Columbia Population Research Center
Advancing Research in Population, Health, and Society
CPRC Working Paper No. 09-04
Child Maltreatment and Crime:
New Evidence from a Sample of Twins
Department of Economics
Department of Economics
Andrew Young School of Policy Studies
Georgia State University
Unpublished manuscript. Do not copy or cite without permission of author(s).
1255 Amsterdam Avenue ▪ New York, New York ▪ 10027
Child Maltreatment and Crime: New Evidence from a Sample of Twins
Columbia University, UCLA and NBER
Department of Economics
420 W 118th St.
New York, NY 10027
Georgia State University and NBER
Department of Economics
Andrew Young School of Policy Studies
Georgia State University
P.O. Box 3992
Atlanta, GA 30302-3992
We would like to thank Neil Guterman, Christina Paxson, Chris Ruhm, Joyce Tabor, Volkan
Topalli, and participants in seminars at the City University of New York, Columbia’s School
Social Work, Columbia’s Teachers College, 2005 Population Association of America Meetings,
2005 IZA/SOLE Meetings, 2005 SEA Meetings, Georgia State University, MDRC, University of
Illinois at Chicago, and Koc University for helpful comments. Roy Wada provided excellent
research assistance. We gratefully acknowledge funding from the National Institutes of Health
grant number R21 HD050740-01-A2. We are solely responsible for any interpretations or errors.
Child Maltreatment and Crime: New Evidence from a Sample of Twins
Child maltreatment is a major social problem. This study measures the effects of child
maltreatment on crime using data from the National Longitudinal Study of Adolescent Health
(Add Health). We focus on crime because it is one of the most costly potential outcomes of
maltreatment. Our work addresses many limitations of the existing literature. First, we use a
large national sample, and investigate different types of abuse in a similar framework. Second,
we pay careful attention to controlling for possible confounding factors by comparing male twins
and by controlling for differences in genetic endowments that have been linked to aggression and
risk taking behaviors. We find that maltreatment greatly increases the probability of engaging in
crime and that programs to reduce maltreatment would be cost-effective.
Child maltreatment, which includes both child abuse and child neglect, is a major social
problem. In the U.S., the death rate is 2.4 per 100,000 and 1,500 children die every year,
making child maltreatment the leading cause of deaths from injuries in children over a year old
(Institute of Medicine, 1999). Moreover, deaths are only the tip of the iceberg. According to the
U.S. Department of Health and Human Services (1996), over a million children are victims of
maltreatment annually. Paxson and Waldfogel (1999, 2002) show that abuse and neglect are
more common in families of lower socioeconomic status, so that maltreatment may exacerbate
differences in the life chances of rich and poor children.
Maltreatment may have many long-term consequences for survivors. This paper focuses
on the effect of child maltreatment on crime using data from the National Longitudinal Study of
Adolescent Health (Add Health). We focus on crime because it is one of the most socially
costly potential outcomes of maltreatment, yet there is controversy about the extent to which a
“cycle of violence” in which child maltreatment leads to future violence has been substantiated
(c.f. Widom, 1989a).
Our work contributes to the existing literature in three ways. First, we use data from a
national survey that includes a large “control” group of children who were neither maltreated nor
committed crime. Second, in contrast to many studies that control for only a very limited
number of background characteristics, we attack the problem of identifying the effects of
maltreatment by comparing male twins. Since some twins are fraternal, we also control for
genetic markers that have been associated with aggressive and risk-taking behaviors. Third, we
examine the effects of sexual abuse, physical abuse, and neglect in a similar framework.
We show that the estimated effects of maltreatment on crime are large and remarkably
robust to different estimation methods and specifications: Maltreatment approximately doubles
the probability that a child will engage in crime in future. Moreover, while sexual abuse has the
largest effects, neglect and (to a lesser extent) physical abuse are also important predictors of
crime. The large size of the effects and the relatively modest cost of an intervention that has
been shown to be effective in reducing childhood maltreatment suggest that further action to
reduce the prevalence of child maltreatment is warranted.
Many studies have examined the long-term consequences of child maltreatment. The
National Research Council (1993) offers a review and a critique. The main limitation of the
studies they review is that most did not control adequately for potential confounders. It is
possible that households in which children are maltreated have other characteristics that are
associated with negative child outcomes and higher propensities to engage in crime. For
example, in our data, children who were maltreated are more likely to have had a mother with
less than a high school education, more likely to have had a father in jail at the beginning of the
survey, more likely to have been in a family on welfare at the beginning of the survey, and more
likely to have a mother who was a teenager at the time of her child’s birth.
The sensitive nature of child maltreatment makes it difficult to use experimental designs.
It would be unthinkable to divide study children into a control group and a treatment group that
was maltreated. Hence, researchers must find another way to separate the effects of
maltreatment from those of other confounders.
In this section, we attempt to briefly summarize some of the more recent work. A first
group of studies establishes a cross-sectional relationship between past experiences of
maltreatment and other adverse events, and current risky behaviors/outcomes. For example,
Felitti (1998) and Dube et al. (2003a) show that adverse childhood experiences (ACEs) increase
the risk for depressed affect, suicide attempts, multiple sexual partners, sexually transmitted
diseases, smoking, and alcoholism, while other studies show a relationship between ACEs, illicit
drug use, and teen pregnancy (Dube et al., 2003b; Hillis et al., 2004).
