Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Long-term impacts of childhood adversity on health and human capital
(USC Thesis Other)
Long-term impacts of childhood adversity on health and human capital
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
LONG-TERM IMPACTS OF CHILDHOOD ADVERSITY ON HEALTH AND
HUMAN CAPITAL
by
Laura Esperancilla Henkhaus
__________________________________
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(HEALTH ECONOMICS)
August 2019
Copyright 2019 Laura Esperancilla Henkhaus
For everyone who was once a child with a story to be told
i
Acknowledgments
I conducted the research presented here with the advice and support of many
people. Firstly, I am incredibly grateful to my dissertation chair, Darius
Lakdawalla, for his support from the time I first discussed with him my plans
to pursue a PhD. He provided not only technical training through timely advice
but also guidance on how to communicate my research – both of which were
critical to the development of the research presented here and to my
development as a researcher. I thank Darius for his patient attention in
discussion of my goals and thoughtful consideration of how to support my
success.
I am also grateful for the active support of the rest of my dissertation
committee – Geoffrey Joyce, Dana Goldman, and Robynn Cox. Each provided
critical feedback as well as guidance on how to tell a story with my research
for greatest public health and policy impact. I thank Geoff for timely feedback
on my work and for his constant support throughout my graduate training. I
am immensely grateful to Dana for pushing me to continue my research on
childhood sexual abuse despite empirical challenges and for believing in the
importance of this work. Thanks also to Dana for encouraging me to apply for
an AHRQ dissertation grant, which funded the research presented here. I thank
Robynn for providing her expertise in labor economics and social determinants
of health, and I am grateful to her for directing me toward valuable
opportunities for my development as a researcher.
I thank John Romley, who provided feedback on the earliest piece of this
dissertation and continued to lend his expertise and attention to technical
details as a member of my qualifying exam committee. Beyond his service on
my committee, I am grateful to John for being a generous teacher since we first
ii
met – after I had recently earned my undergraduate degree and was working
as a research associate on his team.
In addition, I thank Steven Fox for clinical input on the design of this
research as a member of my qualifying exam committee. I am also grateful to
Gauri Kolhatkar for providing her clinical expertise on childhood trauma and
input on interpretation of findings and implications for health care practice. I
thank Gauri for her reliable and generous support.
Thanks to Rebecca Myerson, Erin Trish, Mireille Jacobsen, and Alice
Chen for feedback during my seminars and advice on the job market. I thank
Sarah Axeen for helping me navigate through grant-writing, the job market
process, and for offering encouraging words as my office neighbor who put up
with many days of heavy sighs.
I am grateful to my friends Emmanuel Drabo and Gwyn Pauley for their
solid support, feedback on my research, and companionship.
I thank my colleagues at the Schaeffer Center for supporting both my
development as a researcher and my well-being—especially Bryan Tysinger,
Jillian Wallis, Laura Gascue, and Patty St. Clair. I am immensely grateful to my
colleague Michelle Ton for being a strong supporter, a good friend, and being
ever-willing to join me in search of a treat. I thank Lucy Broni, our office
building security guard, for being my companion over the weekends and
laughing with me. Thanks to Sara Geiger, Renelle Davis, and Hanh Nguyen for
their kind support and timely assistance.
I also thank my parents, Bob and Coraine, for never satisfying my
curiosities as a child – instead answering my questions with, “What do you
think, Laura?” and leaving me to pursue what I believe is important.
Above all, I thank my sisters, Alicyn and Michelle, for their relentless
love and support.
iii
I gratefully acknowledge funding from a dissertation grant from the
Agency for Healthcare Research and Quality (R36HS026444-01), a Schaeffer-
Amgen fellowship, the USC School of Pharmacy, and the Leonard D. Schaeffer
Center for Health Policy and Economics. I thank Leonard and Pamela Schaeffer
for their generous financial support which enabled the quality of experience I
had as a researcher at the Schaeffer Center including the relationships I built
there.
This research uses data from Add Health, a program 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. Special acknowledgment is
due to 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). No direct
support was received from Grant P01-HD31921 for this analysis.
iv
Contents
Introduction .................................................................................................................................. 1
1.1 Prior literature on long-term impacts of early life conditions ................................ 1
1.2 The Adverse Childhood Experiences Study ..................................................................... 5
1.3 Objectives of this dissertation ............................................................................................... 6
Data ................................................................................................................................................... 9
2.1 Key measures .............................................................................................................................. 10
2.1.1 Adverse childhood experiences .......................................................................................... 10
2.1.2 Family background and geographic information ....................................................... 11
2.2 Sample characteristics ............................................................................................................ 12
2.3 Missing data and imputation ............................................................................................... 12
The lasting consequences of childhood sexual abuse on human capital .......... 13
3.1 Introduction ................................................................................................................................ 14
3.2 Literature...................................................................................................................................... 19
3.3 Sample characteristics ............................................................................................................ 23
3.4 Methods ......................................................................................................................................... 26
3.4.1 Outcomes measures ................................................................................................................. 27
3.4.2 Baseline model ........................................................................................................................... 27
3.4.3 Robustness checks .................................................................................................................... 29
3.4.3.1 Sensitivity analyses .................................................................................................................. 29
3.4.3.2 Bounding effects of childhood sexual abuse ................................................................. 30
3.4.3.3 Sibling fixed effects regression ........................................................................................... 34
3.4.3.4 Falsification test ........................................................................................................................ 36
3.4.4 Heterogeneity analyses .......................................................................................................... 37
3.4.5 Determinants of earnings ...................................................................................................... 37
3.5 Results ........................................................................................................................................... 38
3.5.1 Baseline results .......................................................................................................................... 38
3.5.2 Robustness checks .................................................................................................................... 42
3.5.2.1 Sensitivity analyses .................................................................................................................. 42
3.5.2.2 Bounding effects of childhood sexual abuse ................................................................. 44
v
3.5.2.3 Sibling fixed effects regression ........................................................................................... 48
3.5.2.4 Falsification test ........................................................................................................................ 49
3.5.3 Heterogeneity analyses .......................................................................................................... 50
3.5.4 Determinants of earnings ...................................................................................................... 52
3.6 Discussion and conclusions .................................................................................................. 53
Childhood Abuse and Adult Health: A Comprehensive Examination
across Health Conditions ....................................................................................................... 58
4.1 Introduction ................................................................................................................................ 59
4.2 Sample characteristics ............................................................................................................ 61
4.3 Methods ......................................................................................................................................... 63
4.3.1 Exposure measure: childhood abuse ............................................................................... 64
4.3.2 Outcome measures ................................................................................................................... 64
4.3.3 Baseline model ........................................................................................................................... 65
4.3.4 Robustness checks .................................................................................................................... 66
4.3.4.1 Sensitivity analyses .................................................................................................................. 66
4.3.4.2 Childhood abuse, family socioeconomic status, and health: a
decomposition ............................................................................................................................ 68
4.3.4.3 Sibling fixed effects regression ........................................................................................... 71
4.3.5 Heterogeneity analyses .......................................................................................................... 72
4.4 Results ........................................................................................................................................... 72
4.4.1 Descriptive results .................................................................................................................... 72
4.4.2 Baseline results .......................................................................................................................... 74
4.4.2.1 Health ............................................................................................................................................. 74
4.4.2.2 Health care access ..................................................................................................................... 76
4.4.3 Robustness checks .................................................................................................................... 77
4.4.3.1 Sensitivity analyses .................................................................................................................. 77
4.4.3.2 Childhood abuse, family socioeconomic status, and health:
a decomposition ........................................................................................................................ 78
4.4.3.3 Sibling fixed effects regression ........................................................................................... 79
4.4.4 Heterogeneity analyses .......................................................................................................... 81
4.5 Discussion and conclusions .................................................................................................. 83
The Child Left Behind: Parental Incarceration and Adult Human Capital in
vi
the United States ....................................................................................................................... 86
5.1 Introduction ................................................................................................................................ 87
5.2 Sample characteristics ............................................................................................................ 90
5.3 Methods ......................................................................................................................................... 92
5.3.1 Exposure measure: parental incarceration ................................................................... 92
5.3.2 Outcome measures ................................................................................................................... 93
5.3.3 Baseline model ........................................................................................................................... 93
5.3.4 Robustness checks .................................................................................................................... 95
5.3.4.1 Bounding effects of parental incarceration ................................................................... 95
5.3.4.2 Falsification tests ...................................................................................................................... 98
5.3.5 Heterogeneity analysis ........................................................................................................... 99
5.4 Results ........................................................................................................................................ 100
5.4.6 Baseline results ....................................................................................................................... 100
5.4.7 Robustness checks ................................................................................................................. 103
5.4.7.1 Bounding effects of parental incarceration ................................................................ 103
5.4.7.2 Falsification tests ................................................................................................................... 104
5.4.8 Heterogeneity analysis ........................................................................................................ 106
5.5 Discussion and conclusions ............................................................................................... 106
Concluding remarks .............................................................................................................. 109
6.1 Synthesis of key findings .................................................................................................... 109
6.2 Implications for policy and health care delivery ...................................................... 111
References ................................................................................................................................. 113
Appendices ............................................................................................................................... 122
Appendix A: Chapter 3 appendix tables ...................................................................... 122
Appendix B: Chapter 4 appendix tables....................................................................... 138
Appendix C: Chapter 5 appendix tables ....................................................................... 162
Appendix D: Multiple imputation ................................................................................... 184
1
Introduction
In a growing body of literature, researchers have examined how prenatal and
early life insults and supports impact later-in-life outcomes. Much of the
earlier work focused on prenatal conditions, and this body of literature has
come to be referred to as research on the fetal origins hypothesis (Almond and
Currie 2011). Over the past decade, there has been increasing study of long-
term impacts of conditions and events during childhood (Almond, et al. 2018).
The present dissertation adds to the literature in this latter group, filling an
important gap to understand the effects of childhood adversity – namely,
childhood abuse and parental incarceration – on health and human capital in
adulthood.
1.1 Prior literature on long-term impacts of early life
conditions
Extant literature on childhood circumstances and later-in-life outcomes have
examined several types of early life conditions. Topics studied largely fall into
four main domains: changes in household resources, health interventions,
environmental conditions, parental investments, and adverse childhood
experiences (ACEs). For a more thorough review of all but the last topic on
ACEs, see Almond, Currie, and Duque (2018).
In a large body of literature, researchers have investigated both the
effects of social programs which expanded household resources as well as the
effects of changes in local or household economic conditions. For example,
2
prior work has measured positive effects on children whose families received
food stamps (Hoynes, et al. 2016), conditional cash transfers (Fiszbein and
Schady 2009), and cash to single-mother households through the US Mother’s
Pension program of the early 1930’s (Aizer, et al. 2016). Several researchers
have measured positive impacts on children who received Medicaid (Gaudette,
et al. 2017). In other work, researchers found that preschool at no cost to
families had durable impacts on later childhood and adult outcomes: higher
scores on a standardized academic test at age 14, greater likelihood of adult
employment, and fewer arrests (Conti, et al. 2016, Heckman, et al. 2010,
Heckman, et al. 2013).
Prior research demonstrated large impacts of local economic conditions
on children’s mental well-being but small or no effects of parental job loss on
children’s health and human capital outcomes. On local economic conditions,
research on the effects of recessions in the US has shown that higher
unemployment rates led to worse mental health in children and higher
likelihood of using special education services for emotional problems
(Golberstein, et al. 2016). Research on changes in family economic conditions,
leveraging short panels and an individual fixed effects strategy, has shown that
parental job loss in childhood had no effect on children’s health in the short-
term, as measured by parent-reported child diagnoses, but did lead to greater
probability of the parent respondent designating child health as fair or poor
(Schaller and Zerpa 2019). In other work, exploiting variation in timing when
children were exposed to parental layoff, Hilger found that layoffs during
adolescence reduced annual college enrollment by just one percent and had no
impact on children’s earnings through age 25 (Hilger 2016).
3
Previous work has measured the benefits of childhood health
interventions on both health and human capital outcomes. Fitzsimons and
Vera-Hernández concluded that breast feeding improved child cognition
scores, leveraging variation in provision of breast feeding training by day of
birth in the United Kingdom (Fitzsimons and Vera-Hernández 2013). Others
have found that intensive care for low birth weight babies reduced infant
mortality rates and increased earnings among those who survived to adulthood
(Bharadwaj, et al. 2013). In Kenya, researchers have measured positive effects
of deworming medications for school-going children (Miguel and Kremer
2004).
Evaluations of impacts of environmental policies have shown positive
impacts on children’s human capital outcomes in both developed and less
developed countries. Isen, Rossin-Slater, and Walker have concluded that
children born in areas that experienced pollution reduction due to the US Clean
Air Acts had higher earnings as adults, as a result of the improved air quality
(Isen, et al. 2017). Using data from Mexico, Beach and colleagues showed that
water purification improved educational attainment and later earnings of
surviving children (Beach, et al. 2016).
There is little literature evaluating the impacts of parental investments
on child outcomes. The main barrier to research in this area is disentangling
the effect of an overall high-quality parent from the effect of the particular
investment. In the only study, to my knowledge, which evaluated the impact
of parental investment on child outcomes, Fryer, Levitt, and List, in
unpublished work, concluded that parental attendance at an early-childhood
parenting academy led to improved self-regulation in the children but no better
scores on the Peabody Picture Vocabulary Test (Fryer, et al. 2015). The authors
4
reached this conclusion by using random assignment of financial rewards for
attendance at the parent academy to instrument for parental attendance.
Because parents could earn up to $7,000 per year for attendance at the early-
childhood sessions – in a school district where per capita income was less than
$18,000 – it is unclear whether children’s improved behavioral outcomes
resulted from parental attendance at the academy or increased family
resources.
Over the past two decades, there has been increasing attention from
researchers, clinicians, public health practitioners, and policymakers on
“adverse childhood experiences,” or “ACEs.” There is a large and growing body
of literature on ACEs and later-in-life outcomes, generally focusing on
household-level adversity beyond poverty. While the goal of this literature is
to understand the long-term impacts of childhood adversity and what mediates
negative outcomes, research has largely been correlational due to limited
individual-level data and lack of opportunity for natural experiments. The
primary challenge is to disentangle the effect of the particular experience from
other types of adversity, such as poor family economic conditions, which may
be correlated with probability of facing the experience. Thus, despite a
growing body of descriptive research in public health and social welfare
literature, there is a paucity of literature identifying the impacts of adverse
childhood experiences on children’s outcomes. Among this literature, Saarela
and Elo, studying Finnish adults aged 43-84, found no impacts of forced
migration in childhood on receipt of benefits for short-term sickness, disability
pension for long-term illness, or mortality, exploiting the creation of a new
border during World War II which forced 12 percent of the Finnish population
to migrate (Saarela and Elo 2016). Finlay and Neumark investigated the effect
of never-married motherhood, which might be hypothesized as an adverse
5
condition, on children’s outcomes, exploiting differences in state-incarceration
rates of males in mothers’ “marriage markets” defined by demographics. The
authors concluded that never-married motherhood led to lower high school
dropout rates for Hispanic children (but there was no consistent evidence of
effects on Black children) (Finlay and Neumark 2010). Currie and Tekin, using
sibling fixed effect models and a nationally representative US sample
(employed in this dissertation and discussed below), suggested that childhood
maltreatment led to greater probability of engaging in crime (Currie and Tekin
2012).
1.2 The Adverse Childhood Experiences Study
Increased interest in understanding the long-term impacts of childhood
adversity was spurred by the first publication, in 1998, disseminating results
from the Adverse Childhood Experiences Study, conducted jointly by the US
Center for Disease Control and Prevention and Kaiser Permanente (Felitti, et
al. 1998). The CDC-Kaiser ACE Study relied on medical records from enrollees
in a local Kaiser Health Plan and responses from consenting members to mailed
questionnaires about childhood abuse, neglect, and household adversity.
Specifically, the questionnaire, by the second wave of survey data collection,
assessed ten categories of ACEs: sexual, physical, or emotional abuse; physical
or emotional neglect; having a household member who had been incarcerated,
abused drugs, or was mentally ill; exposure to domestic violence against the
mother; and parental separation. The CDC-Kaiser ACE Study showed, firstly,
that the prevalence of adverse childhood experiences was high – in their
sample from a San Diego, California Kaiser Health Plan, with on average higher
socioeconomic status than reported from nationally representative US data.
6
For example, among the almost 18,000 participants of the CDC-Kaiser ACE
Study, 20.7 percent reported contact childhood sexual abuse (24.7 percent of
women and 16.0 percent of men), 28.3 percent reported physical abuse, 10.6
percent reported emotional abuse, and 4.7 percent reported incarceration of a
household member (Dong, et al. 2003). Secondly, the CDC-Kaiser ACE Study
showed a graded relationship between the number of ACE categories
experienced and risk of mental and physical health outcomes – comparing
participants who reported at least four ACEs to those who reported none
(Felitti, et al. 1998). While the CDC-Kaiser ACE Study clearly showed that
survivors of ACEs had worse health in adulthood, due to the cross-sectional
design and related lack of control for any measure of childhood socioeconomic
status, it is impossible to conclude from this work whether poverty alleviation
or interventions focused on specific childhood experiences would best alleviate
the health burden faced by people who have survived high levels of childhood
adversity.
1.3 Objectives of this dissertation
To fill important gaps in our understanding of the impacts of adverse childhood
experiences, this dissertation has three primary objectives:
(1) To evaluate the impact of childhood sexual abuse on adult educational
attainment and labor market outcomes;
(2) To study the long-term effects of childhood abuse on adult health and to
examine survivors’ access to health care; and
(3) To assess the impact of parental incarceration on adult educational
attainment and labor market outcomes.
7
To address each of these objectives, I used rich survey data from the National
Longitudinal Study of Adolescent to Adult Health (Add Health), which I
describe in Chapter 2.
In Chapter 3, I addressed Objective 1 by constructing likely bounds on
the size of the effects of childhood sexual abuse on high school diploma receipt,
college degree attainment, employment, and earnings in adulthood. To do this,
I implemented partial identification methods developed by Altonji, et al. (2002,
2005) and Oster (2017). This chapter fills a largely neglected area of research
on the economic well-being of survivors of childhood abuse.
In Chapter 4, I addressed Objective 2 by conducting a systematic
assessment of the effects of childhood abuse on adult heath and by examining
survivors’ likelihood of being uninsured and reporting unmet medical need. I
studied health using self-reports of diagnoses, health measurements where
available for some conditions, and reports of all prescription medications used
in the prior four weeks. I collapsed information on health and medication use,
separately, into the following six categories: cancer, cardiometabolic
conditions, gastrointestinal issues, nervous system conditions, and
respiratory/allergic conditions. Four main study attributes are novel here.
First, I examined reports of an extensive set of diagnoses of chronic conditions
and acute symptoms along with all recent prescription medication use in one
sample – a US nationally representative sample. Second, I utilized health
measurements to supplement the survey data and evaluate whether there were
true disparities in health vs. disparities in diagnosis. Third, by using rich,
longitudinal survey data, I controlled for finer levels of childhood
socioeconomic status and neighborhood factors, other childhood adversity, and
conducted robustness checks to evaluate whether childhood abuse was
8
independently predictive of particular health conditions. Fourth, this work
provides the first assessment of health care access of survivors of childhood
abuse.
In Chapter 5, I addressed Objective 3 by calculating bounds on the effects
of parental incarceration on adult education and labor market outcomes,
applying similar methods as used in Chapter 3 to study impacts of childhood
sexual abuse. This work is the first to quantify population-level effects of
parental incarceration on children’s educational attainment, likelihood of
employment, and earnings in adulthood.
I conclude with Chapter 6 by discussing the main results of this
dissertation and policy implications.
9
Data
To answer the questions posed in this dissertation, I utilized restricted-use
data files from the National Longitudinal Study of Adolescent to Adult Health
(Add Health), which recruited a random sample of children in grades 7 to 12
(Harris 2013). Add Health has been previously used to describe relationships
between childhood abuse and adult physical health (Duncan, et al. 2015a,
Fuemmeler, et al. 2009, Gooding, et al. 2014, Haydon, et al. 2011, Richardson,
et al. 2014, Shin and Miller 2012, Suglia, et al. 2014) and mental health (Dunn,
et al. 2013, Exner-Cortens, et al. 2013, Fletcher 2009, Fletcher 2010, Foster, et
al. 2008, Gaston 2016, Hahm, et al. 2010, McLaughlin, et al. 2012, Shafa 2016,
Zeglin, et al. 2015). Recruitment occurred first by sampling secondary schools
from a national database, then by sampling students within the schools. The
sample includes 132 schools – a mix of public and private schools (Harris, et
al. 2006). Some groups were oversampled, such as minority racial groups,
disabled adolescents, and siblings (Harris 2013). Thus, I used survey weights
in analyses so that results reflect a nationally representative sample.
I used data on participants from Waves I (children and parents
interviewed separately; collected 1994-1995 when participants were aged 11-
18 years old) and Wave IV (2008-2009, 24-32 years old
1
) as well as data from
Wave II (1996, 12-19 years old) and Wave III (2001-2002, 18-26 years old) to
replace missing data where possible and appropriate. The full sample which
completed both Wave I and Wave IV interviews consisted of roughly 15,000
individuals. The unweighted response rate for Wave IV was 80 percent.
Analyses from Add Health study staff indicated that nonresponse bias was
1
Fifty-two respondents were aged 33-34 years old at time of Wave IV interview.
10
negligible and that participants in Wave IV were representative of the original
cohort recruited in Wave I (Brownstein, et al. 2011).
2.1 Key measures
2.1.1 Adverse childhood experiences
A key feature of Add Health is that, while the survey was largely administered
through computer-assisted personal interview, questions on sensitive topics
such as childhood maltreatment were completed through audio computer-
assisted self-interview. Self-interview methods have been found to capture
higher rates of sexual and drug-related behaviors than measured from face-to-
face interviews (Midanik and Greenfield 2008, Perlis, et al. 2004).
I designated history of childhood abuse for any individual who reported
childhood sexual abuse, physical abuse, or chronic emotional abuse. I defined
the sample of adults who experienced childhood sexual abuse as those who
reported either that they experienced nonconsensual sexual touching or sexual
relations by an adult caregiver before or forced sexual activity by a non-
caregiver before age 18. Childhood physical abuse reflects report in self-
interview of at least one time to the question, “Before your 18th birthday, how
often did a parent or adult caregiver hit you with a fist, kick you, or throw you
down on the floor, into a wall, or down stairs?” Emotional abuse reflects the
maximum count, at least ten times, to the question in self-interview: “Before
your 18th birthday, how often did a parent or other adult caregiver say things
that really hurt your feelings or made you feel like you were not wanted or
loved?”
11
I defined parental incarceration as report in Wave IV via personal
interview of either biological parent, mother figure, or father figure being
incarcerated while the child was alive but not yet 18-years-old. In constructing
the parental incarceration measure, I categorized responses of “don’t know”
as such rather than designating these values as missing. The “don’t know”
group consists of (i) respondents selecting “don’t know” in the initial question
of whether the parent had been incarcerated, which was largely due to the
parent’s absence in the respondent’s life
2
and (ii) respondents who
acknowledged that a parent had been incarcerated but responded “don’t know”
to questions on age when parent was incarcerated.
2.1.2 Family background and geographic information
I utilized information on race/ethnicity, childhood family socioeconomic
status, and the places where participants went to school. I categorized
race/ethnicity as White non-Hispanic, Black non-Hispanic, Native American
non-Hispanic, Asian/Pacific Islander non-Hispanic, Other non-Hispanic
(includes multi-racial group), and Hispanic. For the remainder of this paper, I
will omit the “non-Hispanic” label for these groups. I used parent-reported
household income and highest parental educational attainment as measures of
family-level socioeconomic status. And, I utilized pseudo-identifiers for the
middle schools that participants attended.
2
The majority of this group also responded “don’t know” to the preceding question of whether
the parent was still alive.
12
2.2 Sample characteristics
The mean age of respondents was 28.8 years when outcomes were assessed in
the present study. In the self-interview module, 14.1 percent of adults reported
history of contact sexual abuse, 17.8 percent reported physical abuse by a
caregiver, and 12.0 percent reported chronic emotional abuse by a caregiver.
Further, 11.7 percent experienced at least part of childhood with a parent
incarcerated. More than one-third of the full sample – 36.2 percent – reported
at least one of these four adverse childhood experiences.
2.3 Missing data and imputation
Because I relied on longitudinal survey data to conduct this research, it was
important to examine the extent and nature of missing data. The rates of
missing data were generally low except for one key variable only available for
children whose parents participated in interviews: childhood household
income. I used multiple imputation for analysis, as described in Appendix D.
The sample who completed Wave I and Wave IV interviews, with survey
weights calculated by Add Health, consists of 15,642 people. The rates of
missing data were less than 2 percent for all variables required to implement
the baseline models described below – except for childhood household income.
The rate missing for childhood household income was 24 percent.
13
The lasting consequences of childhood sexual
abuse on human capital
Abstract
Scientists posit neurobiological mechanisms explaining effects of chronic
childhood stress on physiological systems and cognitive development. Here, I
examine whether there are durable effects of childhood sexual abuse on human
capital. Using the National Longitudinal Study of Adolescent to Adult Health, I
employed a school fixed effects strategy and utilized rich information on
childhood socioeconomic status and neighborhood factors. I implemented
partial identification methods to examine robustness of results to varying
assumptions about remaining selection on unobservables, using information
on selection on observables. In this nationally representative US survey, 14
percent of respondents noted history of contact sexual abuse in childhood. I
provide evidence that childhood sexual abuse led to lower educational
attainment and worse labor market outcomes in adulthood. Results suggest
that the effect of childhood sexual abuse on likelihood of college degree
attainment was smaller for Hispanics and Blacks than for Whites. This study
has important implications for public health and public policy, highlighting the
importance of detection and quality treatment of trauma symptoms. In
particular, results suggest that only treating the mental health symptoms of
survivors of childhood sexual abuse is not enough to reduce disparities in well-
being.
14
3.1 Introduction
Childhood sexual abuse is a public health crisis. Meta-analyses collectively
covering all continents
3
show that childhood sexual abuse is a worldwide
problem and suggest that the global prevalence is 15.0 to 19.7 percent among
females and 7.6 to 8.0 percent among males (Barth, et al. 2013, Pereda, et al.
2009, Sanjeevi, et al. 2018, Stoltenborgh, et al. 2011). In the United States,
history of caregiver childhood sexual abuse was reported by 16.3 percent of
women and 6.7 percent of men in the 2011-2014 surveys of the Behavioral Risk
Factor Surveillance System (BRFSS), among 23 states which participated in the
ACE module (Merrick, et al. 2018). While childhood sexual abuse directly
impacts a great number of people, we know little about the causal impacts on
survivors beyond the immediate trauma. Main barriers to research in this area
have been a lack of longitudinal, individual-level data identifying survivors of
sexual abuse and difficulty overcoming contamination from selection. Yet,
measuring the impact on survivors is crucial to understanding their well-being
across a variety of domains and the societal costs of sexual abuse.
Over the past two decades, there has been increasing attention from
researchers, clinicians, public health practitioners, and policymakers on
“adverse childhood experiences,” or “ACEs.” The increased attention to this
field was spurred by the first publication, in 1998, disseminating results from
the Adverse Childhood Experiences Study, conducted jointly by the US Center
for Disease Control and Prevention and Kaiser Permanente (Felitti, et al. 1998).
The CDC-Kaiser ACE Study relied on medical records from enrollees in a local
Kaiser Health Plan and responses from consenting members to mailed
3
Antarctica was not included.
15
questionnaires about childhood abuse, neglect, and household adversity.
Specifically, the questionnaire, by the second wave of survey data collection,
assessed ten categories of ACEs: sexual, physical, or emotional abuse; physical
or emotional neglect; having a household member who had been incarcerated,
abused drugs, or was mentally ill; exposure to domestic violence against the
mother; and parental separation. The CDC-Kaiser ACE Study showed firstly
that the prevalence of adverse childhood experiences was high – in their
sample from a San Diego, California Kaiser Health Plan, with on average higher
socioeconomic status than reported from nationally representative US data.
For example, among the almost 18,000 participants of the CDC-Kaiser ACE
Study, 20.7 percent reported contact childhood sexual abuse (24.7 percent of
women and 16.0 percent of men), 28.3 percent reported physical abuse, 10.6
percent reported emotional abuse, and 4.7 percent reported incarceration of a
household member (Dong, et al. 2003). Secondly, the CDC-Kaiser ACE Study
showed a graded relationship between the number of ACE categories
experienced and risk of mental and physical health outcomes – comparing
participants who reported at least four ACEs to those who reported none
(Felitti, et al. 1998).
Following the CDC-Kaiser ACE Study, similar questions on adverse
childhood experiences have been added to several surveys such as an optional
module in the state-administered BRFSS and the National Longitudinal Study
of Adolescent to Adult Health (Add Health), with many public health papers
mimicking the method of studying ACEs as a count exposure. Yet, factor
analysis of data from the BRFSS reveal three distinct groups: sexual abuse,
physical or emotional abuse, and household adversity (Ford, et al. 2014).
Further, best practices to support children who have experienced adversity
16
would likely differ for children who were sexually abused versus those who
experienced parental divorce, for example.
In the present study, I focused on sexual abuse in childhood to study this
especially vulnerable population. I investigated whether there were durable
effects of childhood sexual abuse on human capital. Specifically, I examined
the effects of childhood sexual abuse on educational attainment, employment,
and earnings in adulthood. Findings might be useful to clinicians, educators,
and policymakers in a variety of ways. Results might motivate research by
clinicians to develop best practices for screening for trauma symptoms in
children and improved practices for treatment of sexual trauma, addressing
potential consequences of abuse on children’s cognitive development and
productivity. In the education system, results might spur efforts to screen for
symptoms of childhood trauma and to provide counseling opportunities on
school campuses. In addition, results might support policymaker decisions to
allocate resources to prevention of childhood sexual abuse and the treatment
of trauma symptoms.
The primary consequences of childhood sexual abuse on immediate
trauma symptoms are well understood. History of childhood sexual abuse has
been associated with higher rates of anxiety, depression, panic disorder, and
post-traumatic stress disorder (PTSD) in adulthood (Fletcher 2009, Sachs-
Ericsson, et al. 2010). Analyses of nationally representative US data show that
history of childhood sexual abuse was associated with much higher risk of
lifetime suicide attempt in adults (Hoertel, et al. 2015, Pérez-Fuentes, et al.
2013). Despite descriptive evidence that survivors of childhood sexual abuse
had poorer health, whether there are durable, causal effects on human capital
and the magnitude of such effects remain unknown.
17
There are several possible mechanisms through which childhood sexual
abuse might impact education and labor market outcomes. There may be direct
effects on human capital through reduced mental and physical health (Bucci,
et al. 2016, Goodman, et al. 2011). In addition, childhood sexual abuse may
indirectly limit labor market opportunities through higher rates of arrest due
to violence (Herrenkohl and Jung 2016) or illegal drug use (Huang, et al. 2011)
or through a higher rate of teen pregnancy (Anda, et al. 2001). Further,
childhood sexual abuse may lead to preferences for jobs which are not in high-
pressure environments, considering that high rates of survivors already suffer
mental health struggles, and for jobs that are not in male-dominated fields,
given that the perpetrators are predominately male (Black, et al. 2011). Such
jobs are generally in lower-wage categories. This discussion of potential
pathways from childhood sexual abuse to worse human capital outcomes is not
exhaustive, and there may be ways by which childhood sexual abuse, while
having devastating psychological consequences, fosters long-term resiliency.
In the psychology field, the term “post-traumatic growth” describes the
phenomenon that, in some people, terrible adversity has stimulated
remarkable strengthening (Tedeschi, et al. 1998, Woodward and Joseph 2003).
In the empirical literature, a study by Feeney and colleagues using the Irish
Longitudinal Study on Ageing (TILDA) showed that despite still having higher
rates of clinical depression and anxiety in middle-age, the survivors of
childhood sexual abuse also had higher cognition scores (Feeney, et al. 2013).
Higher cognitive performance would promote better human capital outcomes
in the survivors. The ultimate result of potentially many processes spurred by
childhood sexual abuse, as hypothesized above, remains an open and important
question.
18
The main challenge to identifying the causal effects of childhood sexual
abuse is selection bias – which may occur at the neighborhood, family, or child
level. For example, growing up in a neighborhood of lower socioeconomic
status (SES) may cause a child both to be at higher risk of sexual abuse and to
attend lower quality schools, thus diminishing labor market opportunities.
Being born to parents who are negligent or who in general do not invest in
their children may make those children more vulnerable to abuse and suppress
their cognitive development. In addition, having low self-worth or engaging in
risky behaviors as a child may make the child both more likely to experience
childhood sexual abuse and less likely to succeed in school or on the labor
market. Studying childhood sexual abuse is challenging because there is little
opportunity for quasi-experimental design. And, individual fixed effects
regression is not possible because no data exists from community samples with
repeated, prospectively collected assessments of sexual abuse in childhood;
this would be an intervention.
To examine the long-term effects of childhood sexual abuse on human
capital, I used US national survey data (Add Health) and took several
approaches to address and evaluate selection. By utilizing large, population-
level data, I was able to study the heterogeneity of response – including among
men, who are hugely understudied in the literature on sexual trauma. To study
the causal effects of childhood sexual abuse, I controlled for finer levels of
socioeconomic status and school fixed effects, and I calculated likely bounds
on the effects of childhood sexual abuse by making explicit assumptions about
selection on unobservables. I show that my approach sweeps out any bias from
unobservables that are correlated with other forms of adverse childhood
experiences. I implemented partial identification methods developed by
Altonji, et al. (2002, 2005) and Oster (2017) to construct bounds on the effects.
19
I found strong evidence that childhood sexual abuse led to lower rates of high
school diploma and college degree attainment and some evidence that
childhood sexual abuse led to lower rates of full-time employment and lower
earnings.
The remainder of this chapter proceeds as follows. In Section 3.2, I
discuss the literature on childhood abuse and long-term outcomes, focusing on
human capital. In Section 3.3, I present sample characteristics. In Section 3.4,
I describe the empirical strategy, and in Section 3.5, I present results. In
Section 3.6, I conclude and discuss practical implications.
3.2 Literature
A large body of work has explored neurobiological mechanisms explaining
effects of chronic childhood stress on physiological systems, behavior, and
cognitive development. Still, the etiologic pathways through which the effects
of intense childhood stress become embedded in the body and brain are not
fully understood. Neuroscience literature describes dysregulated functioning
during a toxic stress response, which is characterized by a prolonged state of
“fight or flight” mode during which executive functions such as control of
attention, inhibition, and emotion regulation are impaired (Bucci, et al. 2016).
In other work examining the negative relationship between childhood poverty
and academic achievement, Hair and colleagues concluded that differences in
childhood brain development played a mediating role. In this prospective
study, 15 to 20 percent of the achievement gap across low- and higher-income
children was explained by differences in development of the frontal and
temporal lobes of the brain, which are responsible for executive functions
including complex learning, memory, and language comprehension (Hair, et
20
al. 2015). While it is unknown to what extent these differences by childhood
socioeconomic status are explained by particular circumstances associated
with poverty, it is known that children from lower income households are at
higher risk for abuse (Merrick, et al. 2018).
The neuroscience literature is accompanied by a growing body of social
science literature attempting to test and quantify some of the posited
relationships. Overall, this literature is plagued with selection problems, and
almost all studies of outcomes of survivors of childhood abuse are
correlational. And, the literature has largely neglected to measure the potential
consequences of childhood abuse on cognitive development, productivity, and
later-life economic well-being. Beyond the lack of or inadequate approaches to
addressing selection problems, the small body of work on childhood abuse and
adult economic welfare suffers from other important limitations, such as use
of abuse measures with low sensitivity and study of narrow populations.
Studies in this body of work either use survey data or official child welfare
records of abuse to identify the exposed group. Comparison of the prevalence
of childhood abuse across survey data and child welfare records reveals that a
small percentage of occurrences of childhood sexual abuse are brought to the
attention of the courts and substantiated (Finkelhor, et al. 2013, Wildeman, et
al. 2014). Thus, child welfare data have poor sensitivity in defining samples
who experienced abuse and in general include more severe cases.
4
There has
been no research measuring the independent effects of childhood abuse on
human capital outcomes in community samples.
4
In Widom et al. (1997), using a childhood sexual abuse case-control sample with cases drawn
from welfare records, face-to-face interviews in adulthood revealed that about 16 percent of
the control group (no official welfare records of child maltreatment) reported having
experienced contact sexual abuse in childhood.
21
Research in the small collection of literature on childhood abuse and
adult educational attainment and labor market outcomes is inconsistent.
Below, I discuss studies of any type of childhood abuse in any country which
made some attempt to address confounding by childhood SES (by matching on
or controlling for parental income, occupation class, or educational
attainment). There is no research, to my knowledge, which examines the
relationship between childhood abuse and adult education or labor market
outcomes in community samples while controlling for childhood SES and
neighborhood factors.
There is no evidence that childhood sexual abuse causes lower
educational attainment. While a study of the 1958 British Birth Cohort found
that sexual abuse was associated with lower educational attainment (Geoffroy,
et al. 2016), Boden and colleagues, using New Zealand data from the
Christchurch Health and Development Study (CHDS), found no significant
relationship between child sexual abuse and educational achievement once
inter-parental violence, parental drug use, and childhood poverty were
included as covariates (Boden, et al. 2007). In the US, Foster and colleagues
found that adolescent verbal intimate partner violence but not sexual abuse
before grade 6 was associated with lower probability of high school
graduation, using the public sample
5
of the same US survey data employed in
the present study (Add Health) (Foster, et al. 2008). Currie and Widom found
that women with court-substantiated records of child maltreatment (physical
abuse, sexual abuse, or physical neglect) in 1967-1971 had completed half of a
grade less than controls from the same Midwestern county matched on age,
5
The public sample of the National Longitudinal Study of Adolescent to Adult Health (Add
Health) contains roughly one-third of the full sample who participated in home interviews.
22
sex, race/ethnicity, and neighborhood via school or birth hospital (Currie and
Widom 2010).
The few studies on childhood abuse and adult employment status are
inconsistent but show worse outcomes of childhood abuse when sexual abuse
was studied. Barrett and colleagues, using data on middle-aged Irish men from
TILDA, concluded that sexual abuse was associated with higher ratios of
potential working life spent out of the labor force due to sickness or disability
(Barrett, et al. 2014). Currie and Widom, studying substantiated abuse and
neglect cases in the Midwestern US, concluded that maltreatment was
associated with lower probability of employment (Currie and Widom 2010).
Among studies of childhood abuse and adult earnings, only Currie and
Widom addressed potential confounding by childhood neighborhood
circumstances and SES in some way. They found that women with official
records of childhood maltreatment earned much less than controls matched on
demographics and neighborhood, but there were no statistical differences
detected among men across those with and without records of maltreatment.
Women in the maltreatment group earned $5,584 less per year than matched
controls, which was about 23 percent of mean earnings among the matched
pairs (Currie and Widom 2010). No research has yet studied the relationship
between childhood sexual abuse and adult earnings among US men. Studies
using national data from Ireland (TILDA) and New Zealand (CHDS) found no
association between childhood sexual abuse and adult earnings (Barrett, et al.
2014, Fergusson, et al. 2013). The present study focuses on the US context.
23
3.3 Sample characteristics
In the self-interview module, 14.1 percent of adults reported history of contact
sexual abuse in childhood. Among this group who reported childhood sexual
abuse, 54.0 percent experienced abuse by an adult caregiver, and 60.3 percent
experienced abuse by a non-caregiver. There was a substantial difference in
prevalence of childhood sexual abuse across sex. Among females, the
prevalence of childhood sexual abuse was 20.6 (s.e., 0.7) percent, and among
males, the prevalence was 7.8 (0.5) percent. There was a smaller difference,
however, by childhood household income. The prevalence of childhood sexual
abuse was 19.7 (1.0) percent among those who grew up in the bottom quintile
of household income and 12.1 (0.7) percent among those who grew up in the
top quintile of household income.
