related to violent offending for victims of
low and high frequency physical abuse in Models 2 and 3. In particular, self-
control reduces violent offending across both groups of victims, where inci-
dence rate ratios (IRR) from Models 2 and 3 indicate that a one unit increase
in self-control decreases the rate of violent offending by 7% (IRR = .93) for
those who experienced a low frequency of physical abuse, and by 4% (IRR =
.96) for those who experienced a high frequency of physical abuse.11 In addi-
tion, low depression is negatively related to violent offending for both low
frequency (IRR = .95) and high frequency (IRR = .93) physical abuse vic-
tims. Being married (IRR = .48), attending college (IRR = .53), and graduat-
ing college (IRR = .29) also emerged as protective factors against violence,
but only for individuals who experienced a high frequency of physical abuse
(see Model 3). Thus, while the protective effects of marriage and educational
attainment are specific to victims of high frequency physical abuse, the
effects of self-control and depression appear to be protective for both groups
of physical abuse victims examined in Models 2 and 3. However, upon closer
Wright et al. 1273
examination, differences can be detected.12 More specifically, the protective
effects of self-control on violence are weaker for individuals who experi-
enced a high frequency of physical abuse (z = |3.33|, p < .01). In contrast, the
effects of low depression do not vary by frequency of physical abuse.
Table 1. Negative Binomial Regression Models Predicting Violent Offending by
Frequency of Physical Abuse.
Violent Offending in Early Adulthood
No Physical Abuse in
Low Frequency Physical
High Frequency Physical
Model 1 Model 2 Model 3
b (SE)z b (SE)Z b (SE)z
Self-control −0.06 (.01) −6.10** −0.07 (.01) −5.78** −0.04 (.02) −2.06*
−0.04 (.02) −2.52* −0.05 (.02) −2.12* −0.07 (.02) −2.81**
Self-esteem −0.03 (.03) −1.02 −0.03 (.05) −0.66 0.04 (.03) 1.19
−0.04 (.03) −1.44 −0.07 (.04) −1.94 0.07 (.04) 1.83
Marriage −0.52 (.23) −2.29* 0.26 (.33) 0.79 −0.74 (.34) −2.19*
Job satisfaction 0.30 (.24) 1.27 −0.52 (.21) −1.24 0.18 (.26) 0.71
Mentorship −0.02 (.16) −0.16 0.12 (.23) 0.52 −0.11 (.26) −0.44
Religiosity −0.03 (.02) −1.41 −0.01 (.03) −0.18 −0.03 (.04) −0.91
−0.24 (.21) −1.14 0.29 (.30) 0.99 −0.17 (.29) −0.59
Some college −0.87 (.25) −3.44** 0.15 (.31) 0.49 −0.64 (.29) −2.24*
College grad −0.76 (.29) −2.62** 0.04 (.46) 0.08 −1.24 (.46) −2.68**
−0.04 (.11) −0.33 0.03 (.14) 0.18 0.21 (.18) 1.17
0.02 (.08) 0.19 0.06 (.07) 0.77 0.01 (.06) 0.11
Age −0.11 (.04) −2.57* −0.10 (.06) −1.52 −0.14 (.07) −2.13*
Male 1.10 (.22) 4.98** 1.69 (.26) 6.45** 0.89 (.23) 3.84**
Black 0.72 (.20) 3.68** 0.73 (.25) 2.95** 0.87 (.26) 3.35**
Hispanic 0.32 (.21) 1.56 0.38 (.27) 1.37 0.04 (.29) 0.14
Other minority 0.04 (.26) 0.14 0.18 (.46) 0.40 0.09 (.34) 0.25
F test 15.97** 17.68** 9.62**
N9,303 1,856 1,957
Note. Entries are unstandardized partial regression coefficients (b), clustered robust standard errors in
parentheses, and z tests. Coefficients and standard errors for verbal intelligence are multiplied by 10 for
ease of interpretation.
*p < .05. **p < .01 (two-tailed test).
1274 Journal of Interpersonal Violence 34(6)
Table 2. Negative Binomial Regression Models Predicting Violent Offending by
Frequency of Sexual Abuse.