While suggestive, these relationships do not necessarily imply that ACEs cause risky
behaviors. If, for example, poverty is associated with ACEs then the fact that people with ACEs
have higher rates of criminal activity could actually reflect a causal relationship between poverty
and involvement in crime. Moreover, many ACE studies aggregate maltreatment with other
forms of household dysfunction rather than trying to separately identify the effect of
A second group of studies is more closely related to the current study in that they control
for family background factors such as poverty by using samples of twins in which one twin was
maltreated and the other was not. Nelson et al. (2002), Kendler et al. (2000) and Dinwiddie et
al. (2000) use this design to examine the effects of child sexual abuse on future psychiatric
problems. The first two studies conclude that maltreated twins are more likely than their non-
maltreated twins to suffer negative outcomes. However, Dinwiddie finds no differences
between maltreated and non-maltreated twins. Our study examines neglect and physical abuse
as well as sexual abuse, and focuses on effects on crime.
Widom (1989b) is one of the best known studies of the link between child maltreatment
and crime. She matched a sample of 908 children with substantiated cases of maltreatment
processed by the courts from 1967 to 1971 to controls who were selected to be similar in terms
of age, sex, race, and socioeconomic status. She finds that being abused or neglected as a child
increases an individual’s risk for an arrest as a juvenile by 53 percent and increases the
probability of arrest as an adult by 38 percent.
Several other prospective longitudinal studies focus on groups of “at risk” children and
follow them over time, comparing children who come to the attention of child welfare authorities
to those who do not (see Stouthamer-Loeber et al., 2001 and Ireland, Smith, and Thornberry,
2006). Like Widom’s, these studies deal only with substantiated cases. Administrative data on
maltreatment and criminality capture only a fraction of all abuses and criminal behaviors because
not all maltreatment or criminal activity are reported to or captured by government agencies.
Administrative data sets often contain only limited information on important mediating
factors such as family income and socio-economic background. Matching on a small number of
observable traits provides no guarantee that the controls are really similar to the “experimental”
group in terms of unmeasured as well as measured characteristics. Furthermore, families about
whom there are official records may be those more likely to come to the attention of official
agencies, and thus may be an unrepresentative sample of families in which child abuse occurs
(Smith and Thornberry, 1995). Lansdale et al. (2002) and (2007) address this problem by using
maternal reports of maltreatment. But these reports may be suspect if the mother is the
perpetrator, given that the researchers are mandated to report any ongoing abuse.
Our study is the first to examine the effects of child maltreatment on crime using a
national sample of twins. The twin-comparison design offers a compelling way to control for
unobserved family-level characteristics. In an effort to control for possible differences between
fraternal twins that might be associated both with outcomes and how they are treated, we also
include controls for genetic variations that have been associated with risky or aggressive
behaviors. A third strength of our study is that, like Widom, we examine the effects of physical
abuse, neglect, and sexual abuse in a similar framework.
The Add Health survey was specifically designed to investigate adolescents' health and
risk behaviors. It is considered the largest and most comprehensive survey of adolescents ever
undertaken. It began with an in-school questionnaire administered to children in sampled
schools between September 1994 and April 1995. A random sample of students from each
school completed more detailed in-home interviews between April and December 1995. A
second interview wave took place in 1996, and the third wave was administered between August
2001 and April 2002. 1
The Add Health over-sampled twins, and in this study we focus only on male twins
because it is well known that males are more likely than females to engage in serious crime
(Lanctot and LeBlanc, 2002). Out of 484 twins, there are a total of 176 identical (monozygotic)
twins. Sample sizes for some of our empirical analyses will be slightly less than these numbers,
however, because of missing data in key variables for at least one twin in some twin pairs.
Measures of Maltreatment
In Wave III, respondents answered questions about the way they were treated by their
parents or other adults who took care of them before they were in the 6th grade. Specifically,
they were asked whether and how often:
1 The Add Health is a program project designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris,
and funded by a grant P01-HD31921 from the National Institute of Child Health and Human Development, with
cooperative funding from 17 other agencies. Special acknowledgment is due 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
(firstname.lastname@example.org). More information on the design of Add Health can be found at
1. Parents (or other adult care-givers) had not taken care of their basic needs, such as keeping
them clean or providing food or clothing.
2. Parents (or other adult care-givers) slapped, hit, or kicked them.
3. Parents (or other adult care-givers) had touched them in a sexual way, forced them to touch
him or her in a sexual way, or forced them to have sexual relations.
The definitions of child maltreatment given at the Child Welfare Information Gateway website,
which is a service of the Children's Bureau, Administration for Children and Families, U.S.
Department of Health and Human Services (http://www.childwelfare.gov/can/defining/) are
consistent with the abuse questions asked in the Add Health. It is important to emphasize that neglect is
an important and very common form of child maltreatment, one that our results suggest can have long-
Several researchers have studied the validity of self-reported data on child maltreatment
and have concluded that, if collected properly, they are valid (Allen, Leadbeater, and, Aber,
1994; Dembo et al. 1991). In our data, the respondents listened to pre-recorded questions on
sensitive topics through earphones and entered their answers directly on laptops in order to
maintain confidentiality and to minimize the potential for interviewer or other third party
influence. In order to obtain accurate responses about the timing of events, subjects were
prompted with a calendar that gave the dates of many important events. Mocan and Tekin (2005,
2006) and Tekin and Markowitz (2008) provide evidence that rates of many of the risky
behaviors reported in the Add Health are consistent with other sources. Widom and Shepard
(1996) found good that official reports of physical abuse predict official reports of violent
criminal behavior, just as self-reports of physical abuse predict self-reports of violent behavior.