There were important differences in characteristics across the children
who eventually experienced sexual abuse and those who did not. Table 1 below
displays sample characteristics for those who did and did not report childhood
sexual abuse in Add Health. The children who experienced sexual abuse were
from more disadvantaged family backgrounds, in terms of lower household
income and lower rates of living with both biological parents. In addition, they
were more likely to experience other types of abuse or parental incarceration,
compared to those who did not experience childhood sexual abuse. There were
also differences in childhood health across these groups. The sexual abuse
group had higher rates of physical disability, learning disability or ADHD, and
depressive symptoms compared to the group reporting no childhood sexual
abuse. Still in adulthood, those who had experienced childhood sexual abuse
had much higher rates of depressive symptoms compared to those who had not
been sexually abused as children: 42 vs. 26 percent. In addition, Table 1 shows
24
the large differences in unadjusted means for education and labor market
outcomes.
25
Table 1. Sample characteristics across childhood sexual abuse history
(standard errors in parentheses)
(1) (2) (3)
No Yes
N=13,309 N=2,198
Female 0.454 (0.008) 0.718 (0.017) ***
Race/ethnicity
White NH 0.774 (0.024) 0.729 (0.031)
***
Black NH 0.147 (0.02) 0.184 (0.027)
Native American NH 0.009 (0.003) 0.014 (0.006)
Asian/Pacific Islander NH 0.028 (0.007) 0.023 (0.008)
Other NH 0.032 (0.003) 0.057 (0.009)
Hispanic 0.11 (0.016) 0.125 (0.016)
Biological parents in household
Mom and dad 0.585 (0.014) 0.404 (0.018)
***
Mom only 0.331 (0.011) 0.471 (0.017)
Dad only 0.045 (0.003) 0.049 (0.007)
Neither 0.039 (0.004) 0.076 (0.009)
Childhood household income, median
(2010$)
$58,782 $47,026
–
Child health and cognition
Physical disability 0.022 (0.002) 0.038 (0.007) ***
Educational challenge 0.155 (0.011) 0.193 (0.013) **
Depressive symptoms 0.282 (0.009) 0.43 (0.018) ***
Adverse childhood experiences
Sexual abuse 0.000 1.000 –
Physical abuse 0.147 (0.005) 0.368 (0.016) ***
Emotional abuse 0.097 (0.004) 0.261 (0.012) ***
Parental incarceration 0.104 (0.007) 0.199 (0.012) ***
Adult outcomes
Depressive symptoms 0.256 (0.008) 0.416 (0.016) ***
High school diploma 0.857 (0.01) 0.737 (0.016) ***
College degree 0.332 (0.017) 0.193 (0.015) ***
Full-time employment 0.738 (0.008) 0.621 (0.015) ***
Adult earnings, median (2010$) $30,414 $21,785 –
NOTE.— Column (1) displays variable means for the sample who reported no childhood sexual
abuse, with the median value displayed where noted. Column (2) displays this information for the
sample who reported childhood sexual abuse.
NH, Non-Hispanic
* p<0.05, ** p<0.01, *** p<0.001
26
While there were important differences across the group who did and
did not experience childhood sexual abuse, the overlapping groups who
experienced childhood sexual abuse, physical abuse, emotional abuse, or
parental incarceration were generally similar to each other by background
characteristics. (The exception is a higher proportion of females in the group
who experienced childhood sexual abuse.) See Appendix Table A1 for
descriptive statistics by type of adverse childhood experience.
3.4 Methods
I employed several strategies to evaluate whether there was a causal impact of
childhood sexual abuse on education or labor market outcomes. First, in the
baseline specification I implemented school fixed effects regression models
controlling for family-level SES and other adverse childhood experiences,
among other potential confounders. Second, I evaluated sensitivity of results
to inclusion of specific, potential confounders not preferred in the baseline
specification because these factors may also be caused by sexual abuse or are
not available for the full sample. Third, I examined robustness of results by
constructing bounds under varying assumptions about the importance of
unobservable confounders. Fourth, I implemented sibling conditional fixed
effects regressions to parse out unobserved characteristics of the family
environment. Fifth, I executed a falsification test to evaluate whether there
were pre-existing differences in cognitive ability among those who never and
who eventually experienced childhood sexual abuse. Lastly, I examined the
heterogeneity of response to childhood sexual abuse by sociodemographic
characteristics. I conclude this section by decomposing the effect of childhood
27
sexual abuse by current (adult) characteristics which might be caused by
sexual abuse.
3.4.1 Outcomes measures
I studied young adult outcomes: having a high school diploma, a college degree,
employment, full-time employment, and earnings level. I considered an
individual to be employed if working at least 10 hours per week, the minimum
amount of time inquired about in Add Health. I defined full-time employment
status as working at least 35 hours per week. Earnings included any wages,
bonuses, and self-employment.
3.4.2 Baseline model
In my baseline approach to study the effects of childhood sexual abuse on adult
human capital outcomes, I used a school fixed effects strategy while controlling
for demographics, childhood socioeconomic status, disability, and other
adverse childhood experiences. In the baseline specification below, let Y
i
denote the outcome for individual i. Y
i
is either a binary variable for having a
high school diploma, having a college degree, currently employed, currently
employed full-time, or a continuous measure of earnings level. The parameter
of interest is α in the following equation:
f(Y
i
) = 𝛼 ·ChildhoodSexualAbuse
i
+ X
i
′β + ε
i
(3.1)
28
ChildhoodSexualAbuse
i
is a dummy variable equal to one for persons who
reported childhood sexual abuse in Add Health. X
i
is a vector of controls and
includes a constant. Other components of X
i
include demographics (age when
outcome measured, sex, and race/ethnicity), childhood household
socioeconomic status (highest parental education level and household income),
school fixed effects, childhood physical disability, and other adverse childhood
experiences (physical abuse, emotional abuse, and parental incarceration).
Lastly, the model allows an individual, idiosyncratic error term ε
i
.
By including school fixed effects, I was able to parse out unobserved
characteristics of the school and school-neighborhood environment that
affected both the likelihood of experiencing childhood sexual abuse and the
likelihood of success in school and on the labor market. For example,
descriptive statistics showed that a child in a low-income household was more
likely to experience sexual abuse. And, we know that schools in low-income
neighborhoods, on average, provide lower quality education than is available
to children who attend schools in higher income neighborhoods. Interpreting
these baseline results causally requires the assumption that there is no
unobserved heterogeneity across children at the same schools which influence
both their likelihood of sexual abuse and their education and labor market
outcomes. In the following sections, I address the possibility of remaining
unobserved confounding at the family or individual level.
I used ordinary least squares (OLS) to model binary outcomes and
replicate these models with logit and probit regression in ancillary analyses. I
examined annual earnings with a two-part model (Dow and Norton 2003,
Roodman 2009) to include people with zero earnings. The first part was a
probit model estimating the probability of having positive earnings, and the
29
second part was a generalized linear model (log link function and gamma
distribution) estimating earnings levels, using observations with positive
earnings.
3.4.3 Robustness checks
3.4.3.1 Sensitivity analyses
In sensitivity analyses, I evaluated robustness of results to inclusion of
additional potential confounders: caregiver neglect, educational challenge, and
childhood depressive symptoms. I did not include these potential confounders
in the baseline specification for two reasons. First, neglect was only evaluated
during Wave III, thus this measure was missing for 20 percent of the full Wave
I/Wave IV sample. Second, educational challenge (defined below) and
depressive symptoms might be outcomes of childhood sexual abuse, thus
inclusion as covariates would produce underestimates of the true effects of
childhood sexual abuse.
I defined caregiver neglect as physical or supervisory neglect.
Educational challenge included learning disability and attention-
deficit/hyperactivity disorder (ADHD), grouped together due to the language
of the survey question on learning disabilities. Such disability might cause
heightened risk of childhood sexual abuse or be a misdiagnosis of trauma
symptoms. I used a binary variable for childhood depressive symptoms
constructed from a five-item instrument adapted from the Center for
Epidemiologic Studies Depression Scale (Orme, et al. 1986, Radloff 1977).
30
3.4.3.2 Bounding effects of childhood sexual abuse
In this section, I calculated bounds on the effects of childhood sexual abuse
under varying assumptions about the degree of selection on unobservables
relative to selection on observables, following partial identification methods
developed by Altonji, et al. (2002, 2005) and Oster (2017). Partial
identification allows the researcher to recover bounds on the estimated
treatment effects in contexts where unobservable variables may cause
confounding. This method relies on the premise that confounding on
observables provides insight into the influence of confounding on
unobservables.
In this context, I assessed whether unobserved factors, such as low
parental investment in children, fully explained the results. I adopted Oster’s
approach for linear models (Oster 2017). This method involves using
information from observable confounders to bound the likely effect of
unobservable confounders. Thus, key parameters are: (i) the amount of
selection on unobservables relative to selection on observables and (ii) the
amount of outcome variance that would be explained by the unobservable
confounders. The first value, the selection parameter, is denoted δ below, and
the second value informs the value R
max
, which represents the R
2
from the
hypothetical regression including unobservable confounders on the right-
hand-side. I construct effect bounds for varying values of δ from 0 to 1 and
make the following alternative assumptions about the value R
max
:
(i) R
max
= 1.3𝑅 ̃
, where 𝑅 ̃
is the R
2
from equation (3.1) above with the
full set of observed controls, and
(ii) R
max
= 2𝑅 ̃
.
31
Below, I describe the approach to construct bounds on the effects of
childhood sexual abuse. To start, consider a modified version of equation (3.1)
above, omitting the individual subscripts i:
Y = 𝛼 ̈·ChildhoodSexualAbuse + X ′𝛽 ̈ + W
2
+ η (3.2)
The new term, W
2
, represents unobservable factors which determine the
human capital outcome Y and are correlated with likelihood of childhood
sexual abuse but not correlated with any of the observable confounders. For
example, unobserved level of parental investment in children will be contained
in W
2
if the partial correlation with childhood sexual abuse, conditional on the
observed controls, is non-zero. By definition as a confounder, cov(W
2
, Y) ≠ 0
and cov(W
2
, ChildhoodSexualAbuse) ≠ 0. An additional requirement of W
2
is
that cov(W
2
, W
1
) = 0, denoting W
1
=X’𝛽 ̈ . This orthogonality requirement implies
that W
1
captures observables in addition to any confounding from
unobservables which are correlated with the observables. Now, the parameter
𝛼 ̈ is the true effect of childhood sexual abuse on outcome Y. The goal is to
estimate bounds for 𝛼 ̈, but the coefficients on the observed variables in W
1
should not be interpreted causally. Recall that X includes physical and
emotional abuse, thus while elements of W
2
are correlated with childhood
sexual abuse, they must be uncorrelated with these other types of childhood
abuse. Lastly, η is the individual idiosyncratic error term which contains
unobservables which determine the outcome Y but are uncorrelated with
childhood sexual abuse, controls, or W
2
.
The ratio of selection on unobservables to selection on observables is
defined as
δ=
𝜎 2,𝑐𝑠𝑎 𝜎 2
2
/
𝜎 1,𝑐𝑠𝑎 𝜎 1
2
32
where σ
j,csa
= cov(W
j
, ChildhoodSexualAbuse) and σ
j
2
= var(W
j
) for j ∈ {1, 2}.
The denominator of this ratio – the level of selection on observables – can be
readily computed from the data. The numerator – the selection on
unobservables – is identified with a restriction on the parameter 𝛼 ̈, the effect
of childhood sexual abuse, and a restriction on R
max
. Here, I choose to compute
bounds on the effect 𝛼 ̈, thus I assume various values for the selection ratio δ
and R
max
.
I calculated bounds on effect sizes for a set of R
max
values informed by
the R
2
from the regression with the full set of observed controls. In most cases,
it is unlikely that R
max
= 1, i.e., that the treatment/exposure, controls, and the
unobservable component W
2
fully explain the outcome, due to measurement
error in the outcome or other idiosyncratic variation in the outcome (Oster
2017). In the present case, there were likely important determinants of human
capital outcomes which were determined after Wave I but not caused by these
childhood factors. For example, all cases of depression not caused by childhood
sexual abuse (or controls such as the other adverse childhood experiences) will
be included in an individual’s η error term.
The first R
max
condition above, R
max
= 1.3𝑅 ̃
, was proposed by Oster from
calculations with data from randomized studies. From her set of results from
randomized studies published in top economics journals which reported
uncontrolled and controlled estimates, she calculated that when holding fixed
δ=1, 90 percent of the results would survive if R
max
= 1.3𝑅 ̃
(Oster 2017).
6
I
6
To “survive” means both that the identified set does not include zero and that the identified
set is within 2.8 standard errors of the fully controlled estimate. The sample of results selected
in Oster (2017) come from randomized studies in top-five economics journals over a six-year
period. Across this set of 65 results, she found that 90 percent would survive a cutoff of R
max
= 1.3𝑅 ̃
, and only 40 percent would survive a cutoff of R
max
= 1 when fixing δ=1. Among non-
33
present the main results of the bounding exercise under this condition: R
max
=
1.3𝑅 ̃
. Note that the second R
max
condition, R
max
= 2𝑅 ̃
, implies that observables
and unobservables explain the same amount of variance in the outcome. These
choices for R
max
were informed by a review of literature citing Oster (2017).
7
Interpreting the ratio δ here is difficult given the large amount of
selection on observables. In general, if the estimated treatment (or exposure)
effect erodes to zero only when δ > 1, this is considered evidence that at least
part of the estimated effect is real (Altonji, et al. 2002, Altonji, et al. 2005,
Oster 2017). However, the true ratio of selection on unobservables to selection
on observables, δ, is likely less than 1 and may be much less than 1, in particular
when selection on observables is substantial as in the present case. I calculated
bounds for the effect size under the conditions δ=0, 0.25, 0.5, 1.
I calculated bounds on the effects of childhood sexual abuse on the
following outcomes: high school diploma receipt, college degree attainment,
full-time employment, and log earnings. Because there was no relationship
between childhood sexual abuse and probability of positive earnings (as will
be found from the first part of the two-part model of earnings implemented in
the fully controlled baseline model), I proceeded with these methods of
calculating effect bounds in OLS models by substituting log earnings for
earnings.
randomized results from the same set of economics journals, 45 percent survived the cutoff
R
max
= 1.3𝑅 ̃
(fixing δ=1).
7
For example, see Álvarez et al. (2018); Anger et al. (2017); Beach et al. (2017); Ebenstein et
al. (2017); El-Mallakh et al. (2018); Gambaro et al. (2018); Jha (2015); Lavy et al. (2018); Lee
et al. (2018); Lehne et al. (2018); Lin et al. (2018); Valbuena et al. (2018); Walther (2018)
among others.
34
3.4.3.3 Sibling fixed effects regression
I addressed the possibility of confounding from unobserved family
characteristics by measuring the association between non-caregiver childhood
sexual abuse and the human capital outcomes while using fixed effects
regression to parse out unobservable characteristics at the family level. I
restricted the key exposure variable to non-caregiver childhood sexual abuse
to mitigate measurement error from discordant reports of caregiver sexual
abuse across siblings which could reflect in one sibling a failure to report
history of caregiver sexual abuse in the survey rather than true differences in
experience of caregiver sexual abuse. Potential unobserved confounders at the
family level might include poor parental supervision, which might make the
child more vulnerable to sexual abuse while also indicating that the parent was
not spending time investing in the child.
Specifically, I implemented conditional sibling fixed effects models of
the following form:
Y
is
= 𝛼 ̇·NC_ChildhoodSexualAbuse
is
+ β
1
age
is
+ β
2
sex
is
+ μ
s
+ ν
i
(3.3)
where NC_ChildhoodSexualAbuse
is
equals 1 if individual i in sibling group s
reported non-caregiver childhood sexual abuse, μ
s
represents the sibling fixed
effect, and I control for individual age and sex. I excluded individuals who
reported caregiver sexual abuse, thus the parameter 𝛼 ̇ represents differences
between siblings who experienced non-caregiver childhood sexual abuse and
siblings who experienced no childhood sexual abuse.
The advantage of the sibling fixed effects model is removal from 𝛼 ̇, the
estimated effect of non-caregiver childhood sexual abuse, any unobserved
confounding at the family level that is constant across siblings. There might
35
still be, however, important differences across siblings due to changing family
environment across births or other differences due to family dynamics. I did
not use this strategy in the baseline model due to substantially reduced power
and concerns of sample representativeness. Conditional fixed effects
regression requires variation within sibling groups for each measure included
in the model. The sample with siblings from families in which at least one child
experienced sexual abuse and at least one child did not was much smaller than
the full sample, reducing the sample size from roughly 15,000 individuals to
about 3,000 individuals. But also, this group was a selected sample in which
one child in every family experienced non-caregiver childhood sexual abuse.
The results from sibling fixed effect models would underestimate the true
effects of non-caregiver childhood sexual abuse if the comparison sibling
experienced secondary trauma or also experienced sexual abuse but reported
no sexual abuse in the Add Health survey.
Because I employed a sibling fixed effects model rather than controlling
for a set of specific family-level characteristics, I had no missing data problem
here. Hence, I implemented the analysis with the original analytic file –
without multiply imputed data values. As before, I used OLS to model high
school diploma receipt, college degree attainment, and full-time work status.
Here, I modelled log earnings with OLS (after finding from baseline results of
the two-part model of earnings that there was no relationship between
childhood sexual abuse and probability of positive earnings). Because there
were so few individuals per group (siblings per family), including sibling fixed
effects in the probit regression within a two-part model might lead to
inconsistent estimates.
36
3.4.3.4 Falsification test
A concern for causal interpretation of results is that children who were
sexually abused grew up in more disadvantaged households, on average, thus
global disadvantage or a particular source of disadvantage associated with
childhood sexual abuse might be driving the results. Adults who were sexually
abused as children might have worse educational and labor market outcomes
because their parents did not invest in them and not because of the abuse
experience itself. To evaluate the hypothesis that factors other than childhood
sexual abuse drove the results, I assessed whether there were pre-existing
deficits in human capital. To do this, I exploited the longitudinal nature of the
survey and the timing of exposure to childhood sexual abuse. I examined
differences in scores on the Add Health Picture Vocabulary Test, which was
administered in Wave I, in regressions including the full control set from the
baseline specification detailed above. The goal was to assess whether Wave I
test scores differed between the group who reported childhood sexual abuse
which occurred at an age after their Wave I interview and the group who
reported no childhood sexual abuse. Thus here, the childhood sexual abuse
measure is a three-category variable distinguishing (i) participants who
reported no childhood sexual abuse (reference group), (ii) participants who
reported childhood sexual abuse that occurred after Wave I but no sexual abuse
before Wave I, and (iii) participants who reported sexual abuse that occurred
before Wave I. I imposed a one-year buffer period – that is, I excluded
observations with reports of childhood sexual abuse at an age within one-year
of Wave I interview. In addition, I evaluated whether child vocabulary score
was a valid surrogate outcome of human capital by assessing whether
37
vocabulary scores were predictive of the studied adult outcomes. To do this, I
regressed each outcome on vocabulary score in OLS models.
3.4.4 Heterogeneity analyses
I studied the heterogeneity of the relationship between childhood sexual abuse
and human capital outcomes by sociodemographic characteristics. In separate
regressions, I modified the baseline model by interacting childhood sexual
abuse with sex, race, and log childhood household income. I also replicated the
baseline model in the subsample of males and the subsample of females. In
addition, I implemented the same specifications described above with
interactions for sociodemographic characteristics to model current depressive
symptoms. The goal was to explore whether disparities in the childhood sexual
abuse-human capital relationship might be explained by differential rates of
current depressive symptoms.
In addition, I investigated whether caregiver vs. non-caregiver sexual
abuse led to different outcomes. To do this, I modified the baseline regressions
by splitting the childhood sexual abuse measure into two variables: caregiver
sexual abuse and non-caregiver sexual abuse (each before age 18). Recall from
the presentation of sample characteristics that some children experienced
sexual abuse by both types of perpetrator, thus these two groups were not
mutually exclusive.
3.4.5 Determinants of earnings
Finally, I studied the extent to which current (adult) factors, which might be
caused by childhood sexual abuse, explained the relationship between
38
childhood sexual abuse and adult earnings. To do this, I modified the baseline
two-part model of earnings to include participant educational attainment and
current depressive symptoms as covariates in separate regressions, and I
examined changes in the coefficient on childhood sexual abuse.
3.5 Results
3.5.1 Baseline results
Baseline results showed that survivors of childhood sexual abuse had lower
educational attainment and worse labor market outcomes. For high school
diploma attainment, the unadjusted difference in means was -11.2 percentage
points, as shown in Table 2, column (1) below as the OLS coefficient on
childhood sexual abuse from the model without controls. Controlling for
demographics, family socioeconomic status, school fixed effects, childhood
physical disability, other types of childhood abuse, and parental incarceration
reduced the estimate to -7.7 percentage points.
8
Thus, observables explained
31 percent of the raw difference in rate of high school diploma receipt across
those who did and did not experience childhood sexual abuse. The unadjusted
difference in rates of college degree attainment, -14.0 percentage points,
reduced by 40 percent in the fully adjusted model to -8.4 percentage points. In
replications of these models with probit and logit regression, I found that the
average marginal effects and standard error estimates were identical or nearly
8
Allowing that some children might have attended schools outside of their home
neighborhoods, in ancillary analyses I additionally controlled for information from the Census
block group of children’s home residence: percent of adults without a high school degree.
Results were nearly identical to estimates reported here from the school fixed effects models.
39
identical to the OLS results. See complete regression results from the fully
controlled models in Appendix Tables A5 and A6.
Observed characteristics explained over half of the differences in labor
market outcomes. The unadjusted difference in employment was -8.6
percentage points, which reduced to -3.9 percentage points in the fully
controlled model. Results for full-time employment were larger: -11.6
percentage points difference in the unadjusted model, which reduced to -5.3
percentage points in the fully adjusted model. While the unadjusted difference
in earnings was -$9,281, or 31 percent of median earnings for the full sample,
the fully adjusted difference was -$4,390, or 14 percent of median earnings.
See complete regression results from the fully controlled model in Appendix
Tables A7-A9.
40
Table 2. Estimates of the effects of childhood sexual abuse on education and labor
market outcomes
Average marginal effects (standard errors in parentheses)
(1) (2) (3) (4) (5) (6)
Controls None Demographics
a
Col. 2 +
household SES
b
Col. 3 +
school FE
c
Col. 4 +
disability
d
Col. 5 +
other ACEs
e
A. High school diploma; mean (s.d.): 0.828 (0.378)
OLS -0.112*** -0.123*** -0.100*** -0.090*** -0.088*** -0.077***
(0.014) (0.015) (0.014) (0.013) (0.014) (0.014)
R
2
0.014 0.029 0.099 0.150 0.152 0.157
Adjusted
R
2
0.013 0.029 0.098 0.141 0.143 0.148
B. College degree; mean (s.d.): 0.302 (0.459)
OLS -0.140*** -0.159*** -0.109*** -0.100*** -0.098*** -0.084***
(0.014) (0.013) (0.012) (0.011) (0.011) (0.012)
R
2
0.011 0.041 0.209 0.259 0.259 0.265
Adjusted
R
2
0.010 0.040 0.208 0.251 0.252 0.257
C. Employment; mean (s.d.): 0.805 (0.396)
OLS -0.086*** -0.061*** -0.053*** -0.050** -0.049** -0.039*
(0.015) (0.015) (0.015) (0.015) (0.015) (0.015)
R
2
0.005 0.021 0.033 0.068 0.069 0.069
Adjusted
R
2
0.005 0.021 0.032 0.058 0.059 0.059
D. Full-time employment; mean (s.d.): 0.706 (0.455)
OLS -0.116*** -0.075*** -0.066*** -0.063*** -0.063*** -0.053***
(0.015) (0.015) (0.015) (0.015) (0.015) (0.015)
R
2
0.007 0.036 0.044 0.092 0.093 0.093
Adjusted
R
2
0.007 0.036 0.043 0.083 0.083 0.083
E. Earnings; mean (s.d.): $35,025 (44,170); median: $30,414
2PM -$9,281*** -$5,907*** -$3,959*** -$4,496*** -$4,461*** -$4,390***
(1,322) (1,415) (1,431) (1,066) (1,070) (1,143)
41
NOTE.— Sample size was N=14,741 for each regression here except for those in Panel E for earnings
(N=13,938).
a
Demographic controls included age when outcome was measured; sex as female or male; and
race/ethnicity as White Non-Hispanic, Black Non-Hispanic, Native American Non-Hispanic,
Asian/Pacific Islander Non-Hispanic, Other Non-Hispanic (includes multi-racial), or Hispanic.
b
Household SES controls included log of childhood household income and highest parental
educational attainment as (i) less than high school, (ii) GED, (iii) high school diploma, (iv) vocational
school after high school, (v) some college, (vi) college graduate, or (vii) beyond 4-yr college.
c
School fixed effects: schools were the primary sampling unit.
d
Disability represents childhood physical disability.
e
Other ACEs included childhood physical abuse, emotional abuse, and parental incarceration.
* p<0.05, ** p<0.01, *** p<0.001
While in unadjusted models physical abuse and emotional abuse each
were associated with worse education and labor market outcomes, most of
these associations disappeared in the fully controlled models. A few results
survived: physical abuse predicted 3.0 (s.e., 1.3) percentage points lower
likelihood of college degree; emotional abuse predicted 3.1 (1.5) percentage
points lower likelihood of any employment and 3.8 (1.8) percentage points
lower likelihood of full-time employment. See results in Appendix Tables A2
and A3. However, this does not mean that there was no effect of physical or
emotional abuse on other outcomes but that I could not detect any differences
when controlling for other types of abuse. That is, sexual, physical, and
emotional abuse might capture similar constructs of which sexual abuse is a
higher or more prolonged dose. It is notable, however, that while family
background and health of children who experienced each type of abuse were
similar, sexual abuse was uniquely predictive of lower rates of high school
diploma receipt and earnings in fully adjusted models.
42
3.5.2 Robustness checks
3.5.2.1 Sensitivity analyses
Estimates of the effects of childhood sexual abuse were robust to inclusion of
the additional potentially confounding factors: physical or supervisory neglect,
educational challenge, and childhood depressive symptoms.
Notably, the estimated effects of educational challenge on outcomes
were larger in magnitude than the estimates for childhood sexual abuse. But
still, inclusion of educational challenge as a control in these ancillary analyses
reduced the estimated effects of childhood sexual abuse by less than one
percentage point. For example, educational challenge decreased likelihood of
high school diploma receipt by -11.6 percentage points. Educational challenge
was more important in determining high school diploma outcome than was
childhood sexual abuse, yet inclusion of educational challenge as a covariate
hardly reduced the estimated effect of childhood sexual abuse: from -7.7
percentage points in the baseline specification with full control set (6) above
to -7.1 percentage points here. Results for college degree, full-time
employment, and earnings were similarly robust, hardly changing upon
inclusion of these theoretically important confounders. See results in Table 3
below.
43
Table 3. Sensitivity analyses: estimates of the effects of childhood sexual
abuse on education and labor market outcomes
Average marginal effects (standard errors in parentheses)
(1) (2) (3) (4)
A. High school diploma; mean (s.d.): 0.828 (0.378)
Childhood sexual abuse
-0.077***
(0.013)
-0.074***
(0.014)
-0.071***
(0.014)
-0.074***
(0.013)
Neglect
-0.024
(0.013)
Educational challenge
-0.116***
(0.014)
Childhood depressive
symptoms
-0.051***
(0.011)
R
2
0.156 0.166 0.169 0.161
Adjusted R
2
0.147 0.154 0.159 0.151
N 14,741 12,215 14,741 14,741
B. College degree; mean (s.d.): 0.302 (0.459)
Childhood sexual abuse
-0.083***
(0.012)
-0.080***
(0.014)
-0.076***
(0.012)
-0.080***
(0.012)
Neglect
-0.027
(0.017)
Educational challenge
-0.152***
(0.017)
Childhood depressive
symptoms
-0.049***
(0.010)
R
2
0.265 0.269 0.277 0.267
Adjusted R
2
0.256 0.259 0.269 0.258
N 14,741 12,215 14,741 14,741
C. Full-time work; mean (s.d.): 0.706 (0.455)
Childhood sexual abuse
-0.051**
(0.016)
-0.071***
(0.018)
-0.047**
(0.016)
-0.048**
(0.016)
Neglect
0.017
(0.016)
Educational challenge
-0.084***
(0.015)
Childhood depressive
symptoms
-0.046***
(0.011)
44
R
2
0.093 0.089 0.098 0.095
Adjusted R
2
0.083 0.076 0.088 0.085
N 14,741 12,215 14,741 14,741
D. Earnings; mean (s.d.): $35,025 (44,170); median: $30,414
Childhood depressive
symptoms
-$4,325***
(1,178)
-$4,065**
(1,305)
-$4,038**
(1,266)
$4,137**
(1,288)
Neglect
$929
(1,318)
Educational challenge
-$8,797***
(888)
Childhood depression
-$4,570***
(847)
N 14,741 10,735 13,510 13,510
NOTE.— Each regression included the full control set described in notes to Table 3:
demographics, childhood SES, school fixed effects, physical disability, and other ACEs. Binary
outcomes were modeled with OLS and the average marginal effects on earnings were
estimated from a two-part model. Sample size was smaller in regressions including neglect
because this measure was only included in the Wave III interview in which not all of the core
sample participated.
* p<0.05, ** p<0.01, *** p<0.001
3.5.2.2 Bounding effects of childhood sexual abuse
Based on the assessment of robustness to selection on unobservables in this
section, I conclude that there is strong evidence that childhood sexual abuse
caused lower educational attainment and some evidence that childhood sexual
abuse caused worse labor market outcomes. Figure 1 below displays estimates
of the effect of childhood sexual abuse under varying assumptions about the
ratio of selection on unobservables to selection on observables, δ, and the total
amount of variance in the outcome that would be explained once controlling
for the unobserved confounders in a hypothetical regression. This latter value,
R
max
, varies across the horizontal axis. The estimated effect of childhood sexual
abuse varies across the vertical axis. A separate line of parameter estimates is
45
plotted for each value of δ, and the dots on each line mark the R
max
values 1.3𝑅 ̃
and 2𝑅 ̃
, from left to right.
46
Figure 1. Bounds on the effects of childhood sexual abuse under varying
assumptions about selection on unobservables
NOTE.— The figures above depict the average marginal effect of childhood sexual abuse under
varying assumptions about the importance of unobservables. Models controlled for age when
outcome was measured, sex, race, highest parental educational attainment, childhood
household income, school fixed effects, childhood disability, and other adverse childhood
-0.10
-0.08
-0.06
-0.04
-0.02
0.00
0.20 0.25 0.30 0.35
Average marginal effect of childhood sexual abuse
R
max
A. High school diploma
mean (s.d.): 0.828 (0.378)
-0.10
-0.08
-0.06
-0.04
-0.02
0.00
0.30 0.40 0.50 0.60
R
max
B. College degree
mean (s.d.): 0.302 (0.459)
-0.06
-0.04
-0.02
0.00
0.02
0.04
0.10 0.12 0.14 0.16 0.18 0.20
Average marginal effect of childhood sexual abuse
R
max
C. Full-time work
mean (s.d.): 0.708 (0.455)
-0.18
-0.16
-0.14
-0.12
-0.10
-0.08
-0.06
-0.04
-0.02
0.00
0.02
0.04
0.20 0.25 0.30 0.35 0.40
R
max
D. Log earnings
Earnings mean (s.d.): $37,962
(44,861); median: $31,428
δ = 1
δ = 0.75
δ = 0.5
δ = 0.25
δ = 0
1.3𝑅 ̃
2𝑅 ̃
1.3𝑅 ̃
2𝑅 ̃
1.3𝑅 ̃
2𝑅 ̃
1.3𝑅 ̃
2𝑅 ̃
δ = 1
δ = 0.75
δ = 0.5
δ = 0.25
δ = 0
δ = 1
δ = 0.75
δ = 0.5
δ = 0.25
δ = 0
δ = 1
δ = 0.75
δ = 0.5
δ = 0.25
δ = 0
47
experiences (physical abuse, emotional abuse, parental incarceration). The points marked
represent, from left to right: R
max
= 1.3𝑅 ̃
and R
max
= 2𝑅 ̃
. When δ = 0, R
max
must equal 𝑅 ̃
, but the
average marginal effect for δ = 0 is plotted over the same values of R
max
for ease of comparison
with the estimates under assumptions of δ >0.
a
Modelled as log earnings, among those with any earnings. OLS models of the binary outcome
“having any earnings” showed no relationship between childhood sexual abuse and having
positive earnings. Descriptive statistics for earnings noted here are provided for the sample
with positive earnings.
The Oster bounds (R
max
= 1.3𝑅 ̃
) suggest that childhood sexual abuse led
to 6.6 to 7.7 percentage points lower likelihood of high school diploma receipt,
6.0 to 8.4 percentage points lower likelihood of college degree attainment, 2.8
to 5.3 percentage points lower likelihood of full-time employment, and 11.2 to
17.6 percent lower earnings among those with any earnings. In relative terms,
these magnitudes translate to 38.4 to 44.8 percent greater likelihood of high
school dropout, 19.8 to 27.8 percent lower likelihood of college degree
attainment, and 4.0 to 7.5 percent lower likelihood of full-time employment.
9
Notice in Figure 1 that the bounds on the effect of childhood sexual abuse
were negative for high school diploma receipt and college degree attainment
even when allowing as much selection on unobservables as on observables
(δ=1) while also allowing these unobserved confounders to explain as much of
the outcome variance as do the observables (R
max
= 2𝑅 ̃
). For full-time
employment and earnings, estimates of the effect of childhood sexual abuse
remained negative under the condition R
max
= 1.3𝑅 ̃
for all values of δ. Under
the condition R
max
= 2𝑅 ̃
, results survived through the value δ=0.5 for full-time
employment and through δ=0.75 for log earnings. See Figure 1 for bounds on
the effects of childhood sexual abuse for all combinations of R
max
and δ
conditions.
9
I translated the average marginal effects to results in relative terms by dividing the average
marginal effect by the outcome mean.
48
3.5.2.3 Sibling fixed effects regression
Results from the sibling fixed effects models showed that, even when parsing
out the impacts of unobserved characteristics of the family environment, there
remained a negative relationship between non-caregiver childhood sexual
abuse and each education and labor market outcome. These estimates,
however, are less precise. Only the estimated effects on high school diploma
receipt and any employment remained significant here. It is worth reiterating
that the sibling fixed effects sample was not only much smaller but also a very
selected sample. These models could only be implemented across children from
families in which at least one child experienced sexual abuse and at least one
other child did not.
Results from sibling fixed effects models with any childhood sexual
abuse as the key exposure variable were qualitatively similar. In particular,
these estimates of the effects of childhood sexual abuse were similar in
magnitude to those obtained from the baseline specification and fell within the
Oster bounds (R
max
= 1.3𝑅 ̃
, δ = 0 to δ = 1) presented above. See results in
Appendix Table A4.
Table 4. Sibling fixed effects estimates from OLS: non-caregiver childhood
sexual abuse
(standard errors in parentheses)
Outcome
(1) (2) (3) (4) (5)
High
school
diploma
College
degree Employment
Full-time
employment
Log
earnings
49
Non-caregiver
sexual abuse
-0.108*
(0.051)
-0.072
(0.043)
-0.137*
(0.06)
-0.052
(0.062)
-0.076
(0.156)
N 3,081 3,063 3,080 3,080 2,712
NOTE.— This table displays the estimated average marginal effects and standard errors for the effect
of non-caregiver childhood sexual abuse, as measured in models with sibling conditional fixed effects
and controlling for individual age and sex. Conditional fixed effects regression exploits variation
within the group (siblings), thus sibling groups for which there was no variation in the outcome or
right-hand-side variables were dropped from the regression. Sample sizes varied across outcomes for
this reason and also for log earnings due to use only of observations with positive earnings.
* p<0.05, ** p<0.01, *** p<0.001
3.5.2.4 Falsification test
I did not detect any pre-existing differences in cognitive ability. The Wave I
vocabulary scores were no different for the group who eventually experienced
childhood sexual abuse after Wave I and the group who reported no childhood
sexual abuse. See Table 5 below. Results from separate regressions validated
Wave I vocabulary score as an appropriate measure for the falsification test.
Firstly, vocabulary score served as a surrogate measure for the adult human
capital outcomes studied: higher scores predicted better education and labor
market outcomes. Secondly, childhood sexual abuse before Wave I predicted
lower vocabulary scores in the Wave I tests, and this result was significant at
the 5 percent level.
Table 5. Falsification test: childhood sexual abuse and Wave I vocabulary
score
Outcome: vocabulary score
a
mean (s.d.): 101.3 (14.6)
OLS results
Timing of sexual abuse Coefficient (s.e.)
Before Wave I -1.1*
50
(0.5)
After Wave I (but before age 18) 0.0
(1.1)
Never ref.
N 13,575
NOTE.— This table displays the coefficients and standard errors from an OLS regression corresponding
to equation (3.1), including the full control set. To reduce contamination in the control group resulting
from possibly imprecise recall of age when first sexually abused, participants who reported sexual abuse
at an age within one-year of the Wave I interview were excluded from the falsification sample.
Key: ref., reference group; s.d., standard deviation; s.e., standard error
a
Vocabulary score is the age-normalized score from the Add Health Picture Vocabulary Test.
* p<0.05, ** p<0.01, *** p<0.001
3.5.3 Heterogeneity analyses
Heterogeneity analyses revealed some differences in the relationship between
childhood sexual abuse and human capital outcomes by sociodemographic
characteristics. Analysis by race showed that OLS estimates of the effects of
childhood sexual abuse were smaller for some minorities, compared to results
for Non-Hispanic Whites. The negative estimate of the effect of childhood
sexual abuse on college degree attainment was smaller in magnitude for Non-
Hispanic Blacks and Hispanics compared to Non-Hispanic Whites. For Non-
Hispanic Asians, there was no effect of childhood sexual abuse on likelihood of
full-time employment, on average. There was some heterogeneity of results by
childhood household income. For college degree attainment, the estimate for
childhood sexual abuse was larger in magnitude for adults from higher income
families. Childhood household income, however, did not moderate the effect of
childhood sexual abuse on high school diploma receipt or full-time
employment. See full results in Appendix Tables A5-A9.
51
Through inclusion of an interaction term between childhood sexual
abuse and sex, I cannot detect any differences by sex in the relationship
between child sexual abuse and human capital outcomes. However, I can
confirm that sexual abuse predicts lower educational attainment and earnings
in men. No prior literature has demonstrated a relationship between childhood
sexual abuse and education or labor market outcomes among US men. See
results by sex in Table 6 below.
Table 6. Heterogeneity by sex
Average marginal effects of childhood sexual abuse (standard errors in parentheses)
Outcome
(1) (2) (3) (4) (5)
High
school
diploma
College
degree Employment
Full-time
employment Earnings
Female
-0.079***
(0.016)
-0.083***
(0.016)
-0.027
(0.020)
-0.038*
(0.018)
-$2,673
(1,653)
Male
-0.112***
(0.030)
-0.089***
(0.018)
-0.043
(0.027)
-0.044
(0.032)
-$6,234*
(2,503)
NOTE.— Columns (1) – (4) display OLS estimates and column (5) displays average marginal effects
from a two-part model – with standard errors in parentheses. The baseline regression described above,
but excluding school fixed effects and sex, were implemented separately in the subsample of females
and subsample of males.
* p<0.05, ** p<0.01, *** p<0.001
None of the disparities by sociodemographic characteristics described
here were explained by differential rates of current depressive symptoms.
Neither sex, nor race, nor childhood household income modified the
relationship between childhood sexual abuse and adult depressive symptoms.
Non-caregiver sexual abuse was at least as important as caregiver
sexual abuse in predicting worse education and labor market outcomes.
Splitting up the measure of childhood abuse by identity of the perpetrator
52
showed that the coefficients on non-caregiver sexual abuse were much larger
in magnitude than the coefficients on caregiver sexual abuse. For example,
estimates of the effect of non-caregiver sexual abuse on education outcomes
were more than double the estimates for caregiver sexual abuse. None of the
differences in coefficients, however, were statistically significant.