Violent Offending in Early Adulthood
No Sexual Abuse in
Low Frequency Sexual
High Frequency Sexual
Model 1 Model 2 Model 3
b (SE)z b (SE)z b (SE)Z
Self-control −0.06 (.01) −7.00** −0.04 (.01) −2.76** −0.08 (.04) −2.22**
Low depression −0.05 (.02) −3.42** −0.02 (.03) −0.85 −0.18 (.06) −2.73**
Self-esteem −0.02 (.02) −0.83 −0.06 (.06) −0.91 −0.01 (.03) −0.39
−0.01 (.02) −0.64 −0.06 (.05) −1.07 −0.03 (.02) −1.71
Marriage −0.45 (.18) −2.44* 0.28 (.42) 0.65 −1.16 (1.35) −0.86
Job satisfaction −0.16 (.19) −0.85 −0.67 (.36) −1.86 −0.44 (.42) −1.06
Mentorship 0.05 (.13) 0.37 −0.08 (.29) −0.27 0.70 (.38) 1.87
Religiosity −0.04 (.02) −2.16* 0.05 (.04) 1.19 −0.15 (.08) −1.92
High school grad −0.09 (.18) −0.48 0.13 (.35) 0.37 −1.07 (.56) −1.90
Some college −0.65 (.19) −3.36** 0.51 (.40) 1.27 −0.71 (.57) −1.24
College grad −0.64 (.25) −2.58* −0.06 (.51) −0.11 −5.02 (.83) −6.08**
Financial hardship 0.04 (.10) 0.40 −0.01 (.21) −0.05 1.28 (.68) 1.87
−0.01 (.05) −0.21 −0.01 (.12) −0.03 0.23 (.13) 1.72
Age −0.11 (.04) −3.25** −0.07 (.08) −0.82 −0.11 (.20) −0.55
Male 1.19 (.16) 7.52** 1.49 (.30) 4.71** 2.27 (.72) 3.16**
Black 0.79 (.15) 5.23** 0.52 (.30) 1.75 0.60 (.51) 1.19
Hispanic 0.30 (.18) 1.61 0.26 (.47) 0.56 1.69 (.73) 2.30*
Other minority 0.06 (.22) 0.30 0.95 (.77) 1.22 −0.76 (.47) −1.63
F test 25.90** 4.60** 5.23**
N12,510 379 227
Note. Entries are unstandardized partial regression coefficients (b), clustered robust standard errors in
parentheses, and z tests. Coefficients and standard errors for verbal intelligence are multiplied by 10 for
ease of interpretation.
*p < .05. **p < .01 (two-tailed test).
Childhood Sexual Abuse
Table 2 presents findings with respect to the frequency of sexual abuse vic-
timization during childhood. As seen in Models 2 and 3, several statistically
significant protective effects emerge. Self-control reduces violent offending
among individuals who experienced a low frequency of sexual abuse (IRR =
.96) and a high frequency of sexual abuse (IRR = .92) during childhood.
Having low depression (IRR = .84) and graduating from college (IRR = .01)
Wright et al. 1275
are also negatively related to violence, but only among individuals who expe-
rienced a high frequency of sexual abuse (see Model 3). Similar to the pattern
of findings observed with respect to physical abuse in Table 1, the effects of
self-control on violent offending vary between high frequency and low
Table 3. Negative Binomial Regression Models Predicting Violent Offending by
Frequency of Physical and Sexual Abuse.
Violent Offending in Early adulthood
No Physical and
Sexual Abuse in
Physical and Sexual
Physical and Sexual
Model 1 Model 2 Model 3
b (SE)z b (SE)z b (SE)z
Self-control −0.06 (.01) −6.99** −0.03 (.01) −2.56* −0.05 (.02) −2.44*
−0.05 (.01) −3.43** −0.02 (.03) −0.56 −0.12 (.04) −3.28**
Self-esteem −0.02 (.02) −0.81 −0.04 (.06) −0.68 0.09 (.06) 1.59
−0.01 (.02) −0.62 −0.05 (.06) −0.85 −0.03 (.08) −0.44
Marriage −0.45 (.18) −2.45* −0.58 (.39) 1.50 0.10 (.55) 0.18
Job satisfaction −0.18 (.19) 0.94 −0.23 (.38) −0.61 −0.95 (.42) −2.24*
Mentorship 0.05 (.13) 0.37 0.02 (.27) 0.09 0.10 (.39) 0.26
Religiosity −0.04 (.02) −2.19* 0.09 (.05) 1.84 −0.04 (.05) −0.87
−0.09 (.18) −0.48 0.12 (.36) 0.32 −0.66 (.37) −1.78
Some college −0.66 (.19) −3.40** 0.62 (.33) 1.90 −0.57 (.44) −1.30
College grad −0.65 (.25) −2.60** 0.01 (.51) 0.02 −5.15 (.89) −5.77**
0.04 (.10) 0.42 −0.04 (.23) −0.19 0.32 (.25) 1.24
−0.01 (.08) 0.30 0.08 (.16) 0.50 0.21 (.12) 1.76
Age −0.10 (.05) −0.28 −0.09 (.08) −1.03 0.05 (.09) 0.58
Male 1.20 (.16) 7.58** 1.21 (.31) 3.84** 1.91 (.37) 5.17**
Black 0.79 (.15) 5.33** 0.24 (.33) 0.72 0.31 (.43) 0.73
Hispanic 0.29 (.18) 1.60 0.19 (.43) 0.45 0.35 (.37) 0.97
Other minority 0.07 (.22) 0.30 1.01 (.70) 1.44 −0.04 (.54) −0.09
F test 26.06** 2.57** 4.52**
N12,617 240 259
Note. Entries are unstandardized partial regression coefficients (b), clustered robust standard errors in
parentheses, and z tests. Coefficients and standard errors for verbal intelligence are multiplied by 10 for
ease of interpretation.