A limitation our study shares with most others is that it is based on retrospective reports
of maltreatment. It is possible that people tend to forget past abuse as they grow older. We have
investigated “forgetting” directly by examining whether the older people in the sample were less
likely to report childhood abuse than those who were 18. We find no evidence that this is the
A potentially more serious problem is that people with negative outcomes may be more
likely to report childhood maltreatment. For example, they may blame past maltreatment for
their current problems. In this case, Ordinary Least Squares (OLS) estimates will tend to find
“effects” of maltreatment that are too large. Measurement error of this type could also bias twin
fixed effects models. Suppose that one twin is more likely to report bad things in general. Then
this twin will be more likely to report both maltreatment and crime.
We investigated this problem by examining responses to a series of questions that should
have been answered in the same way by both twins. These questions included whether or not the
father was in jail at Wave I; how far the two twins lived away from each other; how often the
twins saw each other; how often the twins talked to each other; and how often the twins fought
with each other. We found few significant correlations between differences in the twins’ reports
about maltreatment and differences in reports about these other variables. These findings
provide some evidence against the hypothesis that one twin is just “more negative” than the
In the third wave of Add Health, respondents answered detailed questions about their
relationships with their parents. Specifically, they answered questions on whether (1) they
enjoyed spending time with parents, (2) parents were warm and loving toward them, (3) they felt
close to their parents, and (4) they received money or any other significant payment from their
parents within the past 12 months. We compared these measures to the reports of maltreatment
as a way to both assess the internal validity of our data and to test the reliability of our
maltreatment measures. For all four measures of closeness to parents, respondents who reported
maltreatment also reported that they had more distant relationships with their parents for all of
our maltreatment variables.
Table 1 shows the fraction of respondents reporting various forms of maltreatment. The
incidence of reports of particular types of maltreatment varies widely. Roughly fifteen percent
of the sample report that their parents ever failed to meet their basic needs. Only 3.9% of the
sample report any form of sexual abuse, while almost a third report that their parents ever hit
them. These rates are broadly consistent with other studies. For example, Scher et al. (2004)
use a sample of 967 adults from a community survey in Memphis Tennessee to determine the
prevalence of retrospectively reported child maltreatment and find that 19% reported physical
abuse, 18% reported neglect, and 5% reported sexual abuse. Still, in most of our work we will
use a measure of whether the child was hit 10 or more times in order to indicate more severe or
persistent physical maltreatment. Only 5.8% of respondents report physical abuse this severe.
Note that the fraction of the twins sample reporting maltreatment is similar to those of the full
Add Health male sample and the full male sibling sub-sample.
Column (2) of Table 1 shows the fraction of twins who have different reports of
maltreatment. This column shows that discrepant reports are quite common, which is necessary
if we are to identify effects of abuse using twin comparisons. As discussed above, it is not
uncommon for some children in a household to be maltreated while others are unharmed. The
question is whether this has anything to do with pre-existing characteristics of the child? By
focusing on males, we eliminate gender as a potential explanation for why parents would
maltreat one twin and not another. We control for birth weight (another characteristic that often
varies between twins) directly in our models.
In the case of non-identical twins, we will also control for differences in genetic markers
that have been associated with aggressive or risky behavior, as discussed further below. Note
that in the case of identical twins, there are no genetic differences.
In Wave III of the Add Health, DNA samples were collected from participants who were
siblings using cheek swabs. Every person has two versions of each gene which are referred to as
alleles. While most portions of most genes are the same across individuals, differences in the
building blocks of the alleles may occur in some fragments and these differences are inherited by
the children on a random basis. Alleles are often referred to by the length of the fragments
involved. Variations in the coding of alleles are believed to be linked to some variations in
human outcomes. There are many previous studies that associate certain types of individual
behaviors with these variations (See Caspi et al., 2002; Rowe et al., 1998; Ding et al., 2006; Guo
and Tong, 2006). These studies have guided us in incorporating genetic markers in our empirical
analyses. We created control variables corresponding to different genetic patterns as described
in Appendix 1.
Measures of Crime
The crime questions in Add Health are similar to those found in other surveys and to
official definitions of “crime” found in government sources such as the Bureau of Justice
Statistics. We focus on whether the respondent committed any hard drug crime, or any non-drug
crime in the past 12 months, and on three particular crimes that have relatively high incidence in
our data: Whether the respondent damaged property, whether he assaulted someone, or whether
he committed robbery. We created binary variables for these variables as follows: Any non-
drug = 1 if committed any of the non-drug crimes in the past 12 months, and = 0 otherwise;
Damaged Property = 1 if deliberately damaged property that belonged to someone else in the
past 12 months, and = 0 otherwise ; Assault = 1 if pulled a knife on someone, shot someone, or
badly hurt someone in the past 12 months, and = 0 otherwise; Any Hard Drug = 1 if used any
hard drugs (heroin, crystal meth, cocaine, LSD, ecstasy, inhalants, PCP, mushrooms, etc.) in the
past 12 months, and = 0 otherwise; and Robbery = 1 if used or threatened to use a weapon to get
something from someone else in the past 12 months, and = 0 otherwise.