10
Table 7. Perpetration by caregivers vs. non-caregivers
Average marginal effects (standard errors in parentheses)
Outcome
(1) (2) (3) (4) (5)
High school
diploma
College
degree Employment
Full-time
employment Earnings
Caregiver
sexual abuse
-0.036*
(0.018)
-0.054***
(0.015)
-0.002
(0.019)
-0.019
(0.021)
-$3,057*
(1,432)
Non-
caregiver
sexual abuse
-0.082***
(0.015)
-0.090***
(0.017)
-0.068***
(0.018)
-0.076***
(0.021)
-$5,699**
(1,630)
NOTE.— Columns (1) – (4) display OLS estimates and column (5) displays average marginal effects
from a two-part model – with standard errors in parentheses. Each regression controls for
demographics, childhood socioeconomic status, physical disability, childhood physical abuse,
emotional abuse, parental incarceration, and includes school FEs as in the baseline model. N=14,741
* p<0.05, ** p<0.01, *** p<0.001
3.5.4 Determinants of earnings
Decomposing the effect of childhood sexual abuse on earnings, I found that the
negative relationship persisted when controlling for current depressive
symptoms and educational attainment in separate regressions. The average
marginal effect of childhood sexual abuse on earnings was -$4,390, or 14.0
10
It is also noteworthy that the language of the questions on sexual abuse were different for
abuse perpetrated by caregivers and by non-caregivers, and respondents may have interpreted
the questions differently. Caregiver sexual abuse included sexual touching (by the caregiver
or forced touching of the caregiver) and forced sexual relations. Non-caregiver sexual abuse
included any sexual activity that was forced in a physical or non-physical way.
53
percent of median earnings, in the fully controlled baseline model above. Once
controlling for depressive symptoms, this estimate fell in magnitude to -$3,813
($1,275), or 12.1 percent of median earnings, remaining significant at the 1
percent level. In a separate regression, when additionally controlling for
educational attainment, the average marginal effect of childhood sexual abuse
reduced to -$3,106 (1,284), or 9.9 percent of median earnings, remaining
significant at the 5 percent level. Thus, disparities in educational attainment
explained only a minority (29 percent) of the disparity in earnings between
those who were and were not sexually abused in childhood.
3.6 Discussion and conclusions
In the present study, I investigated whether childhood sexual abuse had
durable consequences on human capital. I examined outcomes in young
adulthood in the US setting, using nationally representative, longitudinal
survey data. In this paper, I measured negative effects of childhood sexual
abuse on education and labor market outcomes when controlling for a rich set
of measures of childhood socioeconomic status including school fixed effects,
showed that these results persisted once sweeping out unobservables
correlated with other adverse childhood experiences, and showed that these
results were robust to varying assumptions about the remaining amount of
selection on unobservables. Sibling fixed effects estimates confirmed that the
negative relationship between childhood sexual abuse and education and labor
market outcomes persisted when parsing out unobserved family-level
characteristics (constant across siblings). In conclusion, results suggested that
childhood sexual abuse led to 38 to 45 percent greater likelihood of high school
dropout, 20 to 28 percent lower likelihood of college degree attainment, 4 to 8
54
percent lower likelihood of full-time employment, and 11 to 18 percent lower
earnings. Disparities in earnings might widen with age. Data from the Current
Population Survey (CPS) showed that high school dropouts aged 25 years and
older earned about 43 percent less than the average person in this age group,
when considering their higher rates of unemployment (U.S. Bureau of Labor
Statistics 2018). Considering the bounds from the present study on the effect
of childhood sexual abuse on likelihood of high school dropout along with data
from the CPS on the earnings penalty of dropping out of high school (U.S.
Bureau of Labor Statistics 2018), the aggregate productivity loss due to
childhood exposure to parental incarceration totals roughly $38 billion to $45
billion per year.
11
This paper is the first, to my knowledge, to provide evidence that
childhood sexual abuse causes lower educational attainment and worse labor
market outcomes in adulthood. Results from previous studies of child
maltreatment (Currie and Widom 2010) or childhood physical abuse (Covey,
et al. 2013) might be driven by effects of sexual abuse. In addition, this paper
is the first to establish that childhood sexual abuse predicts lower educational
attainment and worse labor market outcomes among US men. The present
work is also the first to study heterogeneity of response to childhood sexual
abuse on human capital outcomes by sociodemographic factors and identity of
the perpetrator. I find that effect sizes are smaller in magnitude for more
disadvantaged groups. The gap in college degree attainment across survivors
and non-survivors of childhood sexual abuse was smaller among adults who
were raised in lower income households as well as among Non-Hispanic Blacks
11
Inputs of this calculation also include the prevalence of childhood sexual abuse measured in
the present study and the size of the working age population, reported by Federal Reserve
Bank of St. Louis (2019).
55
and Hispanics versus Non-Hispanic Whites. This study did not explain the
reason for the disparity in effect sizes across demographics, but differences
might be due to greater resilience among those already exposed to adversity
or due to diminishing marginal effects of disadvantage for other reasons.
Beyond providing evidence of the consequences of childhood sexual
abuse on human capital, this study using nationally representative data also
showed that prevalence is high especially among women. About 21 percent of
women and 8 percent of men reported childhood sexual abuse in the self-
interview module of Add Health. The prevalence of adverse childhood
experiences was roughly consistent with reports from other US data, with
differences corresponding to variations in inclusiveness of the criteria (Dong,
et al. 2003, Merrick, et al. 2018). The Add Health data used in the present study
did, however, reveal higher rates of childhood sexual abuse among women
than reported to interviewers via telephone in the Behavioral Risk Factor
Surveillance System (BRFSS), in which childhood sexual abuse was restricted
to cases in which the perpetrator was at least five years older. In the 23 states
participating in the ACE module of the BRFSS at some point from 2011-2014,
about 16 percent of women and 7 percent of men reported experiences of
childhood sexual abuse (Merrick, et al. 2018).
The present study highlights that poverty alleviation is not enough to
prevent the disparities in economic outcomes that survivors of childhood
sexual abuse face. Firstly, descriptive results showed that childhood sexual
abuse is not just a problem in lower SES communities. Prevalence among those
who grew up in the top quintile of household income was 12 percent (versus
20 percent among those who grew up in the bottom quintile of household
income). Secondly, I provided evidence of a causal effect of childhood sexual
56
abuse on human capital – not conflated with the effects of childhood poverty.
Thirdly, the negative effect of childhood sexual abuse on college degree
attainment was even larger for those from higher income households.
Practitioners and policymakers could respond in a number of ways to
findings that childhood sexual abuse has durable consequences on human
capital. The present work highlights a critical need for identifying trauma
symptoms and treating trauma. In particular, results suggest a need for
detection of trauma symptoms and better catering to survivors of sexual abuse
within the education system. Due to federal laws mandating that teachers
report suspected child maltreatment to an appropriate agency, teachers
receive training to recognize physical signs of abuse. There are, however, no
best practices for identifying students suffering psychological trauma, such as
from sexual abuse. Yet, emotional disturbance is an eligible reason for special
education services under federal law, the Individuals with Disabilities
Education Act (Public Law 101-476). Educators and clinicians could collaborate
to develop trainings for teachers to recognize trauma symptoms.
Some states have led the development of policy addressing childhood
trauma. Illinois legislation, effective June 2017, requires social and emotional
screening for children when entering elementary, middle, and high school (SB
565, Public Act 99-0927). Effective July 2017 in Vermont, H.508 (Act 43)
established an Adverse Childhood Experiences Working Group to assess
existing resources and propose structures to advance evidence-based
approaches to support children experiencing trauma. The Act requires the
state’s Agency of Human Services to present a plan for “integration of
evidence-informed and family-focused prevention, intervention, treatment,
57
and recovery services for individuals affected by adverse childhood
experiences” by January 2019.
Childhood sexual abuse has durable impacts on human capital. There
remains a huge need for identifying effective approaches to prevent, detect,
and treat symptoms of childhood sexual abuse. In particular, results of the
present study highlight the need for development of quality interventions
which address the potential impact on the survivor’s cognitive development
and productivity. Policymakers could respond to findings by funding research
to identify effective approaches and quality support services for survivors of
childhood sexual abuse. Considering the magnitude of reduced earnings
measured in this nationally representative sample, this study could support
decisions for resource allocation for such services.
58
Childhood Abuse and Adult Health: A
Comprehensive Examination across Health
Conditions
Abstract
A growing body of literature suggests that the impacts of childhood adversity
may be wide-ranging and long-lasting. Here, I used detailed, nationally
representative survey data from the United States to examine the effects of
childhood abuse on adult health and to evaluate whether survivors of
childhood abuse face barriers to health care access. Firstly, childhood abuse
was prevalent: 1 in 3 adults reported history of childhood abuse. After
controlling for demographics, childhood socioeconomic status, the
neighborhoods they grew up in, and other childhood adversity, I found that
survivors of childhood abuse had higher risks of cardiometabolic conditions,
nervous system conditions, respiratory or allergic conditions, and cancer along
with higher risk of recent gastrointestinal symptoms. Meanwhile, survivors of
childhood abuse were also more likely to report unmet medical needs. This
study highlights the immediate need for development of best practices for
detection and quality treatment of childhood trauma and its sequelae.
59
4.1 Introduction
Childhood abuse can have lasting mental health consequences. History of
childhood abuse is associated with higher rates of anxiety, depression, and
post-traumatic stress disorder (PTSD) in adulthood (Fletcher 2009, Sachs-
Ericsson, et al. 2010). And, survivors of childhood abuse have much higher risk
of lifetime suicide attempt (Hoertel, et al. 2015, Pérez-Fuentes, et al. 2013).
A growing body of literature suggests that the consequences of
childhood trauma may extend beyond mental health. Scientists have posited
neurobiological mechanisms explaining effects of chronic childhood stress on
physiological systems, behavior, and cognitive development. The stress
response includes activation of the hypothalamic-pituitary-adrenal (HPA) axis
and the sympatho-adrenomedullary (SAM) axis, which interact with
components of the central nervous system including the brain centers
responsible for decision-making and emotional reactions, appetite regulation,
and response to rewards. When the brain and body are in a state of “fight-or-
flight,” the stress response may include inhibition of vegetative functions such
as digestion and immunity as the body prepares to address a threat (Bucci, et
al. 2016). Chronic activation of the stress response system – without adequate
mediation such as through support from others – has been associated with
impaired immune system functioning, which can lead to heightened
vulnerability to the common cold, developing allergies or asthma, and certain
tumors (Bucci, et al. 2016, Elenkov and Chrousos 1999). In addition, having
chronically elevated levels of the stress hormone cortisol has been indirectly
associated with the development of the individual components of metabolic
60
syndrome: hypertension, obesity, glucose intolerance, and abnormal lipid
levels (Bucci, et al. 2016).
12
Social scientists have tested some of these posited relationships from
childhood adversity to later-in-life outcomes. Extant literature has found that
adult survivors of childhood abuse had worse cardiometabolic health.
Survivors had higher rates of obesity (Duncan, et al. 2015b, Fuemmeler, et al.
2009, Ranchod, et al. 2016, Richardson, et al. 2014), diabetes (Duncan, et al.
2015a, Rich-Edwards, et al. 2010), hypertension (Ford and Browning 2014),
and metabolic syndrome (Lee, et al. 2014), compared to rates in adults who
suffered no abuse or less severe abuse. Survivors also had worse health
behaviors: they had higher rates of heavy drinking (Shin, et al. 2013), smoking
(De Von Figueroa-Moseley, et al. 2010, Roberts, et al. 2008), and substance
abuse (Exner-Cortens, et al. 2013, Jonson-Reid, et al. 2012). The studies cited
here controlled for some measure of childhood socioeconomic status. There
were inconsistencies, however, in results across the literature by population.
While a larger body of literature than cited here has studied the
relationships between childhood adversity and adult health outcomes, the
majority of papers failed to address confounding by childhood socioeconomic
status. And, the literature has largely overlooked study of acute health
conditions. There has been no systematic study of adult health or health care
use of survivors of childhood abuse. A major challenge in studying the effects
of childhood abuse is that there have been no known natural experiments, to
date, which have manipulated the probability that a child would be abused.
12
For more comprehensive reviews of the neurobiology of toxic stress, see Bucci et al. (2016)
and Burke et al. (2017).
61
In the present paper, I conducted the first systematic study of childhood
abuse and adult health and the first examination of childhood abuse and adult
health care access. Four main study attributes are novel here. First, I examined
reports of an extensive set of diagnoses of chronic conditions and acute
symptoms along with all recent prescription medication use in one sample – a
US nationally representative sample. Second, I utilized health measurements
to supplement the survey data and evaluate whether there were true
disparities in health vs. disparities in diagnosis. Third, by using rich,
longitudinal survey data, I controlled for finer levels of childhood
socioeconomic status and neighborhood factors, other childhood adversity, and
conducted robustness checks to evaluate whether childhood abuse was
independently predictive of particular health conditions. Fourth, this paper
provides the first assessment of health care access of survivors of childhood
abuse.
The remainder of this chapter proceeds as follows. In Section 4.2, I
describe sample characteristics. In Section 4.3, I describe the empirical
strategy. In Section 4.4, I present results. Finally, in Section 4.5, I conclude and
discuss implications for policy, health care delivery, and future research.
4.2 Sample characteristics
Childhood abuse was prevalent. In the self-interview module, 14.1 percent of
adults reported history of contact sexual abuse, 17.8 percent reported physical
abuse by a caregiver, and 12.0 percent reported chronic emotional abuse by a
caregiver. Almost one-third of the full sample – 30.7 percent – reported history
of at least one type of childhood abuse.
62
Children who experienced abuse were more disadvantaged by family
socioeconomic status and health. They lived in lower income households, on
average, were more likely to have grown up with a parent incarcerated at some
point during childhood, and relatedly, were less likely to live with both
biological parents. In addition, the group of children who experienced abuse
had higher rates of physical disability and depressive symptoms. See Table 1
for descriptive statistics.
While there were important differences in characteristics across the
children who experienced childhood abuse and those who did not, the
(overlapping) groups of children who experienced each type of abuse or
parental incarceration were generally similar to each other. In each ACE group,
the rate of experiencing each other ACE was at least 20 percent. See Appendix
Table A1 for sample characteristics broken down by individual ACEs.
Table 8. Childhood characteristics: sample means across childhood abuse history
a
(standard errors in parentheses)
Abuse No abuse
N=4,573 N=10,096
Female 0.565 (0.011) 0.460 (0.009) ***
Race/ethnicity
White NH 0.754 (0.026) 0.776 (0.024)
**
Black NH 0.162 (0.022) 0.148 (0.021)
Native American NH 0.011 (0.005) 0.009 (0.003)
Asian/Pacific Islander NH 0.026 (0.007) 0.028 (0.007)
Other NH 0.048 (0.005) 0.029 (0.003)
Hispanic 0.123 (0.017) 0.108 (0.017)
Biological parents in household
Mom and dad 0.431 (0.015) 0.620 (0.013)
***
Mom only 0.443 (0.013) 0.307 (0.011)
Dad only 0.063 (0.006) 0.038 (0.003)
Neither 0.063 (0.006) 0.035 (0.004)
63
Childhood household income, median
(2010$)
$47,026 $58,782
–
Child health and cognition
Physical disability 0.031 (0.004) 0.020 (0.002) ***
Learning challenge 0.173 (0.01) 0.158 (0.012)
Depressive symptoms 0.391 (0.012) 0.265 (0.009) ***
Adverse childhood experiences
Sexual abuse 0.472 (0.011) 0.000 –
Physical abuse 0.576 (0.014) 0.000 –
Emotional abuse 0.382 (0.011) 0.000 –
Parental incarceration 0.198 (0.011) 0.081 (0.006) ***
NH, Non-Hispanic
a
Except where noted as “median.”
*p<0.05, **p<0.01, ***p<0.001
4.3 Methods
I used several approaches to study the effects of childhood abuse on adult
health and examine health care access of survivors. First, I calculated the
unadjusted means for all health and health care access outcomes. Second, I
implemented school fixed effects models controlling for family socioeconomic
status to study the relationship between childhood abuse and adult health – for
those outcomes for which unadjusted differences across the group reporting
and not reporting childhood abuse were significant at the 5 percent level.
Third, I examined robustness of results to inclusion of additional controls for
childhood health and environment not preferred in the main model. Fourth, I
implemented sibling fixed effects regression models. Fifth, I examined
confounding by childhood socioeconomic status. Finally, I studied the
heterogeneity of the relationship between childhood abuse and adult health by
sociodemographic characteristics.
64
4.3.1 Exposure measure: childhood abuse
In this chapter, I focus on any childhood abuse: sexual, physical, or chronic
emotional abuse. Whereas in Chapter 3, I found that the effect sizes for
childhood sexual abuse dominated the associations measured between the
other types of abuse (physical and emotional) and the human capital outcomes,
the pattern did not replicate for health outcomes. That is, preliminary analyses
showed that no type of childhood abuse was consistently the most important
in predicting health outcomes, across the conditions studied. To explain this
difference in results, I hypothesize that childhood sexual abuse is a more
prolonged dose of trauma, compared to physical and emotional abuse in
childhood. While both health and human capital can accumulate and
depreciate, there is a latent period between harmful exposures and
development of health conditions—in particular, chronic conditions. In
contrast, a decline in mental health, for example, can have immediate impact
on leaving or losing a job. The data used here support the hypothesis that
childhood sexual abuse is a more prolonged dose of trauma. In adulthood,
survivors of childhood sexual abuse have significantly higher risk of suicide
attempt, compared to those who experienced physical or emotional abuse (see
Appendix Table A1).
4.3.2 Outcome measures
I studied outcomes reported in young adulthood during Wave IV: health
conditions, recent prescription medication use, and health care access. In
consultation with two clinicians, I grouped health conditions and medications
into the following groups: (1) cancer, (2) cardiometabolic, (3) gastrointestinal,
(4) infectious, (5) respiratory/allergic, and (6) nervous system conditions. In
65
the group with a cardiometabolic condition, I included participants with
undiagnosed diabetes or hypertension based on health measurements taken
during the survey. I used reports of recent symptoms to identify respondents
with gastrointestinal or infectious (cold or flu) symptoms in the last two
weeks. I identified respondents who recently used medications in these six
groups through their reports of all prescription medications used in the last
four weeks. See Appendix Table B1 for more details on the component
conditions and medications for each group.
To study heath care access, I examined self-reports of uninsured status,
unmet medical need, and health decline due to the unmet need. Uninsured
reflects status at time of survey while unmet medical need reflected self-
reported need in the past twelve months.
4.3.3 Baseline model
To study the association between childhood abuse and adult health outcomes,
I used a school fixed effects strategy while controlling for demographics,
childhood socioeconomic status, and parental incarceration. In the baseline
specification below, let Y
i
denote the outcome for individual i. Y
i
is a binary
variable for a health condition, recent medication use, or a health care access
outcome. The parameter of interest is α in the following equation:
f (Y
i
)= 𝛼 ·ChildhoodAbuse
i
+ X
i
′β + ε
i
(4.1)
ChildhoodAbuse
i
is a dummy variable equal to one for individuals who
reported childhood sexual, physical, or chronic emotional abuse. X
i
is a vector
of controls and included a constant. Other components of X
i
included
demographics (age when outcome measured, sex, and race/ethnicity),
66
childhood household socioeconomic status (highest parental education level
and household income), school fixed effects, and parental incarceration.
Lastly, ε
i
represents individual, idiosyncratic error. I implemented models with
logit regression and replicated with ordinary least squares (OLS).
The school fixed effects sweep out the effects of any unobserved
confounders which are characteristics of the school environment. For example,
perhaps a child attending a school in a lower income, higher crime
neighborhood is both more likely to experience abuse and less likely to receive
quality health education and have access to nurses at school. School fixed
effects remove the effect of going to a particular school with these
characteristics from the estimate of the association between childhood abuse
and the outcome. I controlled for parental incarceration rather than including
in the key exposure variable as adverse childhood experiences more generally
because childhood abuse and parental incarceration are very different. While
parental incarceration can also cause substantial childhood stress, in addition,
parental incarceration is a huge economic shock to the household (Cox 2018,
Grinstead, et al. 2001).
4.3.4 Robustness checks
4.3.4.1 Sensitivity analyses
The main concern for causal interpretation is that unobserved parent or child
behavior both make a child more likely to be abused and to develop poorer
health. For example, a neglectful parent might be more likely to abuse the
child, might be less likely to take the child to the doctor for health
67
maintenance, and might be less likely to teach the child healthy behaviors.
Thus, in sensitivity analyses, I examined the robustness of results for the
health outcomes to inclusion of additional covariates relating to childhood
environment, health, and health care access. In addition, I studied to what
extent results were explained by current characteristics. The additional
controls noted in this section were not preferred in the baseline model either
because (i) the information came from a different survey section, thereby
limiting the sample, (ii) the factor included here as a control could be an
outcome of childhood abuse, such as depressive symptoms, or (iii) the variable
might provide marginal additional information and be collinear with
covariates in the baseline model.
In the set of sensitivity analyses addressing potential confounding by
other childhood experiences or family or neighborhood environment, I
additionally controlled for physical or supervisory neglect, attending a well
child doctor visit in the past twelve months (prior to Wave I interview),
continuous health insurance coverage for the past twelve months, heavy
parental drinking, cigarette use by a household member, and neighborhood
education information at the Census block group level.
13
To address the concern
that unobserved child-level characteristics explain the results, I controlled for
traits which might impede adoption of health habits and might lead to higher
risk of childhood abuse: learning challenge and depressive symptoms.
Learning challenge included both learning disability and diagnosis of
attention-deficit/hyperactivity disorder due to the language of the survey
questions. I identified childhood depressive symptoms by using a five-item
13
Specifically, the neighborhood education variable was the percentage of residents aged 25
years or older without a high school degree.
68
instrument of the Center for Epidemiological Studies Depression Scale (Orme,
et al. 1986, Radloff 1977), which was included in the Wave I survey.
In other sensitivity analyses, I controlled for factors measured at the
same time as the outcome: current insured status, currently a smoker, ever a
smoker, and month of interview. I controlled for current insured status
because insured status might be differential across the groups who did and did
not report abuse (which is studied in this paper) while also affecting both the
likelihood of having a diagnosed health condition or likelihood of prescription
medication use. I controlled for smoker status to examine to what extent this
risk factor might explain the relationship between childhood abuse and adult
health. And, I included dummy variables for month of interview to allow for
seasonal patterns of health status for the outcomes assessed in recent periods:
gastrointestinal symptoms in the past two weeks; cold, flu, or fever symptoms
in the past two weeks; and prescription medication use in the past four weeks.
4.3.4.2 Childhood abuse, family socioeconomic status,
and health: a decomposition
In this section, I examined confounding by childhood socioeconomic status.
The goal of this exercise was to determine whether the association between
childhood abuse and adult health was explained by family socioeconomic
status. A concern for causal inference is that the estimates for childhood abuse
from the models presented above might be confounded by unobserved
variation in childhood socioeconomic status and not represent true effects of
childhood abuse. If this is the case, then controlling for parental education
should reduce the size of the coefficients on childhood abuse.
69
Here, I measured the amount of the association between childhood
abuse and adult health which was explained by family socioeconomic status
and the amount of the association between family socioeconomic status and
adult health which was explained by childhood abuse. To do this, I
implemented a set of three regressions specified below. I discretized childhood
socioeconomic status as a measure of low parental education to study
movements in the coefficient on this variable rather than on a collection of
measures of socioeconomic status included in the baseline regression (e.g.,
household income, school fixed effects). Specifically, LowParentalEducation is
a binary variable which indicates that the parents’ highest education level was
below college degree.
14
I did not use household income as the one indicator of
socioeconomic status because childhood abuse could cause lower income
through parental incarceration, whereas parental educational attainment by
the time of their children’s adolescence is a less mutable measure. In each
model, I controlled for current age and sex. I did not control for race/ethnicity,
which might be a proxy for family socioeconomic status.
In equation (4.2) below, I regressed each outcome on childhood abuse
to measure this association (while controlling for age and sex). In equation
(4.3), I regressed the outcome on the binary measure of low parental education
level. In equation (4.4), the unrestricted model, I simultaneously included
childhood abuse and low parental incarceration on the right-hand side, while
controlling for age and sex as in the restricted models.
f (Y
i
)= 𝛼 ̇·ChildhoodAbuse
i
+ 𝛽 1
̇ ·Age
i
+ 𝛽 2
̇ ·Sex
i
+ 𝜈 ̇ i
(4.2)
14
While parental education below college degree is described as “low parental education” for
brevity here, the average child has parents without college degrees. In this cohort, 70 percent
of children had parents with education level below a college degree.
70
f (Y
i
)= 𝛾 ̇·LowParentalEducation
i
+𝛽 1
̈ ·Age
i
+ 𝛽 2
̈ ·Sex
i
+ 𝜈 ̈ i
where LowParentalEducation
is
{
1 𝑖𝑓 𝑛𝑒𝑖𝑡 ℎ𝑒𝑟 𝑝𝑎𝑟𝑒𝑛𝑡 𝑐𝑜𝑚𝑝𝑙𝑒𝑡𝑒𝑑 𝑎 𝑐𝑜𝑙𝑙𝑒𝑔𝑒 𝑑𝑒𝑔𝑟𝑒𝑒 0 𝑜𝑡 ℎ𝑒𝑟𝑤𝑖𝑠𝑒
(4.3)
f (Y
i
)= 𝛼 ⃛·ChildhoodAbuse
i
+ 𝛾⃛·LowParentalEducation
i
+ 𝛽 1
⃛
·Age
i
+
𝛽 2
⃛
·Sex
i
+ 𝜈⃛
i
(4.4)
I tested formally whether the coefficients from the full model were
significantly smaller than the coefficients from the restricted model. That is, I
evaluated whether parental education level explained away any of the
difference in health outcomes by childhood abuse history and whether
childhood abuse history explained any of the difference in health outcomes by
parental education level. To do this, I tested whether the differences in
coefficients 𝛼 ̇ – 𝛼 ⃛ and 𝛾 ̇ – 𝛾⃛ were significantly greater than zero, following
Clogg, Petkova, and Haritou (1995) to compute the standard error of the
difference in coefficients and then evaluating significance with a right-tailed
t-test. I also report the absolute difference in coefficients from the restricted
to unrestricted models and the ratio of the difference in coefficients to the
magnitude (absolute value) of the coefficient from the unrestricted model. For
example, I calculated the ratio (𝛼 ̇ – 𝛼 ⃛)/|𝛼 ̇ |, which is the fraction of the
association between childhood abuse and adult health which was explained by
parental education.
71
4.3.4.3 Sibling fixed effects regression
I used sibling fixed effects regression models to address potential confounding
at the family level. Here, I asked a different question: is non-caregiver
childhood sexual abuse correlated with health or access outcomes when
parsing out family-level characteristics? I focussed on non-caregiver sexual
abuse rather than caregiver sexual, physical, or emotional abuse because
variation in caregiver abuse experience between siblings might more likely
reflect measurement error rather than true differences in these exposures. For
these analyses, I excluded participants who experienced caregiver sexual
abuse to avoid comparison between a sibling who reported non-caregiver
sexual abuse and a sibling who reported caregiver sexual abuse but no non-
caregiver sexual abuse. Specifically, I estimated 𝛼 ̇ in the OLS conditional fixed
effects regression below for individual i in sibling group s, controlling for
individual age and sex. The new term, μ
is
, is the component which is specific
to the sibling group, and η
i
is the individual error term.
f (Y
is
)= 𝛼 ̃·NonCaregiverSexualAbuse
is
+ β
1
·Age
is
+ β
2
·Sex
is
+ μ
is
+ η
i
(4.5)
The purpose of the sibling fixed effects regression model was to parse out any
unobserved heterogeneity due to family environmental characteristics to
which the siblings were equally exposed, such as having neglectful parents
who did not promote healthy behaviors. However, estimating the effect of non-
caregiver childhood sexual abuse in a sibling fixed effects regression assumes
no effect on the sibling, through secondary trauma or other reasons. Thus, 𝛼 ̃
might provide a conservative estimate of the effect of non-caregiver childhood
sexual abuse. Because fixed effects regression only exploits the variation
within the sibling group and because the sample of participants with a sibling
72
in Add Health was less than one-quarter of the full sample, estimates would be
less precise. Nevertheless, examining estimates from sibling fixed effects
regression is a useful exercise to determine whether there remained a
relationship between the exposure and outcome once parsing out unobserved
family-level factors such as parental quality.
4.3.5 Heterogeneity analyses
I examined the heterogeneity of the relationship between childhood abuse and
health and health care access outcomes by sociodemographic characteristics.
To do this, I replicated the baseline model three times per outcome to interact
childhood abuse with (1) sex, (2) race/ethnicity, and (3) log of childhood
household income in separate regressions. Beyond the condition categories
studied in baseline models, here I also examined specific health outcomes
measured during the survey in efforts to disentangle whether there were true
disparities in health status vs. disparities in diagnosed status. I studied two
additional health outcomes: high cholesterol, as identified from blood samples
drawn during the survey, and obesity, as identified by body mass index (BMI)
from height and weight measured at the time of survey.
4.4 Results
4.4.1 Descriptive results
The unadjusted results showed that individuals who had suffered childhood
abuse were more likely to suffer poorer health in adulthood – across every
73
category of health. However, they only had higher rates of recent medication
use for cardiometabolic, infectious, and nervous system conditions. Higher
rates of health issues among those who experienced childhood abuse without
corresponding increased rates of medication use might be due to lack of access.
Survivors reported higher rates of being uninsured, having an unmet medical
need in the past 12 months, and experience of worsening of health due to the
unmet need. See Table 2 below.
Table 9. Outcome means by childhood abuse status
(1) (2) (3)
Full sample
Childhood
abuse No report
Diff.
col. 2
vs. 3
N=14,741 N=4,573 N=10,096
Mean (s.e.)
A. Diagnosis/symptoms
Cancer 0.013
(0.001)
0.020
(0.003)
0.010
(0.001)
***
Cardiometabolic 0.332
(0.007)
0.351
(0.012)
0.323
(0.007)
*
Gastrointestinal 0.090
(0.003)
0.240
(0.010)
0.196
(0.007)
***
Infectious 0.236
(0.005)
0.130
(0.008)
0.073
(0.003)
***
Nervous system 0.336
(0.008)
0.316
(0.010)
0.254
(0.007)
***
Respiratory/allergic 0.272
(0.006)
0.266
(0.008)
0.222
(0.006)
***
B. Medication use
Cancer 0.007
(0.001)
0.008
(0.002)
0.006
(0.001)
Cardiometabolic 0.054
(0.003)
0.062
(0.005)
0.051
(0.003)
*
Gastrointestinal 0.031
(0.003)
0.035
(0.004)
0.029
(0.003)
74
Infectious 0.068
(0.003)
0.084
(0.006)
0.061
(0.003)
***
Nervous system 0.130
(0.005)
0.175
(0.009)
0.111
(0.005)
***
Respiratory/allergic 0.044
(0.003)
0.046
(0.004)
0.043
(0.003)
C. Health care access
Uninsured 0.224
(0.007)
0.264
(0.011)
0.207
(0.008)
***
Unmet medical need 0.249
(0.006)
0.338
(0.009)
0.211
(0.007)
***
Health decline due to unmet need 0.102
(0.004)
0.161
(0.007)
0.077
(0.004)
***
NOTE.— See Appendix Table B1 for a list of health conditions and medications within each group.
Unmet medical need was assessed over the past 12 months. The full sample size noted in column
(1) exceeds the sum across columns (2) and (3) due to missing observations for the childhood
abuse measure.
*p<0.05, **p<0.01, ***p<0.001
4.4.2 Baseline results
4.4.2.1 Health
The positive, significant association between childhood abuse and poorer
health persisted in adjusted models (for each condition and for each
medication group for which there were differences in unadjusted means by
abuse status). Average marginal effects from logit models indicate that
childhood abuse predicted 11.4 percentage points greater likelihood of
diagnosis of a nervous system condition, 5.5 percentage points greater
likelihood of diagnosis of a respiratory or allergic condition, 3.9 percentage
points greater likelihood of a cardiometabolic condition, 1.0 percentage point
greater likelihood of cancer diagnosis, 4.6 percentage points greater likelihood
of recent gastrointestinal symptoms, and 3.3 percentage points greater
75
likelihood of recent cold or flu symptoms. Results from OLS were similar (see
Table 3 below). Survivors’ greater likelihood of using medications for
cardiometabolic, infections, and nervous system conditions persisted in these
adjusted models. Translating all results to relative terms as the average
marginal effect percentage of the outcome mean, childhood abuse was
associated with 33.9 percent greater likelihood of diagnosis of a nervous
system condition by young adulthood and 38.4 percent greater likelihood of
recent medication use for an associated condition, 20.2 percent greater
likelihood of diagnosis of a respiratory or allergic condition, 11.7 percent
greater likelihood of diagnosis of a cardiometabolic condition and 27.7 percent
greater likelihood of recent medication use for an associated condition, 78.2
percent greater likelihood of cancer diagnosis, 51.1 percent greater likelihood
of recent gastrointestinal issues, 14.0 percent greater likelihood of recent cold
or flu symptoms, and 20.5 percent greater likelihood of recent medication use
for an infectious condition (not just cold or flu). Results survived multiple
comparison adjustment (see Appendix Table B2).
Table 10. Childhood abuse and adult health: baseline results
Coefficients on childhood abuse (standard errors in parentheses) [average marginal effects in
brackets]
A. Diagnosis/symptoms
Cancer
Cardio-
metabolic
Gastro-
intestinal Infectious
Nervous
system
Respirator
y/ allergic
Logit 0.567** 0.184** 0.547*** 0.186** 0.544*** 0.282***
(0.184) (0.06) (0.089) (0.057) (0.054) (0.059)
[0.010] [0.039] [0.046] [0.033] [0.114] [0.055]
OLS 0.008** 0.039** 0.048*** 0.034*** 0.116*** 0.055***
(0.003) (0.013) (0.008) (0.010) (0.012) (0.012)
R
2
0.026 0.056 0.039 0.025 0.096 0.037
Outcome
mean
0.013 0.332 0.090 0.236 0.336 0.272
B. Medication use
76
Cancer
Cardio-
metabolic
Gastro-
intestinal Infectious
Nervous
system
Respirator
y/ allergic
Logit – 0.272* – 0.218* 0.437*** –
– (0.106) – (0.096) (0.076) –
– [0.015] – [0.014] [0.050] –
OLS – 0.014* – 0.014* 0.051*** –
– (0.006) – (0.006) (0.009) –
R
2
– 0.022 – 0.029 0.043 –
Outcome
mean
0.007 0.054 0.031 0.068 0.130 0.044
NOTE.— See Appendix Table B1 for a list of health conditions and medications within each group.
Health categories for which there were no differences in unadjusted means are omitted. Sample
size was N = 14,037.
*p<0.05, **p<0.01, ***p<0.001
4.4.2.2 Health care access
Adults who experienced childhood abuse had poorer access to health care.
Childhood abuse predicted 6.8 percentage points greater likelihood of being
uninsured, 9.5 percentage points greater likelihood of reporting an unmet
medical need, and 6.9 percentage points greater likelihood of reporting
degraded health due to the unmet need (from logit models). See logit and OLS
results in Table 4. Translated to relative terms, childhood abuse was associated
with 30.3 percent greater likelihood of being uninsured in young adulthood,
38.2 percent greater likelihood of reporting an unmet medical need, and 67.7
percent greater likelihood of reporting subsequent decline in health.
77
Table 11. Childhood abuse and adult health care access: results from the
baseline model
Coefficients on childhood abuse (standard errors in parentheses) [average marginal effects
in brackets]
Uninsured
Unmet medical
need
Health decline due
to unmet need
Logit 0.312** 0.600*** 0.746***
(0.070) (0.054) (0.078)
[0.068] [0.095] [0.069]
OLS 0.050*** 0.113*** 0.072***
(0.012) (0.011) (0.008)
R
2
0.078 0.051 0.039
Outcome mean 0.224 0.249 0.102
NOTE.— Sample size was N = 14,037.
*p<0.05, **p<0.01, ***p<0.001
4.4.3 Robustness checks
4.4.3.1 Sensitivity analyses
Results were robust across sensitivity analyses additionally controlling for
other types of childhood adversity, characteristics of the childhood
environment, insured status, well child visit attendance, and adult
characteristics.
The most important factor in determining adult health in these models
was childhood mental health. Having depressive symptoms in childhood was
associated with 4.1 (s.e., 1.2) percentage points greater likelihood of recent
cold or flu symptoms and 3.0 (0.7) percentage points greater likelihood of
recent gastrointestinal issues. The association between childhood depressive
symptoms and adult health did not, however, explain the relationships
78
between childhood abuse and adult health. Estimates of the association
between childhood abuse and adult health remained fairly stable across
sensitivity analyses. See Appendix Table B3.
4.4.3.2 Childhood abuse, family socioeconomic status,
and health: a decomposition
Childhood abuse was independently predictive of worse health. Parental
education level did not explain away the association between childhood abuse
and adult health. If the estimates of the effect of childhood abuse on adult
health presented above were confounded by unobserved variation in childhood
socioeconomic status, then controlling for parental education should erode the
estimates for childhood abuse. The coefficients on childhood abuse, however,
were not reduced – with one exception. Low parental education explained just
6.9 percent of the association between childhood abuse and having a
cardiometabolic condition in adulthood. For each other health outcome, there
were no significant reductions in the association between childhood abuse and
health from the restricted model to the full model controlling for parental
education.
This exercise also demonstrated that childhood abuse partially
explained the association between childhood socioeconomic status and adult
health. For example, childhood abuse explained 16.7 percent of the association
between low parental education and cancer diagnosis. See results in Appendix
Table B4.
Notably, low parental education did explain small fractions of the
association between childhood abuse and health care access outcomes.
79
Specifically, coefficients on childhood abuse from the restricted model to the
model additionally controlling for low parental education decreased by 9.1
percent for uninsured status, by 2.5 percent for unmet medical need, and by
1.8 percent for degraded health due to unmet medical need. In perspective,
childhood abuse explained much more of the association between low parental
education and unmet medical needs: 14.8 percent for unmet medical need and
19.3 percent for degraded health due to unmet medical needs.
4.4.3.3 Sibling fixed effects regression
In general, sibling fixed effects estimates of the association between non-
caregiver sexual abuse and adult health were larger in magnitude but less
precise compared to the baseline estimates of the association between
childhood abuse and these same health outcomes. The only estimates which
remained significant here showed that non-caregiver sexual abuse was
associated with 47.9 percent greater likelihood of diagnosis of a nervous
system condition and 68.0 percent greater likelihood of diagnosis of a
respiratory/allergic condition. Results for recent cold or flu symptoms and
recent use of medication for an infectious condition reduced to zero in the
sibling fixed effects models. Substantial power was lost across the baseline and
sibling fixed effects regressions, with sample size reducing from 14,037 for the
baseline models to 3,390 for the sibling fixed effects models. See results in
Table 5.
80
Table 12. Non-caregiver sexual abuse and adult health: results from sibling
conditional fixed effects regression
Coefficients on non-caregiver sexual abuse (standard errors in parentheses)
A. Diagnosis/symptoms
Cancer
Cardio-
metabolic
Gastro-
intestinal Infectious
Nervous
system
Respirator
y/ allergic
OLS 0.021 0.048 0.053 -0.054 0.161* 0.185**
(0.048) (0.078) (0.051) (0.071) (0.073) (0.072)
Outcome
mean
0.013 0.332 0.090 0.236 0.336 0.272
B. Medication use
Cancer
Cardio-
metabolic
Gastro-
intestinal Infectious
Nervous
system
Respirator
y/ allergic
OLS – 0.030 – 0.000 0.028 –
– (0.023) – (0.019) (0.023) –
Outcome
mean
0.007 0.054 0.031 0.068 0.130 0.044
NOTE.— See Appendix Table B1 for a list of health conditions and medications within each group.
Health categories for which there were no differences in unadjusted means are omitted. Sample
size was N=3,390.
*p<0.05, **p<0.01, ***p<0.001
In the sibling fixed effects models, the magnitude of the association
between non-caregiver sexual abuse and uninsured status shrunk (relative to
results from baseline estimates for any childhood abuse) while estimates for
unmet medical needs increased. The magnitude of the association between
childhood abuse and adult uninsured status reduced substantially in the sibling
fixed effects model and lost significance. There remained large discrepancies
in rates of unmet medical needs and worsening of health due to unmet needs
between siblings who experienced non-caregiver childhood sexual abuse and
siblings who did not. Here, non-caregiver childhood sexual abuse was
associated with 13.7 percentage points greater likelihood of reporting an
unmet medical need in the past 12 months and 19.0 percentage points greater
likelihood of reporting worsening of health due to the unmet need. See results
81
in Table 6. These results for unmet medical need and accompanying health
decline might reflect their greater health burden. It is important to
acknowledge, however, that these measures of unmet need and health decline
were subjective and could reflect differential reporting rates given same health
status and health care use.