*p < .05. **p < .01 (two-tailed test).
1276 Journal of Interpersonal Violence 34(6)
frequency abuse victims. Specifically, self-control has a significantly weaker
effect on violent offending among individuals who experienced a high fre-
quency of childhood sexual abuse (z = |2.43|, p < .05).
Childhood Physical and Sexual Abuse
The models presented in Table 3 assess violent offending across groups of
individuals who experienced a combination of both physical and sexual
abuse during childhood. In keeping with the pattern of findings thus far, self-
control reduces violence among individuals who experienced low (IRR =
.97) and high (IRR = .95) frequencies of both physical and sexual abuse (see
Models 2 and 3). In addition, low depression (IRR = .89), job satisfaction
(IRR = .39), and graduating from college (IRR = .01) also emerge as protec-
tive factors against violent offending, but these effects are specific to indi-
viduals who experienced high frequencies of abuse (see Model 3). Although
self-control reduced violence among both groups of victims assessed in Table
3, invariance tests revealed that the effects of self-control on violent offend-
ing are weaker among high frequency abuse victims (z = |2.60|, p < .01).
Despite the robustness of the results to listwise deletion and selection effects
(see Notes 3 and 10), a series of supplemental analyses were conducted (not
shown in table form). In particular, models were estimated separately for men
and women to determine whether the findings were sensitive to using a
mixed-gender sample (e.g., Topitzes et al., 2012). Indeed, men are responsi-
ble for perpetrating the majority of violent offenses, and it is possible that the
impact of protective factors vary by gender (e.g., Belknap, 2015). Accordingly,
gender-specific analyses were estimated for subtypes of victims where the
sample sizes were large enough to accommodate all covariates in a stable
way. This excluded groups of victims who experienced high frequencies of
sexual abuse, and those who experienced low and high frequencies of both
physical and sexual abuse.13
These supplemental analyses revealed that findings were generally similar
across men and women. Consistent with the results presented previously,
self-control reduced violent offending across all groups of male and female
victims assessed, and low depression was linked to lower violence among
high frequency victims of physical abuse. Nevertheless, some differences
arose with respect to females who experienced three or more instances of
physical abuse. For such women, going to college and graduating college
were no longer related to violent offending (compare with Table 1, Model 3).
Wright et al. 1277
Despite these differences, the findings remained similar across both male and
female victims of child abuse. Taken altogether, there is a great deal of het-
erogeneity in violent offending among victims of physical and sexual abuse,
and several prosocial individual and social factors can help explain why some
victims of child abuse are more likely to engage in violence in early adult-
hood than others.
The debate over the existence of a cycle of violence has gone on for decades.
Such a reality led Thornberry and colleagues (2012, p. 145) to lament “there
are almost as many review pieces as there are original studies.” One thing
seems clear: Abused children are at an increased risk for perpetrating future
violence as adults. The strength of this relationship can be debated for another
several decades, but doing so misses an opportunity to understand why future
violence among previously victimized children is not an absolute certainty.
Most abused children do not complete the cycle, and we know surprisingly
little about why that is the case—especially when it comes to identifying mal-
leable protective factors. The current study sought to build on the limited
literature by examining the protective factors that reduce violent offending
among young adults who were previously abused in childhood. Our work
here leads to three broad conclusions.
First, a number of protective factors reduced the likelihood of violent
offending among victims of child abuse. The most consistent and strongest of
these factors was the individual protective factor of self-control. This finding
is consistent with the broader criminological literature that documents a
strong relationship between low self-control and criminal behavior (Pratt &
Cullen, 2000), as well as that which links victimization, low self-control, and
future criminal behavior (Turanovic & Pratt, 2013). We add to this literature
by finding that self-control may reduce offending for victims of abuse. Stated
differently, building self-control among abused children may be a way of
breaking the cycle of violence. And although self-control was a consistent
predictor in all models, we note that it was a weaker predictor of violence in
some subsamples (especially among individuals who were abused more fre-
quently). The individual protective factor of low depression and the social
protective factors of job satisfaction, attending college, and graduating from
college emerged as protective factors in one or more of our models. The cycle
of violence is not inevitable, and these protective factors offer ways it may be
Second, certain factors may be protective for some forms and extents of
abuse but not others. Social protective factors were more protective among
1278 Journal of Interpersonal Violence 34(6)
those young adults who had been frequently abused as children. Indeed, no
social protective factors were significant in any models examining individu-
als who experienced a low frequency of abuse. This pattern is somewhat at
odds with research that suggests that social factors are more critical for resil-
ience in nonabused children (and perhaps less serious cases of abused chil-
dren) whereas personality characteristics and self-esteem processes are more
important for resilience in abused children (Cicchetti & Rogosch, 1997). Our
results suggest that self-control in particular is less protective among children
who were more frequently abused, and factors such as education and job
satisfaction mattered more for these victims (see also Jaffee et al., 2007). And
even among the social factors, the protective impact varied depending on the
type, frequency, and comorbidity of abuse. Marriage was protective among
victims who had a high frequency of physical abuse but not among other
victims. Job satisfaction was protective among victims who were both physi-
cally and sexually abused at a high frequency but not among victims experi-
encing either type of abuse separately at a high frequency. The broader
implication of this pattern of results is that remaining resilient to the negative
consequences of abuse is not a “one size fits all” endeavor.