The means of the crime variables are presented in Table 2 by whether or not respondents
suffered various types of maltreatment. The table indicates that across almost every domain,
children who suffered maltreatment are much more likely to have engaged in crime as those who
did not. The differences for non-drug crimes are greatest for sexual abuse, but the incidence of
crime is also high for those who report that their basic needs were not met (i.e. that they were
We estimate models of the form:
(1) Crimeij= β0 + β1*Maltreatmentij + β2*Birthweightij + β3* SNPij + β4*PairIDij + εij,
where Crimeij is one of the crime variables for male twin i in family j, Maltreatment indicates
one of the indicators for maltreatment, Birthweight is measured in grams, SNP is a vector of
genetic markers as described above, and PairID is a unique identifier for each twin pair. These
models control for all of the common elements of family background that are shared by the
twins. Still, twin studies may underestimate the effects of abuse for two reasons. First, to the
extent that there is random error in reports of abuse, estimates from twin models will be biased
towards zero. Second, the occurrence of maltreatment in a household may traumatize all
children in the household. Hence, although the effects we find are large, there is some reason to
believe that the true effects may be even larger.
In order to examine the hypothesis that the probability of engaging in crime increases
with the experience or multiple forms of maltreatment, we created binary indicators for
experiencing multiple types of maltreatment. About 11 percent of our twin sample experienced at
least one type of maltreatment and another 5.2 percent experienced two types of maltreatment.
We created three binary indicators for “no maltreatment”, “one type of maltreatment” and “two
types of maltreatment” and used these as the measures of maltreatment. There is no one in the
twin sample who experienced all three types of maltreatment.
Twin studies highlight an important question, which is whether it is common for parents
to maltreat one twin and not the other, and whether maltreatment is related to characteristics of
the child. Hamilton-Giachristis and Browne (2005) examine siblings in families in which police
reports of maltreatment were filed for an index child. They found that in 37 percent of families,
maltreatment was directed at all siblings, while in the remaining families, maltreatment was
directed at some children but not others. They find little evidence that this phenomena was
related in any consistent way to characteristics of the index child such as the presence of a
disability or medical condition. Many authors have argued that in families with child
maltreatment, it is common to have one child be singled out as a target or “scape-goat” (Young,
Jaffee et al. (2004) examine pairs of identical twins in which one was maltreated and the
other was not. Again, they find little evidence that the maltreatment was related to observable
characteristics of the index child. Campbell, Bogen, and Berger (2006) report on a survey of
physicians with expertise in treating victims of abuse. When asked whether signs of abuse were
“often”, “occasionally”, or “almost never” likely to be present in the siblings of index children
upon examination, these experts selected “occasionally” as the appropriate response.
In summary, there is considerable evidence that while siblings of abused children are at
higher risk of maltreatment than other children, there are many families in which only some
siblings are abused. And it is not easy to predict, on the basis of observable characteristics,
which child will be maltreated. These findings lend support to the idea of using twin
comparisons to identify the effects of maltreatment.
Our main results are shown in Table 3. This table shows quite different estimates of the
effects of different forms of maltreatment. Having parents who ever failed to meet ones basic
needs greatly increases the probability of committing both non-drug and drug crimes. Sexual
abuse also has large effects (though not on hard drug use). In contrast, having a parent who
struck, hit or kicked them increases the probability of hard drug use, but does not have a
significant effect on the other categories of crime in these models. It is possible that the
imprecise estimated effects for physical abuse reflect lack of consensus about what constitutes
physical abuse (e.g. some may include punishments such as spanking while others may not).
We have also estimated these models using only identical twins. We have only 150 pairs
of male MZ twins with complete data, and relatively few with differential maltreatment: For
example 8 pairs have discordant reports for physical abuse (>10 times); 10 for sexual abuse; 16
for failing to meet basic needs; and 16 for any maltreatment. Therefore, it is remarkable that we
still obtain significant effects of maltreatment on many types of crime even in these models,
though we do not attach a lot of weight to the point estimates. In particular, estimates for any
non-drug crime, assault, and robbery are robust to this change of samples.
Table 4 compares estimates from models in which we focus on whether the child suffered
only one type of maltreatment, to estimates in which the child suffered two or more types of
maltreatment. For example, if sexual abuse and neglect both cause crime separately, one might
expect that a person who experienced both sexual abuse and neglect would be at a higher risk of
committing crime than a person who had only one of these experiences.
Our data show that joint experiences of maltreatment are relatively common. For
example, the percentage of individuals whose basics needs were not satisfied is 84 percent
among those who experienced sexual abuse, while it is only 13 percent among the others. About
11 percent of our twin sample experienced at least one type of maltreatment and another 5.2
percent experienced two types of maltreatment. We created three binary indicators for “no
maltreatment”, “one type of maltreatment” and “two types of maltreatment”. There is no one in
the twin sample who experienced all three types of maltreatment. As shown in Table 4, the
results provide evidence that the probability of engaging in crime increases if a person suffers
multiple forms of maltreatment.
Discussion and Conclusions
In order to determine the social cost of child maltreatment, we need to quantify its effects
on important outcomes. This paper focuses on the effects on crime. We find that child
maltreatment roughly doubles the probability that an individual engages in many types of crime.
