ESSAYS IN LABOR AND HEALTH
by
MARY KOPRIVA
A DISSERTATION
Presented to the Department of Economics
and the Division of Graduate Studies of the University 
of Oregon in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
June 2021
DISSERTATION APPROVAL PAGE
Student: Mary Kopriva
Title: Essays in Labor and Health
This dissertation has been accepted and approved in partial fulfillment of the
requirements for the Doctor of Philosophy degree in the Department of Economics
by:
Edward Rubin Co-Chair
Benjamin Hansen Co-Chair
Alfredo Burlando Core Member
Jonathan Davis Core Member
Clare Evans Institutional Representative
and
Andy Karduna Interim Vice Provost for Graduate Studies
Original approval signatures are on file with the University of Oregon Division of 
Graduate Studies.
Degree awarded June 2021
ii
©c 2021 Mary Kopriva
iii
DISSERTATION ABSTRACT
Mary Kopriva
Doctor of Philosophy
Department of Economics
June 2021
Title: Essays in Labor and Health
This dissertation consists of three essays:
“The Effect of Women’s Access to Free Health Care on Breastfeeding
Practices: Evidence from Armenia” examines how women’s decisions regarding
breastfeeding respond to increased maternal access to medical care. Specifically,
I examine the effects of the Armenian “For You, Women” program which grants
women access to free health care services for a period of one month each year.
Employing a differences-in-differences specification, I explore how the prevalence
and duration of breastfeeding are affected by the program. I find that access to free
women’s health care leads to a decline in the likelihood of breastfeeding.
In “Impacts of the Relocation Program on Native American Migration
and Fertility,” I estimate the historical migratory and fertility effects of the
US Relocation Program. Between 1952 and 1973, the US federal government
attempted to move Native Americans off reservations and to urban areas under
the promises of financial assistance and job training. Using the variation in which
cities were targeted by the program, I employ a difference-in-differences strategy
and estimate that the Relocation Program significantly increased the Native
iv
American population in target cities. I also find evidence that second-generation
Native American women living in cities have a substantially lower fertility rate
than Native American women living on tribal land. Jointly, these findings indicate
that this federal program substantially shifted the spatial distribution of the Native
American population in the US throughout the 20th century.
Finally, “Right to Carry Laws and Intimate Partner Homicide,” examines
to what extend right to carry laws impact intimate partner homicide rates. I
employ generalized difference-in-differences, event study, and synthetic control
methods specifications to examine the impact of right to carry laws on annual,
state-level intimate partner homicide rates. I find no evidence of an impact of
right to carry laws on female victim intimate partner homicide rates. While there
are some specifications that suggest a decrease in male victim firearm intimate
partner homicide rates, these results are limited and fluctuating across model
specifications. Overall, the estimates are noisy but suggest little evidence of an
impact, highlighting the importance of more formalized sensitivity analysis for
testing the robustness of results.
v
CURRICULUM VITAE
NAME OF AUTHOR: Mary Kopriva
GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED:
University of Oregon, Eugene, OR
Creighton University, Omaha, NE
DEGREES AWARDED:
Doctor of Philosophy, Economics, 2021, University of Oregon
Master of Science, Economics, 2017, University of Oregon
Bachelor of Science, Economics & Mathematics, 2014, Creighton University
AREAS OF SPECIAL INTEREST:
Health Economics
Labor Economics
Gender Economics
GRANTS, AWARDS AND HONORS:
Best Third Year Field Paper Award, UO Department of Economics, 2019
Graduate Teaching Fellowship, University of Oregon, 2017-2021
Kleinsorge Economics Fellowship, University of Oregon, 2016-2017
Distinctive Scholar Award, University of Oregon, 2016-2016
Fulbright U.S. Student Grant, Fulbright Program, 2014-2015
PUBLICATIONS:
Briggs, K. and Wade, M. (2014). More is Better: Evidence that Joint
Patenting Leads to Quality Innovation. Applied Economics, 46(35): 4370-
4379
vi
ACKNOWLEDGEMENTS
Thank you to my committee. In particular, thank you to Ed Rubin and Ben
Hansen for your guidance and encouragement throughout the dissertation writing
process, and thank you to Alfredo Burlando, Jonathan Davis, and Clare Evans for
your helpful comments and fresh perspectives.
Thank you to Kristie Briggs for fostering my love of research as an
undergraduate and for encouraging me to pursue economics further.