Third, the dynamic protective factors in our model suggest specific pro-
gramming implications for those who may be interested in intervening in the
lives of abused children. The pattern of self-control being significant in all of
our models is consistent with research documenting self-regulation as a pro-
tective factor in adaptation by childhood victims (Cicchetti & Rogosch,
1997). Based on this finding, emotion and behavior regulation training may
be beneficial toward reducing the likelihood of future violence perpetration
among victims of child abuse (Haskett et al., 2006; Topitzes et al., 2013;
Topitzes et al., 2012). The self-control criminological literature suggests a
number of promising ways to increase and strengthen self-control (Piquero,
Jennings, & Farrington, 2010). Beyond the self-control finding, our models
suggest additional protective factors that may be promoted based on type and
extent of abuse—a conclusion that affirms that different types of victims may
respond to treatment differently (Cicchetti, 2004). Individual resources alone
may not be enough to promote resilience among young adults who experi-
enced multiple forms and higher frequencies of abuse (Jaffee et al., 2007),
and it is encouraging that our results suggest a number of social protective
factors may break the cycle of violence for those who experienced more
severe forms of abuse.
We believe our findings have an even greater importance when consider-
ing the full extent of child abuse in America. It cannot be assumed that all
abused children will come to the attention of social service agencies and
receive the appropriate programming. Indeed, the very reason we used
Wright et al. 1279
nationally representative, self-report data was to capture these hidden victims
who likely would never be identified as abused. It seems pertinent, then, that
some of the policy implications from our findings are those that can be
emphasized among the general population—with the added benefit being that
they could potentially break the cycle of violence among abused children
specifically. For example, a focus on eliminating truancy and encouraging
high school completion—a necessary step before attaining the additional pro-
tective factors of higher education—may be especially critical for youth who
were previously abused (Johnson, Wright, & Strand, 2012; see Tanaka,
Georgiades, Boyle, & MacMillan, 2015).
While we were able to shed light on the processes that shape whether
victims of abuse can remain resilient, we also hope that our work prompts
additional research that may address some of the things we could not. For
example, although our data allow us to examine abuse that may have gone
undetected and provide a sufficient number of cases to look at specific sub-
types of abuse, a prospective longitudinal study within a community sam-
ple is typically regarded as the gold standard for this type of research
(Thornberry et al., 2012). In addition, we have the same actor report on
both abuse and violence, and retrospective recall bias could introduce addi-
tional threats to the validity of our findings (Heller et al., 1999; Widom,
1989b; see also Note 4). The simple truth is that there are going to be
strengths and weaknesses to any type of approach taken. Both child abuse
and adult violence could be underreported when using official records, and
asking children about their possible abuse presents serious ethical and
We have also used a relatively limited measure of resilience given our
focus on the cycle of violence. It is possible that these adults are not truly
resilient in other areas such as cognitive and emotional functioning (McGloin
& Widom, 2001). Relatedly, our measure of childhood victimization could
include elements such as exposure to violence (Sousa et al., 2011), our mea-
sure of violence could include elements such as family violence and intimate
partner violence (Tomsich, Jennings, Richards, Gover, & Powers, 2017), and
future studies could also include additional individual, familial, and commu-
nity protective factors identified in the literature (Caspi et al., 2002). Finally,
it is likely that the victimization, protective factors, and resilience linkages
operate differently across race and ethnicity. This is an important avenue for
future research, but that research should not simply consider race or ethnicity
as risk or protective factors. Instead, future work should examine how modi-
fiable risk factors are conditioned by race and ethnicity. This approach may
take us closer toward understanding how to break the cycle of violence
among those children who may be most at risk.