It is useful to put our results into perspective by comparing it to other estimates of the effects of
factors related to crime. For example, using time series data from New York, Corman and
Mocan (2005) find that a 1 percentage point decline in unemployment generates only a 2.2
percentage point decline in burglaries. They also find that a 10 percent increase in the minimum
wage leads to about 3.5 percent decrease in robberies in New York City. Grogger and Willis
(2000) find that the introduction of crack cocaine in metropolitan areas was responsible for a 20
percent increase in murder rates, a fifteen percent increase in rapes, and a 27 percent increase in
Using individual-level data from the first two waves of Add Health, Mocan and Tekin
(2006) find that having access to a gun at home increases the propensity to commit a variety of
crimes by about 30 percent among adolescents. Duggan (2001) calculates that the decreases in
gun ownership over the 1990s can explain up to a third of the decline in crime over the same
period. Finally, Bingenheimer et al. (2005) find that exposure to firearm violence approximately
doubles the probability that an adolescent will engage in serious violence over the subsequent
Hence, the effects of maltreatment are large relative to those of many environmental risk
factors for crime, and are perhaps comparable to those of guns. The large size of these effects
suggests that maltreatment may generate large costs to society in the form of crime.
Table 5 considers a crude translation of some of the effects measured in this paper into
dollar terms, using estimates of the costs of crime from the literature. We use estimates of the
baseline incidence of robbery and assault from the full male sample of the Add Health. Table 5
shows that these figures are quite similar to those shown for male twins in Table 2.
The estimated effects of maltreatment on the costs of crime are very sensitive to the
estimated cost per crime, which varies widely between studies. Lochner and Moretti (2004) use
estimates of the cost of crime that take account of the impacts on victims, as well as the costs of
incarceration. Their estimates do not take account of other costs, such as the costs to society of
avoiding crime, and thus should probably be regarded as lower bounds. Cohen (2004) derives
estimates based on surveys of “willingness to pay” for crime reduction. These estimates may
well be upper bounds given that people are not required to pay anything when they fill in these
surveys. The largest estimated costs of crime are for murder, which is such a rare crime that we
cannot look at changes in its incidence using our data.
The Lochner and Moretti estimates suggest that the assault and robbery induced by abuse
costs society about $8.8 billion per year. The Cohen estimates suggest a much larger figure of
$68.6 billion. These figures must be compared to the costs of preventing maltreatment.
Unfortunately, few intervention programs have been proven to be effective in rigorous studies.
An exception is Olds et al. (1999) who report that home visits by nurses that start in the infancy
and continue through two years can reduce the incidence of substantiated cases of maltreatment
by 50 percent. At a cost of about $4,000 per child, the steady-state cost of providing this service
to all children would be about $14 billion per year (assuming that there are roughly 3.5 children
born each year). Given that the additional assaults and robberies induced by abuse are only one
of the social costs of maltreatment, these estimates suggest that a home visiting program like
Olds’ could pay for itself in terms of reducing social costs, even using the most conservative
estimates of the costs of crime. If we attach some benefit to improving the lives of children
(beyond the value we attach to saving taxpayer money) then the cost-benefit analysis looks even
Figures in Unicef (2003) suggest that there are very large variations in deaths due to
maltreatment between European countries, with rates of only 0.1 to 0.3 per 100,000 in Spain,
Greece, Italy, Ireland, and Norway. This suggests that there are other aspects of social
organization which are protective for children, and that it might be worthwhile studying the large
differences between countries to gain a better understanding of these factors.
In summary, we find that being maltreated greatly increases the probability of engaging
in many types of crime and that the probability of engaging in crime increases with the
experience of multiple forms of maltreatment, so that criminal behavior increases not only with
the incidence of maltreatment but also with the severity of maltreatment. A rough cost-benefit
calculation suggests that a home visiting program that prevented maltreatment would be cost
effective, though there may well be other ways to reduce the appalling toll associated with child
Allen, J. P., Leadbeater, B. J., & Aber, J. L. (1994). The Development of Problem Behavior
Syndromes in At-Risk Adolescents. Development and Psychopathology, 6, 323–342.
Bingenheimer, Jeffrey B., Robert T. Brennan, and Felton J. Earls (2005). “Firearm Violence
Exposure and Serious Violent Behavior,” Science 308(5726): 1323-1326.
Campbell, Kristine, Debra Bogen, and Rachel Berger. “The Other Children: A Survey of Child
Abuse Physicians on the Medical Evaluation of Children Living with a Physically
Abused Child,” Archives of Adolescent Medicine, 160, Dec. 2006.
Caspi A., J, McClay, T.E. Moffitt, J. Mill, J. Martin, I.W. Craig, A. Taylor, and R. Poulton.
“Role of Genotype in Cycle of Violence in Maltreated Children,” Science
Cohen, Mark A.; Roland T. Rust, Sara Stehen, Simon T. Tidd. “Willingness-To-Pay for Crime
Control Programs“, Criminology, 42(1), 2004, 89-109.
Corman, Hope, and Naci Mocan. “Carrots, Sticks, and Broken Windows,” Economica,
November 2005, 72, pp. 655-681.
Dembo, R., Williams, L., Schmeidler, J., Wish, E. D., Getreu, A., & Berry, E. (1991). Juvenile
Crime and Drug Abuse: A Prospective Study of High Risk Youth. Journal of Addictive
Diseases, 11, 5–31.
Ding, Weili, Steven F. Lehrer, J. Niels Rosenquist, and Janet Audrain-McGovern. “The Impact
of Poor Health on Education: New Evidence Using Genetic Markers,” National Bureau
of Economic Research (NBER) Working Paper No. 12304, June 2006.