Thank you to my family. Thank you to my mom for the constant love and
support. Thanks also to my husband, Cody, and my sisters, Elizabeth, Sonya, and
Coleen, for always being there for me no matter what. I am so thankful for all your
advice, your humor, and your patience.
Finally, thank you to the University of Oregon Department of Economics
and in particular my fellow graduate students for your constructive and supportive
feedback as well as your friendship and community.
vii
For my husband, Cody.
viii
TABLE OF CONTENTS
Chapter Page
I. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
II. THE EFFECT OF WOMEN’S ACCESS TO FREE HEALTH CARE ON
BREASTFEEDING PRACTICES: EVIDENCE FROM ARMENIA . 4
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
III. IMPACTS OF THE RELOCATION PROGRAM ON NATIVE AMERICAN
MIGRATION AND FERTILITY . . . . . . . . . . . . . . . . . . . . . 38
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
ix
Chapter Page
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Fertility Differences . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Tribal Land Population Counterfactuals . . . . . . . . . . . . . . . 65
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
IV. RIGHT TO CARRY LAWS AND INTIMATE PARTNER HOMICIDE . 70
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Concealed Carry Laws and Intimate Partner Homicide . . . . . . . 75
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
V. CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
APPENDIX: ADDITIONAL TABLES AND FIGURES . . . . . . . . . . . . . 98
REFERENCES CITED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
x
LIST OF FIGURES
Figure Page
1. Map of Armenia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2. Armenia Breastfeeding Trends By Treatment Status 2000-2014 . . . . . . 17
3. Event Study: Likelihood Ever Breastfed . . . . . . . . . . . . . . . . . . . 25
4. Triple Difference Estimates by Month . . . . . . . . . . . . . . . . . . . . 26
5. Percent of Children Breastfed within the First Hour . . . . . . . . . . . . 34
6. Percent of Children Fed Something Other Than Breastmilk in First 3 Days 34
7. Type of Food Given to Children in First 3 Days . . . . . . . . . . . . . . 35
8. Percent of Children Fed Infant Formula in Last 24 Hours . . . . . . . . . 36
9. Map of Relocation Program Target MSAs . . . . . . . . . . . . . . . . . . 50
10. Native American Population Trends by MSA Type . . . . . . . . . . . . . 51
11. Estimates of the Relocation Program’s Migration Effects over Time . . . 56
12. Target City Population Trends by Race . . . . . . . . . . . . . . . . . . . 58
13. Native American Population Trends for Target MSAs versus Tribal Land 59
14. Migration Trends for Native Americans Born in Tribal States . . . . . . . 61
15. General Fertility Rate by Race and Current Residence . . . . . . . . . . . 63
16. General Fertility Rate by Age . . . . . . . . . . . . . . . . . . . . . . . . 64
17. Counterfactual Native American Population on Tribal Land . . . . . . . . 67
18. Intimate Partner Homicides per 100,000 by Victim Sex . . . . . . . . . . 78
19. Number of States that have Adopted Right to Carry Laws over Time . . 79
20. Specification Chart for the Impact of Right to Carry Laws on the Rate of Intimate
Partner Homicides with a Firearm: Female Victims . . . . . . . . . 87
21. Event Study Estimates for Female Victims, Weapon Type: Firearm . . . 88
xi
Figure Page
22. Specification Chart for the Impact of Right to Carry Laws on the Rate of Intimate
Partner Homicides with a Firearm: Male Victims . . . . . . . . . . 89
23. Event Study Estimates for Male Victims, Weapon Type: Firearm . . . . . 90
24. Partially Pooled SCM for Male Victims, Weapon Type: Firearm . . . . . 91
A1. Pre-Treatment Coefficient Estimates by Month of Birth . . . . . . . . . . 99
A2. Target City Native American Population Trends . . . . . . . . . . . . . . 100
A3. Native American Population Trends by MSA Type, Log Scale . . . . . . . 101
A4. Partially Pooled SCM for Female Victims, Weapon Type: Firearm . . . . 102
A5. Adoption of Right to Carry Laws over Time . . . . . . . . . . . . . . . . 103
A6. Individual State SCM Estimates Female Victims, Weapon Type: Firearm 104
A7. Individual State SCM Estimates Male Victims, Weapon Type: Firearm . 106
xii
LIST OF TABLES
Table Page
1. Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2. Effect of Access to Free Maternal Healthcare on the Number of Months Spent
Breastfeeding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3. Extensive and Intensive Margins . . . . . . . . . . . . . . . . . . . . . . . 23
4. Heterogeneity by Gender on Months Breastfed . . . . . . . . . . . . . . . 27
5. Heterogeneity by Gender Extensive and Intensive Margins . . . . . . . . . 28
6. Placebo Regressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