1280 Journal of Interpersonal Violence 34(6)
No one can dispute that the effects of child abuse (including future vio-
lence) are substantial, and work moving forward should continue to try to
understand the potential cycle of violence that abuse may set in motion. That
said, rather than simply examining whether that cycle occurs, it is important
that future research also examine how that cycle may be broken. In recent
years, the field of criminology has seen a renewed interest in explaining the
negative cases that do not conform to theoretical expectations (Sullivan,
2011; Wright & Bouffard, 2016). We began our work by stating that we were
going to look at the negative cases in the cycle of violence—those individuals
who could be expected to perpetrate future violence given their previous
abuse, yet do not. The truth is that these are not actually the negative cases as
most abused children do not go on to future violence. These cases deserve
more scholarly attention, and they may be able to teach us about the protec-
tive factors necessary to break the cycle of violence for those less fortunate.
The authors wish to thank Travis Pratt for his helpful comments on a previous draft.
This research uses data from Add Health, a program project directed by Kathleen
Mullan Harris and designed by J. Richard Udry, Peter S. Bearman, and Kathleen
Mullan Harris at the University of North Carolina at Chapel Hill, and funded by Grant
P01-HD31921 from the Eunice Kennedy Shriver National Institute of Child Health
and Human Development, with cooperative funding from 23 other federal agencies
and foundations. No direct support was received from Grant P01-HD31921 for this
analysis. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle
for assistance in the original design. Information on how to obtain the Add Health data
files is available on the Add Health website (http://www.cpc.unc.edu/addhealth).
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research,
authorship, and/or publication of this article.
The author(s) received no financial support for the research, authorship, and/or publi-
cation of this article.
1. We conceive of resilience as “successful adaptation or development during or
following adverse conditions” (Masten & Wright, 1998, p. 10). We use the term
Wright et al. 1281
“resilience” within the cycle of violence to connote a successful adaptation in
young adulthood (i.e., being less likely to engage in violence given one’s expo-
sure to childhood victimization) rather than “resiliency,” which suggests a per-
sonality trait (McGloin & Widom, 2001, p. 1022).
2. For more information on the National Longitudinal Study of Adolescent to Adult
Health (Add Health) sample, research design, survey content, and data quality,
3. To determine the robustness of the findings, supplemental analyses were con-
ducted using listwise deletion to handle missing data. The results closely mir-
rored those observed using multiple imputation, and findings were the same in
terms of sign and significance.
4. It is possible that retrospective accounts of child abuse might result in over- or
underreporting abuse as respondents might “forget” or redefine their experiences
in light of later life circumstances and their current situation. Based on previous
research, we suspect that abuse is underreported in this sample (Widom & Morris,
1997; Widom & Shepard, 1996). Nevertheless, Add Health prevalence estimates
of child physical and sexual abuse are generally consistent with those from other
national surveys (see, for example, Boney-McCoy & Finkelhor, 1995).
5. We focus here on the type and frequency of abuse as no additional information
was captured on the contexts surrounding each instance of childhood victimiza-
tion (i.e., when and where it occurred, who the perpetrator was, and whether
there were injuries) that could tap into dimensions of severity or chronicity.
6. We do not explicitly examine neglect as a form of child victimization. We agree
with other scholars who argue that child neglect is a more general problem of
inadequate parenting (Dodge, Bates, & Pettit, 1990) and that it reflects an act
of omission rather than commission (Cicchetti & Rogosch, 1997). Indeed, we
examine here the cycle of violence with respect to childhood victimization:
Violence inflicted upon children will result in them inflicting violence on others
in adulthood. Nevertheless, given that neglect represents the most commonly
reported type of child maltreatment, we include it as an important explanatory
variable in all regression models.
7. According to Sweeten (2012, p. 554), variety scores are the preferred way to
measure criminal offending because they “possess high reliability and validity,
and are not compromised by high frequency non-serious items in the scale.”
8. The negative binomial model has the same mean structure as the Poisson model,
but it includes an extra parameter to model overdispersion (see Hilbe, 2011). If
applied to overdispersed data, Poisson regression can result in underestimated
standard errors and spuriously large z values (Long & Freese, 2006). Because
there is overdispersion in violent offending (our dependent variable), the nega-
tive binomial model is a good fit for the data.
9. The Add Health sampling weights are used to address potential bias originat-
ing from differential sampling probabilities and attrition, and to prevent the
underestimation of standard errors. For more detailed guidelines on analyzing
1282 Journal of Interpersonal Violence 34(6)
the Add Health data, see http://www.cpc.unc.edu/projects/addhealth/data/guides/
10. Additional models were estimated to rule out issues with sample selection bias.
As individuals within each subsample of victims were not selected by ran-
dom assignment, it is possible that the results could be biased in several ways
(Bushway, Johnson, & Slocum, 2007). Accordingly, models were estimated using
setpoisson in Stata 13 (Miranda, 2012). This involved the estimation of a probit
model with exclusion restrictions for selection into each subsample, which was
estimated simultaneously with a second-stage Poisson regression model predict-
ing violent offending. The setpoisson model forces overdispersion in the depen-
dent variable to guard against the underestimation of standard errors observed in
standard Poisson regression. The findings remained the same in terms of sign and
significance, and likelihood ratio tests of independent equations for these models
indicated that sample selection was not a detectable source of bias.