Dinwiddie, S. A.C. Heath, M.P. Dunne, K.K. Bucholz, P.A. Madden, W.S. Slutske, L.J. Bierut,
D.B. Statham, N.G. Martin. “Early Sexual Abuse and Lifetime Psychopathology: A Co-
Twin-Control Study,” Psychology and Medicine, 30 #1, 41-52, Jan. 2000.
Dube, Shanta,Vincent Felitti, Maxia Dong, W.H. Giles, and Robert Anda. “The Impact of
Adverse childhood Experiences on Health Problems: Evidence from Four Birth Cohorts
Dating Back to 1900,” Preventive Medicine, 37 #3, 2003a, 268-277.
Dube, Shanta, Vincent Felitti, Maxia Dong, Daniel Chapman, Wayne Giles, and Robert Anda.
“Childhood Abuse, Neglect and Household Dysfunction and the Risk of Illicit Drug Use:
The Adverse Childhood Experiences Study,” Pediatrics, 111 #3, 2003b, 654-572.
Duggan, Mark. “More Guns, More Crime,” Journal of Political Economy, 109:5 (October 2001),
Felitti, Vincent J., Robert Anda, Dale Nordenberg, David Williamson, Alison Spitz, Valerie
Edwards, Mary Koss, and James Marks. “Relationship of Childhood Abuse and
Household Dysfunction to Many of the Leading Causes of Death in Adults,” American
Journal of Preventive Medicine, 14 #4, 1998, 245-258.
Grogan-Kaylor, Andrew, and Melanie D. Otis. “The Effect of Childhood Maltreatment on Adult
Criminality: A Tobit Regression Analysis,” Child Maltreatment, Vol. 8, No. 2, May
Grogger, Jeffrey, and Michael Willis. “The Emergence of Crack Cocaine and the Rise in Urban
Crime Rates,” Review of Economics and Statistics, November 2000.
Guo, Guang, and Yuying Tong. “Age at First Sexual Intercourse, Genes, and Social Context:
Evidence from Twins and the Dopamine D4 Receptor Gene,” Demography, Vol. 43, No.
4, November 2006, 747-769.
Hamilton-Giachritsis and Kevin Browne. “A Retrospective Study of Risk to Siblings in Abusing
Families,” Journal of Family Psychology, 19 #4, 629-624, 2005.
Hillis, S.D., R.F. Anda, S.R. Dube, V.J. Felitti, P.A. Marchbanks, J.S. Marks. “The
Association Between Adverse Childhood Experiences and Adolescent Pregnancy, Long Term
Psychosocial Consequences, and Fetal Death,” Pediatrics, 113 #2, 2004, 320-327.
Institute of Medicine. Reducing the Burden of Injury, Richard Bonnie (ed.) (Washington D.C.:
National Academy Press) 1999.
Ireland, T, C. Smith, and T. Thornberry. “Developmental Issues in the Impact of Child
Maltreatment on Later Delinquency and Drug Use,” Criminology, 40, 2002, 359-399.
Jaffee, S.R., A. Caspi, T.E. Moffitt, M. Polo-Tomas, A. Taylor. “The Limits of Child Effects:
Evidence for Genetically Mediated Child Effects on Corporal Punishment but Not on
Physical Maltreatment,” Developmental Psychology, 40 #6, 1047-58.
Lansford, Jennifer, Kenneth Dodge, Gregory Pettit, John Bates, J.Crozier, and J. Kaplow. “A 12-
year Prospetive Study of the Long-Term Effects of Early Child Physical Maltreatment on
Psychological, Behavioral, and Academic Problems in Adolescence,” Archives of
Pediatrics and Adolescent Medicine, 156, 2002, 824-830.
Lansford, Jennifer, Shari Miller-Johnson, Lisa Berlin, Kenneth Dodge, John Bates, and Gregory
Pettit. “Early Physical Abuse and Later Violent Delinquency: A Prospective Longitudinal
Study,” Child Maltreatment, 12, 2007, 233-.
Kendler, Kenneth, Cynthia Bulik, Judy Silberg, John Hettema, John Myers, Carol Prescott.
“Childhood Sexual Abuse and Adult Psychiatric and Substance Use disorders in Women:
An Epidemiological and Co-twin Control Analysis,” Archives of General Psychiatry, 57,
Lochner, Lance and Enrico Moretti. “The Effect of Education on Crime: Evidence from Prison
Inmates, Arrests, and Self Reports,” American Economic Review, 2004, 94(1), pages 155-
Mocan, H. Naci, and Erdal Tekin. “Guns and Juvenile Crime,” Journal of Law and Economics, 9:2, pp.
507-532, October 2006.
Mocan, H. Naci, and Erdal Tekin. “Drug Use and Juvenile Crime: Evidence from a Panel of
Siblings and Twins,” in Economics of Substance Use: Individual Behavior, Social
Interactions, Markets, and Politics, Bjorn Lindgren and Michael Grossman (eds.),
Amsterdam: Elsevier, 2005, pp.91-120.
National Research Council. “Understanding Child Abuse and Neglect,” (Washington D.C.:
National Academy Press) 1993.
Nelson, E.C., A.C. Heath, P.A. Madden, M.L. Cooper, S.H. Dinwiddie, K.K. Bucholz, A.
Glowinski, T. McLaughlin, M.P. Dunne, D.J. Statham, N.G. Martin. “Association
Between Self-reported Childhood Sexual Abuse and Adverse Psychosocial Outcomes:
Results from a Twin Study,” Archives of General Psychiatry, 59 #2, 139-145, Feb. 2002.