7. Women’s Employment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
8. Baseline Native American Population Response to the Relocation Program 55
9. Native American Population Response to the Relocation Program, Alternative
Control Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
10. Doubly Robust DiD using Inverse Probability of Treatment Weighting . . 60
11. RTC Laws and Non-Firearm Intimate Partner Homicide Rates . . . . . . 92
A1. Effect of Access to Free Maternal Healthcare on Breastfeeding for the Full
Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
A2. Impact of Right to Carry Laws on Rates of Handgun Sales . . . . . . . . 102
xiii
CHAPTER I
INTRODUCTION
In “The Effect of Women’s Access to Free Health Care on Breastfeeding
Practices: Evidence from Armenia,” I examine the impacts of women’s increased
access to health care on breastfeeding outcomes. More specifically, I analyze how
the “For You, Women” program in Armenia, which extended free health care to
women for one month each year, affected two main breastfeeding outcomes, namely:
the duration of breasting and the likelihood of ever breastfeeding. Using the timing
of the program’s implementation and the geographic variation in where services
were located, I employ a difference-in-differences specification to identify a causal
impact of the program on breastfeeding.
I find that there is a roughly 5% decrease in the likelihood of ever
breastfeeding associated with the increased access to health care. I find no impact
of the program on the duration of breastfeeding for those who choose to breastfeed.
These findings are robust to a placebo check and are unlikely to be driven by
changes in female employment. Even with Armenia’s strong son-preference, I find
no strong evidence of a differential impact for male versus female children though
this may be due to a lack of precision among the disaggregated estimates. One
mechanism that is likely contributing to this decrease is the documented lack of
supportive practices for breastfeeding in the region’s medical centers, which I verify
using data from the Demographic and Health Surveys.
The second essay, “Impacts of the Relocation Program on Native American
Migration and Fertility,” analyzes the migratory and fertility impacts of the
decades-long US federal Relocation Program, which attempted to relocate Native
1
Americans off of rural reservations into urban centers by offering minimal financial
assistance to facilitate the move and promising employment upon arrival in the
new city. I employ a difference-in-differences methodology, comparing cities that
were targeted by the program to cities that were not, to examine the impact of
the Relocation Program on Native American migration and provide descriptive
evidence of the impact on second-generation fertility.
I find that the program significantly increased the Native American
population in cities targeted by the program relative to other US cities. Specially,
I find that the program caused more than 30% of the baseline targeted population
to relocate to target cities, constituting a major shift in the spatial distribution of
the Native American population. These findings are robust to the use of different
subsets of the control group and to a doubly robust difference-in-differences
specification based on propensity score matching techniques. Additionally, I find
descriptive evidence that the fertility rate for second-generation Native American
women living in cities is about 50% lower than that of Native American women
living on tribal lands.
Finally, the third essay, “Right to Carry and Intimate Partner Homicide,”
estimates the effect of right to carry laws on rates of intimate partner homicide.
I use three methods to analyze this relationship—difference-in-differences, event
study, and synthetic control methods. In my main specifications, I analyze the
intimate partner homicide rates for male and female victims separately and focus
only on homicides committed using a firearm. I consider rates of non-firearm
intimate partner homicides by victim sex as an additional check.
I find no evidence for an impact of right to carry laws on female victim
intimate partner homicide rates and only weak evidence of an effect for male
2
victims. The results are highly sensitive to defensible researcher choices about
model specification related to which sample is used, which methodology is
employed, and which fixed effects and controls are included. Overall, these results
suggest little to no impact of right to carry laws on intimate partner homicides.
Further, these findings highlight the need for more formal sensitivity analysis as
standard practice in social science research and particularly within the right to
carry literature where my findings illustrate the severity of the problem.