11. Exponentiating the negative binomial regression coefficient gives us the incident
rate ratio (IRR).
12. Comparisons between models were conducted using the test statistic recom-
mended by Brame, Paternoster, Mazerolle, and Piquero (1998) for maximum-
likelihood regression coefficients, where
13. Specifically, there were 65 males who experienced a high frequency of sexual
abuse, 90 females who experienced low frequencies of both physical and sexual
abuse, and 96 males who experienced a high frequency of both physical and
sexual abuse during childhood. Given that females are overrepresented in the
high frequency sexual abuse groups, additional research with a larger number of
sexual abuse victims could examine whether the protective factors are gendered
for these subgroups.
Agnew, R. (2006). General strain theory: Current status and directions for future research.
In F. T. Cullen, J. P. Wright, & K. R. Blevins (Eds.), Taking stock: The status of
criminological theory (pp. 101-123). New Brunswick, NJ: Transaction Publishers.
Allison, P. D. (2000). Multiple imputation for missing data: A cautionary tale.
Sociological Methods & Research, 8, 301-309.
Baumeister, R. F., Campbell, J. D., Krueger, J. I., & Vohs, K. D. (2003). Does high
self-esteem cause better performance, interpersonal success, happiness, or health-
ier lifestyles? Psychological Science in the Public Interest, 4, 1-44.
Belknap, J. (2015). The invisible woman: Gender, crime, and justice (4th ed.).
Belmont, CA: Wadsworth.
Boney-McCoy, S., & Finkelhor, D. (1995). Psychosocial sequelae of violent vic-
timization in a national youth sample. Journal of Consulting and Clinical
Psychology, 63, 726-736.
Boxer, P., & Sloan-Power, E. (2013). Coping with violence: A comprehensive frame-
work and implications for understanding resilience. Trauma, Violence, & Abuse,
Wright et al. 1283
Brame, R., Paternoster, R., Mazerolle, P., & Piquero, A. R. (1998). Testing the equal-
ity of maximum-likelihood regression coefficients between two independent
equations. Journal of Quantitative Criminology, 14, 245-261.
Bushway, S., Johnson, B. D., & Slocum, L. A. (2007). Is the magic still there? The use
of the Heckman two-step correction for selection bias in criminology. Journal of
Quantitative Criminology, 23, 151-178.
Cameron, A. C., & Trivedi, P. K. (2013). Regression analysis of count data (2nd ed.).
New York, NY: Cambridge University Press.
Caspi, A., McClay, J., Moffitt, T. E., Mill, J., Martin, J., Craig, I. W., . . . Poulton, R.
(2002). Role of genotype in the cycle of violence in maltreated children. Science,
Chen, P., & Chantala, K. (2014). Guidelines for analyzing Add Health data. Chapel
Hill: Carolina Population Center, University of North Carolina
Cicchetti, D. (2004). An odyssey of discovery: Lessons learned through three decades
of research on child maltreatment. American Psychologist, 59, 731-741.
Cicchetti, D., & Rogosch, F. A. (1997). The role of self-organization in the promo-
tion of resilience in maltreated children. Development and Psychopathology, 9,
Cloninger, C. R. (1987). A systematic method for clinical description and classification
of personality variants: A proposal. Archives of General Psychiatry, 44, 573-588.
Currie, J., & Tekin, E. (2011). Understanding the cycle: Childhood maltreatment and
future crime. Journal of Human Resources, 47, 509-549.
Dodge, K. A., Bates, J. E., & Pettit, G. S. (1990). Mechanisms in the cycle of vio-
lence. Science, 250, 1678-1683.
DuMont, K. A., Widom, C. S., & Czaja, S. J. (2007). Predictors of resilience in abused
and neglected children grown-up: The role of individual and neighborhood char-
acteristics. Child Abuse & Neglect, 31, 255-274.
Ensel, W. M. (1986). Measuring depression: The CES-D scale. In N. Lin, A. Dean, &
W. M. Ensel (Eds.), Social support, life events, and depression (Vol. 1, pp. 51-
68). Orlando, FL: Academic Press.
Grych, J., Hamby, S., & Banyard, V. (2015). The resilience portfolio model: Understanding
healthy adaptation in victims of violence. Psychology of Violence, 5, 343-354.
Hackman, J. R., & Oldham, G. R. (1975). Development of the job diagnostic survey.
Journal of Applied Psychology, 60, 159-170.
Harris, K. M. (2009). The National Longitudinal Study of Adolescent Health (Add
Health), Waves I & II, 1994-1996; Wave III, 2001-2002; Wave IV, 2007-2008.
Chapel Hill: Carolina Population Center, University of North Carolina.
Harris, K. M. (2011). Design features of Add Health. Chapel Hill: Carolina Population
Center, University of North Carolina.