Olds, David, Charles Henderson, Harriet Kitzman, John Eckenrode, Robert Cole, and Robert
Tatelbaum. “Prenatal and Infancy Home Visitation by Nurses: Recent Findings,” in The
Future of Children, v9 #1, 1999, 44-65.
Paxson, Christine and Jane Waldfogel. “Parental Resources and Child Abuse and Neglect,
Papers and Proceedings of the American Economic Association, Vol. 89, No. 2, pp. 239-
244, May 1999.
Paxson, Christine and Jane Waldfogel. “Work, Welfare and Child Maltreatment,” Journal of
Labor Economics, 20(3), pp. 435474, 2002.
Rowe, C. David, C. Stever, J. M. Gard, H.H. Cleveland, M. L. Sanders, A. Abramowitz, S. T.
Kozol, J. H. Mohr, S. L. Sherman, and I. D. Waldman. “The Relation of the Dopamine
Transporter Gene (DAT1) to Symptoms of Internalizing Disorders in Children,”
Behavior Genetics, vol. 28, no. 3. 1998, pp. 215-225.
Scher, Christine, David Forde, John McQuaid, Murray Stein. “Prevalence and Demographic
Correlates of Childhood Maltreatment in an Adult Community Sample,” Child Abuse and
Neglect, 28 #2, 2004, 167-180.
Smith, C. and T. O. Thornberry. “The Relationship between Childhood Maltreatment and
Adolescent Involvement in Delinquency,” Criminology, 1995, 33: 451-77.
Stouthamer-Loeber, M, R. Loeber, D. Homish, and E. Wei. “Maltreatment of Boys and the
Development of Disruptive and Delinquent Behavior,” Development and
Psychopathology, 13, 2001, 941-955.
Tekin, Erdal, and Sara Markowitz. ”Suicidal Behavior and Labor Market Productivity,” NBER
Working Paper No. 11238. Forthcoming in Southern Economic Journal, 2008.
Unicef, Child Maltreatment Deaths in Rich Nations, Innocenti Report Card #5, (Unicef:
Innocenti Research Center, Florence) September, 2003.
Widom, C.S. (1989a). Does Violence Beget Violence? A Critical Examination of the Literature,
Psychological Bulletin, 106, 3-28.
Widom, C.S. (1989b). The Cycle of Violence, Science, 244: 160-166.
Widom, C. S. (1998). Childhood victimization: Early adversity and subsequent psychopathology,
in B. P. Dohrenwend (Ed.) Adversity, Stress, and Psychopathology. New York: Oxford
University Press (pp. 81-95).
Widom, C. S., and Shepard, R. L. (1996). Accuracy of Adult Recollections of Childhood
Victimization: Part 1. Childhood Physical Abuse. Psychological Assessment, 8, 412-421.
Young, Marjorie (1976). “Multiple Correlates of Abuse: A Systems Approach to the Etiology of
Child Abuse,” Journal of Pediatric Psychology, Spring, pp. 56-61.
Fraction of Add Health Sample Reporting Maltreatment
Type of Maltreatment
Ever basic needs unsatisfied 0.151 0.119
Ever hit, spanked, kicked etc. 0.279 0.364
Hit etc. > 10 times 0.058 0.079
Ever sexual abuse 0.039 0.071
Ever any maltreatment 0.337 0.329
Any maltreatment using > 10 times cutoffs
for physical abuse 0.205 0.164
Number of observations 484
Source: Authors’ calculations from the Add Health.
calc ns from the A alth.
Table 2: Means of Crime Variables by Type of Maltreatment
Source: Authors’ ulatio dd He
Variable Name Full
Sample Never Any
Maltreatment Ever Fail to
Meet Needs Physical
Abuse>10X Ever Sexual
Any Non-Drug 0.234 0.180 0.444 0.429 0.684
Robbery 0.031 0.016 0.123 0.036 0.368
Assault 0.106 0.055 0.329 0.214 0.579
Damaged Property 0.135 0.114 0.208 0.179 0.421
Any Hard Drugs 0.151 0.132 0.178 0.357 0.211
Effects of Different Types of Maltreatment on Outcomes
Maltreatment a Basic needs
abuse>10X Sexual Abuse
Any Non-drug 0.538*** 0.518*** 0.079 0.503***
(0.124) (0.140) (0.170) (0.191)
Damaged Property 0.288*** 0.278** 0.082 0.387**
(0.101) (0.114) (0.133) (0.150)
Assault 0.305*** 0.285*** -0.098 0.328**
(0.093) (0.104) (0.126) (0.142)
Any Hard drug 0.383*** 0.270** 0.442*** 0.057
(0.094) (0.110) (0.123) (0.149)
Robbery 0.183*** 0.249*** 0.082 0.499***
(0.054) (0.059) (0.071) (0.070)
N 292 302 302 308
a Using >10X cut off for physical abuse.
Notes: Each cell of the table shows the coefficient on the maltreatment indicator from a different regression.
Standard errors are in parentheses. A *, **, *** indicates significance at 90%, 95%, 99% respectively.