Table 13. Non-caregiver sexual abuse and adult health care access: results
from sibling conditional fixed effects regression
Coefficients (standard errors in parentheses)
Uninsured
Unmet medical
need
Health decline due
to unmet need
OLS 0.030 0.137* 0.190***
(0.069) (0.062) (0.054)
Outcome mean 0.224 0.249 0.102
NOTE.— Sample size was N = 3,049.
*p<0.05, **p<0.01, ***p<0.001
4.4.4 Heterogeneity analyses
Sex, race, and childhood SES modified the relationships between childhood
abuse and several health and health care access outcomes. In general, health
disparities by childhood abuse status were larger for women vs. men and for
Whites vs. other racial groups. In particular, women survivors of childhood
abuse vs. male counterparts were at higher risks of diagnosis of a nervous
system condition, diagnosis of a respiratory/allergic condition, and having
recent gastrointestinal symptoms. Results by race showed that, given
childhood abuse, Asians/Pacific Islanders were less likely than Whites to be
diagnosed with a nervous system condition, to be diagnosed with cancer, and
to report recent cold or flu symptoms. Black survivors of childhood abuse were
82
also less likely than White survivors to be diagnosed with a nervous system
condition. While survivors of “other” race were less likely than White
survivors to be diagnosed with cancer, they were more likely than Whites to
be diagnosed with a respiratory/allergic condition. There were also differences
by childhood household income: individuals from higher vs. lower income
families were more likely to be diagnosed with nervous system conditions and
more likely to be obese.
There were some disparities in uninsured status and reporting of unmet
medical needs. Given being a survivor of childhood abuse, Native Americans
vs. Whites were less likely to be uninsured. Women survivors reported unmet
medical needs and degraded health due to unmet medical need at higher rates
compared to male survivors. See results from these heterogeneity analyses in
Appendix Tables B5-B18.
It is difficult to disentangle whether childhood abuse was even worse
for the health of females vs. males and Whites vs. other races or if they were
only more likely to be diagnosed. For example, higher rates of diagnosis of
nervous system conditions in female vs. male survivors of childhood abuse and
in Whites vs. Black and Asian/Pacific Islander survivors might reflect stigma
barriers to mental health diagnoses among males and minority racial groups.
Results for obesity from measured BMI and for recent acute symptoms – cold
or flu symptoms and gastrointestinal issues – suggest, however, that there may
be some real differences in the association between childhood abuse history
and health by sociodemographic characteristics.
83
4.5 Discussion and conclusions
Nearly one in three children were abused in childhood. Such high prevalence
has implications for policy and large-scale changes in health care delivery.
Survivors of childhood abuse faced greater risks of poor health in adulthood –
both for psychological and somatic conditions. In this paper, I found that young
adults who had experienced sexual, physical, or chronic emotional abuse in
childhood were more likely to have been diagnosed with nervous system
conditions, respiratory or allergic conditions, cardiometabolic conditions, and
cancer. Survivors were also more likely to report recent gastrointestinal
symptoms.
Specifically, survivors of childhood abuse were 34 percent more likely
to be diagnosed with a nervous system condition and 38 percent more likely to
report recent medication use for an associated condition, 20 percent more
likely to be diagnosed with a respiratory or allergic condition, 12 percent more
likely to have a cardiometabolic condition and 28 percent more likely to report
recent medication use for an associated condition, 78 percent more likely to be
diagnosed with cancer, and 51 percent more likely to report recent
gastrointestinal symptoms. These health disparities were not explained by
childhood socioeconomic status (or unobservables correlated with parental
education level). And, all results for the health outcomes abovementioned in
this section were robust to sensitivity analyses controlling separately for
several potential confounders relating to childhood health as well as
household- and neighborhood-level risk factors. In general, health outcomes
among survivors of childhood abuse were worse for females vs. males and for
Whites vs. other racial groups.
84
This paper is the first, to my knowledge, to measure positive links
between childhood abuse and domains of physical health beyond
cardiometabolic health, while controlling for childhood socioeconomic status.
Results are consistent with prior literature which measured positive
associations between childhood abuse and various cardiometabolic conditions
(Duncan, et al. 2015a, Duncan, et al. 2015b, Ford and Browning 2014,
Fuemmeler, et al. 2009, Lee, et al. 2014, Ranchod, et al. 2016, Rich-Edwards,
et al. 2010, Richardson, et al. 2014). In a recent statement by the American
Heart Association, the authors reviewed the evidence on the influence of
childhood adversity on cardiometabolic outcomes and prescribed further
research to inform the development of early interventions. The authors
suggested future investigations to identify the mechanisms behind the
association and to identify both vulnerability factors leading to greater risk
that adverse childhood experiences lead to poorer cardiometabolic health and
resiliency factors protecting against poorer cardiometabolic health (Suglia, et
al. 2018). The present study suggests that in future work, researchers should
devote attention to health status in a more comprehensive fashion to study
what factors increase and decrease risks of poorer health among people who
have suffered adverse childhood experiences.
This paper is also the first, to my knowledge, to study health care access
of adults who experienced childhood abuse. While this population had a higher
health burden, they also had poorer access to health care. In baseline models,
childhood abuse was associated with 30 percent greater likelihood of being
uninsured, 38 percent greater likelihood of reporting unmet medical needs in
the past 12 months, and 68 percent greater likelihood of reporting an
accompanying decline in health. Higher rates of being uninsured might be the
result of lower rates of full-time employment, which I found in Chapter 3. I do
85
not, however, conclude here that childhood abuse caused greater risk of being
uninsured in young adulthood because estimates were substantially smaller
and insignificant in a sibling fixed effects regression model of non-caregiver
sexual abuse. However, the large associations between childhood abuse and
unmet medical needs, which did survive sibling fixed effects regression
models, should direct attention to the quality of health care available to
address the needs of survivors of childhood abuse.
Limitations in the present paper remain. First, I studied outcomes in
young adulthood, and results might be different at older ages. Health
disparities might diminish or widen over time. Second, the study period
predates major health reform – the Affordable Care Act of 2010 – which might
have reduced gaps in health care access across those who did and did not suffer
childhood abuse.
Survivors of childhood abuse had worse health and worse access to
health care. The present study provided evidence suggesting that childhood
abuse led to worse health in adulthood beyond poorer mental well-being.
Further, this paper demonstrated that adult survivors of childhood abuse were
at much greater risk of reporting unmet medical needs. This paper is highly
relevant to current health policy discussions by highlighting a vulnerable
population which might stand to lose from roll-backs of insurance protections
and cuts in funding to public insurance programs and safety net providers.
Further research is needed to identify effective approaches to detecting cases
of childhood abuse and treating symptoms so that potential long-term health
consequences are avoided.
86
The Child Left Behind: Parental Incarceration and
Adult Human Capital in the United States
Abstract
Parental incarceration can be not only a huge economic shock to the family but
also a source of psychological stress on family members. The effects of this
stress – whether through economic disadvantage or psychological stress –
might be long-lasting. Neuroscientists and psychiatrists describe how chronic
stress in childhood may impair cognitive development. Exposure to parental
incarceration is particularly prevalent in the United States, where over 7
percent of children have lived with a parent who was incarcerated during their
childhood. In this paper, I investigate whether incarceration has long-term
human capital consequences on children in the US. I provide evidence at the
population level that parental incarceration causes lower rates of high school
diploma receipt and likely causes lower rates of full-time employment in young
adulthood. This work adds to the body of evidence documenting an
intergenerational transmission of socioeconomic disadvantage and has
important implications for social policy. Within the education system, results
might motivate improved support for children’s socio-emotional health.
87
5.1 Introduction
The United States has the world’s highest incarceration rate (Wagner and
Sawyer 2018).
15
Accordingly, many children in the US have been exposed to
parental incarceration. Data from 2011-2012 show that about 7 percent of US
children had lived with a parent who was incarcerated during their lifetime
(Murphey and Cooper 2015). Black children have been exposed to parental
incarceration at higher rates. The likelihood of experiencing parental
incarceration by age 14 was 4 percent for White children and 25 to 28 percent
for Black children – in the cohort of children born in the US in 1990 (Wildeman
2009).
Parental incarceration might exact economic and psychological
consequences on children, which could have durable impacts on children’s
educational attainment and labor market outcomes in adulthood. Extant
literature show that children exposed to parental incarceration do face
heightened risks of hardship during childhood. They have higher rates of
school expulsion or suspension (Johnson 2009), household food insecurity
(Cox and Wallace 2016), and homelessness (Wildeman 2014). Parental
incarceration almost always comes with huge economic cost to families.
Beyond loss in income, families also face large financial costs from court fines,
inmate room and board, and visitation fees (Cox 2018, Grinstead, et al. 2001).
Economic strain on families might lead to reduced investment in children, as
families face trade-offs between spending resources to maintain contact with
the incarcerated family member and spending resources to support children
through academic enrichment or counseling, for example. At the same time,
15
Among nations with populations of at least 500,000.
88
children might struggle with focus and learning in school due to distress from
stigma of having an incarcerated parent, from missing the incarcerated parent,
or from unmet physical needs. Parental incarceration might also increase
children’s likelihood of engaging in crime due to economic need or through a
role model effect, thereby limiting employment opportunities through arrest
history.
On the other hand, it is also possible that parental incarceration allows
improvements in children’s education and eventual labor market outcomes. If
the parent is abusive or otherwise a problematic household figure, children
might benefit from removal of the parent from the home. For families in which
an incarcerated family member returns to the home, children might benefit if
incarceration is rehabilitative. Parental incarceration could also result in
better outcomes for children by discouraging participation in criminal
behaviors through their observation of the cost of crime. In fact, extant
literature suggests that there might be positive impacts of male incarceration
on children. Finlay and Neumark (2010) found that never-married
motherhood, driven by higher rates of incarceration of males in their
demographic and state marriage market, decreased rates of high school
dropout for Hispanic children.
The impact of parental incarceration on children certainly varies by
case, and long-term effects of parental incarceration at the population level
are ambiguous. The main challenge to measuring effects of parental
incarceration is that the groups exposed and not exposed to incarceration are
very different. In addition to the racial differences mentioned above, parents
in households where one has been incarcerated have lower educational levels
and lower occupational attainment (Mears and Siennick 2016). Importantly, in
89
studying effects of incarceration on children, context matters. Across nations
and localities, children of the incarcerated might have different outcomes due
to varying criminal justice, child welfare, and safety net policies.
In the present study, I used nationally representative data to measure the
impact of parental incarceration on children’s educational attainment and
labor market outcomes in the United States during young adulthood. I applied
partial identification methods to bound the likely effects of parental
incarceration. To do this, I examined robustness of results to different
assumptions about remaining selection on unobservables, using information
on the amount of selection on observables. My work is distinct from studies of
the spillover effects of incarceration which exploit policies of random
assignment of cases to judges, using administrative data.
16
These studies
measure the effects of incarceration of only the marginally incarcerated person
who was accused, for whom the likelihood that the judge delivered a sentence
of jail or prison time was believed to have been the deciding factor of whether
the accused was incarcerated or not. This method of analysis excludes the least
and most severe cases.
The remainder of this chapter is organized as follows. In Section 5.2, I
describe sample characteristics. In Section 5.3, I describe the empirical
strategy. In Section 5.4, I present results. Finally, in Section 5.5, I conclude and
discuss implications for public policy and future research.
16
For example, see Aizer et al. (2015) in the context of Chicago, IL on the effects of own
incarceration during childhood on human capital; Dobbie et al. (2018) in the context of Sweden
on the effects of parental incarceration on teen pregnancy, teen crime, and employment in
early adulthood; and Bhuller et al. (2018) in the context of Norway on the effects of parental
incarceration on school grades and being charged with a crime.
90
5.2 Sample characteristics
In this nationally representative sample, 11.7 (s.e., 0.6) percent of adults
reported having experienced at least part of childhood with a biological parent
or primary caregiver incarcerated. Exposure to parental incarceration was
usually due to incarceration of the father figure: 10.3 (0.6) percent of the full
sample experienced incarceration of a father figure while 2.3 (0.2) percent of
the sample experienced incarceration of a mother figure.
Those exposed to parental incarceration experienced other types of
childhood adversity at higher rates. They were twice as likely to experience
each type of childhood abuse: sexual, physical, and emotional. Median
household income for children who experienced parental incarceration was
about 55 percent of median income of households in which no child
experienced parental incarceration. There were stark differences by race.
Among the sample exposed to parental incarceration in childhood, the percent
of respondents who were Black was nearly double the rate in the group never
exposed to parental incarceration in childhood. Children who experienced
parental incarceration had much higher rates of depressive symptoms but no
difference in rate of educational challenges (learning disability or ADHD). The
large disparity in rates of depressive symptoms persisted in adulthood, when
those who had been exposed to parental incarceration had lower educational
attainment and worse labor market outcomes. See descriptive statistics in
Table 1 below.
91
Table 14. Sample means across parental incarceration history
a
(standard errors in parentheses)
No parental
incarceration
Parental
incarceration
N=11,551 N=1,717
Female 0.491 (0.008) 0.503 (0.018)
Race/ethnicity
White NH 0.788 (0.023) 0.677 (0.035)
***
Black NH 0.134 (0.019) 0.246 (0.036)
Native American NH 0.008 (0.002) 0.018 (0.01)
Asian/Pacific Islander NH 0.032 (0.008) 0.008 (0.003)
Other NH 0.032 (0.003) 0.047 (0.01)
Hispanic 0.109 (0.017) 0.135 (0.021)
Childhood household income,
median (2010$)
$61,721 $33,800
Child health and cognition
Child depressive symptoms 0.295 (0.009) 0.391 (0.016) ***
Educational challenge 0.157 (0.012) 0.168 (0.014)
Adverse childhood experiences
Parental incarceration 0.000 1.000
Sexual abuse 0.120 (0.005) 0.240 (0.014) ***
Physical abuse 0.149 (0.006) 0.340 (0.018) ***
Emotional abuse 0.101 (0.004) 0.226 (0.014) ***
Adult outcomes
Adult depressive symptoms 0.261 (0.008) 0.376 (0.016) ***
High school diploma 0.873 (0.009) 0.682 (0.017) ***
College degree 0.352 (0.017) 0.124 (0.012) ***
Full-time employment 0.737 (0.008) 0.652 (0.016)
***
Adult earnings, median (2010$) $30,414 $22,304
NH, Non-Hispanic.
a
Except where noted as “median.” * p<0.05, ** p<0.01, *** p<0.001
92
5.3 Methods
I employed multiple strategies to evaluate whether there was a causal impact
of parental incarceration on children’s eventual education or labor market
outcomes. First, in the baseline specification I implemented school fixed effects
regression models controlling for family-level SES and other adverse childhood
experiences (ACEs), among other confounders. Second, I examined robustness
of results by constructing bounds under varying assumptions about the
importance of unobservable confounders. Third, I tested for selection with two
sets of falsification tests.
5.3.1 Exposure measure: parental incarceration
In this chapter, I focus on parental incarceration during childhood. I defined
parental incarceration as report in Wave IV via personal interview of either
biological parent, mother figure, or father figure being incarcerated while the
child was alive but not yet 18-years-old. In constructing the parental
incarceration measure, I categorized responses of “don’t know” as such rather
than designating these values as missing. The “don’t know” group consists of
(i) respondents selecting “don’t know” in the initial question of whether the
parent had been incarcerated, which was largely due to the parent’s absence
in the respondent’s life
17
and (ii) respondents who acknowledged that a parent
had been incarcerated but responded “don’t know” to questions on age when
parent was incarcerated.
17
The majority of this group also responded “don’t know” to the preceding question of whether
the parent was still alive.
93
5.3.2 Outcome measures
I studied young adult outcomes: having a high school diploma, a college degree,
employment, full-time employment, and earnings level. I considered an
individual to be employed if working at least 10 hours per week, the minimum
amount of time inquired about in Add Health. I defined full-time employment
status as working at least 35 hours per week. Earnings included any wages,
bonuses, and self-employment.
5.3.3 Baseline model
In my baseline approach to study the effects of parental incarceration on adult
human capital outcomes, I used a school fixed effects strategy while controlling
for demographics, childhood socioeconomic status, and other adverse
childhood experiences. In the baseline specification below, let Y
i
denote the
outcome for individual i. The parameter of interest is α in the following
equation:
f (Y
i
)= 𝛼 ·ParentalIncarceration
i
+ X
i
′β + ε
i
(5.1)
ParentalIncarceration
i
is a dummy variable equal to one for persons who
reported having a parent incarcerated at any time during ages 0-17. X
i
is a
vector of controls and includes a constant. Other components of X
i
include
demographics (age when outcome measured, sex, and race/ethnicity),
childhood household socioeconomic status (highest parental education level
and household income), school fixed effects, and other adverse childhood
experiences (sexual abuse, physical abuse, emotional abuse, and “don’t know”
if parent incarcerated). Lastly, the model allows an individual, idiosyncratic
error term ε
i
.
94
By including school fixed effects, I am able to parse out unobserved
characteristics of the school and school neighborhood environments that
affected both the likelihood of parental incarceration and the likelihood of
success in school and on the labor market. For example, descriptive statistics
showed that a child in a low-income neighborhood (not shown) is more likely
to experience parental incarceration. And, we know that schools in low-income
neighborhoods, on average, provide lower quality education than is available
to children who attend schools in higher income neighborhoods. Interpreting
these baseline results causally requires the assumption that there is no
unobserved heterogeneity across children at the same schools which influence
both their likelihood of parental incarceration and their education and labor
market outcomes. In the following sections, I address the possibility of
remaining unobserved confounding.
I used ordinary least squares (OLS) to model binary outcomes and
replicate these models with logit and probit regression in ancillary analyses. I
analyzed annual earnings with a two-part model (Dow and Norton 2003,
Roodman 2009) to include people with zero earnings. The first part was a
probit model estimating the probability of having positive earnings, and the
second part was a generalized linear model (log link function and gamma
distribution) estimating earnings levels, using observations with positive
earnings.
95
5.3.4 Robustness checks
5.3.4.1 Bounding effects of parental incarceration
In this section, I calculated bounds on the effects of parental incarceration
under varying assumptions on the degree of selection on unobservables
relative to selection on observables, following partial identification methods
developed by Altonji, et al. (2002, 2005) and Oster (2017). Partial
identification allows the researcher to recover bounds on the estimated
treatment effects in contexts where unobservable variables may cause
confounding. This method relies on the premise that confounding on
observables provides insight into the influence of confounding on
unobservables.
In this context, I assessed whether unobserved factors, such as low
parental investment in children, fully explained the results. I adopted Oster’s
approach for linear models (Oster 2017). This method involves using
observable confounders to bound the likely effect of unobservable
confounders. Thus, key parameters are: (i) the amount of selection on
unobservables relative to selection on observables and (ii) the amount of
outcome variance that would be explained by the unobservable confounders.
The first value, the selection parameter, is denoted δ below, and the second
value informs the value R
max
, which represents the R
2
from the hypothetical
regression including unobservable confounders on the right-hand-side. I
constructed effect bounds for varying values of δ from 0 to 1. I made the
following alternative assumptions about the value R
max
:
96
(i) R
max
= 1.3𝑅 ̃
, where 𝑅 ̃
is the R
2
from equation (5.1) above with the
full set of observed controls, and
(ii) R
max
= 2𝑅 ̃
.
Below, I describe the approach to construct bounds on the effects of
parental incarceration. To start, consider a modified version of equation (5.1)
above, omitting the individual subscripts i:
Y = 𝛼 ̈·ParentalIncarceration + X ′𝛽 ̈ + W
2
+ η (5.2)
The new term, W
2
, represents unobservables which determine the human
capital outcome Y and are correlated with parental incarceration but not
correlated with any of the observable confounders. For example, unobserved
level of parental investment in children will be contained in W
2
if the partial
correlation with parental incarceration, conditional on the observed controls,
is non-zero. By definition as a confounder, cov(W
2
, Y) ≠ 0 and cov(W
2
,
ParentalIncarceration) ≠ 0. An additional requirement of W
2
is that cov(W
2
,
W
1
) = 0, denoting W
1
=X’𝛽 ̈ . This orthogonality requirement implies that W
1
captures observables in addition to any confounding from unobservables
which are correlated with the observables. Now, the parameter on parental
incarceration, 𝛼 ̈, is the true effect on outcome Y. The goal is to estimate bounds
for 𝛼 ̈, but the coefficients on the observed variables in W
1
should not be
interpreted causally. Recall that X includes sexual, physical, and emotional
abuse along with an indicator for the group designated as “don’t know”
whether a parent was incarcerated during childhood. Thus while elements of
W
2
are correlated with parental incarceration, they must be uncorrelated with
these other types of childhood adversity. Lastly, η is the individual
idiosyncratic error term which contains unobservables which determine the
outcome Y but are uncorrelated with parental incarceration, controls, or W
2
.
97
The ratio of selection on unobservables to selection on observables is
defined as
δ=
𝜎 2,𝑝 𝜎 2
2
/
𝜎 1,𝑝 𝜎 1
2
,
where σ
j,p
= cov(W
j
, ParentalIncarceration) and σ
j
2
= var(W
j
) for j ∈ {1, 2}. The
denominator of this ratio – the level of selection on observables – can be readily
computed from the data. The numerator – the selection on unobservables – is
identified with a restriction on the parameter 𝛼 ̈, the effect of parental
incarceration, and a restriction on R
max
. Here, I chose to compute bounds on
the effect 𝛼 ̈, thus I assume various values for the selection ratio δ and R
max
.
I calculated bounds on effect sizes for a set of R
max
values informed by
the R
2
from the regression with the full set of observed controls. In most cases,
it is unlikely that R
max
= 1, i.e., that the treatment/exposure, controls, and the
unobservable component W
2
fully explain the outcome, due to measurement
error in the outcome or other idiosyncratic variation in the outcome (Oster
2017). In the present case, there were likely important determinants of human
capital outcomes which were determined after Wave I but not caused by these
childhood factors. For example, all cases of depression not caused by parental
incarceration (or controls such as the other adverse childhood experiences)
would be included in an individual’s η error term.
The first R
max
condition above, R
max
= 1.3𝑅 ̃
, was proposed by Oster from
calculations with data from randomized studies. From her set of results from
randomized studies published in top economics journals which reported
uncontrolled and controlled estimates, she calculated that when holding fixed
98
δ=1, 90 percent of the results would survive if R
max
= 1.3𝑅 ̃
(Oster 2017).
18
I
present the main results of the bounding exercise under this condition: R
max
=
1.3𝑅 ̃
. Note that the second R
max
condition, R
max
= 2𝑅 ̃
, implies that the
unobserved confounders explained as much variance in the outcome as do the
observables.
Interpreting the ratio δ here is difficult given the large amount of
selection on observables. In general, if the estimated treatment (or exposure)
effect erodes to zero only when δ > 1, this is considered evidence that at least
part of the estimated effect is real (Altonji, et al. 2002, Altonji, et al. 2005,
Oster 2017). However, the true ratio of selection on unobservables to selection
on observables, δ, is likely less than 1 and may be much less than 1, in particular
when selection on observables is substantial as in the present case. I calculated
bounds for the effect size under the conditions δ=0, 0.25, 0.5, 1.
I calculated bounds on the effects of parental incarceration on the
following outcomes: high school diploma receipt, college degree attainment,
and full-time employment. I did not compute bounds for log earnings because
results from equation (5.1) showed no effect of parental incarceration on
earnings in the sample with positive earnings.
5.3.4.2 Falsification tests
I implemented direct tests of selection in two ways. In the first set of
falsification tests, I evaluated whether outcomes were worse among those who
18
To “survive” means both that the identified set does not include zero and that the identified
set is within 2.8 standard errors of the fully controlled estimate. The sample of results selected
in Oster (2017) come from randomized studies in top-five economics journals over a six-year
period. Across this set of 65 results, she found that 90 percent would survive a cutoff of R
max
= 1.3𝑅 ̃
, and only 40 percent would survive a cutoff of R
max
= 1.
99
had a parent who was incarcerated but not during their childhood (compared
to those who never had a parent incarcerated), which would suggest that
results might be driven by unobserved family qualities rather than direct
experience of exposure to parental incarceration. As a second test, I examined
whether there was a pre-existing difference in cognitive ability by exploiting
child age when parent was incarcerated and the timing of the Wave I survey
when a picture vocabulary test was administered.
I used data on the participant’s age when parent was first incarcerated
and when last released. I constructed five groups: (i) parent incarcerated and
last released before child was born, (ii) parent incarcerated after child’s birth
but before age 18, (iii) parent first incarcerated after child turned 18, (iv)
parent never incarcerated (the reference group), and (v) “don’t know,”
reflecting either a “don’t know” response for timing of parental incarceration
or “don’t know” response to whether a parent had ever been incarcerated. If
either group (i) or (iii) had worse outcomes compared to the group whose
parents were never incarcerated, conditional on the full control set, then there
would be evidence of unaddressed selection bias. That is, such a result would
suggest that unobserved family qualities and not direct exposure to parental
incarceration caused worse adult human capital outcomes.
5.3.5 Heterogeneity analysis
I examined the heterogeneity of the response to parental incarceration on
human capital outcomes by sociodemographic and incarceration
characteristics. Specifically, I measured differences by child sex, race, and
childhood household income; whether the father figure or mother figure was
incarcerated, and by duration of exposure to parental incarceration. I studied
100
differences by sociodemographic characteristics by interacting age, sex, and
race with parental incarceration in separate regressions. I investigated
whether there were differences by paternal vs. maternal incarceration in two
ways. First, I included separate variables for incarceration by a father figure
and incarceration by a mother figure (not mutually exclusive). Second, I
deconstructed the parental incarceration variable into mutually exclusive
groups: father and mother figures incarcerated, father figure incarcerated
only, mother figure incarcerated only, no parental incarceration, and “don’t
know” whether a parent was incarcerated during own ages 0-17. I repeated the
first set of regressions including separate measures for paternal and maternal
incarceration while interacting child sex with each incarceration measure.
Lastly, I evaluated whether the response to parental incarceration varied by
duration of exposure by replacing the binary measure for any parental
incarceration with the number of years between child age when a parental
figure was first incarcerated and age when a parental figure was last released.
I top-coded this value at 18 years, focusing only on parental incarceration
during childhood as in all other sections of this paper.
5.4 Results
5.4.6 Baseline results
Baseline results showed that adults who had been exposed to parental
incarceration in childhood had worse education and labor market outcomes.
The unadjusted differences in means for the education variables were large: -
18.5 percentage points for high school diploma and -23.6 percentage points for
101
college degree. The difference in unadjusted mean for full-time work was -8.1
percentage points and for earnings was -$9,766, or 32.1 percent of median
earnings for this young adult sample. In the fully adjusted models, the
differences in educational attainment and labor market outcomes persisted but
reduced substantially. The regressions with the full control set showed that
the estimates of the average marginal effect (AME) of parental incarceration
were -10.0 percentage points for high school diploma attainment, -7.8
percentage points for college degree attainment, -4.3 percentage points for
full-time work status, and -$3,645 for earnings (12.0 percent of median
earnings). See results in Table 2. Results for full-time employment and
earnings were driven by lower likelihood of having any earnings. Because, for
those with positive earnings, parental incarceration was not associated with
earnings level, I did not construct bounds on effects on earnings below.
For each outcome, more than 55 percent of the reduction in average
marginal effects for parental incarceration were explained by household SES.
However, this was not just due to parental incarceration causing lower
household income. Sensitivity analyses omitting household income from
control set (3) – that is, controlling for household SES only through highest
parental education – showed that parental education contributes to roughly
half of the reduction between the uncontrolled and fully controlled estimates
for parental incarceration.
102
Table 15. Estimates of the effects of parental incarceration on education and
labor market outcomes
Average marginal effects (standard errors in parentheses)
Controls None
a
Demographics
b
Col. 2 +
household
SES
c
Col. 3 +
school FE
d
Col. 4 + other
ACEs
e
(1) (2) (3) (4) (5)
A. High school diploma; mean (s.d.): 0.828 (0.378)
OLS -0.185*** -0.176*** -0.122*** -0.109*** -0.100***
(0.017) (0.018) (0.019) (0.018) (0.017)
R
2
0.034 0.044 0.101 0.153 0.155
Adjusted
R
2
0.034 0.044 0.100 0.144 0.146
B. College degree; mean (s.d.): 0.302 (0.459)
OLS -0.236*** -0.217*** -0.106*** -0.092*** -0.078***
(0.017) (0.017) (0.013) (0.014) (0.013)
R
2
0.032 0.055 0.209 0.259 0.264
Adjusted
R
2
0.032 0.055 0.208 0.251 0.256
C. Employment; mean (s.d.): 0.805 (0.396)
OLS -0.063*** -0.059*** -0.037* -0.039* -0.030*
(0.015) (0.015) (0.015) (0.015) (0.016)
R
2
0.005 0.023 0.033 0.076 0.068
Adjusted
R
2
0.005 0.023 0.032 0.067 0.058
D. Full-time employment; mean (s.d.): 0.706 (0.455)
OLS -0.081*** -0.076*** -0.055*** -0.052** -0.043*
(0.016) (0.016) (0.016) (0.016) (0.017)
R
2
0.006 0.039 0.045 0.097 0.093
Adjusted
R
2
0.005 0.038 0.044 0.088 0.083
E. Earnings; mean (s.d.): $35,025 (44,170); median: $30,414
2PM -$9,766*** -$8,450*** -$4,262*** -$3,766*** -$3,632**
(1,318) (1,230) (1,221) (1,109) (1,101)
103
NOTE.— Sample size was N=14,741 for each regression here.
a
The only right-hand side variables are parent incarcerated during childhood and “don’t know”
if parent incarcerated when aged 0-17 years.
b
Demographic controls include age when outcome measured; sex as female or male; and
race/ethnicity as White Non-Hispanic, Black Non-Hispanic, Native American Non-Hispanic,
Asian/Pacific Islander Non-Hispanic, Other Non-Hispanic (includes multi-racial), or Hispanic.
c
Household SES controls include log of childhood household income and highest parental
educational attainment as (i) less than high school, (ii) GED, (iii) high school diploma, (iv)
vocational school after high school, (v) some college, (vi) college graduate, or (vii) beyond 4-yr
college.
d
School fixed effects: schools were the primary sampling unit.
e
Other ACEs include childhood sexual, physical, and emotional abuse.
* p<0.05, ** p<0.01, *** p<0.001
5.4.7 Robustness checks
5.4.7.1 Bounding effects of parental incarceration
Evaluation of robustness of results to varying assumptions about remaining
selection on unobservables provided strong support for real, negative effects
of parental incarceration on high school diploma receipt and some support for
effects on college degree attainment and full-time employment. Under the
Oster condition R
max
= 1.3𝑅 ̃
, the estimates remained negative for all values of
δ. See the AMEs of parental incarceration plotted in Figure 1 below. Translated
to relative terms, bounds for R
max
= 1.3𝑅 ̃
suggested that parental incarceration
led to 40.1 to 58.1 greater risk of high school dropout, 14.4 to 25.8 lower
likelihood of obtaining a college degree, and 4.3 to 6.1 lower likelihood of
working full-time. While results for high school diploma attainment were
robust even to the strict condition R
max
= 2𝑅 ̃
, results for college degree
attainment and full-time work were less robust when allowing the highest
levels of relative selection on unobservables.
104
5.4.7.2 Falsification tests
Falsification tests were successful – showing that parental incarceration
occurring only before the child was born or only after the child became an adult
-0.12
-0.10
-0.08
-0.06
-0.04
-0.02
0.00
0.02
0.04
0.06
0.20 0.22 0.24 0.26 0.28 0.30 0.32
AME of parental incarceration
R
max
A. High school diploma
sample mean (s.d.): 0.828 (0.378)
-0.08
-0.06
-0.04
-0.02
0.00
0.02
0.04
0.06
0.30 0.35 0.40 0.45 0.50 0.55
R
max
B. College degree
sample mean (s.d.): 0.302 (0.459)
-0.05
-0.04
-0.03
-0.02
-0.01
0.00
0.01
0.10 0.12 0.14 0.16 0.18 0.20
AME of parental incarceration
R
max
C. Full-time work
sample mean (s.d.): 0.708 (0.455)
NOTE. — I used linear probability models
controlling for age when outcome
measured, sex, race, highest parental
educational attainment, childhood
household income, school fixed effects, and
other adverse childhood experiences
(sexual abuse, physical abuse, emotional
abuse). The points marked represent, from
left to right: R
max
= 1.3𝑅 ̃
and R
max
= 2𝑅 ̃
. I did
not perform this exercise for earnings
because, among those with positive
earnings, parental incarceration was not
associated with earnings level.
AME, average marginal effect.
δ = 1
δ = 0.75
δ = 0.5
δ = 0.25
δ = 0
δ = 1
δ = 0.75
δ = 0.5
δ = 0.25
δ = 0
δ = 1
δ = 0.75
δ = 0.5
δ = 0.25
δ = 0
Figure 1. Bounds on the effects of parental incarceration on outcomes of adult children
105
did not affect the outcomes. See results in Table 3 below. And, conditional on
the control set, there was no pre-existing difference in childhood vocabulary
scores. See Appendix Table C1.
Table 16. Falsification test: timing of parental incarceration and education
and labor market outcomes
Average marginal effects (standard errors in parentheses)
Outcome
High school
diploma
College
degree
Full-time
employment Earnings
Timing of parental
incarceration
Before birth -0.036
(0.043)
-0.028
(0.048)
-0.010
(0.056)
$1,047
(2,848)
Age 0 - 17 -0.104***
(0.017)
-0.079***
(0.013)
-0.042**
(0.016)
-$3,861***
(1,112)
Age 18+ -0.058
(0.035)
-0.036
(0.031)
0.038
(0.031)
-$3,008
(2,227)
Don’t know if parent ever
incarcerated
a
-0.076***
(0.018)
-0.054***
(0.016)
-0.038
(0.021)
-$3,365*
(1,569)
Never ref. ref. ref. ref.
N 13,563 13,503 13,560 13,041
NOTE.— This table displays the average marginal effects and standard errors for the regressions
corresponding to equation (1), including the full control set (OLS for binary outcomes, two-part model
for earnings).
a
“Don’t know” indicates response of “don’t know” either to the questions on age when parent was
first incarcerated and last released (among those acknowledging parental incarceration) or to any
question on whether biological parents or parental figure (when not the biological parent) had ever
been incarcerated.
* p<0.05, ** p<0.01, *** p<0.001
106
5.4.8 Heterogeneity analysis
Negative responses to parental incarceration were largest for Whites, females,
and children from higher income households. Only the larger disparity for
Whites was consistent across all education and labor market outcomes. Given
parental incarceration, females and individuals from higher income
households had lower rates of college degree attainment, but neither child sex
nor childhood household income modified the relationships between parental
incarceration and any other education or labor market outcome. The difference
by female status was not just due to incarceration of parent of same sex. Given
having only a father figure incarcerated, women were less likely than men to
have obtained a college degree. See results in Appendix Tables C2-C6.
There was a dose-response relationship between length of exposure to
parental incarceration in childhood and adult human capital outcomes.
Specifically, one additional year of exposure to parental incarceration
predicted 0.9 (0.2) percentage points lower likelihood of high school diploma
receipt, 0.3 (0.1) percentage points lower likelihood of college degree
attainment, and $233 (110) lower earnings.
5.5 Discussion and conclusions
This paper provides, to my knowledge, the first evidence at the US population
level that parental incarceration led to substantial increases in high school
dropout rates, with some support for negative effects on likelihood of college
degree attainment and employment in young adulthood. Results suggested that
parental incarceration led to 40 to 58 percent greater likelihood of high school
dropout, 14 to 26 percent lower likelihood of college degree attainment, and 4
107
to 6 percent lower likelihood of full-time employment. Among those working,
parental incarceration was not associated with any difference in earnings
level. Small and possibly nonexistent disparities in employment outcomes may
be due to the young age of the adult sample: ages 24 to 32. Data from the
Current Population Survey (CPS) show that high school dropouts aged 25 and
older earn about 43 percent less than the average person in this age group,
when considering their higher rates of unemployment (U.S. Bureau of Labor
Statistics 2018). Considering the bounds from the present study on the effect
of parental incarceration on likelihood of high school dropout along with data
from the CPS on the earnings penalty of dropping out of high school (U.S.
Bureau of Labor Statistics 2018), the aggregate productivity loss due to
childhood exposure to parental incarceration totals roughly $33 billion to $48
billion per year.
19
Results from the present study of adult outcomes of parental
incarceration are consistent with two findings from literature examining
childhood outcomes in the Fragile Families and Child Wellbeing Study. Firstly,
the children of incarcerated parents had worse education outcomes, on
average. Secondly, the disparities in outcomes between those exposed and not
exposed to parental incarceration were larger for groups of higher
socioeconomic status and for females. Using the sample of children from
“fragile families,” three-quarters of whom were born to unmarried parents
(Reichman, et al. 2001), researchers found that, in elementary school, parental
incarceration was associated with greater likelihood of suspension or
expulsion (Jacobsen 2019); greater likelihood of early grade retention (Turney
19
Inputs of this calculation also include the prevalence of exposure to parental incarceration
measured in the present study and the size of the working age population, reported by Federal
Reserve Bank of St. Louis (2019).
108
and Haskins 2014); and lower math, reading, and memory scores (Haskins
2016). However, for children with lower risk of parental incarceration, defined
by propensity score, the negative associations between parental incarceration
and reading scores, math scores, and frequency of externalizing behaviors
were larger (Turney 2017, Turney and Wildeman 2015). Specifically,
disparities in reading scores between those exposed and not exposed to
parental incarceration were largest among Whites and girls, while the
association was not significant among Blacks or Hispanics (Turney 2017).
Parental incarceration was associated with lower math scores among girls but
not among boys (Turney 2017).
The present work adds to the body of evidence documenting an
intergenerational transmission of socioeconomic disadvantage and has
important implications for criminal justice policy and safety net programs for
children. When evaluating the costs and benefits of incarceration policies,
policymakers should include the economic costs of incarceration on children.
109
Concluding remarks
6.1 Synthesis of key findings
In this dissertation, I examined the durable impacts of adverse childhood
experiences on health and human capital. I investigated this question in three
chapters. In all three chapters, I used nationally representative US survey data
from the National Longitudinal Study of Adolescent to Adult Health. Firstly, in
Chapter 3, I measured the impacts of childhood sexual abuse on educational
attainment and labor market outcomes in adulthood, using partial
identification methods developed by Altonji, et al. (2002, 2005) and Oster
(2017) to construct likely bounds on these effects. Results suggest that
childhood sexual abuse led to 38 to 45 percent greater likelihood of high school
dropout, 20 to 28 percent lower likelihood of college degree attainment, 4 to 8
percent lower likelihood of full-time employment, and 11 to 18 percent lower
earnings in young adulthood. Population data on the earnings penalty to high
school dropout (U.S. Bureau of Labor Statistics 2018) suggest that these
disparities in labor market outcomes might widen over time. The cost of
childhood sexual abuse in productivity loss of survivors alone totals roughly
$33 billion to $48 billion per year.
Secondly, in Chapter 4, I pursued two objectives relating to health of
survivors of childhood abuse. I studied the impacts of childhood abuse on
health in adulthood and examined health care access of survivors of childhood
abuse in young adulthood. To do this, I used reports of diagnoses, acute
symptoms, recent prescription medication use, and some health
measurements. Addressing the first objective, I found that survivors of
110
childhood abuse had higher rates of cancer, cardiometabolic conditions,
nervous system conditions, respiratory/allergic conditions, and recent
gastrointestinal symptoms – when controlling for demographics, childhood
socioeconomic status, the neighborhoods they grew up in, and parental
incarceration. These results were robust to a battery of sensitivity analyses
controlling for additional measures of childhood adversity, environment, and
health care access and use. In addition, these results on elevated health risks
among the aforementioned five domains were consistent in a sibling fixed
effect model of the effect of non-caregiver sexual abuse on adult health
outcomes. Investigating the second objective here, I found that survivors of
childhood abuse were more likely to report unmet medical needs. An important
caveat here is that reports of unmet need were subjective.