Haskett, M. E., Nears, K., Ward, C. S., & McPherson, A. V. (2006). Diversity in
adjustment of maltreated children: Factors associated with resilient functioning.
Clinical Psychology Review, 26, 796-812.
Heller, S. S., Larrieu, J. A., D’Imperio, R., & Boris, N. W. (1999). Research on resil-
ience to child maltreatment: Empirical considerations. Child Abuse & Neglect,
1284 Journal of Interpersonal Violence 34(6)
Herrenkohl, T. I., Huang, B., Tajima, E. A., & Whitney, S. D. (2003). Examining the
link between child abuse and youth violence: An analysis of mediating mecha-
nisms. Journal of Interpersonal Violence, 18, 1189-1208.
Hilbe, J. M. (2011). Negative binomial regression (2nd ed.). New York, NY:
Cambridge University Press.
Hussey, J. M., Chang, J. J., & Kotch, J. B. (2006). Child maltreatment in the United
States: Prevalence, risk factors, and adolescent health consequences. Pediatrics,
Jaffee, S. R., Caspi, A., Moffitt, T. E., Polo-Tomas, M., & Taylor, A. (2007).
Individual, family, and neighborhood factors distinguish resilient from non-
resilient maltreated children: A cumulative stressors model. Child Abuse &
Neglect, 31, 231-253.
Johnson, C. L., Wright, K. A., & Strand, P. S. (2012). Transitions of truants:
Community truancy board as a turning point in the lives of adolescents. Journal
of Juvenile Justice, 1, 34-51.
Kempe, C. H., Silverman, F. N., Steele, B. F., Droegemueller, W., & Silver, H.
K. (1962). The battered-child syndrome. Journal of the American Medical
Association, 181, 17-24.
Kreider, R. M., & Simmons, T. (2003, October). Marital status: 2000 (Census 2000
Brief). Washington, DC: U.S. Census Bureau.
Kuhl, D. C., Warner, D. F., & Wilczak, A. (2012). Adolescent violent victimization
and precocious union formation. Criminology, 50, 1089-1127.
Long, J. S., & Freese, J. (2006). Regression models for categorical and limited depen-
dent variables using Stata (2nd ed.). College Station, TX: Stata Press.
Malvaso, C. G., Delfabbro, P., & Day, A. (2018). The maltreatment–offending
association: A systematic review of the methodological features of prospective
and longitudinal studies. Trauma, Violence, & Abuse, 19, 20-34. doi:10.1177/
Masten, A. S., & Wright, M. O. (1998). Cumulative risk and protection models of
child maltreatment. Journal of Aggression, Maltreatment & Trauma, 2, 7-30.
Maxfield, M. G., & Widom, C. S. (1996). The cycle of violence: Revisited 6 years
later. Archives of Pediatrics & Adolescent Medicine, 150, 390-395.
McGloin, J. M., & Widom, C. S. (2001). Resilience among abused and neglected
children grown up. Development and Psychopathology, 13, 1021-1038.
Miranda, A. (2012). Setpoisson: Stata module (Mata) to estimate a selection endoge-
nous treatment Poisson model by maximum simulated likelihood. Retrieved from
Noll, J. G. (2005). Does childhood sexual abuse set in motion a cycle of vio-
lence against women? What we know and what we need to learn. Journal of
Interpersonal Violence, 20, 455-462.
Piquero, A. R., Jennings, W. G., & Farrington, D. P. (2010). On the malleability of
self-control: Theoretical and policy implications regarding a general theory of
crime. Justice Quarterly, 27, 803-834.
Pratt, T. C., & Cullen, F. T. (2000). The empirical status of Gottfredson and Hirschi’s
general theory of crime: A meta-analysis. Criminology, 38, 931-964.
Wright et al. 1285
Radloff, L. S. (1977). The CES-D scale: A self-report depression scale for research in
the general population. Applied Psychological Measurement, 1, 385-401.
Radloff, L. S. (1991). The use of the Center for Epidemiologic Studies Depression Scale
in adolescents and young adults. Journal of Youth and Adolescence, 20, 149-166.
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton
Sampson, R. J., & Laub, J. H. (1993). Crime in the making: Pathways and turning
points through life. Cambridge, MA: Harvard University Press.
Siennick, S. E. (2007). The timing and mechanisms of the offending-depression link.
Criminology, 45, 583-615.
Sousa, C., Herrenkohl, T. I., Moylan, C. A., Tajima, E. A., Klika, J. B., Herrenkohl,
R. C., & Russo, M. J. (2011). Longitudinal study on the effects of child abuse
and children’s exposure to domestic violence, parent-child attachments, and anti-
social behavior in adolescence. Journal of Interpersonal Violence, 26, 111-136.
Sullivan, C. J. (2011). The utility of the deviant case in the development of crimino-
logical theory. Criminology, 49, 905-920.