Effects of Multiple Experiences of Maltreatment a on Outcomes
Any Non-drug Damaged
Property Assault Any Hard drug Robbery
Only one type of 0.398*** 0.140 0.214** 0.435*** 0.045
Maltreatment b (0.134) (0.109) (0.101) (0.099) (0.050)
Two types of 0.583*** 0.391*** 0.291** 0.367*** 0.464***
Maltreatment (0.173) (0.140) (0.130) (0.127) (0.065)
a Using >10X cut off for physical abuse.
b Omitted category is “no-maltreatment”
Notes: Each cell of the table shows the coefficient on the maltreatment indicator from a different regression. Standard errors are in parentheses.
A *, **, *** indicates significance at 90%, 95%, 99% respectively.
Estimated Annual Costs of Maltreatment in Terms of Increases in Costs of Crime
a) Costs from Lochner and Moretti (2004)
Estimated Per Person Cost of
Cost Per Prevalence Effect Abuse Cost of Abuse
Crime Crime in Add Health Twin FE Abuse (millions)
Assault $13,884 .105 .294 $429 $7,715
Robbery $13,139 .028 .169 $62 $1111
b) Costs from Cohen (2004)
Estimated Per Person Cost of
Cost Per Prevalence Effect of Cost of Abuse
Crime Crime in Add Health Abuse - OLS Abuse (millions)
Assault $81,900 .105 .294 $2528 $45,509
Robbery $271,440 .028 .169 $1284 $23,120
Notes: Lochner and Moretti (2004) present estimates of the social costs of crime and include
incarceration costs. Cohen (2004) derives estimates from a survey asking willingness to pay for
Estimated effects of abuse on probability of crime are from Table 3, column 6. Annual per person
costs are obtained by multiplying the cost by the estimated effect, and then multiplying by the
incidence of any abuse (.23). Total cost of abuse is estimated based on 39 million people aged
20-29 in 2000. Costs converted to millions $2006.
Appendix 1: Defining Genetic Variables
The Add Health Biomarker Team targeted the following genes that have been specifically
hypothesized to be linked to risky behaviors in previous studies: the dopamine transporter
(DAT1), the dopamine D4 receptor (DRD4), monoamine oxidase A promoter (MAOA), the
serotonin transporter (SLC6A4), the dopamine D2 receptor (DRD2), and the Cytochrome P450
gene (CYP2A6). More detailed information on the procedures of DNA extraction and
genotyping of the Add Health DNA sample can be found at
We created variables corresponding to different alleles of these genes. For example, the
Dopamine Transporter gene (DAT1) has 7, 9, 10, 11 repeat alleles (in addition to some other
much rarer patterns). The 10-repeat allele has a fragment size of 480, while the next most
frequent, the 9-repeat allele has a fragment size of 440. We looked at all the combinations of
fragment lengths in our data, and included an indicator for all combinations with over 30
observations. A few people had missing genetic information (a maximum of 11, depending on
the measure) and so we also created dummy variables for missing data. Our results are not
sensitive to the way the genetic information is coded.
The Table shows the categories we created for each of the six genes:
GENE Description Number of Cases
DRD4_475 =1 if both alleles are 475, =0 otherwise 468
DRD4_619 =1 if both alleles are 619, =0 otherwise 47
DRD4_475_619 =1 if allele A is 475 and allele B is 619, =0 otherwise 260
DRD4_427 =1 if either allele is 427, =0 otherwise 72
DRD4_379_475 =1 if allele A is 379 and allele B is 475, =0 otherwise 123
DRD4_379_619 =1 if allele A is 379 and allele B is 619, =0 otherwise 39
=1 if allele A is 475 and allele B is not 475 or 619, =0
DRD4_ALL_OTHERS a =1 if none of the above, =0 otherwise 34
DRD4_MISSING =1 if both alleles are missing, =0 otherwise 11
Total number of cases 1117
DAT1_440 =1 if both alleles are 440, =0 otherwise 57
DAT1_480 =1 if both alleles are 480, =0 otherwise 639
DAT1_440_480 =1 if allele A is 440 and allele B is 480, =0 otherwise 374
DAT1_ALL_OTHERS a =1 all the others, =0 otherwise 42
DAT1_MISSING =1 if both allele are missing, =0 otherwise 5
Total number of cases 1117
MAOA351 =1 if both alleles are 351, =0 otherwise 609
MAOA321 =1 if both alleles are 321, =0 otherwise 459
MAOA_ALL_OTHERS a =1 both alleles are either 291, 336, or 381, =0
MAOA_MISSING =1 if both alleles are missing , =0 otherwise 15
Total number of cases 1117
(Dopamine D2 Receptor)
DRD2_178 =1 if both alleles are 178, =0 otherwise 601
DRD2_304 =1 if both alleles are 304, =0 otherwise 90
DRD2_178_304 a =1 if allele A is 178 and allele B is 304 414
DRD2_MISSING =1 if both alleles are missing, =0 otherwise 12
Total number of cases 1117
(Cytochrome P450 2A6)
CYP_1 =1 if both alleles are 1, =0 otherwise 1056
CYP_1_2 a =1 if allele A is 1 and allele B is 2, =0 otherwise 51
CYP_MISSING =1 if both alleles are missing, =0 otherwise 10
Total number of cases 1117
SLC484 =1 if both alleles are 484, =0 otherwise 229
SLC528 =1 if both alleles are 528, =0 otherwise 359
SLC484_528 a =1 if allele A is 484 and allele B is 528 521
SLC_MISSING =1 both alleles are missing, =0 otherwise 8
Total number of cases 1117
a Omitted category in the regressions.