Finally, in Chapter 5, I measured the likely effects of parental
incarceration on educational attainment and adult labor market outcomes,
implementing partial identification methods described in Chapter 3. Results
suggest that parental incarceration led to about 40 to 58 percent greater
likelihood of high school dropout, 14 to 26 percent lower likelihood of college
degree attainment, and 4 to 6 percent lower likelihood of full-time
employment. Among those working, parental incarceration was not associated
with any difference in earnings level. When applying national estimates on the
earnings penalty to high school dropout (U.S. Bureau of Labor Statistics 2018)
to results from the present study on the impact of parental incarceration of
high school dropout, a back-of-the-envelope calculation suggests that the
aggregate productivity loss due to childhood exposure to parental
incarceration totals roughly $33 billion to $48 billion per year.
111
6.2 Implications for policy and health care delivery
This dissertation has important implications for public policy and health care
delivery. Research presented here highlight the need for development and
implementation of effective approaches to detect and heal childhood trauma.
This work also adds to the body of literature documenting an intergenerational
transmission of socioeconomic disadvantage, with implications for safety net
programs and other supports for children of disadvantaged family
backgrounds.
Chapters 3 and 4 showed that survivors of childhood abuse suffered
worse education, labor market, and health outcomes in adulthood. These
findings highlight the importance of detection of childhood trauma to enable
early intervention – so that durable consequences to survivors might be
avoided. Potential policy responses include requirements for routine screening
for children’s socio-emotional health status. Illinois, in 2017, became the first
state to require socio-emotional health screenings as part of school entry
exams – mandating screenings before children begin elementary, middle, and
high school. There remains, however, lack of consensus on how best to screen
for socio-emotional health status in children. Future research should include
experimental evaluations of screening protocols.
Findings from Chapters 3 and 4 suggest that treating only the mental
health symptoms of childhood abuse are not enough to eliminate disparities in
well-being. The collection of results, showing that survivors of childhood abuse
had elevated risks both for particular health conditions and for being
uninsured, should draw attention toward improving quality of and access to
health care for survivors of trauma. Results on the particular health conditions
112
suffered by survivors of childhood abuse could inform screening protocols,
especially in integrated behavioral health care settings. These results also have
implications for education policy, suggesting a need for improving support for
children with emotional disturbance, an eligible condition for receipt of special
education services.
Chapter 5 showed that childhood exposure to parental incarceration led
to huge increases in the likelihood of high school dropout. It is unclear to what
extent these results were driven by psychological stress on the child versus a
sharp decline in family economic resources. The stress mechanism would
further support screening for children’s socio-emotional health and improving
quality of support services through the education and health care systems, as
discussed above. If findings were at least partially driven by the economic
resources mechanism, these results would demonstrate a need for stronger
safety net programs for children.
Taken together, the results from this dissertation highlight the
immediate need for development of best practices to screen for children’s
socio-emotional health and to support children’s ability to thrive in the face of
adversity. Evaluations of screening protocols must consider the effectiveness
in detecting cases of childhood trauma along with the financial costs and
potential costs of re-traumatization and stigma. Both health care delivery and
public policy should be leveraged to support individuals who have had adverse
childhood experiences to lead healthy and productive lives.
113
References
Aizer, Anna, and Joseph J. Doyle Jr. 2015. "Juvenile Incarceration, Human Capital, and
Future Crime: Evidence from Randomly Assigned Judges." Quarterly Journal of
Economics 130 (2): 759-803.
Aizer, Anna, Shari Eli, Joseph Ferrie, and Adriana Lleras-Muney. 2016. "The Long-
Run Impact of Cash Transfers to Poor Families." American Economic Review 106
(4): 935-71.
Almond, D., and J. Currie. 2011. "Killing Me Softly: The Fetal Origins Hypothesis."
Journal of Economic Perspectives 25 (3): 153-72.
Almond, Douglas, Janet Currie, and Valentina Duque. 2018. "Childhood Circumstances
and Adult Outcomes: Act Ii." Journal of Economic Literature 56 (4): 1360-446.
Altonji, Joseph G, Todd E Elder, and Christopher R Taber. 2002. "Selection on
Observed and Unobserved Variables: Assessing the Effectiveness of Catholic
Schools." Northwestern University. Unpublished. Available from:
https://economics.yale.edu/sites/default/files/files/Workshops-
Seminars/Econometrics/altonji-040428.pdf. Accessed August 6, 2018.
Altonji, Joseph G., Todd E. Elder, and Christopher R. Taber. 2005. "Selection on
Observed and Unobserved Variables: Assessing the Effectiveness of Catholic
Schools." Journal of Political Economy 113 (1): 151-84.
Álvarez, Begoña, and Fernando Ramos Palencia. 2018. "Human Capital and Earnings in
Eighteenth-Century Castile." Explorations in Economic History 67: 105-33.
Anda, R. F., V. J. Felitti, D. P. Chapman, J. B. Croft, D. F. Williamson, J. Santelli, P.
M. Dietz, and J. S. Marks. 2001. "Abused Boys, Battered Mothers, and Male
Involvement in Teen Pregnancy." Pediatrics 107 (2): E19.
Anger, Silke, Georg Camehl, and Frauke Peter. 2017. "Involuntary Job Loss and Changes
in Personality Traits." Journal of Economic Psychology 60: 71-91.
Barrett, Alan, Yumiko Kamiya, and Vincent O’Sullivan. 2014. "Childhood Sexual
Abuse and Later-Life Economic Consequences." Journal of Behavioral and
Experimental Economics 53: 10-16.
Barth, J., L. Bermetz, E. Heim, S. Trelle, and T. Tonia. 2013. "The Current Prevalence
of Child Sexual Abuse Worldwide: A Systematic Review and Meta-Analysis."
International Journal of Public Health 58 (3): 469-83.
Beach, Brian, Joseph Ferrie, Martin Saavedra, and Werner Troesken. 2016. "Typhoid
Fever, Water Quality, and Human Capital Formation." The Journal of Economic
History 76 (1): 41-75.
Beach, Brian, and W. Walker Hanlon. 2017. "Coal Smoke and Mortality in an Early
Industrial Economy." The Economic Journal 128 (615): 2652-75.
Benjamini, Yoav, and Yosef Hochberg. 1995. "Controlling the False Discovery Rate: A
Practical and Powerful Approach to Multiple Testing." Journal of the Royal
Statistical Society. Series B (Methodological): 289-300.
114
Bharadwaj, Prashant, Katrine Vellesen Løken, and Christopher Neilson. 2013. "Early
Life Health Interventions and Academic Achievement." American Economic Review
103 (5): 1862-91.
Bhuller, Manudeep, Gordon B. Dahl, Katrine V. Loken, and Magne Mogstad. 2018.
"Intergenerational Effects of Incarceration." AEA Papers and Proceedings 108: 234-
40.
Black, Michele C, Kathleen C Basile, Matthew J Breiding, Sharon G Smith, Mikel L
Walters, Melissa T Merrick, and Mark R Stevens. 2011. "The National Intimate
Partner and Sexual Violence Survey: 2010 Summary Report." Atlanta, GA: National
Center for Injury Prevention and Control, Centers for Disease Control and
Prevention 19: 39-40.
Boden, J. M., L. J. Horwood, and D. M. Fergusson. 2007. "Exposure to Childhood Sexual
and Physical Abuse and Subsequent Educational Achievement Outcomes." Child
Abuse Negl 31 (10): 1101-14.
Brownstein, Naomi, William D Kalsbeek, Joyce Tabor, Pamela Entzel, Eric Daza, and
Kathleen Mullan Harris. 2011. "Non-Response in Wave Iv of the National
Longitudinal Study of Adolescent Health." Carolina Population Center, University
of North Carolina: Chapel Hill.
Bucci, M., S. S. Marques, D. Oh, and N. B. Harris. 2016. "Toxic Stress in Children and
Adolescents." Adv Pediatr 63 (1): 403-28.
Burke, N. N., D. P. Finn, B. E. McGuire, and M. Roche. 2017. "Psychological Stress in
Early Life as a Predisposing Factor for the Development of Chronic Pain: Clinical
and Preclinical Evidence and Neurobiological Mechanisms." J Neurosci Res 95 (6):
1257-70.
Clogg, Clifford C, Eva Petkova, and Adamantios Haritou. 1995. "Statistical Methods for
Comparing Regression Coefficients between Models." American Journal of
Sociology 100 (5): 1261-93.
Conti, Gabriella, James J. Heckman, and Rodrigo Pinto. 2016. "The Effects of Two
Influential Early Childhood Interventions on Health and Healthy Behaviour." The
Economic Journal 126 (596): F28-F65.
Covey, H. C., S. Menard, and R. J. Franzese. 2013. "Effects of Adolescent Physical
Abuse, Exposure to Neighborhood Violence, and Witnessing Parental Violence on
Adult Socioeconomic Status." Child Maltreat 18 (2): 85-97.
Cox, Robynn. 2018. "Mass Incarceration, Racial Disparities in Health, and Successful
Aging." Generations 42 (2): 48-55.
Cox, Robynn, and Sally Wallace. 2016. "Identifying the Link between Food Security and
Incarceration." Southern Economic Journal 82 (4): 1062-77.
Currie, Janet, and Erdal Tekin. 2012. "Understanding the Cycle: Childhood Maltreatment
and Future Crime." Journal of Human Resources 47 (2): 509-49.
Currie, Janet, and Cathy Spatz Widom. 2010. "Long-Term Consequences of Child Abuse
and Neglect on Adult Economic Well-Being." Child Maltreat 15 (2): 111-20.
De Von Figueroa-Moseley, C., J. M. Abramson, and G. C. Williams. 2010. "College
Women: History of Childhood Abuse and Its Relationship to Smoking." Violence
Against Women 16 (11): 1242-51.
115
Dobbie, Will, Hans Grönqvist, Susan Niknami, Mårten Palme, and Mikael Priks. 2018.
"The Intergenerational Effects of Parental Incarceration." National Bureau of
Economic Research Working Paper Series No. 24186.
Dong, Maxia, Robert F. Anda, Shanta R. Dube, Wayne H. Giles, and Vincent J. Felitti.
2003. "The Relationship of Exposure to Childhood Sexual Abuse to Other Forms of
Abuse, Neglect, and Household Dysfunction During Childhood." Child Abuse Negl
27 (6): 625-39.
Dow, William H., and Edward C. Norton. 2003. "Choosing between and Interpreting the
Heckit and Two-Part Models for Corner Solutions." Health Services and Outcomes
Research Methodology 4 (1): 5-18.
Duncan, A. E., W. F. Auslander, K. K. Bucholz, D. L. Hudson, R. I. Stein, and N. H.
White. 2015a. "Relationship between Abuse and Neglect in Childhood and Diabetes
in Adulthood: Differential Effects by Sex, National Longitudinal Study of
Adolescent Health." Prev Chronic Dis 12: E70.
Duncan, A. E., C. E. Sartor, M. Jonson-Reid, M. A. Munn-Chernoff, M. A.
Eschenbacher, E. W. Diemer, E. C. Nelson, et al. 2015b. "Associations between
Body Mass Index, Post-Traumatic Stress Disorder, and Child Maltreatment in Young
Women." Child Abuse Negl 45: 154-62.
Dunn, Erin C., Katie A. McLaughlin, Natalie Slopen, Jonathan Rosand, and Jordan
W. Smoller. 2013. "Developmental Timing of Child Maltreatment and Symptoms
of Depression and Suicidal Ideation in Young Adulthood: Results from the National
Longitudinal Study of Adolescent Health." Depression and Anxiety 30: 955-64.
Ebenstein, Avraham, Maoyong Fan, Michael Greenstone, Guojun He, and Maigeng
Zhou. 2017. "New Evidence on the Impact of Sustained Exposure to Air Pollution
on Life Expectancy from China’s Huai River Policy." Proceedings of the National
Academy of Sciences 114 (39): 10384-89.
El-Mallakh, Nelly, Mathilde Maurel, and Biagio Speciale. 2018. "Arab Spring Protests
and Women's Labor Market Outcomes: Evidence from the Egyptian Revolution."
Journal of Comparative Economics 46 (2): 656-82.
Elenkov, Ilia J., and George P. Chrousos. 1999. "Stress Hormones, Th1/Th2 Patterns,
Pro/Anti-Inflammatory Cytokines and Susceptibility to Disease." Trends in
Endocrinology & Metabolism 10 (9): 359-68.
Exner-Cortens, D., J. Eckenrode, and E. Rothman. 2013. "Longitudinal Associations
between Teen Dating Violence Victimization and Adverse Health Outcomes."
Pediatrics 131: 71-78.
Federal Reserve Bank of St. Louis. "Working Age Population: Aged 15-64: All Persons
for the United States." https://fred.stlouisfed.org/series/LFWA64TTUSM647S.
Feeney, Joanne, Yumiko Kamiya, Ian H Robertson, and Rose Anne Kenny. 2013.
"Cognitive Function Is Preserved in Older Adults with a Reported History of
Childhood Sexual Abuse." J Trauma Stress 26 (6): 735-43.
Felitti, V. J., R. F. Anda, D. Nordenberg, D. F. Williamson, A. M. Spitz, V. Edwards,
M. P. Koss, and J. S. Marks. 1998. "Relationship of Childhood Abuse and
Household Dysfunction to Many of the Leading Causes of Death in Adults. The
Adverse Childhood Experiences (Ace) Study." Am J Prev Med 14 (4): 245-58.
116
Fergusson, D. M., G. F. McLeod, and L. J. Horwood. 2013. "Childhood Sexual Abuse
and Adult Developmental Outcomes: Findings from a 30-Year Longitudinal Study
in New Zealand." Child Abuse Negl 37 (9): 664-74.
Finkelhor, D., H. A. Turner, A. Shattuck, and S. L. Hamby. 2013. "Violence, Crime, and
Abuse Exposure in a National Sample of Children and Youth: An Update." JAMA
Pediatrics 167 (7): 614-21.
Finlay, Keith, and David Neumark. 2010. "Is Marriage Always Good for Children?
Evidence from Families Affected by Incarceration." Journal of Human Resources 45
(4): 1046-88.
Fiszbein, Ariel, and Norbert R Schady. Conditional Cash Transfers: Reducing Present
and Future Poverty. The World Bank, 2009.
Fitzsimons, Emla, and Marcos Vera-Hernández. 2013. "Food for Thought? Breastfeeding
and Child Development." IFS Working Papers.
Fletcher, Jason M. 2009. "Childhood Mistreatment and Adolescent and Young Adult
Depression." Social Science and Medicine 68: 799-806.
———. 2010. "The Effects of Intimate Partner Violence on Health in Young Adulthood in
the United States." Social Science and Medicine 70: 130-35.
Ford, Derek C., Melissa T. Merrick, Sharyn E. Parks, Matthew J. Breiding, Leah K.
Gilbert, Valerie J. Edwards, Satvinder S. Dhingra, John P. Barile, and William
W. Thompson. 2014. "Examination of the Factorial Structure of Adverse Childhood
Experiences and Recommendations for Three Subscale Scores." Psychology of
Violence 4 (4): 432-44.
Ford, Jodi L., and Christopher R. Browning. 2014. "Effects of Exposure to Violence with
a Weapon During Adolescence on Adult Hypertension." Ann Epidemiol 24: 193-98.
Foster, H., J. Hagan, and J. Brooks-Gunn. 2008. "Growing up Fast: Stress Exposure and
Subjective 'Weathering' in Emerging Adulthood." J Health Soc Behav 49: 162-77.
Fryer, Roland G., Jr., Steven D. Levitt, and John A. List. 2015. "Parental Incentives and
Early Childhood Achievement: A Field Experiment in Chicago Heights." National
Bureau of Economic Research Working Paper Series No. 21477.
Fuemmeler, B. F., E. Dedert, F. J. McClernon, and J. Beckham. 2009. "Adverse
Childhood Events Are Associated with Obesity and Disordered Eating: Results from
a U.S. Population-Based Survey of Young Adults." Behavior Genetics 22: 329-33.
Gambaro, Ludovica, Jan Marcus, and Frauke Peter. 2018. "School Entry, Afternoon
Care, and Mothers’ Labour Supply." Empirical Economics: 1-35.
Gaston, Shytierra. 2016. "The Long-Term Effects of Parental Incarceration: Does Parental
Incarceration in Childhood or Adolescence Predict Depressive Symptoms in
Adulthood?". Criminal Justice and Behavior 43: 1056-75.
Gaudette, Étienne, Gwyn C. Pauley, and Julie M. Zissimopoulos. 2017. "Lifetime
Consequences of Early-Life and Midlife Access to Health Insurance: A Review."
Medical Care Research and Review: 1077558717740444.
Geoffroy, M. C., S. Pinto Pereira, L. Li, and C. Power. 2016. "Child Neglect and
Maltreatment and Childhood-to-Adulthood Cognition and Mental Health in a
Prospective Birth Cohort." J Am Acad Child Adolesc Psychiatry 55 (1): 33-40 e3.
117
Golberstein, Ezra, Gilbert Gonzales, and Ellen Meara. 2016. "Economic Conditions and
Children's Mental Health." National Bureau of Economic Research Working Paper
Series No. 22459.
Gooding, Holly C., Carly Milliren, Katie A. McLaughlin, Tracy K. Richmond, Sabra
L. Katz-Wise, Janet Rich-Edwards, and S. Bryn Austin. 2014. "Child
Maltreatment and Blood Pressure in Young Adulthood." Child Abuse and Neglect
38: 1747-54.
Goodman, Alissa, Robert Joyce, and James P. Smith. 2011. "The Long Shadow Cast by
Childhood Physical and Mental Problems on Adult Life." Proceedings of the
National Academy of Sciences 108 (15): 6032-37.
Grinstead, Olga, Bonnie Faigeles, Carrie Bancroft, and Barry Zack. 2001. "The
Financial Cost of Maintaining Relationships with Incarcerated African American
Men: A Survey of Women Prison Visitors." Journal of African American Men 6 (1):
59-69.
Hahm, H. C., Y. Lee, A. Ozonoff, and M. J. Van Wert. 2010. "The Impact of Multiple
Types of Child Maltreatment on Subsequent Risk Behaviors among Women During
the Transition from Adolescence to Young Adulthood." J Youth Adolesc 39 (5): 528-
40.
Hair, Nicole L., Jamie L. Hanson, Barbara L. Wolfe, and Seth D. Pollak. 2015.
"Association of Child Poverty, Brain Development, and Academic Achievement."
JAMA Pediatrics 169 (9): 822-29.
Harel, Ofer. 2009. "The Estimation of R
2
and Adjusted R
2
in Incomplete Data Sets Using
Multiple Imputation." Journal of Applied Statistics 36 (10): 1109-18.
Harris, K. M., C. T. Halpern, A. Smolen, and B. C. Haberstick. 2006. "The National
Longitudinal Study of Adolescent Health (Add Health) Twin Data." Twin Res Hum
Genet 9 (6): 988-97.
Harris, Kathleen Mullan. 2013. "The Add Health Study: Design and Accomplishments."
Carolina Population Center, University of North Carolina at Chapel Hill.
Unpublished. Available from:
http://www.cpc.unc.edu/projects/addhealth/documentation/guides/DesignPaperWII
V.pdf. Accessed August 6, 2018.
Haskins, Anna R. 2016. "Beyond Boys’ Bad Behavior: Paternal Incarceration and
Cognitive Development in Middle Childhood." Social Forces 95 (2): 861-92.
Haydon, Abigail A., Jon M. Hussey, and Carolyn Tucker Halpern. 2011. "Childhood
Abuse and Neglect and the Risk of Stds in Early Adulthood." Perspectives on Sexual
and Reproductive Health 43: 16-22.
Heckman, J., S. H. Moon, R. Pinto, P. Savelyev, and A. Yavitz. 2010. "Analyzing Social
Experiments as Implemented: A Reexamination of the Evidence from the Highscope
Perry Preschool Program." Quant Econom 1 (1): 1-46.
Heckman, J., R. Pinto, and P. Savelyev. 2013. "Understanding the Mechanisms through
Which an Influential Early Childhood Program Boosted Adult Outcomes." Am Econ
Rev 103 (6): 2052-86.
118
Herrenkohl, T. I., and H. Jung. 2016. "Effects of Child Abuse, Adolescent Violence, Peer
Approval and Pro-Violence Attitudes on Intimate Partner Violence in Adulthood."
Crim Behav Ment Health 26 (4): 304-14.
Hilger, Nathaniel G. 2016. "Parental Job Loss and Children's Long-Term Outcomes:
Evidence from 7 Million Fathers' Layoffs." American Economic Journal: Applied
Economics 8 (3): 247-83.
Hoertel, N., S. Franco, M. M. Wall, M. A. Oquendo, S. Wang, F. Limosin, and C.
Blanco. 2015. "Childhood Maltreatment and Risk of Suicide Attempt: A Nationally
Representative Study." J Clin Psychiatry 76 (7): 916-23; quiz 23.
Hoynes, Hilary, Diane Whitmore Schanzenbach, and Douglas Almond. 2016. "Long-
Run Impacts of Childhood Access to the Safety Net." American Economic Review
106 (4): 903-34.
Huang, Shi, Edward Trapido, Lora Fleming, Kristopher Arheart, Lee Crandall,
Michael French, Shandey Malcolm, and Guillermo Prado. 2011. "The Long-
Term Effects of Childhood Maltreatment Experiences on Subsequent Illicit Drug Use
and Drug-Related Problems in Young Adulthood." Addictive Behaviors 36: 95-102.
Isen, Adam, Maya Rossin-Slater, and W Reed Walker. 2017. "Every Breath You Take—
Every Dollar You’ll Make: The Long-Term Consequences of the Clean Air Act of
1970." Journal of Political Economy 125 (3): 848-902.
Jacobsen, Wade C. 2019. "The Intergenerational Stability of Punishment: Paternal
Incarceration and Suspension or Expulsion in Elementary School." Journal of
Research in Crime and Delinquency 0 (0): 0022427819829794.
Jha, Saumitra. 2015. "Financial Asset Holdings and Political Attitudes: Evidence from
Revolutionary England *." The Quarterly Journal of Economics 130 (3): 1485-545.
Johnson, Rucker. 2009. "Ever-Increasing Levels of Parental Incarceration and the
Consequences for Children." In Do Prisons Make Us Safer? The Benefits and Costs
of the Prison Boom, eds. Steven Raphael and Michael A. Stoll. 177-206.
Jonson-Reid, M., P. L. Kohl, and B. Drake. 2012. "Child and Adult Outcomes of Chronic
Child Maltreatment." Pediatrics 129 (5): 839-45.
Lavy, Victor, and Edith Sand. 2018. "The Effect of Social Networks on Students'
Academic and Non-Cognitive Behavioural Outcomes: Evidence from Conditional
Random Assignment of Friends in School." The Economic Journal 0 (0).
Lee, C., V. Tsenkova, and D. Carr. 2014. "Childhood Trauma and Metabolic Syndrome in
Men and Women." Soc Sci Med 105: 122-30.
Lee, Sun Youn, and Fumio Ohtake. 2018. "Is Being Agreeable a Key to Success or Failure
in the Labor Market?". Journal of the Japanese and International Economies 49: 8-
27.
Lehne, Jonathan, Jacob N. Shapiro, and Oliver Vanden Eynde. 2018. "Building
Connections: Political Corruption and Road Construction in India." Journal of
Development Economics 131: 62-78.
Lin, Dajun, Randall Lutter, and Christopher J. Ruhm. 2018. "Cognitive Performance
and Labour Market Outcomes." Labour Economics 51: 121-35.
119
McLaughlin, K. A., M. L. Hatzenbuehler, Z. Xuan, and K. J. Conron. 2012.
"Disproportionate Exposure to Early-Life Adversity and Sexual Orientation
Disparities in Psychiatric Morbidity." Child Abuse and Neglect 36: 645-55.
Mears, Daniel P., and Sonja E. Siennick. 2016. "Young Adult Outcomes and the Life-
Course Penalties of Parental Incarceration." Journal of Research in Crime and
Delinquency 53 (1): 3-35.
Merrick, M. T., D. C. Ford, K. A. Ports, and A. S. Guinn. 2018. "Prevalence of Adverse
Childhood Experiences from the 2011-2014 Behavioral Risk Factor Surveillance
System in 23 States." JAMA Pediatrics.
Midanik, L. T., and T. K. Greenfield. 2008. "Interactive Voice Response Versus
Computer-Assisted Telephone Interviewing (Cati) Surveys and Sensitive Questions:
The 2005 National Alcohol Survey." J Stud Alcohol Drugs 69 (4): 580-8.
Miguel, Edward, and Michael Kremer. 2004. "Worms: Identifying Impacts on Education
and Health in the Presence of Treatment Externalities." Econometrica 72 (1): 159-
217.
Murphey, David, and P. Mae Cooper. 2015. "Parents Behind Bars: What Happens to Their
Children." 1-22. Bethesda, MD: Child Trends.
Nguyen, Cattram D., John B. Carlin, and Katherine J. Lee. 2017. "Model Checking in
Multiple Imputation: An Overview and Case Study." Emerging Themes in
Epidemiology 14 (1): 8.
Orme, J. G., J. Reis, and E. J. Herz. 1986. "Factorial and Discriminant Validity of the
Center for Epidemiological Studies Depression (Ces-D) Scale." J Clin Psychol 42
(1): 28-33.
Oster, Emily. 2017. "Unobservable Selection and Coefficient Stability: Theory and
Evidence." Journal of Business & Economic Statistics: 1-18.
Pereda, Noemí, Georgina Guilera, Maria Forns, and Juana Gómez-Benito. 2009. "The
Prevalence of Child Sexual Abuse in Community and Student Samples: A Meta-
Analysis." Clinical Psychology Review 29 (4): 328-38.
Pérez-Fuentes, Gabriela, Mark Olfson, Laura Villegas, Carmen Morcillo, Shuai Wang,
and Carlos Blanco. 2013. "Prevalence and Correlates of Child Sexual Abuse: A
National Study." Compr Psychiatry 54 (1): 16-27.
Perlis, Theresa E, Don C Des Jarlais, Samuel R Friedman, Kamyar Arasteh, and
Charles F Turner. 2004. "Audio‐Computerized Self‐Interviewing Versus Face‐to‐
Face Interviewing for Research Data Collection at Drug Abuse Treatment
Programs." Addiction 99 (7): 885-96.
Radloff, Lenore Sawyer. 1977. "The Ces-D Scale: A Self-Report Depression Scale for
Research in the General Population." Applied psychological measurement 1 (3): 385-
401.
Ranchod, Y. K., I. E. Headen, L. C. Petito, J. K. Deardorff, D. H. Rehkopf, and B. F.
Abrams. 2016. "Maternal Childhood Adversity, Prepregnancy Obesity, and
Gestational Weight Gain." Am J Prev Med 50 (4): 463-9.
Reichman, Nancy E, Julien O Teitler, Irwin Garfinkel, and Sara S McLanahan. 2001.
"Fragile Families: Sample and Design." Child Youth Serv Rev 23 (4-5): 303-26.
120
Rich-Edwards, J. W., D. Spiegelman, E. N. Lividoti Hibert, H. J. Jun, T. J. Todd, I.
Kawachi, and R. J. Wright. 2010. "Abuse in Childhood and Adolescence as a
Predictor of Type 2 Diabetes in Adult Women." Am J Prev Med 39 (6): 529-36.
Richardson, A. S., W. H. Dietz, and P. Gordon-Larsen. 2014. "The Association between
Childhood Sexual and Physical Abuse with Incident Adult Severe Obesity across
13 Years of the National Longitudinal Study of Adolescent Health." Pediatric
Obesity 9: 351-61.
Roberts, M., B. F. Fuemmeler, F. J. McClernon, and J. Beckham. 2008. "Association
between Trauma Exposure and Smoking in a Population-Based Sample of Young
Adults." Journal of Adolescent Health 42: 266-74.
Roodman, David. 2009. "Estimating Fully Observed Recursive Mixed-Process Models
with Cmp."
Rubin, Donald B. Multiple Imputation for Nonresponse in Surveys. New York: Wiley,
1987.
Rubin, Donald B. 1996. "Multiple Imputation after 18+ Years." Journal of the American
Statistical Association 91 (434): 473-89.
Saarela, J. M., and I. T. Elo. 2016. "Forced Migration in Childhood: Are There Long-Term
Health Effects?". SSM Popul Health 2: 813-23.
Sachs-Ericsson, N., M. D. Gayman, K. Kendall-Tackett, D. A. Lloyd, A. Medley, N.
Collins, E. Corsentino, and K. Sawyer. 2010. "The Long-Term Impact of
Childhood Abuse on Internalizing Disorders among Older Adults: The Moderating
Role of Self-Esteem." Aging Ment Health 14 (4): 489-501.
Sanjeevi, Jerusha, Daniel Houlihan, Kelly A. Bergstrom, Moses M. Langley, and
Jaxson Judkins. 2018. "A Review of Child Sexual Abuse: Impact, Risk, and
Resilience in the Context of Culture." Journal of Child Sexual Abuse 27 (6): 622-41.
Schaller, Jessamyn, and Mariana Zerpa. 2019. "Short-Run Effects of Parental Job Loss
on Child Health." American Journal of Health Economics 5 (1): 8-41.
Schomaker, Michael, and Christian Heumann. 2018. "Bootstrap Inference When Using
Multiple Imputation." Statistics in Medicine 37 (14): 2252-66.
Shafa, Vanessa. 2016. "Family Relationship Factors and Their Effects on Resilience in
Adolescents Maltreated by Adult Caregivers." Alliant International University.
Shin, S. H., and D. P. Miller. 2012. "A Longitudinal Examination of Childhood
Maltreatment and Adolescent Obesity: Results from the National Longitudinal Study
of Adolescent Health (Addhealth) Study." Child Abuse Negl 36 (2): 84-94.
Shin, S. H., D. P. Miller, and M. H. Teicher. 2013. "Exposure to Childhood Neglect and
Physical Abuse and Developmental Trajectories of Heavy Episodic Drinking from
Early Adolescence into Young Adulthood." Drug Alcohol Depend 127: 31-38.
Stoltenborgh, M., M. H. van Ijzendoorn, E. M. Euser, and M. J. Bakermans-
Kranenburg. 2011. "A Global Perspective on Child Sexual Abuse: Meta-Analysis
of Prevalence around the World." Child Maltreat 16 (2): 79-101.
Suglia, S. F., C. J. Clark, R. Boynton-Jarrett, N. R. Kressin, and K. C. Koenen. 2014.
"Child Maltreatment and Hypertension in Young Adulthood." BMC Public Health
14: 1149.
121
Suglia, S. F., K. C. Koenen, R. Boynton-Jarrett, P. S. Chan, C. J. Clark, A. Danese, M.
S. Faith, et al. 2018. "Childhood and Adolescent Adversity and Cardiometabolic
Outcomes: A Scientific Statement from the American Heart Association."
Circulation 137 (5): e15-e28.
Tedeschi, Richard G, Crystal L Park, and Lawrence G Calhoun. Posttraumatic Growth:
Positive Changes in the Aftermath of Crisis. Personality and Clinical Psychology
Series. Mahwah (NJ): Lawrence Erlbaum Associates, Inc., Publishers, 1998.
Turney, K. 2017. "The Unequal Consequences of Mass Incarceration for Children."
Demography 54 (1): 361-89.
Turney, Kristin, and Anna R Haskins. 2014. "Falling Behind? Children’s Early Grade
Retention after Paternal Incarceration." Sociology of Education 87 (4): 241-58.
Turney, Kristin, and Christopher Wildeman. 2015. "Detrimental for Some?
Heterogeneous Effects of Maternal Incarceration on Child Wellbeing." Criminology
& Public Policy 14 (1): 125-56.
U.S. Bureau of Labor Statistics. "Unemployment Rates and Earnings by Educational
Attainment." https://www.bls.gov/emp/chart-unemployment-earnings-
education.htm.
Valbuena, Javier, and Yu Zhu. 2018. "The Returns to Uk Degrees for Foreign-Educated
Graduates." British Educational Research Journal 44 (3): 440-62.
Wagner, Peter, and Wendy Sawyer. 2018. "States of Incarceration: The Global Context
2018." Northampton, MA: Prison Policy Initiative.
Walther, Selma. 2018. "Noncooperative Decision Making in the Household: Evidence from
Malawi." Journal of Development Economics 134: 428-42.
White, I. R., P. Royston, and A. M. Wood. 2011. "Multiple Imputation Using Chained
Equations: Issues and Guidance for Practice." Statistics in Medicine 30 (4): 377-99.
Widom, Cathy Spatz, and Suzanne Morris. 1997. "Accuracy of Adult Recollections of
Childhood Victimization, Part 2: Childhood Sexual Abuse." Psychological
assessment 9 (1): 34.
Wildeman, C., N. Emanuel, J. M. Leventhal, E. Putnam-Hornstein, J. Waldfogel, and
H. Lee. 2014. "The Prevalence of Confirmed Maltreatment among Us Children,
2004 to 2011." JAMA Pediatr 168 (8): 706-13.
Wildeman, Christopher. 2009. "Parental Imprisonment, the Prison Boom, and the
Concentration of Childhood Disadvantage." Demography 46 (2): 265-80.
———. 2014. "Parental Incarceration, Child Homelessness, and the Invisible Consequences
of Mass Imprisonment." The Annals of the American Academy of Political and Social
Science 651 (1): 74-96.
Woodward, C., and S. Joseph. 2003. "Positive Change Processes and Post-Traumatic
Growth in People Who Have Experienced Childhood Abuse: Understanding
Vehicles of Change." Psychol Psychother 76 (Pt 3): 267-83.
Zeglin, R. J., M. R. DeRaedt, and R. P. Lanthier. 2015. "Does Having Children Moderate
the Effect of Child Sexual Abuse on Depression?". J Child Sex Abus 24 (6): 607-26.
122
Appendices
Appendix A: Chapter 3 appendix tables
Table A 1. Sample means across adverse childhood experiences
a
(standard errors in parentheses)
(1) (2) (3) (4) (5)
None of
studied
ACEs
n=9,790
Sexual
abuse
n=2,198
Physical
abuse
n=2,876
Emotional
abuse
n=1,871
Parental
incarcera-
tion
n=1,893
Female
0.463
(0.008)
0.719
(0.016)
0.493
(0.013)
0.586
(0.016)
0.504
(0.016)
Race/ethnicity
White NH
0.769
(0.025)
0.721
(0.029)
0.732
(0.029)
0.800
(0.024)
0.663
(0.035)
Black NH
0.146
(0.020)
0.193
(0.027)
0.169
(0.023)
0.120
(0.018)
0.261
(0.036)
Native American NH
0.008
(0.002)
0.014
(0.006)
0.013
(0.006)
0.011
(0.005)
0.018
(0.010)
Asian/Pacific Islander NH
0.036
(0.008)
0.024
(0.008)
0.037
(0.010)
0.034
(0.008)
0.008
(0.003)
Other NH
0.030
(0.003)
0.051
(0.008)
0.048
(0.006)
0.051
(0.008)
0.047
(0.009)
Hispanic
0.111
(0.018)
0.121
(0.015)
0.131
(0.020)
0.098
(0.015)
0.142
(0.021)
US-born
0.947
(0.009)
0.943
(0.009)
0.935
(0.012)
0.955
(0.009)
0.968
(0.007)
Biological parents in
household
Mom and dad
0.646
(0.012)
0.401
(0.017)
0.413
(0.019)
0.391
(0.018)
0.247
(0.018)
Mom only
0.284
(0.011)
0.472
(0.016)
0.442
(0.016)
0.453
(0.015)
0.594
(0.017)
Dad only
0.037
(0.003)
0.050
(0.007)
0.078
(0.008)
0.087
(0.010)
0.063
(0.008)
Neither
0.033
(0.004)
0.077
(0.009)
0.067
(0.009)
0.070
(0.008)
0.095
(0.010)
123
Childhood household
income, median (2010$)
$64,660 $44,087 $48,495 $51,434 $33,800
Child health and cognition
Physical disability
0.019
(0.002)
0.039
(0.006)
0.029
(0.004)
0.035
(0.005)
0.027
(0.005)
Educational challenge
0.163
(0.015)
0.193
(0.013)
0.163
(0.012)
0.161
(0.015)
0.168
(0.014)
Depressive symptoms
0.267
(0.009)
0.432
(0.017)
0.400
(0.014)
0.439
(0.018)
0.384
(0.015)
Vocabulary score 101.7 (0.7) 99.2 (0.7) 101.4 (0.6) 104.5 (0.6) 97.6 (0.7)
Adverse childhood
experiences
Sexual abuse 0.000 1.000
0.293
(0.011)
0.305
(0.013)
0.245
(0.014)
Physical abuse 0.000
0.370
(0.015)
1.000
0.570
(0.018)
0.342
(0.018)
Emotional abuse 0.000
0.260
(0.012)
0.385
(0.013)
1.000
0.226
(0.013)
Parental incarceration 0.000
0.202
(0.014)
0.247
(0.016)
0.239
(0.016)
1.000
Adult outcomes
High school diploma
0.864
(0.011)
0.724
(0.017)
0.784
(0.015)
0.792
(0.018)
0.674
(0.017)
College degree
0.346
(0.018)
0.187
(0.015)
0.232
(0.017)
0.266
(0.022)
0.120
(0.012)
Full-time work
0.733
(0.013)
0.625
(0.014)
0.701
(0.012)
0.662
(0.015)
0.661
(0.014)
Adult earnings, median
(2010$)
$32,442 $22,304 $28,387 $27,373 $23,318
Married
0.440
(0.013)
0.445
(0.019)
0.417
(0.017)
0.436
(0.019)
0.379
(0.020)
Depressive symptoms
0.233
(0.009)
0.418
(0.016)
0.384
(0.014)
0.396
(0.016)
0.376
(0.016)
Suicidal ideation in past year
0.045
(0.004)
0.157
(0.011)
0.126
(0.008)
0.134
(0.011)
0.118
(0.009)
NOTES. Columns (2) – (5) represent overlapping populations, as some people experience multiple ACEs.
Key. NH, Non-Hispanic
a
Except where noted as “median.”
124
Table A 2. Estimates of the effects of childhood physical abuse on education and
labor market outcomes
Average marginal effects (standard errors in parentheses)
(1) (2) (3) (4) (5) (6)
Controls None
Demographics
a
Col. 2 +
household
SES
b
Col. 3 +
school FE
c
Col. 4 +
disability
d
Col. 5 +
other ACEs
e
A. High school diploma; mean (s.d.): 0.828 (0.378)
OLS -0.058*** -0.055*** -0.037** -0.032** -0.032** -0.003
(0.012) (0.013) (0.012) (0.011) (0.011) (0.011)
R
2
0.004 0.017 0.091 0.144 0.146 0.157
Adjusted
R
2
0.003 0.017 0.090 0.135 0.137 0.148
B. College degree; mean (s.d.): 0.302 (0.459)
OLS -0.099*** -0.094*** -0.057*** -0.057*** -0.056*** -0.030*
(0.014) (0.013) (0.012) (0.012) (0.012) (0.013)
R
2
0.005 0.033 0.205 0.256 0.257 0.265
Adjusted
R
2
0.005 0.032 0.204 0.248 0.249 0.257
C. Employment; mean (s.d.): 0.805 (0.396)
OLS -0.034** -0.034** -0.027* -0.031** -0.031** -0.010
(0.011) (0.011) (0.011) (0.011) (0.011) (0.011)
R
2
0.001 0.019 0.032 0.067 0.068 0.069
Adjusted
R
2
0.001 0.019 0.030 0.058 0.058 0.059
D. Full-time employment; mean (s.d.): 0.706 (0.455)
OLS -0.030* -0.029* -0.022 -0.026* -0.025* 0.001
(0.012) (0.012) (0.012) (0.011) (0.012) (0.012)
R
2
0.000 0.034 0.042 0.091 0.091 0.093
Adjusted
R
2
0.000 0.034 0.041 0.081 0.082 0.083
E. Earnings; mean (s.d.): $35,025 (44,170); median: $30,414
2PM -$1,464 -$1,514 $376 $339 $348 $1,799
(1,180) (1,147) (1,095) (966) (969) (1,162)
125
NOTE.— Sample size was N=14,741 for each regression here.
a
Demographic controls include age when outcome measured; sex as female or male; and
race/ethnicity as White Non-Hispanic, Black Non-Hispanic, Native American Non-Hispanic,
Asian/Pacific Islander Non-Hispanic, Other Non-Hispanic (includes multi-racial), or Hispanic.
b
Household SES controls include log of childhood household income and highest parental
educational attainment as (i) less than high school, (ii) GED, (iii) high school diploma, (iv)
vocational school after high school, (v) some college, (vi) college graduate, or (vii) beyond 4-yr
college.
c
School fixed effects: schools were the primary sampling unit.
d
Disability represents childhood physical disability.
e
Other ACEs include the remaining adverse childhood experiences not included in control sets (1)
to (5) from the list: childhood sexual abuse, physical abuse, emotional abuse, parental
incarceration.