Sweeten, G. (2012). Scaling criminal offending. Journal of Quantitative Criminology,
Tabachnick, B. G., & Fidell, L. S. (2012). Using multivariate statistics (6th ed.).
Upper Saddle River, NJ: Pearson.
Tanaka, M., Georgiades, K., Boyle, M. H., & MacMillan, H. L. (2015). Child mal-
treatment and educational attainment in young adulthood: Results from the
Ontario Child Health Study. Journal of Interpersonal Violence, 30, 195-214.
Thornberry, T. P., Knight, K. E., & Lovegrove, P. J. (2012). Does maltreatment beget
maltreatment? A systematic review of the intergenerational literature. Trauma,
Violence, & Abuse, 13, 135-152.
Tomsich, E., Jennings, W. G., Richards, T. N., Gover, A. R., & Powers, R. A. (2017).
Childhood physical maltreatment and young adult dating violence: A propen-
sity score matching approach. Journal of Interpersonal Violence, 32, 3475-3496.
Topitzes, J., Mersky, J. P., Dezen, K. A., & Reynolds, A. J. (2013). Adult resilience
among maltreated children: A prospective investigation of main effect and medi-
ating models. Children and Youth Services Review, 35, 937-949.
Topitzes, J., Mersky, J. P., & Reynolds, A. J. (2012). From child maltreatment to
violent offending: An examination of mixed-gender and gender-specific models.
Journal of Interpersonal Violence, 27, 2322-2347.
Turanovic, J. J., & Pratt, T. C. (2013). The consequences of maladaptive coping: Integrating
general strain and self-control theories to specify a causal pathway between victim-
ization and offending. Journal of Quantitative Criminology, 29, 321-345.
Turanovic, J. J., & Pratt, T. C. (2015). Longitudinal effects of violent victimization
during adolescence on adverse outcomes in adulthood: A focus on prosocial
attachments. Journal of Pediatrics, 166, 1062-1069.
Turanovic, J. J., Reisig, M. D., & Pratt, T. C. (2015). Risky lifestyles, low self-
control, and violent victimization across gendered pathways to crime. Journal of
Quantitative Criminology, 31, 183-206.
1286 Journal of Interpersonal Violence 34(6)
Turner, C. F., Ku, L., Rogers, S. M., Lindberg, L. D., Pleck, J. H., & Sonenstein, F. L.
(1998). Adolescent sexual behavior, drug use, and violence: Increased reporting
with computer survey technology. Science, 280, 867-873.
van der Wal, M. F., de Wit, C. A. M., & Hirasing, R. A. (2003). Psychosocial health
among young victims and offenders of direct and indirect bullying. Pediatrics,
Watts, S. J., & McNulty, T. L. (2013). Childhood abuse and criminal behavior: Testing
a general strain theory model. Journal of Interpersonal Violence, 28, 3023-3040.
Widom, C. S. (1989a). The cycle of violence. Science, 244, 160-166.
Widom, C. S. (1989b). Does violence beget violence? A critical examination of the
literature. Psychological Bulletin, 106, 3-28.
Widom, C. S., & Morris, S. (1997). Accuracy of adult recollections of childhood vic-
timization, Part 2: Childhood sexual abuse. Psychological Assessment, 9, 34-46.
Widom, C. S., & Shepard, R. L. (1996). Accuracy of adult recollections of childhood vic-
timization, Part 1: Childhood physical abuse. Psychological Assessment, 8, 412-421.
Wright, K. A., & Bouffard, L. A. (2016). Capturing crime: The qualitative analysis of
individual cases for advancing criminological knowledge. International Journal
of Offender Therapy and Comparative Criminology, 60, 123-145.
Yun, I., Ball, J. D., & Lim, H. (2011). Disentangling the relationship between child
maltreatment and violent delinquency: Using a nationally representative sample.
Journal of Interpersonal Violence, 26, 88-110.
Kevin A. Wright is an associate professor in the School of Criminology and Criminal
Justice at Arizona State University. His primary research interests focus on crimino-
logical theory and correctional policy.
Jillian J. Turanovic is an assistant professor in the College of Criminology and
Criminal Justice at Florida State University. Her research focuses on victimization
and offending over the life course, and the collateral consequences of incarceration.
She received her PhD in 2015 from Arizona State University and is a graduate
research fellow of the National Institute of Justice.
Eryn N. O’Neal is an assistant professor in the College of Criminal Justice at Sam
Houston State University. Her primary research interests include decision making in
sexual assault cases, intimate partner sexual assault, victim decision making, and
Stephanie J. Morse is a master’s student in the School of Criminology and Criminal
Justice at Arizona State University. Her research interests include offender rehabilita-
tion and reentry.
Evan T. Booth is a master’s student in the Josef Korbel School of International
Studies at the University of Denver. His research interests include the effects of child-