* p<0.05, ** p<0.01, *** p<0.001
Table A 3. Estimates of the effects of childhood emotional abuse on education and
labor market outcomes
Average marginal effects (standard errors in parentheses)
(1) (2) (3) (4) (5) (6)
Controls None
Demographics
a
Col. 2 +
household
SES
b
Col. 3 +
school FE
c
Col. 4 +
disability
d
Col. 5 +
other ACEs
e
A. High school diploma; mean (s.d.): 0.828 (0.378)
OLS -0.044** -0.051** -0.041** -0.039* -0.038* -0.012
(0.016) (0.016) (0.015) (0.015) (0.015) (0.014)
R
2
0.002 0.016 0.089 0.141 0.143 0.157
Adjusted
R
2
0.002 0.016 0.088 0.132 0.134 0.148
B. College degree; mean (s.d.): 0.302 (0.459)
OLS -0.048* -0.062*** -0.040** -0.046** -0.045** -0.008
(0.019) (0.018) (0.015) (0.014) (0.014) (0.015)
R
2
0.001 0.030 0.204 0.256 0.257 0.265
Adjusted
R
2
0.001 0.030 0.204 0.248 0.249 0.257
C. Employment; mean (s.d.): 0.805 (0.396)
OLS -0.053*** -0.044** -0.040** -0.047** -0.046** -0.031*
(0.015) (0.015) (0.015) (0.015) (0.015) (0.015)
R
2
0.002 0.020 0.031 0.065 0.066 0.069
Adjusted
R
2
0.002 0.019 0.030 0.056 0.056 0.059
126
D. Full-time employment; mean (s.d.): 0.706 (0.455)
OLS -0.067*** -0.051** -0.047** -0.052** -0.052** -0.038*
(0.017) (0.017) (0.017) (0.016) (0.016) (0.018)
R
2
0.002 0.035 0.042 0.090 0.090 0.093
Adjusted
R
2
0.002 0.034 0.041 0.080 0.081 0.083
E. Earnings; mean (s.d.): $35,025 (44,170); median: $30,414
2PM -$1,616 -$1,282 -$756 -$1,222 -$1,217 -$721
(1,609) (1,303) (1,203) (1,044) (1,044) (1,210)
NOTE.— Sample size was N=14,741 for each regression here.
a
Demographic controls include age when outcome measured; sex as female or male; and
race/ethnicity as White Non-Hispanic, Black Non-Hispanic, Native American Non-Hispanic,
Asian/Pacific Islander Non-Hispanic, Other Non-Hispanic (includes multi-racial), or Hispanic.
b
Household SES controls include log of childhood household income and highest parental
educational attainment as (i) less than high school, (ii) GED, (iii) high school diploma, (iv)
vocational school after high school, (v) some college, (vi) college graduate, or (vii) beyond 4-yr
college.
c
School fixed effects: schools were the primary sampling unit.
d
Disability represents childhood physical disability.
e
Other ACEs include the remaining adverse childhood experiences not included in control sets (1)
to (5) from the list: childhood sexual abuse, physical abuse, emotional abuse, parental
incarceration.
* p<0.05, ** p<0.01, *** p<0.001
127
Table A 4. Sibling conditional fixed effects estimates from OLS
(standard errors in parentheses)
Outcome
(1) (2) (3) (4) (5)
High school
diploma
College
degree Employment
Full-time
employment
Log
earnings
Childhood sexual
abuse
-0.068
(0.040)
-0.060*
(0.028)
-0.063
(0.040)
-0.049
(0.042)
-0.159
(0.105)
N 3,377 3,359 3,376 3,376 2,962
NOTE.— The table shows the estimated average marginal effects and standard errors for the effect
of childhood sexual abuse as measured in models with sibling fixed effects and controlling for
individual age and sex. Conditional fixed effects regression exploits variation within the group
(siblings), thus sibling groups for which there is no variation in the outcome or right-hand-side
variables are dropped from the regression. Sample sizes vary across outcomes for this reason and
also for log earnings due to use only of observations with positive earnings.
* p<0.05, ** p<0.01, *** p<0.001
Table A 5. Baseline results and heterogeneity by sociodemographics, outcome:
high school diploma receipt
(standard errors in parentheses)
(1) (2) (3)
Baseline By race By SES
Childhood sexual abuse -0.077***
(0.014)
-0.091***
(0.017)
-0.070
(0.220)
Physical abuse -0.003
(0.011)
-0.003
(0.011)
-0.003
(0.011)
Emotional abuse -0.012
(0.014)
-0.011
(0.014)
-0.012
(0.014)
Parental incarceration
No ref. ref. ref.
Yes -0.100***
(0.017)
-0.100***
(0.017)
-0.100***
(0.017)
Don't know -0.063***
(0.017)
-
0.064***
(0.017)
-0.063***
(0.017)
Age -0.003
(0.004)
-0.003
(0.004)
-0.003
(0.004)
Female 0.048***
(0.008)
0.048***
(0.008)
0.048***
(0.008)
128
Race/ethnicity
White NH ref. ref. ref.
Black NH 0.032
(0.018)
0.030
(0.018)
0.032
(0.018)
Native American NH -0.019
(0.099)
-0.074
(0.087)
-0.019
(0.099)
Asian or Pacific Islander NH 0.040
(0.024)
0.020
(0.020)
0.040
(0.024)
Other NH 0.002
(0.029)
0.008
(0.028)
0.002
(0.029)
Hispanic -0.019
(0.018)
-0.028
(0.017)
-0.019
(0.018)
Parental education
< High school -0.103***
(0.021)
-0.103***
(0.021)
-0.103***
(0.021)
GED -0.092**
(0.034)
-0.092**
(0.034)
-0.092**
(0.034)
High school graduate ref. ref. ref.
Vocational school + high school 0.036*
(0.015)
0.036*
(0.015)
0.036*
(0.015)
Some college 0.044**
(0.014)
0.044**
(0.013)
0.044**
(0.014)
College graduate 0.067***
(0.013)
0.066***
(0.012)
0.067***
(0.012)
Beyond 4-yr college 0.094***
(0.015)
0.093***
(0.015)
0.094***
(0.015)
Log family income 0.046***
(0.008)
0.047***
(0.008)
0.046***
(0.008)
Physical disability -0.101**
(0.035)
-0.102**
(0.035)
-0.101**
(0.035)
Childhood sexual abuse * White NH ref.
Childhood sexual abuse * Black NH 0.016
(0.033)
Childhood sexual abuse * Native American NH 0.241
(0.191)
Childhood sexual abuse * Asian or Pacific
Islander NH
0.189*
(0.094)
Childhood sexual abuse * Other NH -0.031
(0.070)
Childhood sexual abuse * Hispanic 0.054
(0.043)
Childhood sexual abuse * Log family income -0.001
(0.020)
129
NOTE.— Sample size was N=14,741 for each regression here. All regressions included school fixed
effects and a constant. * p<0.05, ** p<0.01, *** p<0.001
130
Table A 6. Baseline results and heterogeneity by sociodemographics, outcome:
college degree attainment
(standard errors in parentheses)
(1) (2) (3)
Baseline By race By SES
Childhood sexual abuse -0.084***
(0.012)
-0.102***
(0.016)
0.540**
(0.160)
Physical abuse -0.03*
(0.013)
-0.03*
(0.013)
-0.03*
(0.013)
Emotional abuse -0.008
(0.015)
-0.007
(0.015)
-0.009
(0.015)
Parental incarceration
No ref. ref. ref.
Yes -0.078***
(0.013)
-
0.078***
(0.013)
-0.078***
(0.013)
Don't know -0.055**
(0.017)
-0.055**
(0.017)
-0.053**
(0.017)
Age -0.011**
(0.004)
-0.011**
(0.004)
-0.011**
(0.004)
Female 0.084***
(0.011)
0.085***
(0.011)
0.084***
(0.011)
Race/ethnicity
White NH ref. ref. ref.
Black NH -0.004
(0.020)
-0.015
(0.022)
-0.003
(0.020)
Native American NH -0.041
(0.054)
-0.044
(0.069)
-0.041
(0.055)
Asian or Pacific Islander NH 0.101**
(0.037)
0.098*
(0.040)
0.099**
(0.036)
Other NH 0.001
(0.026)
0.011
(0.030)
0.002
(0.026)
Hispanic -0.038
(0.022)
-0.053*
(0.022)
-0.039
(0.022)
Parental education
< High school -0.005
(0.015)
-0.004
(0.015)
-0.005
(0.015)
GED -0.044*
(0.022)
-0.044*
(0.022)
-0.043*
(0.022)
High school graduate ref. ref. ref.
Vocational school + high school 0.022
(0.017)
0.021
(0.017)
0.021
(0.017)
131
Some college 0.059***
(0.014)
0.059***
(0.014)
0.058***
(0.014)
College graduate 0.197***
(0.018)
0.196***
(0.018)
0.195***
(0.018)
Beyond 4-yr college 0.355***
(0.020)
0.354***
(0.020)
0.352***
(0.020)
Log family income 0.072***
(0.008)
0.072***
(0.008)
0.083***
(0.009)
Physical disability -0.090**
(0.027)
-0.091**
(0.027)
-0.091**
(0.027)
Childhood sexual abuse * White NH ref.
Childhood sexual abuse * Black NH 0.063*
(0.029)
Childhood sexual abuse * Native American NH 0.018
(0.112)
Childhood sexual abuse * Asian or Pacific
Islander NH
0.011
(0.092)
Childhood sexual abuse * Other NH -0.051
(0.049)
Childhood sexual abuse * Hispanic 0.085*
(0.034)
Childhood sexual abuse * Log family income -0.059***
(0.015)
NOTE.— Sample size was N=14,741 for each regression here. All regressions included school fixed
effects and a constant. * p<0.05, ** p<0.01, *** p<0.001
Table A 7. Baseline results and heterogeneity by sociodemographics, outcome: any
employment
(standard errors in parentheses)
(1) (2) (3)
Baseline By race By SES
Childhood sexual abuse -0.039*
(0.015)
-0.064**
(0.021)
0.119 (0.213)
Physical abuse -0.010
(0.011)
-0.010
(0.011)
-0.010
(0.011)
Emotional abuse -0.031*
(0.015)
-0.029
(0.015)
-0.031*
(0.015)
Parental incarceration
No ref. ref. ref.
Yes -0.030
(0.016)
-0.030
(0.016)
-0.030
(0.016)
132
Don't know -0.052*
(0.021)
-0.052*
(0.021)
-0.051*
(0.021)
Age -0.004
(0.003)
-0.004
(0.003)
-0.004
(0.003)
Female -0.090***
(0.010)
-0.089***
(0.010)
-0.090***
(0.010)
Race/ethnicity
White NH ref. ref. ref.
Black NH -0.018
(0.022)
-0.027
(0.021)
-0.018
(0.021)
Native American NH -0.070
(0.109)
-0.063
(0.126)
-0.070
(0.108)
Asian or Pacific Islander NH -0.015
(0.033)
-0.044
(0.032)
-0.016
(0.033)
Other NH 0.005
(0.023)
-0.000
(0.028)
0.005
(0.023)
Hispanic 0.056**
(0.020)
0.045*
(0.021)
0.056**
(0.020)
Parental education
< High school -0.068**
(0.023)
-0.068**
(0.023)
-0.068**
(0.023)
GED -0.015
(0.030)
-0.014
(0.030)
-0.015
(0.030)
High school graduate ref. ref. ref.
Vocational school + high school 0.012
(0.019)
0.012
(0.019)
0.011 (0.019)
Some college 0.003
(0.013)
0.004
(0.013)
0.003
(0.013)
College graduate 0.012
(0.015)
0.012
(0.015)
0.012
(0.015)
Beyond 4-yr college -0.004
(0.017)
-0.004
(0.017)
-0.004
(0.016)
Log family income 0.015
(0.010)
0.015
(0.009)
0.018
(0.010)
Physical disability -0.065
(0.040)
-0.065
(0.040)
-0.065
(0.040)
Childhood sexual abuse * White NH ref.
Childhood sexual abuse * Black NH 0.060
(0.032)
Childhood sexual abuse * Native
American NH
-0.018
(0.164)
Childhood sexual abuse * Asian or Pacific
Islander NH
0.271***
(0.060)
133
Childhood sexual abuse * Other NH 0.030
(0.076)
Childhood sexual abuse * Hispanic 0.066
(0.040)
Childhood sexual abuse * Log family
income
-0.015
(0.020)
NOTE.— Sample size was N=14,741 for each regression here. All regressions included school fixed
effects and a constant. * p<0.05, ** p<0.01, *** p<0.001
Table A 8. Baseline results and heterogeneity by sociodemographics, outcome:
full-time employment
(standard errors in parentheses)
(1) (2) (3)
Baseline By race By SES
Childhood sexual abuse -0.053***
(0.015)
-0.069**
(0.021)
-0.050
(0.180)
Physical abuse 0.001
(0.012)
0.002
(0.012)
0.001
(0.012)
Emotional abuse -0.038*
(0.018)
-0.038*
(0.018)
-0.038*
(0.018)
Parental incarceration
No ref. ref. ref.
Yes -0.043*
(0.017)
-0.043*
(0.017)
-0.043*
(0.017)
Don't know -0.047*
(0.022)
-0.047*
(0.022)
-0.047*
(0.022)
Age -0.002
(0.004)
-0.002
(0.004)
-0.002
(0.004)
Female -0.149***
(0.011)
-0.150***
(0.011)
-0.149***
(0.011)
Race/ethnicity
White NH ref. ref. ref.
Black NH -0.042
(0.022)
-0.043*
(0.021)
-0.042
(0.022)
Native American NH -0.059
(0.094)
-0.082
(0.110)
-0.059
(0.094)
Asian or Pacific Islander NH -0.040
(0.035)
-0.063
(0.035)
-0.040
(0.035)
Other NH -0.017
(0.031)
-0.033
(0.033)
-0.017
(0.031)
Hispanic 0.051*
(0.023)
0.047
(0.024)
0.051*
(0.023)
Parental education
134
< High school -0.061**
(0.020)
-0.061**
(0.020)
-0.061**
(0.020)
GED -0.007
(0.031)
-0.006
(0.031)
-0.006
(0.031)
High school graduate ref. ref. ref.
Vocational school + high school 0.009
(0.018)
0.010
(0.019)
0.009
(0.018)
Some college -0.006
(0.014)
-0.005
(0.014)
-0.006
(0.014)
College graduate -0.004
(0.016)
-0.004
(0.016)
-0.004
(0.016)
Beyond 4-yr college -0.016
(0.019)
-0.015
(0.019)
-0.016
(0.019)
Log family income 0.019*
(0.009)
0.019
(0.009)
0.019
(0.010)
Physical disability -0.032
(0.041)
-0.031
(0.041)
-0.032
(0.041)
Childhood sexual abuse * White NH ref.
Childhood sexual abuse * Black NH 0.010
(0.036)
Childhood sexual abuse * Native American NH 0.110
(0.179)
Childhood sexual abuse * Asian or Pacific
Islander NH
0.230*
(0.109)
Childhood sexual abuse * Other NH 0.084
(0.072)
Childhood sexual abuse * Hispanic 0.030
(0.044)
Childhood sexual abuse * Log family income -0.000
(0.017)
NOTE.— Sample size was N=14,741 for each regression here. All regressions included school fixed
effects and a constant. * p<0.05, ** p<0.01, *** p<0.001
Table A 9. Baseline results and heterogeneity by sociodemographics, outcome:
earnings
(standard errors in parentheses)
(1) (2) (3)
Baseline By race By SES
1st part: probit for having positive earnings
Childhood sexual abuse -0.115 (0.071) -0.181
(0.092)
0.991 (0.766)
135
Physical abuse 0.062
(0.066)
0.065
(0.066)
0.063
(0.066)
Emotional abuse -0.076
(0.081)
-0.072
(0.081)
-0.078
(0.080)
Parental incarceration
No ref. ref. ref.
Yes -0.212**
(0.066)
-0.212**
(0.067)
-0.213**
(0.066)
Don't know -0.326**
(0.105)
-0.322**
(0.104)
-0.323**
(0.103)
Age -0.029
(0.019)
-0.030
(0.019)
-0.031
(0.019)
Female -0.511***
(0.063)
-0.509***
(0.064)
-0.512***
(0.063)
Race/ethnicity
White NH ref. ref. ref.
Black NH -0.074
(0.106)
-0.119 (0.112) -0.069
(0.105)
Native American NH -0.205
(0.295)
-0.073
(0.325)
-0.202
(0.294)
Asian or Pacific Islander NH -0.218
(0.180)
-0.302
(0.179)
-0.224
(0.179)
Other NH 0.138 (0.127) 0.087 (0.142) 0.139 (0.127)
Hispanic 0.042 (0.105) 0.048 (0.116) 0.041 (0.104)
Parental education
< High school -0.245*
(0.094)
-0.252**
(0.095)
-0.243*
(0.094)
GED -0.016
(0.140)
-0.017 (0.141) -0.012
(0.140)
High school graduate ref. ref. ref.
Vocational school + high school 0.087
(0.084)
0.089
(0.084)
0.085
(0.084)
Some college 0.114 (0.071) 0.118 (0.071) 0.112 (0.070)
College graduate 0.074 (0.100) 0.073 (0.101) 0.070 (0.100)
Beyond 4-yr college 0.042
(0.083)
0.043
(0.082)
0.033
(0.082)
Log family income 0.024
(0.040)
0.023 (0.041) 0.047
(0.043)
Physical disability -0.206
(0.119)
-0.197
(0.120)
-0.208
(0.119)
Childhood sexual abuse * White NH ref.
Childhood sexual abuse * Black NH 0.276 (0.190)
136
Childhood sexual abuse * Native American
NH
-0.353
(0.601)
Childhood sexual abuse * Asian or Pacific
Islander NH
0.761 (0.414)
Childhood sexual abuse * Other NH 0.229 (0.308)
Childhood sexual abuse * Hispanic -0.013
(0.170)
Childhood sexual abuse * Log family income -0.104
(0.072)
2nd part: GLM for earnings level
Childhood sexual abuse -0.117**
(0.035)
-0.133**
(0.048)
0.110 (0.518)
Physical abuse 0.043
(0.032)
0.043
(0.032)
0.044
(0.032)
Emotional abuse -0.011
(0.035)
-0.007
(0.035)
-0.012
(0.035)
Parental incarceration
No ref. ref. ref.
Yes -0.08*
(0.032)
-0.079*
(0.032)
-0.08*
(0.032)
Don't know -0.051
(0.047)
-0.051
(0.048)
-0.051
(0.047)
Age 0.035***
(0.009)
0.036***
(0.009)
0.035***
(0.009)
Female -0.275***
(0.022)
-0.276***
(0.022)
-0.275***
(0.022)
Race/ethnicity
White NH ref. ref. ref.
Black NH -0.119**
(0.041)
-0.106*
(0.041)
-0.119**
(0.040)
Native American NH -0.365**
(0.129)
-0.293*
(0.144)
-0.365**
(0.129)
Asian or Pacific Islander NH 0.125 (0.080) 0.042
(0.069)
0.124 (0.079)
Other NH -0.074
(0.047)
-0.095*
(0.048)
-0.074
(0.047)
Hispanic -0.006
(0.048)
-0.009
(0.052)
-0.006
(0.048)
Parental education
< High school -0.076
(0.038)
-0.081*
(0.04)
-0.076
(0.038)
GED -0.052
(0.058)
-0.048
(0.058)
-0.052
(0.058)
137
High school graduate ref. ref. ref.
Vocational school + high school 0.030 (0.041) 0.033 (0.041) 0.030 (0.041)
Some college 0.078*
(0.034)
0.080*
(0.034)
0.078*
(0.034)
College graduate 0.122***
(0.035)
0.124***
(0.035)
0.121***
(0.035)
Beyond 4-yr college 0.111**
(0.038)
0.113**
(0.038)
0.110**
(0.038)
Log family income 0.122***
(0.019)
0.122***
(0.019)
0.126***
(0.017)
Physical disability -0.042
(0.064)
-0.039
(0.065)
-0.042
(0.065)
Childhood sexual abuse * White NH ref.
Childhood sexual abuse * Black NH -0.062
(0.083)
Childhood sexual abuse * Native American
NH
-0.394
(0.304)
Childhood sexual abuse * Asian or Pacific
Islander NH
0.609**
(0.228)
Childhood sexual abuse * Other NH 0.103 (0.108)
Childhood sexual abuse * Hispanic 0.034 (0.091)
Childhood sexual abuse * Log family income -0.021
(0.047)
NOTE.— This table displays coefficients and standard errors from a two-part model of earnings. Sample
size was N=14,741 for each regression here. All regressions included school fixed effects and a
constant. * p<0.05, ** p<0.01, *** p<0.001
138
Appendix B: Chapter 4 appendix tables
Table B 1. Health conditions and associated medications, by group
Group Included conditions
Associated medications
(1) (2) (3)
Cancer Cancer Antineoplastics
Cardiometabolic Diabetes;
Heart disease;
Hypertension;
Lipid disorder
Cardiovascular agents;
Coagulation modifiers;
Metabolic agents
Gastrointestinal Acute gastrointestinal
issues (nausea, vomiting,
or diarrhea in last 2
weeks; blood in stool or
urine in last 2 weeks)
Gastrointestinal agents (antacids,
antidiarrheals, digestive enzymes,
gallstone solubilizing agents, GI
stimulants, H2 antagonists, laxatives,
proton pump inhibitors, 5-
aminosalicylates, H. pylori
eradication agents, functional bowel
disorder agents)
Infectious Short-term infection
(cold or flu symptoms,
last 2 weeks; fever, last 2
weeks)
a
Anti-infective agents;
Immunologic agents: interferons;
Respiratory agents (antitussives,
decongestants, expectorants, upper
respiratory combinations)
Nervous system ADHD;
Anxiety or panic disorder;
Depression;
Migraines;
PTSD;
Seizure
Central nervous system agents;
Psychotherapeutic agents
Respiratory/allergic Asthma, chronic
bronchitis or emphysema;
Hay fever
Respiratory agents (antiasthmatic
combinations, antihistamines,
bronchodilators, leukotriene
modifiers, respiratory inhalant
products)
NOTE.— Health conditions were self-reported diagnoses by a doctor, ever, except where recent
time period is noted, which are self-reported acute symptoms. Medications include prescription
medications used in the past 4 weeks. The medication classes noted here reflect mapping to the
Multum Lexicon therapeutic classification system, a three-level hierarchical system developed by
clinicians and used in other health surveys such as the National Health and Nutrition Examination
Survey. With support from clinicians, I grouped all medications into the six categories noted in
column (1), excluding only medications not prescribed to treat specific health conditions (e.g.,
nutritional products, topical agents, vaccines) and those with very low prevalence in this young
adult sample (immunologic agents).
139
a
While Add Health also includes a question about Hepatitis C diagnosis, I exclude this condition
from the infectious group and only study short-term infections because the prevalence of Hepatitis
C is just 0.2 percent in this young adult sample, and these two health states are very different.
Table B 2. Assessing significance with multiple comparison adjustment
(1) (2) (3) (4)
Outcome Uncorrecte
d p-value Rank
Critical
p-value
Significant at
FDR=0.05
Unmet medical need <0.001 1 0.004 √
Condition: Nervous system <0.001 2 0.008 √
Degraded health due to unmet
need
<0.001 3 0.013 √
Condition: Gastrointestinal <0.001 4 0.017 √
Medication: Nervous system <0.001 5 0.021 √
Condition: Respiratory/allergic <0.001 6 0.025 √
Uninsured <0.001 7 0.029 √
Condition: Infectious 0.001 8 0.033 √
Condition: Cancer 0.003 9 0.038 √
Condition: Cardiometabolic 0.003 10 0.042 √
Medication: Cardiometabolic 0.011 11 0.046 √
Medication: Infectious 0.025 12 0.050 √
NOTE.— This table displays results from the Benjamini-Hochberg procedure for controlling the
false discovery rate (FDR).(Benjamini and Hochberg 1995) Column (1) displays the uncorrected p-
value from the individual test of significance of the coefficient on childhood abuse in the logit
model, column (2) displays the test rank by uncorrected p-value in ascending order, column (3)
notes the critical p-value for the test, and column (4) displays a check (√) if the test survives the
adjustment, i.e., if the uncorrected p-value is less than the critical p-value. The critical p-value is
(i/m)Q, where i=rank, m=number of tests, and Q=FDR, which is the rate of false positives allowed
among total number of rejected null hypotheses.
Table B 3. Sensitivity analyses: OLS estimates of the association between childhood
abuse and health
Cancer
Cardio-
metabolic
Gastro-
intestinal Infectious
Nervous
system
Respirator
y/ allergic
A. Diagnosis/
symptoms
Baseline model 0.008**
(0.003)
0.039**
(0.013)
0.048***
(0.008)
0.037***
(0.010)
0.116***
(0.012)
0.055***
(0.012)
+ Childhood
factors
140
Insured 0.008**
(0.003)
0.039**
(0.013)
0.048***
(0.008)
0.033**
(0.010)
0.117***
(0.012)
0.056***
(0.012)
Well child visit 0.008**
(0.003)
0.038**
(0.013)
0.048***
(0.008)
0.035***
(0.010)
0.116***
(0.012)
0.056***
(0.012)
Parent heavy
drinker
0.008**
(0.003)
0.039**
(0.013)
0.049***
(0.008)
0.033**
(0.010)
0.118***
(0.012)
0.054***
(0.012)
Smoker in
household
0.009**
(0.003)
0.039**
(0.013)
0.050***
(0.009)
0.033**
(0.011)
0.116***
(0.013)
0.059***
(0.012)
Physical neglect 0.006
(0.003)
0.042**
(0.015)
0.055***
(0.010)
0.031**
(0.012)
0.116***
(0.013)
0.061***
(0.013)
Learning
challenge
0.010***
(0.003)
0.039**
(0.013)
0.051***
(0.009)
0.031**
(0.011)
0.111***
(0.013)
0.058***
(0.012)
Depressive
symptoms
0.007**
(0.003)
0.037**
(0.013)
0.045***
(0.008)
0.030**
(0.010)
0.108***
(0.011)
0.054***
(0.012)
Neighborhood
education level
a
0.008**
(0.003)
0.039**
(0.013)
0.048***
(0.008)
0.037***
(0.010)
0.115***
(0.012)
0.056***
(0.012)
+ Current factors
Insured 0.008**
(0.003)
0.040**
(0.013)
0.047***
(0.008)
0.034**
(0.010)
0.115***
(0.012)
0.056***
(0.012)
Smoker 0.008**
(0.003)
0.039**
(0.013)
0.045***
(0.008)
0.033**
(0.010)
0.108***
(0.011)
0.056***
(0.012)
Smoker ever 0.008**
(0.002)
0.040**
(0.013)
0.045***
(0.008)
0.033**
(0.010)
0.105***
(0.012)
0.056***
(0.012)
Month of
interview
b
– –
0.048***
(0.008)
0.036***
(0.010)
– –
B. Medication
Baseline model
–
0.014*
(0.006)
–
0.014*
(0.006)
0.051***
(0.009)
–
+ Childhood
factors
Insured
–
0.014*
(0.006)
–
0.014*
(0.006)
0.051***
(0.009)
–
Well child visit
–
0.015*
(0.006)
–
0.014*
(0.006)
0.051***
(0.009)
–
Parent heavy
drinker
–
0.015**
(0.006)
–
0.015*
(0.006)
0.051***
(0.009)
–
Smoker in
household
–
0.016*
(0.006)
–
0.012
(0.007)
0.051***
(0.010)
–
Physical neglect
–
0.014*
(0.007)
–
0.011
(0.007)
0.052***
(0.010)
–
Learning
challenge
–
0.015*
(0.006)
–
0.011
(0.006)
0.050***
(0.010)
–
Depressive
symptoms
–
0.014*
(0.006)
–
0.013*
(0.006)
0.049***
(0.009)
–
Neighborhood
education level
a
–
0.013*
(0.006)
–
0.016*
(0.006)
0.050***
(0.010)
–
+ Current factors
141
Insured
–
0.016**
(0.006)
–
0.016*
(0.006)
0.054***
(0.009)
–
Smoker
–
0.017**
(0.006)
–
0.014*
(0.006)
0.047***
(0.009)
–
Smoker ever
–
0.017**
(0.006)
–
0.015*
(0.006)
0.048***
(0.009)
–
Month of
interview
b
–
0.014*
(0.006)
–
0.015*
(0.006)
0.051***
(0.009)
–
NOTE.— See Appendix Table B1 for a list of health conditions and medications within each group.
All regressions controlled for current age, sex, race/ethnicity, log of childhood household income,
highest parental educational attainment, school fixed effects, and parental incarceration. Standard
errors were clustered at the school level. Health categories for which there were no differences
in unadjusted means are omitted. Sample sizes were N = 14,741 for the baseline models and others
here except for the following cases: N=13,062 (+ smoker in household), N=11,988 (+ physical
neglect), and N=13,084 (learning challenge), N=14,732 (+ current smoker, + smoker ever).
a
Neighborhood education level is the percent of residents aged 25+ years in the child’s Census block
group of residence who do not have a high school degree.
b
Dummy variables for month of
interview were included only in models of outcomes assessed in a recent period versus diagnosis
ever: gastrointestinal symptoms in past 2 weeks; cold, flu, or fever symptoms in past 2 weeks; and
prescription medication use in past 4 weeks. *p<0.05, **p<0.01, ***p<0.001
142
Table B 4. Childhood abuse, family socioeconomic status, and health
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
Model 1 Model 2 Model 3 d ratio
Model 1 Model 2 Model 3 d ratio
Model 1 Model 2 Model 3 d ratio
A. Diagnosis/symptoms
Cancer Cardiometabolic Gastrointestinal
Predictor
Childhood
abuse
0.0087
0.0085 0.0002 0.023 ns 0.0405
0.0377 0.0028 0.069 * 0.0482
0.048 0.0002 0.004 ns
(0.0031)
(0.0031) (0.0002)
(0.0117)
(0.0118) (0.0013)
(0.0077)
(0.0077) (0.0002)
Low
parental
education
0.0036 0.0031 0.0006 0.167 **
0.0454 0.0429 0.0025 0.055 *
0.0056 0.0025 0.0031 0.554 ***
(0.0026) (0.0026) (0.0002)
(0.0113) (0.0114) (0.0011)
(0.0071) (0.0071) (0.0002)
Infectious Nervous system Respiratory/allergic
Childhood
abuse
0.0382
0.0379 0.0003 0.008 ns 0.1298
0.1302 -0.0004 -0.003 ns 0.0555
0.0586 -0.0031 -0.056 ns
(0.0107)
(0.0108) (0.001)
(0.0118)
(0.0119) (0.0007)
(0.0113)
(0.0113) (0.0004)
Low
parental
education
0.007 0.0045 0.0025 0.357 **
0.0021 -0.0064 0.0085 4.048 ***
-0.0425 -0.0463 0.0038 0.089 ***
(0.0104) (0.0104) (0.0009)
(0.0116) (0.0115) (0.0011)
(0.0114) (0.0114) (0.0008)
B. Medication use
Cardiometabolic Infectious Nervous system
Childhood
abuse
0.0112
0.0109 0.0003 0.027 ns 0.0175
0.0169 0.0006 0.034 ns 0.0597
0.0597 0.000 0.000 ns
(0.0058)
(0.0058) (0.0008)
(0.0066)
(0.0066) (0.0006)
(0.009)
(0.009) (0.0002)
Low
parental
education
0.0051 0.0044 0.0007 0.137 ns
0.0105 0.0094 0.0011 0.105 *
0.0045 0.0006 0.0039 0.867 ***
(0.0055) (0.0055) (0.0007)
(0.0058) (0.0058) (0.0005)
(0.0086) (0.0085) (0.0004)
C. Health care access
Uninsured Unmet medical need Health decline
Childhood
abuse
0.0674
0.0613 0.0061 0.091 *** 0.1263
0.1232 0.0031 0.025 *** 0.0793
0.0779 0.0014 0.018 **
(0.0108)
(0.0108) (0.001)
(0.0112)
(0.0113) (0.0008
)
(0.0084)
(0.0084) (0.0006)
Low
parental
education
0.096 0.092 0.004 0.042 ***
0.0547 0.0466 0.0081 0.148 ***
0.0264 0.0213 0.0051 0.193 ***
(0.0096) (0.0096) (0.0003)
(0.0103) (0.0103) (0.0008
)
(0.0072) (0.0071) (0.0004)
NOTE.— Columns (1) to (3) display coefficients on the predictors noted and standard errors in parentheses from OLS models of the corresponding condition, medication use, or health care
access outcome. Low parental education was defined as less than college degree. Each model controlled for current age and sex and had sample size N=14,299. Column (4) displays the
difference in coefficients between the restricted model (1) or (2) and the full model (3) along with the standard error of this difference, calculated following (Clogg, et al. 1995). Column (5)
displays the ratio of the difference in coefficients to the magnitude of the coefficient from the restricted model. Column (6) notes whether the coefficient from the full model is significantly
smaller than the coefficient from the restricted model.
ns, not significant, *p<0.05, **p<0.01, ***p<0.001
143
Table B 5. Heterogeneity by sociodemographics for: cancer diagnosis
(1) (2) (3) (4)
Childhood abuse 0.008**
(0.003)
0.006
(0.003)
0.006
(0.003)
-0.008
(0.038)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.005
(0.007)
0.005
(0.007)
0.005
(0.007)
0.005
(0.007)
Don't know 0.003
(0.006)
0.003
(0.006)
0.003
(0.006)
0.003
(0.006)
Female 0.011***
(0.003)
0.010**
(0.003)
0.011***
(0.003)
0.011***
(0.003)
Age -0.001
(0.001)
-0.001
(0.001)
-0.001
(0.001)
-0.001
(0.001)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH -0.007
(0.005)
-0.007
(0.005)
-0.009
(0.006)
-0.007
(0.005)
Native American NH -0.012
(0.008)
-0.012
(0.008)
-0.012
(0.011)
-0.012
(0.008)
Asian or Pacific Islander NH 0.001
(0.009)
0.001
(0.009)
0.008
(0.012)
0.001
(0.009)
Other NH -0.010*
(0.005)
-0.010*
(0.005)
-0.004
(0.006)
-0.010*
(0.005)
Hispanic 0.003
(0.007)
0.003
(0.007)
-0.004
(0.006)
0.002
(0.007)
Parental education
< High school 0.004
(0.005)
0.004
(0.005)
0.005
(0.005)
0.004
(0.005)
GED 0.004
(0.009)
0.004
(0.009)
0.004
(0.009)
0.004
(0.009)
High school graduate ref. ref. ref. ref.
Vocational school + high school 0.003
(0.004)
0.003
(0.004)
0.003
(0.004)
0.003
(0.004)
Some college 0.006
(0.004)
0.006
(0.004)
0.005
(0.004)
0.006
(0.004)
College graduate 0.000
(0.003)
0.000
(0.003)
0.000
(0.003)
0.000
(0.003)
Beyond 4-yr college 0.003
(0.005)
0.003
(0.005)
0.003
(0.005)
0.003
(0.005)
Log household income 0.003
(0.002)
0.003
(0.002)
0.003
(0.002)
0.003
(0.002)
Childhood abuse * Female
0.003
(0.005)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
0.005
(0.008)
144
Childhood abuse * Native American
NH
-0.001
(0.010)
Childhood abuse * Asian or Pacific
Islander NH
-0.023*
(0.010)
Childhood abuse * Other NH
-0.015*
(0.007)
Childhood abuse * Hispanic
0.018
(0.011)
Childhood abuse * Log household
income
0.002
(0.004)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1)
displays the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
Table B 6. Heterogeneity by sociodemographics for: cardiometabolic conditions
(1) (2) (3) (4)
Childhood abuse 0.039**
(0.013)
0.041*
(0.018)
0.031
(0.017)
-0.173
(0.182)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.003
(0.020)
0.003
(0.020)
0.002
(0.020)
0.004
(0.020)
Don't know 0.000
(0.023)
0.001
(0.023)
0.001
(0.023)
-0.000
(0.023)
Female -0.144***
(0.012)
-0.143***
(0.014)
-0.144***
(0.012)
-0.144***
(0.012)
Age 0.020***
(0.005)
0.020***
(0.005)
0.020***
(0.005)
0.020***
(0.005)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH 0.052*
(0.020)
0.051*
(0.020)
0.050*
(0.022)
0.051*
(0.020)
Native American NH 0.106
(0.088)
0.106
(0.088)
0.074
(0.113)
0.104
(0.088)
Asian or Pacific Islander NH 0.003
(0.028)
0.003
(0.028)
0.010
(0.035)
0.002
(0.028)
Other NH 0.044
(0.025)
0.044
(0.025)
0.020
(0.034)
0.044
(0.025)
Hispanic 0.028
(0.026)
0.028
(0.026)
0.014
(0.029)
0.027
(0.026)
Parental education
< High school -0.041*
(0.020)
-0.041*
(0.020)
-0.040*
(0.020)
-0.042*
(0.020)
GED 0.050
(0.039)
0.050
(0.039)
0.050
(0.039)
0.050
(0.039)
High school graduate ref. ref. ref. ref.
145
Vocational school + high school -0.035
(0.021)
-0.035
(0.021)
-0.034
(0.021)
-0.034
(0.021)
Some college -0.016
(0.020)
-0.016
(0.020)
-0.015
(0.020)
-0.015
(0.020)
College graduate -0.046*
(0.020
-0.046*
(0.020)
-0.046*
(0.020)
-0.045*
(0.019)
Beyond 4-yr college -0.027
(0.022)
-0.028
(0.022)
-0.028
(0.022)
-0.026
(0.022)
Log household income 0.000
(0.009)
0.000
(0.009)
0.000
(0.009)
-0.007
(0.011)
Childhood abuse * Female
-0.005
(0.022)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
0.004
(0.037)
Childhood abuse * Native American
NH
0.097
(0.146)
Childhood abuse * Asian or Pacific
Islander NH
-0.023
(0.070)
Childhood abuse * Other NH
0.063
(0.061)
Childhood abuse * Hispanic
0.041
(0.035)
Childhood abuse * Log household
income
0.020
(0.017)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1) displays
the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
Table B 7. Heterogeneity by sociodemographics for: high cholesterol (measured)
(1) (2) (3) (4)
Childhood abuse -0.013
(0.008)
-0.013
(0.012)
-0.016
(0.011)
-0.087
(0.107)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.009
(0.013)
0.009
(0.013)
0.009
(0.012)
0.009
(0.013)
Don't know -0.005
(0.016)
-0.005
(0.016)
-0.005
(0.016)
-0.005
(0.016)
Female -0.073***
(0.008)
-0.073***
(0.009)
-0.073***
(0.008)
-0.073***
(0.008)
Age 0.000
(0.003)
0.000
(0.003)
0.000
(0.003)
0.000
(0.003)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH -0.046**
(0.015)
-0.046**
(0.015)
-0.051***
(0.015)
-0.047**
(0.015)
146
Native American NH 0.052
(0.064)
0.052
(0.064)
0.007
(0.070)
0.051
(0.064)
Asian or Pacific Islander NH 0.037
(0.032)
0.037
(0.032)
0.023
(0.030)
0.037
(0.032)
Other NH -0.033
(0.019)
-0.033
(0.019)
-0.034
(0.024)
-0.033
(0.019)
Hispanic 0.007
(0.016)
0.007
(0.016)
0.010
(0.021)
0.006
(0.016)
Parental education
< High school -0.007
(0.017)
-0.007
(0.017)
-0.007
(0.017)
-0.007
(0.017)
GED 0.022
(0.027)
0.022
(0.027)
0.022
(0.027)
0.022
(0.027)
High school graduate ref. ref. ref. ref.
Vocational school + high school -0.006
(0.014)
-0.006
(0.014)
-0.006
(0.014)
-0.006
(0.014)
Some college 0.004
(0.013)
0.004
(0.013)
0.004
(0.013)
0.004
(0.013)
College graduate -0.008
(0.015)
-0.008
(0.015)
-0.008
(0.015)
-0.007
(0.015)
Beyond 4-yr college -0.009
(0.015)
-0.009
(0.015)
-0.010
(0.015)
-0.009
(0.015)
Log household income 0.003
(0.006)
0.003
(0.006)
0.003
(0.006)
0.001
(0.007)
Childhood abuse * Female 0.001
(0.014)
Childhood abuse * White NH ref.
Childhood abuse * Black NH 0.015
(0.017)
Childhood abuse * Native American
NH
0.138
(0.125)
Childhood abuse * Asian or Pacific
Islander NH
0.047
(0.060)
Childhood abuse * Other NH 0.003
(0.041)
Childhood abuse * Hispanic -0.010
(0.027)
Childhood abuse * Log household
income
0.007
(0.010)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1) displays
the baseline results. Sample size was N = 12,999, smaller than in study of other outcomes because
not all survey respondents participated in the blood draw for biomarker measurement. * p<0.05,
** p<0.01, *** p<0.001
147
Table B 8. Heterogeneity by sociodemographics for: obesity (measured)
(1) (2) (3) (4)
Childhood abuse 0.021
(0.013)
0.014
(0.019)
0.036*
(0.016)
-0.390*
(0.186)
Parental incarceration
No ref. ref. ref. ref.
Yes -0.023
(0.020)
-0.023
(0.020)
-0.022
(0.020)
-0.021
(0.020)
Don't know -0.015
(0.025)
-0.016
(0.025)
-0.016
(0.025)
-0.016
(0.025)
Female 0.015
(0.012)
0.010
(0.014)
0.015
(0.012)
0.015
(0.012)
Age 0.003
(0.005)
0.003
(0.005)
0.003
(0.005)
0.003
(0.005)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH 0.070**
(0.022)
0.071**
(0.022)
0.082**
(0.026)
0.069**
(0.022)
Native American NH 0.199
(0.113)
0.198
(0.113)
0.283**
(0.104)
0.194 (0.113)
Asian or Pacific Islander NH -0.120***
(0.032)
-0.120***
(0.032)
-0.111**
(0.037)
-0.121***
(0.032)
Other NH 0.034
(0.030)
0.034
(0.030)
0.074
(0.040)
0.034
(0.030)
Hispanic 0.040
(0.022)
0.040
(0.022)
0.047
(0.026)
0.039
(0.022)
Parental education
< High school -0.019
(0.024)
-0.019
(0.024)
-0.020
(0.024)
-0.020
(0.024)
GED -0.017
(0.040)
-0.017
(0.040)
-0.018
(0.040)
-0.018
(0.040)
High school graduate ref. ref. ref. ref.
Vocational school + high school -0.031
(0.023)
-0.031
(0.023)
-0.032
(0.023)
-0.030
(0.023)
Some college -0.036
(0.020)
-0.036
(0.020)
-0.036
(0.020)
-0.035
(0.020)
College graduate -0.070**
(0.023)
-0.069**
(0.023)
-0.069**
(0.023)
-0.068**
(0.023)
Beyond 4-yr college -0.114***
(0.022)
-0.114***
(0.022)
-0.114***
(0.022)
-0.111***
(0.022)
Log household income -0.013
(0.010)
-0.013
(0.010)
-0.013
(0.010)
-0.026*
(0.012)
Childhood abuse * Female 0.015
(0.023)
Childhood abuse * White NH ref.
Childhood abuse * Black NH -0.035
(0.036)
148
Childhood abuse * Native American
NH
-0.252
(0.234)
Childhood abuse * Asian or Pacific
Islander NH
-0.029
(0.065)
Childhood abuse * Other NH -0.101
(0.053)
Childhood abuse * Hispanic -0.023
(0.044)
Childhood abuse * Log household
income
0.038*
(0.017)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1) displays
the baseline results. Sample size was N = 14,060, slightly smaller than in study of other outcomes
because not all survey respondents participated in the anthropometric measurement. * p<0.05, **
p<0.01, *** p<0.001
Table B 9. Heterogeneity by sociodemographics for: recent gastrointestinal symptoms
(1) (2) (3) (4)
Childhood abuse 0.048***
(0.008)
0.030**
(0.011)
0.051***
(0.011)
0.026
(0.111)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.009
(0.011)
0.009
(0.011)
0.008
(0.011)
0.009
(0.011)
Don't know 0.021
(0.014)
0.021
(0.014)
0.021
(0.014)
0.021
(0.014)
Female 0.051***
(0.008)
0.041***
(0.008)
0.050***
(0.008)
0.051***
(0.008)
Age -0.001
(0.003)
-0.001
(0.003)
-0.001
(0.003)
-0.001
(0.003)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH -0.025**
(0.009)
-0.024*
(0.009)
-0.014
(0.011)
-0.025**
(0.009)
Native American NH 0.008
(0.036)
0.007
(0.037)
-0.040
(0.032)
0.008
(0.037)
Asian or Pacific Islander NH -0.054***
(0.016)
-0.054***
(0.016)
-0.046**
(0.016)
-0.054***
(0.016)
Other NH 0.029
(0.019)
0.029
(0.019)
0.011
(0.022)
0.029
(0.019)
Hispanic 0.004
(0.014)
0.005
(0.014)
0.007
(0.015)
0.004
(0.014)
Parental education
< High school -0.006
(0.012)
-0.006
(0.012)
-0.005
(0.012)
-0.006
(0.013)
GED 0.021
(0.025)
0.022
(0.025)
0.022
(0.025)
0.021
(0.025)
149
High school graduate ref. ref. ref. ref.
Vocational school + high school 0.007
(0.015)
0.008
(0.015)
0.008
(0.015)
0.007
(0.015)
Some college 0.019
(0.010)
0.019
(0.010)
0.019
(0.010)
0.019
(0.010)
College graduate -0.002
(0.010)
-0.002
(0.010)
-0.002
(0.010)
-0.002
(0.010)
Beyond 4-yr college 0.005
(0.013)
0.005
(0.013)
0.006
(0.013)
0.005
(0.013)
Log household income 0.006
(0.006)
0.006
(0.006)
0.006
(0.006)
0.005
(0.007)
Childhood abuse * Female
0.034*
(0.016)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
-0.033
(0.018)
Childhood abuse * Native American
NH
0.146
(0.100)
Childhood abuse * Asian or Pacific
Islander NH
-0.025
(0.025)
Childhood abuse * Other NH
0.045
(0.043)
Childhood abuse * Hispanic
-0.006
(0.024)
Childhood abuse * Log household
income
0.002
(0.010)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1) displays
the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
Table B 10. Heterogeneity by sociodemographics for: recent cold or flu symptoms
(1) (2) (3) (4)
Childhood abuse 0.034**
(0.010)
0.023
(0.017)
0.029*
(0.012)
0.016
(0.147)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.018
(0.018)
0.018
(0.018)
0.017
(0.018)
0.018
(0.018)
Don't know 0.036
(0.021)
0.036
(0.021)
0.036
(0.021)
0.036
(0.021)
Female 0.032**
(0.010)
0.026*
(0.012)
0.032**
(0.010)
0.032**
(0.010)
Age -0.003
(0.004)
-0.003
(0.004)
-0.003
(0.004)
-0.003
(0.004)
Race/ethnicity
White NH ref. ref. ref. ref.
150
Black NH -0.022
(0.018)
-0.021
(0.018)
-0.035
(0.023)
-0.022
(0.018)
Native American NH -0.073
(0.064)
-0.074
(0.063)
-0.121
(0.075)
-0.073
(0.064)
Asian or Pacific Islander NH -0.007
(0.033)
-0.006
(0.033)
0.029
(0.037)
-0.007
(0.033)
Other NH 0.035
(0.026)
0.035
(0.026)
0.024
(0.030)
0.035
(0.026)
Hispanic 0.026
(0.020)
0.026
(0.020)
0.023
(0.022)
0.026
(0.020)
Parental education
< High school -0.018
(0.020)
-0.018
(0.021)
-0.018
(0.020)
-0.019
(0.020)
GED 0.004
(0.035)
0.005
(0.035)
0.004
(0.035)
0.004
(0.035)
High school graduate ref. ref. ref. ref.
Vocational school + high school 0.002
(0.024)
0.002
(0.024)
0.001
(0.024)
0.002
(0.024)
Some college -0.005
(0.015)
-0.005
(0.015)
-0.006
(0.015)
-0.005
(0.015)
College graduate -0.011
(0.016)
-0.011
(0.016)
-0.011
(0.016)
-0.011
(0.016)
Beyond 4-yr college 0.009
(0.020)
0.009
(0.020)
0.009
(0.020)
0.009
(0.020)
Log household income -0.004
(0.009)
-0.004
(0.009)
-0.005
(0.009)
-0.005
(0.011)
Childhood abuse * Female
0.019
(0.022)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
0.036
(0.028)
Childhood abuse * Native American
NH
0.144
(0.139)
Childhood abuse * Asian or Pacific
Islander NH
-0.115*
(0.052)
Childhood abuse * Other NH
0.028
(0.050)
Childhood abuse * Hispanic
0.008
(0.032)
Childhood abuse * Log household
income
0.002
(0.014)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1) displays
the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
151
Table B 11. Heterogeneity by sociodemographics for: nervous system condition
diagnosis
(1) (2) (3) (4)
Childhood abuse 0.116***
(0.012)
0.075***
(0.016)
0.135***
(0.015)
-0.222
(0.141)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.085***
(0.017)
0.084***
(0.017)
0.084***
(0.017)
0.086***
(0.017)
Don't know 0.055*
(0.027)
0.054
(0.027)
0.054*
(0.027)
0.054*
(0.027)
Female 0.146***
(0.011)
0.122***
(0.012)
0.144***
(0.011)
0.146***
(0.011)
Age 0.006
(0.004)
0.006
(0.004)
0.006
(0.004)
0.006
(0.004)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH -0.147***
(0.018)
-0.144***
(0.018)
-0.120***
(0.020)
-0.148***
(0.018)
Native American NH -0.063
(0.066)
-0.067
(0.065)
-0.114
(0.076)
-0.067
(0.066)
Asian or Pacific Islander NH -0.216***
(0.022)
-0.216***
(0.022)
-0.181***
(0.027)
-0.217***
(0.022)
Other NH 0.009
(0.027)
0.008
(0.027)
-0.008
(0.032)
0.009
(0.027)
Hispanic -0.084***
(0.025)
-0.084***
(0.025)
-0.068**
(0.024)
-0.085***
(0.025)
Parental education
< High school -0.048*
(0.019)
-0.047*
(0.019)
-0.047*
(0.019)
-0.048*
(0.019)
GED 0.052
(0.042)
0.054
(0.042)
0.053
(0.042)
0.052
(0.042)
High school graduate ref. ref. ref. ref.
Vocational school + high school -0.014
(0.019)
-0.013
(0.018)
-0.014
(0.019)
-0.014
(0.018)
Some college -0.004
(0.014)
-0.004
(0.014)
-0.004
(0.014)
-0.004
(0.014)
College graduate -0.025
(0.018)
-0.025
(0.018)
-0.024
(0.018)
-0.024
(0.018)
Beyond 4-yr college -0.004
(0.022)
-0.003
(0.022)
-0.003
(0.022)
-0.002
(0.022)
Log household income 0.007
(0.008)
0.008
(0.008)
0.008
(0.008)
-0.004
(0.009)
Childhood abuse * Female
0.079***
(0.022)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
-0.079**
(0.028)
152
Childhood abuse * Native American
NH
0.153
(0.143)
Childhood abuse * Asian or Pacific
Islander NH
-0.111*
(0.048)
Childhood abuse * Other NH
0.040
(0.056)
Childhood abuse * Hispanic
-0.046
(0.033)
Childhood abuse * Log household
income
0.031*
(0.013)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1) displays
the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
Table B 12. Heterogeneity by sociodemographics for: respiratory/allergic condition
diagnosis
(1) (2) (3) (4)
Childhood abuse 0.055***
(0.012)
0.026
(0.016)
0.058***
(0.015)
0.021
(0.150)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.036
(0.019)
0.036
(0.019)
0.035
(0.019)
0.036
(0.019)
Don't know 0.056*
(0.023)
0.055*
(0.023)
0.056*
(0.023)
0.056*
(0.023)
Female 0.049***
(0.011)
0.032*
(0.013)
0.048***
(0.011)
0.049***
(0.011)
Age -0.007
(0.004)
-0.007
(0.004)
-0.007
(0.004)
-0.007
(0.004)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH -0.016
(0.017)
-0.014
(0.017)
-0.003
(0.018)
-0.016
(0.017)
Native American NH 0.010
(0.079)
0.007
(0.079)
-0.059
(0.082)
0.010
(0.079)
Asian or Pacific Islander NH -0.041
(0.029)
-0.041
(0.028)
-0.025
(0.034)
-0.041
(0.028)
Other NH 0.064*
(0.030)
0.063*
(0.030)
0.020
(0.036)
0.064*
(0.030)
Hispanic 0.004
(0.022)
0.004
(0.022)
0.007
(0.025)
0.004
(0.022)
Parental education
< High school -0.026
(0.018)
-0.026
(0.018)
-0.025
(0.019)
-0.026
(0.018)
GED 0.031
(0.035)
0.032
(0.035)
0.032
(0.035)
0.031
(0.035)
High school graduate ref. ref. ref. ref.
153
Vocational school + high school 0.012
(0.021)
0.013
(0.021)
0.013
(0.021)
0.012
(0.021)
Some college 0.017
(0.016)
0.017
(0.016)
0.017
(0.016)
0.017
(0.016)
College graduate 0.044**
(0.016)
0.044**
(0.016)
0.044**
(0.016)
0.044**
(0.016)
Beyond 4-yr college 0.013
(0.019)
0.014
(0.019)
0.014
(0.019)
0.014
(0.019)
Log household income 0.024**
(0.008)
0.025**
(0.008)
0.025**
(0.008)
0.023*
(0.010)
Childhood abuse * Female
0.055*
(0.023)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
-0.040
(0.029)
Childhood abuse * Native American
NH
0.209
(0.148)
Childhood abuse * Asian or Pacific
Islander NH
-0.051
(0.059)
Childhood abuse * Other NH
0.110*
(0.053)
Childhood abuse * Hispanic
-0.009
(0.032)
Childhood abuse * Log household
income
0.003
(0.014)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1) displays
the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
Table B 13. Heterogeneity by sociodemographics for: use of medication for a
cardiometabolic condition
(1) (2) (3) (4)
Childhood abuse 0.014*
(0.006)
0.011
(0.009)
0.011
(0.007)
0.016
(0.076)
Parental incarceration
No ref. ref. ref. ref.
Yes -0.004
(0.008)
-0.004
(0.008)
-0.005
(0.008)
-0.004
(0.008)
Don't know -0.004
(0.013)
-0.004
(0.013)
-0.003
(0.013)
-0.004
(0.013)
Female 0.004
(0.005)
0.002
(0.006)
0.004
(0.005)
0.004
(0.005)
Age 0.005**
(0.002)
0.005**
(0.002)
0.005**
(0.002)
0.005**
(0.002)
Race/ethnicity
White NH ref. ref. ref. ref.
154
Black NH 0.003
(0.011)
0.004
(0.011)
0.004
(0.011)
0.003
(0.011)
Native American NH 0.019
(0.024)
0.018
(0.024)
-0.003
(0.021)
0.019
(0.024)
Asian or Pacific Islander NH -0.039**
(0.012)
-0.039**
(0.012)
-0.038**
(0.014)
-0.039**
(0.012)
Other NH -0.002
(0.016)
-0.002
(0.016)
-0.025*
(0.013)
-0.002
(0.016)
Hispanic -0.012
(0.011)
-0.012
(0.011)
-0.014
(0.012)
-0.012
(0.011)
Parental education
< High school 0.010
(0.010)
0.010
(0.011)
0.011
(0.011)
0.010
(0.011)
GED -0.004
(0.016)
-0.004
(0.016)
-0.004
(0.016)
-0.004
(0.016)
High school graduate ref. ref. ref. ref.
Vocational school + high school -0.018
(0.010)
-0.018
(0.010)
-0.017
(0.010)
-0.018
(0.010)
Some college -0.009
(0.008)
-0.009
(0.008)
-0.009
(0.008)
-0.009
(0.008)
College graduate -0.006
(0.010)
-0.006
(0.010)
-0.006
(0.010)
-0.006
(0.010)
Beyond 4-yr college -0.011
(0.011)
-0.011
(0.011)
-0.011
(0.011)
-0.011
(0.011)
Log household income 0.003
(0.004)
0.003
(0.004)
0.003
(0.004)
0.003
(0.005)
Childhood abuse * Female
0.007
(0.012)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
-0.003
(0.014)
Childhood abuse * Native American
NH
0.063
(0.051)
Childhood abuse * Asian or Pacific
Islander NH
0.000
(0.022)
Childhood abuse * Other NH
0.058
(0.038)
Childhood abuse * Hispanic
0.006
(0.017)
Childhood abuse * Log household
income
-0.000
(0.007)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1) displays
the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
155
Table B 14. Heterogeneity by sociodemographics for: use of medication for an infectious
condition
(1) (2) (3) (4)
Childhood abuse 0.014*
(0.006)
0.019*
(0.009)
0.009
(0.008)
0.063
(0.072)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.009
(0.009)
0.009
(0.009)
0.008
(0.009)
0.009
(0.009)
Don't know 0.025
(0.014)
0.026
(0.014)
0.026
(0.014)
0.026
(0.014)
Female 0.033***
(0.005)
0.036***
(0.007)
0.033***
(0.006)
0.033***
(0.005)
Age 0.004
(0.002)
0.004
(0.002)
0.004
(0.002)
0.004
(0.002)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH -0.017
(0.010)
-0.018
(0.010)
-0.025*
(0.012)
-0.017
(0.010)
Native American NH -0.006
(0.030)
-0.005
(0.030)
-0.007
(0.033)
-0.005
(0.030)
Asian or Pacific Islander NH -0.022
(0.012)
-0.022
(0.012)
-0.016
(0.014)
-0.022
(0.012)
Other NH 0.008
(0.016)
0.008
(0.016)
-0.006
(0.017)
0.008
(0.016)
Hispanic -0.002
(0.012)
-0.002
(0.012)
-0.004
(0.014)
-0.002
(0.012)
Parental education
< High school -0.022*
(0.011)
-0.022*
(0.011)
-0.022*
(0.011)
-0.022*
(0.011)
GED -0.018
(0.019)
-0.018
(0.019)
-0.018
(0.019)
-0.018
(0.019)
High school graduate ref. ref. ref. ref.
Vocational school + high school -0.041**
(0.013)
-0.041**
(0.013)
-0.041**
(0.013)
-0.041**
(0.013)
Some college -0.015
(0.011)
-0.015
(0.011)
-0.014
(0.011)
-0.015
(0.011)
College graduate -0.022
(0.011)
-0.022
(0.011)
-0.022
(0.011)
-0.022*
(0.011)
Beyond 4-yr college -0.027*
(0.011)
-0.028**
(0.010)
-0.028**
(0.010)
-0.028**
(0.011)
Log household income 0.008
(0.004)
0.008
(0.004)
0.008
(0.004)
0.010
(0.005)
Childhood abuse * Female
-0.008
(0.012)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
0.024
(0.016)
156
Childhood abuse * Native American
NH
0.003
(0.055)
Childhood abuse * Asian or Pacific
Islander NH
-0.020
(0.020)
Childhood abuse * Other NH
0.037
(0.033)
Childhood abuse * Hispanic
0.007
(0.018)
Childhood abuse * Log household
income
-0.005
(0.007)
NOTE.— This table displays the coefficients and standard errors from OLS models including
school fixed effects and a constant. Standard errors were clustered at the school level. Column
(1) displays the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
Table B 15. Heterogeneity by sociodemographics for: use of medication for a nervous
system condition
(1) (2) (3) (4)
Childhood abuse 0.051***
(0.009)
0.041**
(0.012)
0.057***
(0.012)
-0.140
(0.124)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.033*
(0.013)
0.033*
(0.013)
0.032*
(0.013)
0.034*
(0.013)
Don't know 0.017
(0.014)
0.017
(0.014)
0.017
(0.014)
0.017
(0.014)
Female 0.037***
(0.008)
0.031***
(0.009)
0.036***
(0.008)
0.037***
(0.008)
Age 0.010**
(0.003)
0.010**
(0.003)
0.010**
(0.003)
0.010**
(0.003)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH -0.055***
(0.012)
-0.054***
(0.012)
-0.043**
(0.013)
-0.056***
(0.012)
Native American NH -0.074
(0.052)
-0.075
(0.051)
-0.112***
(0.017)
-0.076
(0.053)
Asian or Pacific Islander NH -0.063***
(0.017)
-0.063***
(0.017)
-0.052**
(0.018)
-0.064***
(0.017)
Other NH -0.003
(0.020)
-0.003
(0.020)
-0.022
(0.027)
-0.003
(0.020)
Hispanic -0.023
(0.020)
-0.023
(0.020)
-0.017
(0.021)
-0.024
(0.020)
Parental education
< High school 0.008
(0.017)
0.008
(0.017)
0.009
(0.017)
0.008
(0.017)
GED -0.013
(0.023)
-0.012
(0.023)
-0.012
(0.023)
-0.013
(0.023)
High school graduate ref. ref. ref. ref.
157
Vocational school + high school -0.007
(0.014)
-0.007
(0.014)
-0.007
(0.014)
-0.006
(0.014)
Some college -0.007
(0.012)
-0.007
(0.012)
-0.007
(0.012)
-0.007
(0.012)
College graduate 0.001
(0.013)
0.001
(0.013)
0.001
(0.013)
0.002
(0.013)
Beyond 4-yr college 0.015
(0.016)
0.016
(0.016)
0.016
(0.016)
0.017
(0.016)
Log household income -0.006
(0.007)
-0.006
(0.007)
-0.006
(0.007)
-0.012
(0.008)
Childhood abuse * Female
0.018
(0.017)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
-0.035
(0.022)
Childhood abuse * Native American
NH
0.115
(0.149)
Childhood abuse * Asian or Pacific
Islander NH
-0.036
(0.031)
Childhood abuse * Other NH
0.046
(0.053)
Childhood abuse * Hispanic
-0.018
(0.022)
Childhood abuse * Log household
income
0.018
(0.011)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1) displays
the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
Table B 16. Heterogeneity by sociodemographics for: uninsured status
(1) (2) (3) (4)
Childhood abuse 0.050***
(0.012)
0.066***
(0.017)
0.066***
(0.014)
-0.194
(0.163)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.033*
(0.016)
0.033*
(0.016)
0.033*
(0.016)
0.034*
(0.016)
Don't know 0.039
(0.022)
0.039
(0.022)
0.039
(0.022)
0.038
(0.022)
Female -0.085***
(0.009)
-0.076***
(0.011)
-0.086***
(0.009)
-0.085***
(0.009)
Age -0.004
(0.004)
-0.004
(0.004)
-0.004
(0.004)
-0.004
(0.004)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH -0.011
(0.017)
-0.012
(0.017)
0.000
(0.019)
-0.012
(0.018)
158
Native American NH 0.124
(0.095)
0.126
(0.095)
0.235*
(0.115)
0.121
(0.094)
Asian or Pacific Islander NH -0.035
(0.026)
-0.035
(0.026)
-0.014
(0.028)
-0.036
(0.027)
Other NH 0.037
(0.024)
0.037
(0.024)
0.046
(0.033)
0.037
(0.024)
Hispanic -0.028
(0.017)
-0.028
(0.017)
-0.010
(0.019)
-0.029
(0.017)
Parental education
< High school 0.049*
(0.019)
0.049*
(0.019)
0.048*
(0.019)
0.048*
(0.019)
GED 0.017
(0.033)
0.016
(0.033)
0.017
(0.033)
0.017
(0.033)
High school graduate ref. ref. ref. ref.
Vocational school + high school -0.019
(0.016)
-0.019
(0.016)
-0.019
(0.016)
-0.019
(0.016)
Some college -0.005
(0.013)
-0.005
(0.013)
-0.005
(0.013)
-0.005
(0.013)
College graduate -0.031*
(0.014)
-0.031*
(0.014)
-0.030*
(0.013)
-0.030*
(0.014)
Beyond 4-yr college -0.061***
(0.016)
-0.061***
(0.016)
-0.060***
(0.016)
-0.059***
(0.016)
Log household income -0.038***
(0.009)
-0.039***
(0.009)
-0.038***
(0.009)
-0.046***
(0.010)
Childhood abuse * Female
-0.029
(0.018)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
-0.032
(0.026)
Childhood abuse * Native American
NH
-0.339*
(0.131)
Childhood abuse * Asian or Pacific
Islander NH
-0.066
(0.062)
Childhood abuse * Other NH
-0.026
(0.060)
Childhood abuse * Hispanic
-0.054
(0.029)
Childhood abuse * Log household
income
0.023
(0.015)
NOTE.— This table displays the coefficients and standard errors from OLS models including
school fixed effects and a constant. Standard errors were clustered at the school level. Column
(1) displays the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
Table B 17. Heterogeneity by sociodemographics for: unmet medical need
(1) (2) (3) (4)
Childhood abuse 0.113***
(0.011)
0.083***
(0.016)
0.117***
(0.013)
-0.169
(0.148)
Parental incarceration
159
No ref. ref. ref. ref.
Yes 0.055**
(0.019)
0.055**
(0.019)
0.055**
(0.019)
0.057**
(0.019)
Don't know 0.065**
(0.020)
0.064**
(0.020)
0.065**
(0.020)
0.064**
(0.020)
Female -0.022
(0.011)
-0.039**
(0.013)
-0.022*
(0.011)
-0.021
(0.011)
Age -0.002
(0.004)
-0.002
(0.004)
-0.002
(0.004)
-0.002
(0.004)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH 0.024
(0.018)
0.026
(0.018)
0.034
(0.021)
0.023
(0.018)
Native American NH -0.024
(0.086)
-0.027
(0.086)
-0.043
(0.084)
-0.028
(0.087)
Asian or Pacific Islander NH -0.009
(0.030)
-0.008
(0.030)
-0.041
(0.028)
-0.010
(0.030)
Other NH 0.059
(0.032)
0.058
(0.032)
0.040
(0.036)
0.059
(0.032)
Hispanic -0.027
(0.018)
-0.027
(0.018)
-0.011
(0.019)
-0.028
(0.018)
Parental education
< High school 0.013
(0.020)
0.013
(0.020)
0.013
(0.020)
0.013
(0.020)
GED -0.064*
(0.029)
-0.063*
(0.028)
-0.064*
(0.029)
-0.065*
(0.029)
High school graduate ref. ref. ref. ref.
Vocational school + high school -0.008
(0.018)
-0.007
(0.018)
-0.007
(0.018)
-0.007
(0.018)
Some college -0.021
(0.014)
-0.021
(0.014)
-0.021
(0.014)
-0.021
(0.014)
College graduate -0.019
(0.014)
-0.018
(0.014)
-0.018
(0.014)
-0.018
(0.014)
Beyond 4-yr college -0.026
(0.018)
-0.026
(0.018)
-0.025
(0.018)
-0.024
(0.018)
Log household income -0.024**
(0.008)
-0.023**
(0.008)
-0.023**
(0.008)
-0.033***
(0.008)
Childhood abuse * Female
0.057*
(0.023)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
-0.028
(0.032)
Childhood abuse * Native American
NH
0.056
(0.144)
Childhood abuse * Asian or Pacific
Islander NH
0.105
(0.063)
Childhood abuse * Other NH
0.046
(0.050)
160
Childhood abuse * Hispanic
-0.045
(0.030)
Childhood abuse * Log household
income
0.026
(0.014)
NOTE.— This table displays the coefficients and standard errors from OLS models including
school fixed effects and a constant. Standard errors were clustered at the school level. Column
(1) displays the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
Table B 18. Heterogeneity by sociodemographics for: health decline due to unmet
medical need
(1) (2) (3) (4)
Childhood abuse 0.072***
(0.008)
0.045***
(0.012)
0.073***
(0.010)
-0.126
(0.106)
Parental incarceration
No ref. ref. ref. ref.
Yes 0.034*
(0.013)
0.034*
(0.013)
0.033*
(0.013)
0.035*
(0.013)
Don't know 0.037*
(0.017)
0.036*
(0.017)
0.038*
(0.017)
0.037*
(0.017)
Female 0.013
(0.007)
-0.002
(0.008)
0.013
(0.007)
0.014*
(0.007)
Age -0.003
(0.003)
-0.003
(0.003)
-0.003
(0.003)
-0.003
(0.003)
Race/ethnicity
White NH ref. ref. ref. ref.
Black NH -0.014
(0.014)
-0.012
(0.014)
-0.007
(0.015)
-0.014
(0.014)
Native American NH -0.001
(0.067)
-0.004
(0.066)
0.002
(0.069)
-0.004
(0.067)
Asian or Pacific Islander NH -0.015
(0.019)
-0.014
(0.019)
-0.015
(0.017)
-0.016
(0.019)
Other NH 0.023
(0.018)
0.023
(0.018)
-0.011
(0.019)
0.023
(0.018)
Hispanic 0.004
(0.015)
0.005
(0.015)
0.009
(0.015)
0.004
(0.015)
Parental education
< High school -0.004
(0.014)
-0.004
(0.014)
-0.004
(0.014)
-0.005
(0.014)
GED -0.039
(0.022)
-0.038
(0.022)
-0.039
(0.022)
-0.039
(0.022)
High school graduate ref. ref. ref. ref.
Vocational school + high school -0.008
(0.010)
-0.007
(0.010)
-0.007
(0.010)
-0.008
(0.010)
Some college -0.009
(0.011)
-0.009
(0.011)
-0.008
(0.011)
-0.008
(0.011)
College graduate -0.004
(0.010)
-0.004
(0.010)
-0.004
(0.010)
-0.004
(0.010)
161
Beyond 4-yr college -0.021
(0.012)
-0.021
(0.012)
-0.021
(0.012)
-0.020
(0.012)
Log household income -0.010
(0.007)
-0.009
(0.007)
-0.010
(0.007)
-0.016*
(0.007)
Childhood abuse * Female
0.051**
(0.017)
Childhood abuse * White NH
ref.
Childhood abuse * Black NH
-0.017
(0.020)
Childhood abuse * Native American
NH
-0.014
(0.110)
Childhood abuse * Asian or Pacific
Islander NH
0.001
(0.045)
Childhood abuse * Other NH
0.085
(0.048)
Childhood abuse * Hispanic
-0.012
(0.025)
Childhood abuse * Log household
income
0.018
(0.010)
NOTE.— This table displays the coefficients and standard errors from OLS models including school
fixed effects and a constant. Standard errors were clustered at the school level. Column (1) displays
the baseline results. Sample size was N = 14,741. * p<0.05, ** p<0.01, *** p<0.001
162
Appendix C: Chapter 5 appendix tables
Table C 1. Falsification test: parental incarceration and Wave I vocabulary
score
Outcome: vocabulary score
a
mean (s.d.): 101.3 (14.6)
OLS results
When parent was first incarcerated Coefficient (s.e.)
Before Wave I -0.6
(0.4)
After Wave I (but before age 18) 2.1
(2.2)
Don’t know if parent ever incarcerated 0.4
(0.7)
Never ref.
N 13,575
NOTE.— The table shows the coefficients and standard errors from an OLS regression corresponding to
equation (1), including the full control set. To reduce contamination in the control group resulting from
possibly imprecise recall of age when parent was first incarcerated, participants who reported that
parent was first incarcerated at an age within one-year of the Wave I interview were excluded from the
falsification sample.
Key: ref., reference group; s.d., standard deviation; s.e., standard error
a
Vocabulary score is the age-normalized score from the Add Health Picture Vocabulary Test.
* p<0.05, ** p<0.01, *** p<0.001
163
Table C 2. Heterogeneity of association with parental incarceration: results for high school diploma receipt
(1) (2) (3) (4) (5) (6) (7) (8)
Parent incarcerated
No
ref. ref. ref. ref.
Yes -0.100***
(0.017)
-0.106***
(0.025)
-0.132***
(0.025)
0.007
(0.206)
Don't know -0.064***
(0.017)
-0.100***
(0.029)
-0.079***
(0.018)
0.181
(0.250)
-0.057***
(0.017)
Childhood sexual abuse -0.079***
(0.013)
-0.080***
(0.013)
-0.079***
(0.013)
-0.078***
(0.013)
-0.086***
(0.013)
-0.078***
(0.014)
-0.078***
(0.014)
-0.080***
(0.014)
Physical abuse -0.003
(0.011)
-0.003
(0.011)
-0.003
(0.011)
-0.003
(0.011)
-0.005
(0.011)
-0.004
(0.011)
-0.003
(0.011)
-0.003
(0.011)
Emotional abuse -0.012
(0.014)
-0.012
(0.014)
-0.013
(0.014)
-0.012
(0.014)
-0.011
(0.014)
-0.012
(0.014)
-0.012
(0.014)
-0.012
(0.014)
Age -0.003
(0.004)
-0.004
(0.004)
-0.003
(0.004)
-0.003
(0.004)
-0.003
(0.004)
-0.003
(0.004)
-0.003
(0.004)
-0.003
(0.004)
Female 0.048***
(0.008)
0.041***
(0.009)
0.047***
(0.008)
0.048***
(0.008)
0.050***
(0.008)
0.048***
(0.008)
0.048***
(0.008)
0.042***
(0.009)
Race/ethnicity
White NH
ref. ref. ref. ref. ref. ref. ref. ref.
Black NH 0.034
(0.018)
0.034
(0.018)
0.010
(0.019)
0.034
(0.018)
0.028
(0.017)
0.034
(0.018)
0.036*
(0.018)
0.035
(0.018)
Native American NH -0.027
(0.099)
-0.026
(0.101)
-0.125
(0.110)
-0.026
(0.098)
-0.049
(0.088)
-0.027
(0.098)
-0.026
(0.099)
-0.026
(0.099)
Asian or Pacific Islander NH 0.043
(0.024)
0.043
(0.024)
0.042
(0.025)
0.044
(0.024)
0.046
(0.024)
0.043
(0.024)
0.043
(0.024)
0.043
(0.024)
Other NH -0.001
(0.029)
0.000
(0.029)
-0.025
(0.026)
-0.000
(0.029)
-0.008
(0.028)
-0.001
(0.029)
-0.001
(0.029)
-0.001
(0.029)
Hispanic -0.017
(0.018)
-0.016
(0.018)
-0.020
(0.017)
-0.016
(0.018)
-0.019
(0.018)
-0.017
(0.018)
-0.017
(0.018)
-0.017
(0.018)
Parental education
< High school -0.103***
(0.021)
-0.103***
(0.021)
-0.103***
(0.021)
-0.103***
(0.021)
-0.105***
(0.021)
-0.103***
(0.021)
-0.102***
(0.021)
-0.103***
(0.021)
164
GED -0.092**
(0.034)
-0.093**
(0.034)
-0.090**
(0.034)
-0.093**
(0.034)
-0.086*
(0.035)
-0.092**
(0.034)
-0.093**
(0.034)
-0.093**
(0.034)
High school graduate
ref. ref. ref. ref. ref. ref. ref. ref.
Vocational school + high school 0.037*
(0.015)
0.036*
(0.015)
0.038*
(0.015)
0.037*
(0.015)
0.039*
(0.016)
0.038*
(0.015)
0.037*
(0.015)
0.038*
(0.015)
Some college 0.045**
(0.014)
0.044**
(0.014)
0.044**
(0.013)
0.045**
(0.013)
0.046***
(0.013)
0.045**
(0.014)
0.045**
(0.013)
0.044**
(0.014)
College graduate 0.066***
(0.013)
0.066***
(0.013)
0.065***
(0.012)
0.066***
(0.012)
0.071***
(0.013)
0.067***
(0.013)
0.067***
(0.013)
0.066***
(0.013)
Beyond 4-yr college 0.094***
(0.015)
0.094***
(0.015)
0.092***
(0.014)
0.093***
(0.014)
0.097***
(0.015)
0.095***
(0.015)
0.094***
(0.015)
0.094***
(0.015)
Log family income 0.047***
(0.008)
0.047***
(0.008)
0.046***
(0.008)
0.051***
(0.008)
0.047***
(0.008)
0.047***
(0.008)
0.047***
(0.008)
0.047***
(0.008)
Parent incarcerated * Female
0.013
(0.031)
Parent incarcerated, don't know
* Female
0.076
(0.042)
Parent incarcerated * White NH
ref.
Parent incarcerated * Black NH
0.117**
(0.039)
Parent incarcerated * Native
American NH
0.238*
(0.120)
Parent incarcerated * Asian or
Pacific Islander NH
-0.045
(0.111)
Parent incarcerated * Other NH
0.194**
(0.069)
Parent incarcerated * Hispanic
-0.023
(0.042)
Parent incarcerated, don't know
* White NH
ref.
Parent incarcerated, don't know
* Black NH
0.026
(0.039)
Parent incarcerated, don't know
* Native American NH
0.359**
(0.114)
Parent incarcerated, don't know
* Asian or Pacific Islander NH
-0.071
(0.063)
165
Parent incarcerated, don't know
* Other NH
-0.035
(0.106)
Parent incarcerated, don't know
* Hispanic
0.099
(0.062)
Parent incarcerated * Log
family income
-0.010
(0.019)
Parent incarcerated, don't know
* Log family income
-0.023
(0.023)
Years of childhood exposure to
parental incarceration
-0.009***
(0.002)
Father incarcerated
No
ref.
ref.
Yes
-0.082***
(0.018)
-0.095***
(0.025)
Don't know
-0.065***
(0.018)
-0.096**
(0.032)
Mother incarcerated
No
ref.
ref.
Yes
-0.099**
(0.034)
-0.060
(0.059)
Don't know
-0.024
(0.046)
-0.039
(0.075)
Parent incarcerated
No
ref.
Don't know
-0.064***
(0.017)
Mother only
-0.168***
(0.038)
Father only
-0.090***
(0.018)
Both parents
-0.099
(0.061)
Father incarcerated * Female
0.026
(0.032)
166
Father incarcerated, don't know
* Female
0.064
(0.043)
Mother incarcerated * Female
-0.074
(0.083)
Mother incarcerated, don't
know * Female
0.037
(0.091)
NOTE.— This table displays results from OLS models of high school diploma receipt. Each regression includes dummy variables for school
fixed effects. Sample size was N=14,741 for each regression here except column (5) including years of exposure to parental incarceration,
for which N=14,633. Column (1) displays results from the baseline regression. * p<0.05, ** p<0.01, *** p<0.001
Table C 3. Heterogeneity of association with parental incarceration: results for college degree attainment
(1) (2) (3) (4) (5) (6) (7) (8)
Parent incarcerated
No
ref. ref. ref. ref.
Yes -0.078***
(0.013)
-0.045*
(0.02)
-0.119***
(0.017)
0.737***
(0.161)
Don't know -0.055**
(0.017)
-0.035
(0.026)
-0.086***
(0.020)
1.029***
(0.198)
-0.045**
(0.017)
Childhood sexual abuse -0.085***
(0.012)
-0.083***
(0.012)
-0.084***
(0.012)
-0.084***
(0.012)
-0.090***
(0.012)
-0.087***
(0.012)
-0.085***
(0.012)
-0.085***
(0.012)
Physical abuse -0.031*
(0.013)
-0.031*
(0.013)
-0.03*
(0.013)
-0.03*
(0.013)
-0.035*
(0.013)
-0.03*
(0.013)
-0.031*
(0.013)
-0.031*
(0.013)
Emotional abuse -0.008
(0.015)
-0.008
(0.015)
-0.008
(0.015)
-0.007
(0.015)
-0.009
(0.015)
-0.009
(0.015)
-0.009
(0.015)
-0.008
(0.015)
Age -0.011**
(0.004)
-0.011**
(0.004)
-0.011**
(0.004)
-0.011**
(0.004)
-0.011**
(0.004)
-0.011**
(0.004)
-0.011**
(0.004)
-0.011**
(0.004)
Female 0.084***
(0.011)
0.095***
(0.012)
0.084***
(0.011)
0.084***
(0.011)
0.086***
(0.011)
0.084***
(0.011)
0.084***
(0.011)
0.094***
(0.012)
Race/ethnicity
White NH
ref. ref. ref. ref. ref. ref. ref. ref.
Black NH -0.003
(0.020)
-0.003
(0.020)
-0.033
(0.023)
-0.002
(0.020)
-0.010
(0.020)
-0.005
(0.020)
-0.004
(0.020)
-0.005
(0.020)
167
Native American NH -0.048
(0.052)
-0.048
(0.052)
-0.068
(0.065)
-0.044
(0.055)
-0.051
(0.055)
-0.045
(0.052)
-0.047
(0.051)
-0.045
(0.052)
Asian or Pacific Islander NH 0.104**
(0.037)
0.104**
(0.037)
0.103**
(0.038)
0.109**
(0.037)
0.106**
(0.037)
0.105**
(0.037)
0.104**
(0.037)
0.105**
(0.037)
Other NH -0.001
(0.026)
-0.002
(0.026)
-0.017
(0.031)
0.003
(0.026)
-0.004
(0.026)
-0.002
(0.026)
-0.001
(0.026)
-0.002
(0.026)
Hispanic -0.036
(0.022)
-0.036
(0.022)
-0.059*
(0.024)
-0.032
(0.022)
-0.038
(0.022)
-0.036
(0.022)
-0.036
(0.022)
-0.036
(0.022)
Parental education
< High school -0.005
(0.015)
-0.006
(0.015)
-0.004
(0.015)
-0.005
(0.015)
-0.005
(0.015)
-0.006
(0.015)
-0.006
(0.015)
-0.006
(0.015)
GED -0.044*
(0.022)
-0.044*
(0.022)
-0.041
(0.022)
0.045*
(0.022)
-0.046*
(0.022)
-0.043*
(0.021)
-0.045*
(0.022)
-0.043*
(0.021)
High school graduate
ref. ref. ref. ref. ref. ref. ref. ref.
Vocational school + high school 0.022
(0.017)
0.022
(0.017)
0.023
(0.017)
0.024
(0.017)
0.024
(0.017)
0.023
(0.017)
0.022
(0.017)
0.023
(0.017)
Some college 0.060***
(0.014)
0.060***
(0.014)
0.060***
(0.014)
0.059***
(0.014)
0.062***
(0.014)
0.060***
(0.014)
0.059***
(0.014)
0.061***
(0.014)
College graduate 0.196***
(0.018)
0.196***
(0.018)
0.195***
(0.018)
0.193***
(0.018)
0.199***
(0.018)
0.196***
(0.018)
0.196***
(0.018)
0.196***
(0.018)
Beyond 4-yr college 0.355***
(0.020)
0.355***
(0.020)
0.353***
(0.020)
0.346***
(0.020)
0.359***
(0.020)
0.357***
(0.020)
0.355***
(0.020)
0.357***
(0.020)
Log family income 0.073***
(0.008)
0.073***
(0.008)
0.073***
(0.008)
0.095***
(0.009)
0.076***
(0.008)
0.074***
(0.008)
0.073***
(0.008)
0.074***
(0.008)
Parent incarcerated * Female
-0.065*
(0.025)
Parent incarcerated, don't know
* Female
-0.043
(0.032)
Parent incarcerated * White NH
ref.
Parent incarcerated * Black NH
0.098**
(0.033)
Parent incarcerated * Native
American NH
0.082
(0.066)
Parent incarcerated * Asian or
Pacific Islander NH
-0.054
(0.132)
168
Parent incarcerated * Other NH
0.108
(0.062)
Parent incarcerated * Hispanic
0.117***
(0.033)
Parent incarcerated, don't know
* White NH
ref.
Parent incarcerated, don't know
* Black NH
0.132**
(0.046)
Parent incarcerated, don't know
* Native American NH
0.105
(0.122)
Parent incarcerated, don't know
* Asian or Pacific Islander NH
-0.144*
(0.072)
Parent incarcerated, don't know
* Other NH
0.025
(0.072)
Parent incarcerated, don't know
* Hispanic
0.072
(0.047)
Parent incarcerated * Log
family income
-0.078***
(0.016)
Parent incarcerated, don't know
* Log family income
-0.102***
(0.019)
Years of childhood exposure to
parental incarceration
-0.003*
(0.001)
Father incarcerated
No
ref.
ref.
Yes
-0.075***
(0.013)
-0.043*
(0.019)
Don't know
-0.042*
(0.017)
-0.024
(0.026)
Mother incarcerated
No
ref.
ref.
Yes
-0.005
(0.029)
0.014
(0.049)
Don't know
-0.108***
(0.030)
-0.107*
(0.048)
Parent incarcerated
169
No
ref.
Don't know
-0.054**
(0.017)
Mother only
-0.074*
(0.037)
Father only
-0.085***
(0.014)
Both parents
0.014
(0.043)
Father incarcerated * Female
-0.064*
(0.025)
Father incarcerated, don't know
* Female
-0.038
(0.033)
Mother incarcerated * Female
-0.035
(0.061)
Mother incarcerated, don't
know * Female
-0.004
(0.069)
NOTE.— This table displays results from OLS models of college degree attainment. Each regression includes dummy variables for school
fixed effects. Sample size was N=14,741 for each regression here except column (5) including years of exposure to parental incarceration,
for which N=14,633. Column (1) displays results from the baseline regression. * p<0.05, ** p<0.01, *** p<0.001
Table C 4. Heterogeneity of association with parental incarceration: results for any employment
(1) (2) (3) (4) (5) (6) (7) (8)
Parent incarcerated
No
ref. ref. ref. ref.
Yes -0.030
(0.016)
-0.038
(0.020)
-0.05*
(0.022)
-0.132
(0.190)
Don't know -0.052*
(0.021)
-0.045
(0.028)
-0.081**
(0.030)
0.267
(0.293)
-0.049*
(0.021)
Childhood sexual abuse -0.04*
(0.015)
-0.04*
(0.015)
-0.04*
(0.015)
-0.039*
(0.015)
-0.042**
(0.015)
-0.040**
(0.015)
-0.04*
(0.015)
-0.040**
(0.015)
Physical abuse -0.010
(0.011)
-0.010
(0.011)
-0.010
(0.011)
-0.010
(0.011)
-0.010
(0.011)
-0.010
(0.011)
-0.010
(0.011)
-0.010
(0.011)
170
Emotional abuse -0.031*
(0.015)
-0.031*
(0.015)
-0.032*
(0.015)
-0.03*
(0.015)
-0.033*
(0.016)
-0.031*
(0.015)
-0.032*
(0.015)
-0.031*
(0.015)
Age -0.004
(0.003)
-0.004
(0.003)
-0.004
(0.003)
-0.004
(0.003)
-0.004
(0.003)
-0.004
(0.003)
-0.004
(0.003)
-0.004
(0.003)
Female -0.090***
(0.010)
-0.090***
(0.010)
-0.090***
(0.010)
-0.090***
(0.010)
-0.088***
(0.010)
-0.090***
(0.010)
-0.090***
(0.010)
-0.089***
(0.010)
Race/ethnicity
White NH
ref. ref. ref. ref. ref. ref. ref. ref.
Black NH -0.017
(0.022)
-0.018
(0.022)
-0.031
(0.022)
-0.018
(0.022)
-0.018
(0.022)
-0.018
(0.022)
-0.017
(0.022)
-0.019
(0.022)
Native American NH -0.075
(0.108)
-0.075
(0.108)
-0.167
(0.126)
-0.075
(0.107)
-0.089
(0.105)
-0.073
(0.108)
-0.074
(0.108)
-0.070
(0.108)
Asian or Pacific Islander NH -0.013
(0.033)
-0.014
(0.033)
-0.021
(0.035)
-0.013
(0.032)
-0.011
(0.033)
-0.013
(0.033)
-0.013
(0.033)
-0.013
(0.033)
Other NH 0.003
(0.023)
0.003
(0.023)
-0.026
(0.027)
0.003
(0.023)
-0.000
(0.024)
0.003
(0.023)
0.003
(0.023)
0.003
(0.023)
Hispanic 0.057**
(0.020)
0.057**
(0.020)
0.049*
(0.023)
0.057**
(0.020)
0.059**
(0.020)
0.057**
(0.020)
0.057**
(0.020)
0.057**
(0.020)
Parental education
< High school -0.068**
(0.023)
-0.068**
(0.023)
-0.066**
(0.023)
-0.068**
(0.023)
-0.072**
(0.023)
-0.068**
(0.023)
-0.068**
(0.023)
-0.068**
(0.023)
GED -0.015
(0.030)
-0.014
(0.030)
-0.012
(0.030)
-0.016
(0.030)
-0.017
(0.031)
-0.014
(0.030)
-0.015
(0.030)
-0.012
(0.030)
High school graduate
ref. ref. ref. ref. ref. ref. ref. ref.
Vocational school + high school 0.012
(0.019)
0.012
(0.019)
0.013
(0.019)
0.012
(0.019)
0.012
(0.019)
0.012
(0.019)
0.012
(0.019)
0.013
(0.019)
Some college 0.004
(0.013)
0.004
(0.013)
0.003
(0.013)
0.004
(0.013)
0.004
(0.013)
0.004
(0.013)
0.003
(0.013)
0.004
(0.013)
College graduate 0.012
(0.015)
0.012
(0.015)
0.011
(0.015)
0.012
(0.015)
0.013
(0.015)
0.012
(0.015)
0.012
(0.015)
0.012
(0.015)
Beyond 4-yr college -0.003
(0.017)
-0.003
(0.017)
-0.005
(0.017)
-0.003
(0.017)
-0.003
(0.017)
-0.003
(0.017)
-0.004
(0.017)
-0.003
(0.017)
Log family income 0.016
(0.010)
0.016
(0.010)
0.016
(0.010)
0.017
(0.009)
0.016
(0.010)
0.016
(0.010)
0.016
(0.010)
0.016
(0.010)
171
Parent incarcerated * Female
0.016
(0.032)
Parent incarcerated, don't know
* Female
-0.016
(0.040)
Parent incarcerated * White NH
ref.
Parent incarcerated * Black NH
0.042
(0.033)
Parent incarcerated * Native
American NH
0.253
(0.170)
Parent incarcerated * Asian or
Pacific Islander NH
0.255***
(0.058)
Parent incarcerated * Other NH
0.133
(0.068)
Parent incarcerated * Hispanic
0.021
(0.046)
Parent incarcerated, don't know
* White NH
ref.
Parent incarcerated, don't know
* Black NH
0.072
(0.048)
Parent incarcerated, don't know
* Native American NH
0.321*
(0.132)
Parent incarcerated, don't know
* Asian or Pacific Islander NH
-0.023
(0.153)
Parent incarcerated, don't know
* Other NH
0.099
(0.087)
Parent incarcerated, don't know
* Hispanic
0.074
(0.054)
Parent incarcerated * Log
family income
0.010
(0.018)
Parent incarcerated, don't know
* Log family income
-0.030
(0.028)
Years of childhood exposure to
parental incarceration
-0.003
(0.002)
Father incarcerated
No
ref.
ref.
172
Yes
-0.031
(0.017)
-0.040
(0.022)
Don't know
-0.057**
(0.020)
-0.001458
Mother incarcerated
No
ref.
ref.
Yes
0.023
(0.039)
0.035
(0.055)
Don't know
-0.053
(0.040)
0.017
(0.043)
Parent incarcerated
No
ref.
Don't know
-0.052*
(0.021)
Mother only
-0.029
(0.042)
Father only
-0.036*
(0.017)
Both parents
0.053
(0.055)
Father incarcerated * Female
0.017
(0.035)
Father incarcerated, don't know
* Female
-0.003
(0.041)
Mother incarcerated * Female
-0.021
(0.072)
Mother incarcerated, don't
know * Female
-0.159
(0.086)
NOTE.— This table displays results from OLS models of any employment. Each regression includes dummy variables for school fixed
effects. Sample size was N=14,741 for each regression here except column (5) including years of exposure to parental incarceration, for
which N=14,633. Column (1) displays results from the baseline regression. * p<0.05, ** p<0.01, *** p<0.001
173
Table C 5. Heterogeneity of association with parental incarceration: results for full-time employment
(1) (2) (3) (4) (5) (6) (7) (8)
Parent incarcerated
No ref. ref. ref. ref.
Yes -0.043*
(0.017)
-0.055*
(0.022)
-0.073**
(0.023)
-0.050
(0.220)
Don't know -0.048*
(0.022)
-0.027
(0.028)
-0.072*
(0.03)
0.299
(0.252)
-0.045*
(0.022)
Childhood sexual abuse -0.054***
(0.015)
-0.054***
(0.015)
-0.054***
(0.015)
-0.053***
(0.015)
-0.052***
(0.015)
-0.054***
(0.015)
-0.054***
(0.015)
-0.054***
(0.015)
Physical abuse 0.001
(0.012)
0.002
(0.012)
0.001
(0.012)
0.001
(0.012)
0.001
(0.013)
0.001
(0.012)
0.001
(0.012)
0.001
(0.012)
Emotional abuse -0.038*
(0.018)
-0.038*
(0.018)
-0.038*
(0.018)
-0.037*
(0.018)
-0.04*
(0.018)
-0.038*
(0.018)
-0.039*
(0.018)
-0.038*
(0.018)
Age -0.002
(0.004)
-0.002
(0.004)
-0.002
(0.004)
-0.002
(0.004)
-0.003
(0.004)
-0.002
(0.004)
-0.002
(0.004)
-0.002
(0.004)
Female -0.150***
(0.011)
-0.149***
(0.012)
-0.149***
(0.011)
-0.149***
(0.011)
-0.148***
(0.012)
-0.150***
(0.011)
-0.149***
(0.011)
-0.148***
(0.012)
Race/ethnicity
White NH ref. ref. ref. ref. ref. ref. ref. ref.
Black NH -0.041
(0.022)
-0.042
(0.022)
-0.053*
(0.024)
-0.042
(0.022)
-0.045*
(0.022)
-0.043
(0.022)
-0.041
(0.022)
-0.043
(0.022)
Native American NH -0.061
(0.094)
-0.061
(0.093)
-0.112
(0.125)
-0.061
(0.093)
-0.071
(0.098)
-0.060
(0.094)
-0.060
(0.094)
-0.059
(0.094)
Asian or Pacific Islander NH -0.039
(0.035)
-0.039
(0.035)
-0.048
(0.037)
-0.038
(0.035)
-0.036
(0.035)
-0.039
(0.035)
-0.039
(0.035)
-0.039
(0.035)
Other NH -0.018
(0.031)
-0.019
(0.031)
-0.036
(0.034)
-0.018
(0.031)
-0.016
(0.030)
-0.018
(0.031)
-0.018
(0.031)
-0.018
(0.031)
Hispanic 0.052*
(0.023)
0.052*
(0.023)
0.028
(0.025)
0.053*
(0.023)
0.053*
(0.023)
0.052*
(0.023)
0.052*
(0.023)
0.052*
(0.023)
Parental education
< High school -0.061**
(0.020)
-0.061**
(0.020)
-0.059**
(0.020)
-0.061**
(0.020)
-0.064**
(0.021)
-0.061**
(0.020)
-0.061**
(0.020)
-0.061**
(0.020)
174
GED -0.006
(0.031)
-0.006
(0.031)
-0.004
(0.031)
-0.007
(0.031)
-0.013
(0.031)
-0.005
(0.031)
-0.008
(0.031)
-0.004
(0.031)
High school graduate ref. ref. ref. ref. ref. ref. ref. ref.
Vocational school + high school 0.009
(0.018)
0.010
(0.018)
0.009
(0.018)
0.010
(0.018)
0.008
(0.018)
0.010
(0.018)
0.009
(0.018)
0.010
(0.018)
Some college -0.006
(0.014)
-0.005
(0.014)
-0.006
(0.014)
-0.005
(0.014)
-0.006
(0.014)
-0.005
(0.014)
-0.006
(0.014)
-0.005
(0.014)
College graduate -0.004
(0.016)
-0.004
(0.016)
-0.005
(0.016)
-0.005
(0.016)
-0.001
(0.016)
-0.004
(0.016)
-0.004
(0.016)
-0.004
(0.016)
Beyond 4-yr college -0.016
(0.019)
-0.015
(0.019)
-0.018
(0.019)
-0.017
(0.019)
-0.016
(0.019)
-0.015
(0.019)
-0.016
(0.019)
-0.015
(0.019)
Log family income 0.019*
(0.009)
0.019*
(0.010)
0.018
(0.009)
0.022*
(0.010)
0.020*
(0.009)
0.019
(0.010)
0.019*
(0.010)
0.019
(0.010)
Parent incarcerated * Female 0.024
(0.036)
Parent incarcerated, don't
know * Female
-0.044
(0.037)
Parent incarcerated * White
NH
ref.
Parent incarcerated * Black NH 0.049
(0.045)
Parent incarcerated * Native
American NH
0.005
(0.147)
Parent incarcerated * Asian or
Pacific Islander NH
0.196
(0.126)
Parent incarcerated * Other
NH
0.076
(0.080)
Parent incarcerated * Hispanic 0.113**
(0.042)
Parent incarcerated, don't
know * White NH
ref.
Parent incarcerated, don't
know * Black NH
0.039
(0.045)
Parent incarcerated, don't
know * Native American NH
0.354**
(0.131)
175
Parent incarcerated, don't
know * Asian or Pacific
Islander NH
-0.001
(0.165)
Parent incarcerated, don't
know * Other NH
0.077
(0.098)
Parent incarcerated, don't
know * Hispanic
0.082
(0.072)
Parent incarcerated * Log
family income
0.001
(0.021)
Parent incarcerated, don't
know * Log family income
-0.033
(0.024)
Years of childhood exposure to
parental incarceration
-0.003
(0.002)
Father incarcerated
No
ref.
ref.
Yes -0.04*
(0.018)
-0.042
(0.025)
Don't know -0.058**
(0.021)
-0.044
(0.027)
Mother incarcerated
No
ref.
ref.
Yes 0.005
(0.040)
-0.025
(0.066)
Don't know -0.033
(0.040)
0.006
(0.053)
Parent incarcerated
No
ref.
Don't know -0.047*
(0.022)
Mother only -0.060
(0.040)
Father only -0.047*
(0.019)
Both parents 0.048
(0.062)
176
Father incarcerated * Female 0.004
(0.039)
Father incarcerated, don't
know * Female
-0.028
(0.040)
Mother incarcerated * Female 0.057
(0.077)
Mother incarcerated, don't
know * Female
-0.090
(0.087)
NOTE.— This table displays results from OLS models of any employment. Each regression includes dummy variables for school fixed
effects. Sample size was N=14,741 for each regression here except column (5) including years of exposure to parental incarceration, for
which N=14,633. Column (1) displays results from the baseline regression. * p<0.05, ** p<0.01, *** p<0.001
Table C 6. Heterogeneity of association with parental incarceration: results for earnings
(1) (2) (3) (4) (5) (6) (7) (8)
1st part: probit for having
positive earnings
Parent incarcerated
No
ref. ref. ref. ref. ref.
Yes -0.212**
(0.066)
-0.331**
(0.122)
-0.289**
(0.087)
-0.555
(0.847)
Don't know -0.328**
(0.104)
-0.567***
(0.158)
-0.407***
(0.112)
0.116
(1.507)
-0.304**
(0.106)
Childhood sexual abuse -0.120
(0.070)
-0.132
(0.071)
-0.125
(0.070)
-0.120
(0.070)
-0.141*
(0.07)
-0.120
(0.070)
-0.120
(0.070)
-0.129
(0.070)
Physical abuse 0.063
(0.066)
0.061
(0.066)
0.066
(0.066)
0.063
(0.066)
0.054
(0.066)
0.063
(0.066)
0.062
(0.066)
0.058
(0.066)
Emotional abuse -0.076
(0.080)
-0.077
(0.080)
-0.082
(0.082)
-0.076
(0.080)
-0.070
(0.080)
-0.076
(0.080)
-0.077
(0.080)
-0.078
(0.080)
Age -0.030
(0.019)
-0.031
(0.019)
-0.031
(0.019)
-0.030
(0.019)
-0.026
(0.020)
-0.029
(0.019)
-0.030
(0.019)
-0.030
(0.019)
Female -0.510***
(0.063)
-0.585***
(0.068)
-0.517***
(0.062)
-0.510***
(0.063)
-0.503***
(0.062)
-0.510***
(0.064)
-0.510***
(0.063)
-0.574***
(0.066)
Race/ethnicity
177
White NH
ref. ref. ref. ref. ref. ref. ref. ref.
Black NH -0.069
(0.107)
-0.064
(0.106)
-0.075
(0.102)
-0.070
(0.107)
-0.084
(0.106)
-0.072
(0.107)
-0.069
(0.107)
-0.067
(0.107)
Native American NH -0.215
(0.294)
-0.217
(0.285)
-0.187
(0.428)
-0.214
(0.293)
-0.258
(0.302)
-0.208
(0.296)
-0.216
(0.294)
-0.216
(0.288)
Asian or Pacific Islander
NH
-0.211
(0.179)
-0.212
(0.180)
-0.248
(0.193)
-0.211
(0.177)
-0.201
(0.179)
-0.209
(0.180)
-0.211
(0.179)
-0.209
(0.181)
Other NH 0.133
(0.127)
0.138
(0.128)
-0.093
(0.133)
0.133
(0.127)
0.114
(0.124)
0.132
(0.126)
0.133
(0.127)
0.138
(0.128)
Hispanic 0.048
(0.105)
0.049
(0.106)
-0.034
(0.102)
0.048
(0.105)
0.042
(0.108)
0.047
(0.105)
0.048
(0.104)
0.051
(0.105)
Parental education
< High school -0.246**
(0.094)
-0.248**
(0.093)
-0.248**
(0.094)
-0.246**
(0.093)
-0.243*
(0.095) -0.02337
-0.246**
(0.094)
-0.251**
(0.094)
GED -0.014
(0.141)
-0.019
(0.137)
-0.008
(0.139)
-0.018
(0.141)
-0.018
(0.145)
-0.012
(0.141)
-0.014
(0.141)
-0.019
(0.138)
High school graduate
ref. ref. 0.000 (.) ref. ref. ref. ref. ref.
Vocational school + high
school
0.087
(0.084)
0.085
(0.084)
0.084
(0.085)
0.087
(0.084)
0.089
(0.084)
0.091
(0.084)
0.087
(0.084)
0.086
(0.084)
Some college 0.115
(0.071)
0.109
(0.070)
0.112
(0.070)
0.116
(0.071)
0.108
(0.071)
0.115
(0.071)
0.115
(0.071)
0.109
(0.071)
College graduate 0.072
(0.100)
0.071
(0.101)
0.067
(0.100)
0.073
(0.100)
0.078
(0.100)
0.073
(0.100)
0.072
(0.100)
0.072
(0.101)
Beyond 4-yr college 0.043
(0.082)
0.040
(0.082)
0.034
(0.082)
0.044
(0.083)
0.044
(0.082)
0.047
(0.083)
0.043
(0.083)
0.043
(0.083)
Log family income 0.025
(0.040)
0.025
(0.040)
0.023
(0.039)
0.024
(0.043)
0.035
(0.040)
0.026
(0.040)
0.026
(0.040)
0.026
(0.040)
Parent incarcerated *
Female
0.189
(0.159)
Parent incarcerated, don't
know * Female
0.415*
(0.205)
Parental incarceration #
White NH
ref.
Parental incarceration #
Black NH
0.196
(0.143)
178
Parental incarceration #
Native American NH
-0.534
(0.604)
Parental incarceration #
Asian or Pacific Islander
NH
0.057
(0.627)
Parental incarceration #
Other NH
1.391**
(0.428)
Parental incarceration #
Hispanic
0.033
(0.206)
Parent incarcerated, don't
know * White NH
ref.
Parent incarcerated, don't
know * Black NH
-0.180
(0.286)
Parent incarcerated, don't
know * Native American
NH
0.000 (.)
Parent incarcerated, don't
know * Asian or Pacific
Islander NH
0.266
(0.540)
Parent incarcerated, don't
know * Other NH
1.015**
(0.365)
Parent incarcerated, don't
know * Hispanic
0.716*
(0.323)
Parent incarcerated * Log
family income
0.033
(0.081)
Parent incarcerated, don't
know * Log family income
-0.043
(0.139)
Years of childhood
exposure to parental
incarceration
-0.015*
(0.006)
Father incarcerated
No
ref.
ref.
Yes
-0.203**
(0.077)
-0.251
(0.143)
Don't know
-0.308**
(0.109)
-0.518**
(0.177)
179
Mother incarcerated
No
ref.
ref.
Yes
-0.052
(0.155)
-0.284
(0.285)
Don't know
-0.201
(0.166)
-0.256
(0.272)
Parent incarcerated
No
ref.
Don't know
-0.328**
(0.104)
Mother only
-0.211
(0.182)
Father only
-0.214**
(0.074)
Both parents
-0.178
(0.222)
Father incarcerated *
Female
0.079
(0.170)
Father incarcerated, don't
know * Female
0.360
(0.219)
Mother incarcerated *
Female
0.366
(0.322)
Mother incarcerated,
don't know * Female
0.092
(0.404)
2nd part: GLM for
earnings level
Parent incarcerated
No
ref. ref. ref. ref. ref.
Yes -0.08*
(0.032)
-0.080
(0.043)
-0.106*
(0.043)
0.228
(0.439)
Don't know -0.051
(0.047)
-0.106*
(0.053)
-0.079
(0.062)
0.609
(0.608)
-0.044
(0.047)
Childhood sexual abuse -0.118**
(0.035)
-0.119***
(0.035)
-0.118**
(0.036)
-0.116**
(0.035)
-0.115**
(0.035)
-0.117**
(0.035)
-0.117**
(0.035)
-0.117**
(0.035)
180
Physical abuse 0.043
(0.032)
0.042
(0.032)
0.044
(0.032)
0.044
(0.032)
0.039
(0.033)
0.042
(0.032)
0.043
(0.032)
0.040
(0.031)
Emotional abuse -0.011
(0.035)
-0.010
(0.035)
-0.009
(0.035)
-0.009
(0.035)
-0.024
(0.035)
-0.010
(0.035)
-0.011
(0.035)
-0.008
(0.034)
Age 0.035***
(0.009)
0.035***
(0.009)
0.036***
(0.008)
0.035***
(0.009)
0.036***
(0.009)
0.035***
(0.008)
0.036***
(0.009)
0.036***
(0.008)
Female -0.275***
(0.022)
-0.283***
(0.025)
-0.274***
(0.022)
-0.274***
(0.022)
-0.275***
(0.022)
-0.275***
(0.022)
-0.275***
(0.022)
-0.282***
(0.024)
Race/ethnicity
White NH
ref. ref. ref. ref. ref. ref. ref. ref.
Black NH -0.119**
(0.040)
-0.118**
(0.041)
-0.144***
(0.041)
-0.119**
(0.041)
-0.124**
(0.042)
-0.119**
(0.041)
-0.117**
(0.041)
-0.117**
(0.041)
Native American NH -0.370**
(0.128)
-0.362**
(0.131)
-0.450**
(0.156)
-0.372**
(0.125)
-0.354*
(0.141)
-0.368**
(0.129)
-0.366**
(0.129)
-0.367**
(0.129)
Asian or Pacific Islander
NH
0.126
(0.080)
0.127
(0.080)
0.106
(0.083)
0.130
(0.080)
0.133
(0.080)
0.126
(0.080)
0.125
(0.080)
0.125
(0.080)
Other NH -0.075
(0.047)
-0.073
(0.047)
-0.076
(0.055)
-0.073
(0.047)
-0.069
(0.046)
-0.075
(0.047)
-0.074
(0.047)
-0.074
(0.047)
Hispanic -0.005
(0.048)
-0.005
(0.048)
-0.011
(0.048)
-0.003
(0.048)
-0.004
(0.049)
-0.005
(0.049)
-0.007
(0.048)
-0.005
(0.048)
Parental education
< High school -0.076*
(0.038)
-0.077*
(0.038)
-0.073
(0.038)
-0.075
(0.039)
-0.087*
(0.039)
-0.077*
(0.039)
-0.075
(0.038)
-0.078*
(0.039)
GED -0.052
(0.058)
-0.054
(0.058)
-0.047
(0.058)
-0.053
(0.058)
-0.058
(0.059)
-0.052
(0.057)
-0.054
(0.057)
-0.056
(0.058)
High school graduate
ref. ref. ref. ref. ref. ref. ref. ref.
Vocational school + high
school
0.031
(0.041)
0.031
(0.041)
0.032
(0.040)
0.032
(0.041)
0.033
(0.041)
0.031
(0.041)
0.030
(0.041)
0.032
(0.041)
Some college 0.079*
(0.034)
0.077*
(0.034)
0.080*
(0.034)
0.079*
(0.034)
0.081*
(0.035)
0.079*
(0.034)
0.078*
(0.035)
0.077*
(0.034)
College graduate 0.121***
(0.035)
0.120***
(0.035)
0.123***
(0.035)
0.120***
(0.035)
0.123***
(0.035)
0.121***
(0.035)
0.121***
(0.035)
0.120***
(0.035)
Beyond 4-yr college 0.111**
(0.038)
0.111**
(0.039)
0.110**
(0.038)
0.107**
(0.038)
0.113**
(0.039)
0.111**
(0.038)
0.111**
(0.039)
0.111**
(0.038)
181
Log family income 0.122***
(0.019)
0.122***
(0.019)
0.123***
(0.019)
0.133***
(0.021)
0.124***
(0.019)
0.122***
(0.019)
0.123***
(0.019)
0.121***
(0.019)
Parent incarcerated *
Female
0.000
(0.059)
Parent incarcerated, don't
know * Female
0.113
(0.093)
Parental incarceration #
White NH
ref.
Parental incarceration #
Black NH
0.042
(0.068)
Parental incarceration #
Native American NH
-0.387
(0.218)
Parental incarceration #
Asian or Pacific Islander
NH
0.475**
(0.174)
Parental incarceration #
Other NH
0.072
(0.126)
Parental incarceration #
Hispanic
0.093
(0.071)
Parent incarcerated, don't
know * White NH
ref.
Parent incarcerated, don't
know * Black NH
0.172
(0.103)
Parent incarcerated, don't
know * Native American
NH
0.581**
(0.179)
Parent incarcerated, don't
know * Asian or Pacific
Islander NH
0.204
(0.238)
Parent incarcerated, don't
know * Other NH
-0.075
(0.154)
Parent incarcerated, don't
know * Hispanic
-0.067
(0.175)
Parent incarcerated * Log
family income
-0.029
(0.042)
182
Parent incarcerated, don't
know * Log family income
-0.062
(0.056)
Years of childhood
exposure to parental
incarceration
-0.005
(0.003)
Father incarcerated
No
ref.
ref.
Yes
-0.065*
(0.032)
-0.072
(0.046)
Don't know
-0.078
(0.043)
-0.092
(0.059)
Mother incarcerated
No
ref.
ref.
Yes
-0.071
(0.062)
-0.034
(0.096)
Don't know
0.012
(0.171)
-0.166
(0.099)
Parent incarcerated
No
ref.
Don't know
-0.051
(0.047)
Mother only
-0.158*
(0.072)
Father only
-0.073*
(0.033)
Both parents
-0.030
(0.105)
Father incarcerated *
Female
0.016
(0.067)
Father incarcerated, don't
know * Female
0.029
(0.083)
Mother incarcerated *
Female
-0.071
(0.121)
Mother incarcerated,
don't know * Female
0.396
(0.348)
183
NOTE.— This table displays coefficients and standard errors from a two-part model of earnings. Each regression included dummy
variables for school fixed effects and a constant. Sample size was N=14,672 for each regression here except column (5) including years
of exposure to parental incarceration, for which N=14,517. Column (1) displays results from the baseline regression. * p<0.05, ** p<0.01,
*** p<0.001
184
Appendix D: Multiple imputation
In all regression analyses, I addressed missing data by implementing multiple
imputation. I report pooled regression results from the multiply imputed data
sets, adjusting standard errors to account for the imputation procedure.
The multiple imputation procedure consists of three steps: (i) impute
missing values m times to create m data sets, (ii) execute analysis models on
each of the m data sets, (iii) use Rubin’s rules to combine the estimates
produced from each of the m data sets in the prior step (Rubin 1987, Rubin
1996). I imputed 25 sets of values for childhood household income.
1
The
imputation procedure uses the distribution of observed data to impute
multiple values. Each imputed value included a random component, reflecting
the uncertainty around the true value (Rubin 1987, Schomaker and Heumann
2018).
I implemented the imputation model with chained equations. Each of
these equations included survey weights. In the imputation model for
childhood family income, I included the covariates from the analysis models
(control set described above) as well as auxiliary variables correlated with
the missing variables. I included variables from the analytic model to
preserve the relationships between the variables of interest (Nguyen, et al.
2017, Rubin 1987, White, et al. 2011). As auxiliary variables in the imputation
model, I included variables for full-time work status of the resident father
and median household income of the Census block group in which the child
lived during Wave I. I considered the following sociodemographic factors as
1
White et al. (2011) suggest a rule of thumb that the number of imputations should be at
least as large as the largest fraction of missing data * 100, across variables. Here, the largest
fraction of missing data is about 0.24, for log-household income.
185
auxiliary variables as well but excluded from the imputation model due to
low correlation with the variables for which values were to be imputed
(correlation < 0.3): full-time work status of resident mother, parental
respondent age, parental respondent race, and parental respondent US-
nativity.
Variance of parameters estimated through multiple imputation is the
sum of three components: variance within, variance between, and an
additional source of sampling variance. Variance within is the arithmetic
mean across sampling variances for an estimate across each of the imputed
data sets – i.e., the sampling variability expected had there been no missing
data. Variance between is the variance of parameter estimates across each
imputed data set, i.e., it captures additional uncertainty that arises from
missing data. Lastly, the additional sampling variance is the variance
between divided by the number of imputations, representing sampling error
of the average coefficient estimates, which is larger for smaller number of
imputations. Thus, the standard error for each parameter estimate obtained
through multiple imputation is the square root of this total variance across
the three components (Rubin 1987).
Average marginal effects from the nonlinear models and R
2
values from
the OLS models were also calculated using Rubin’s rules (Rubin 1996). Due to
the multiple imputation strategy, the R
2
values were not from one model but
rather represent the average R
2
across all imputations, using Fisher’s
transformation. Following Harel (2009), the R
2
from each imputation was
transformed to a correlation, then Fisher’s transformation was used to
convert the correlation to a z-score. The mean z-score was calculated across
imputations then transformed back to an R
2
value. A simulation study by
186
Harel suggests that while R
2
values obtained through this transformation
procedure tend to be biased upward, adjusted R
2
values tend to be biased
downward (Harel 2009). Thus, I report both R
2
measures.
Abstract (if available)
Abstract
A growing body of literature demonstrates that prenatal and early life circumstances can have long-term impacts on individual well-being. To fill important gaps in our understanding of the impacts of adverse childhood experiences, in this dissertation I examined adult health and human capital outcomes of individuals who experienced abuse or were exposed to parental incarceration during childhood. Each paper relied on rich survey data from the National Longitudinal Study of Adolescent to Adult Health (Add Health) and statistical methods to measure relationships between adverse childhood experiences and adult outcomes. Add Health is representative of children who attended middle school in the United States in the mid-1990s. In the first paper, I measured the impact of childhood sexual abuse on educational attainment and labor market outcomes in young adulthood. I employed partial identification methods to construct likely bounds on the effects of childhood sexual abuse, implementing methods developed by Altonji, Elder, Taber (2002, 2005) and Oster (2017) to make formal assumptions about the amount of selection on unobservables. This paper demonstrated that there were durable consequences of childhood sexual abuse on human capital. When controlling for demographics, childhood socioeconomic status, childhood disability, childhood physical abuse, emotional abuse, and parental incarceration, including school fixed effects, and allowing selection on unobservables to be as large as selection on observables, results suggested that childhood sexual abuse led to 38 to 45 percent greater likelihood of high school dropout, 20 to 28 percent lower likelihood of college degree attainment, 4 to 8 percent lower likelihood of full-time employment, and 11 to 18 percent lower earnings in young adulthood. Results for childhood sexual abuse dominated the associations between physical and emotional abuse and outcomes—which were zero or smaller and less precise. In the second paper, I examined health and health care access outcomes of survivors of childhood abuse. After controlling for demographics, childhood socioeconomic status, parental incarceration, and including school fixed effects, I found that survivors of childhood abuse had higher risks of cardiometabolic conditions, nervous system conditions, respiratory or allergic conditions, and cancer along with higher risk of recent gastrointestinal symptoms. Meanwhile, survivors of childhood abuse were also more likely to report unmet medical needs. This study highlights the immediate need for development of best practices for detection and quality treatment of childhood trauma and its sequelae. In the third paper, I measured the effects of parental incarceration on educational attainment and labor market outcomes in young adulthood. Employing similar models to those used in the first paper, results suggested that parental incarceration led to 40 to 58 percent greater likelihood of high school dropout, 14 to 26 percent lower likelihood of college degree attainment, and 4 to 6 percent lower likelihood of full-time employment. Among those working, parental incarceration was not associated with earnings level. This work adds to the body of evidence documenting an intergenerational transmission of socioeconomic disadvantage and has important implications for social policy. Taken together, results of this dissertation should motivate research and resources toward improving prevention of childhood abuse, detection of trauma and adverse circumstances among children, and support for children’s socio-emotional health.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Childhood adversity across generations and its impact on externalizing behavior
PDF
Essays in opioid use and abuse
PDF
The role of social support in the relationship between adverse childhood experiences and addictive behaviors across adolescence and young adulthood
PDF
Childhood cancer survivorship: parental factors associated with survivor's follow-up care behavior and mental health
PDF
Life course implications of adverse childhood experiences: impacts on elder mistreatment, subjective cognitive decline, and caregivers' health
PDF
Multilevel sociodemographic correlates of the health and healthcare utilization of childhood cancer survivors
PDF
The value of novel antihyperlipidemic treatments in the U.S. healthcare system: Reducing the burden of cardiovascular diseases and filling the gap of low adherence in statins
PDF
Building a trauma-informed community to address adverse childhood experiences
PDF
Essays on health and aging with focus on the spillover of human capital
PDF
A sociocultural and developmental approach to intimate partner violence among a sample of Hispanic emerging adults
PDF
Three essays on estimating the effects of government programs and policies on health care among disadvantaged population
PDF
The determinants and measurement of human capital
PDF
Multilevel influences of care engagement and long-term survival among childhood, adolescent, and young adult cancer survivors
PDF
Three essays on economics of early life health in developing countries
PDF
Incarceration trajectories of mothers in state and federal prisons and their relation to the mother’s mental health problems and child’s risk of incarceration
PDF
Intergenerational transfers & human capital investments in children in the era of aging
PDF
Essays on health insurance programs and policies
PDF
Effects of stress and the social environment on childhood asthma in the children' s health study
PDF
Unraveling the threads of tragedy: a public health exploration of adverse childhood experiences as precursors to targeted school violence and protective factors for mitigation strategies
PDF
Understanding allegations of childhood neglect using structured and unstructured administrative data
Asset Metadata
Creator
Henkhaus, Laura Esperancilla
(author)
Core Title
Long-term impacts of childhood adversity on health and human capital
School
School of Pharmacy
Degree
Doctor of Philosophy
Degree Program
Health Economics
Publication Date
01/30/2021
Defense Date
05/03/2019
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
adverse childhood experiences,Child abuse,child maltreatment,health and inequality,Health Economics,Health policy,human capital,OAI-PMH Harvest,parental incarceration
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Lakdawalla, Darius (
committee chair
), Cox, Robynn (
committee member
), Goldman, Dana (
committee member
), Joyce, Geoffrey (
committee member
)
Creator Email
lehenkhaus@gmail.com,lhenkhau@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-201968
Unique identifier
UC11663140
Identifier
etd-HenkhausLa-7702.pdf (filename),usctheses-c89-201968 (legacy record id)
Legacy Identifier
etd-HenkhausLa-7702.pdf
Dmrecord
201968
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Henkhaus, Laura Esperancilla
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
adverse childhood experiences
child maltreatment
health and inequality
human capital
parental incarceration