3
CHAPTER II
THE EFFECT OF WOMEN’S ACCESS TO FREE HEALTH CARE ON
BREASTFEEDING PRACTICES: EVIDENCE FROM ARMENIA
Introduction
Breastfeeding provides health benefits for infants and mothers, particularly
in areas where food and water may be contaminated or scarce. Infants who are
breastfed benefit from decreased mortality as well as reduced morbidity from
diarrhea and respiratory infections. Additionally, breastfeeding has been shown
to protect mothers against breast cancer and, some evidence suggests, ovarian
cancer. The benefits of breastfeeding are particularly pronounced in low- and
middle-income countries with benefits increasing in the duration and exclusivity
of breastfeeding (Victora et al., 2016). Despite the substantial gains documented
from breastfeeding, typical breastfeeding practices across most countries remain
well below the World Health Organization (WHO) recommendations. There
are still gaps in our understanding of the costs, barriers, and challenges that
are contributing to this disparity between recommended and existing levels of
breastfeeding given the known benefits.
This paper explores the extent to which women’s access to free medical care
can influence women’s decisions regarding breastfeeding practices. Specifically, I
examine the extent to which breastfeeding duration and prevalence are affected by
a program offering free clinical visits to women in Armenia for one month each year
beginning on March 8th, International Women’s Day, and running through April
7th, the Armenia holiday of motherhood and beauty. In addition to examining
4
the overall effect of maternal access to free health care on women’s breastfeeding
decisions, I also examine how this program may affect gender asymmetries in
breastfeeding practices given the country’s strong sex-preferences and previous
works suggesting that strong son-preferences can lead to breastfeeding disparities.
The “For You, Women” program is based in the capital city of Yerevan. It
first began in 2009 at a single medical clinic but expanded to several clinics in
the region in response to the program’s success. This study exploits the regional
variation in the program within Armenia to causally identify the effects of women’s
access to free medical care on breastfeeding outcomes. I employ a differences-in-
differences specification using data from the Integrated Living Conditions Survey
on region of residence and year of birth to determine a child’s exposure to the
“For You, Women” campaign. Questions about whether or not the child was
breastfed and for how many months are used as the main measures of breastfeeding
prevalence and duration.
The results of the analysis show a decrease in the likelihood of ever
breastfeeding by roughly 5% in response to the program, while the average length
of time spent breastfeeding, given that the child is breastfed, remains unaffected.
Interestingly, there is no significant difference in these effects when the sample is
broken down by the gender of the child. As a robustness check, I also examine how
women’s employment changes during the time of the study, finding that increased
maternal employment is likely not driving the results.
While these results may seem counterintuitive at first, there are a number of
reasons why increased exposure to health care could be associated with a decreased
likelihood of breastfeeding. The most compelling possible explanation is that the
medical clinics that participate in the “For You, Women” program may not adhere
5
to practices that support breastfeeding. Therefore, women’s increased exposure to
these health care facilities could negatively impact breastfeeding rates. Previous
literature examining low levels of exclusive breastfeeding in Armenia and analyzing
infant formula promotion in the country have shown that Armenian medical
clinics, particularly those in Yerevan, fall short of best practices for promoting
breastfeeding among new mothers. Using data from the Demographic and Health
Survey in Armenia, I verify the findings from the previous literature and provide
additional evidence for a lack of breastfeeding support in the Armenian medical
community as a likely driver of my main results.
Literature Review
Benefits of Breastfeeding
Victora et al. (2016) provide a review of the current literature on the benefits
of breastfeeding. They find consistent evidence that breastfeeding protects against
the risk of infant mortality due to infectious diseases along with strong evidence
of decreased incidences of diarrhea, respiratory infections, and malocclusion.
Further, they find increased intelligence associated with breastfeeding as well as
suggestive evidence that breastfeeding may decrease incidences of obesity and
diabetes. Additionally, nursing mothers are shown to gain protection against breast
cancer with some suggestive evidence of protection against ovarian cancer as well.
Victora et al. also estimate that substantial increases in breastfeeding rates to near
universal levels could prevent 832,000 annual deaths of children under the age of 5
and 20,000 annual deaths of women from breast cancer.
Beyond these overarching benefits, Brandtzaeg (2010) suggests that breast
milk is specifically tailored to protect against infectious agents and antigens present
6
in the mother’s environment, which allows the infant’s immune system to develop
in a way that specially protects him or her from diseases and infections common
to that particular environment. Le Doare et al. (2018) suggest that breast milk
also contains important microbiota, the symbiotic microbial cells contained in all
peoples’ bodies (Ursell et al., 2012). They review the literature on breast milk
microbiota which suggests that breast milk also contributes to the development
of the child’s immune system through the pass-down of microbiota, which aids in
the healthy development of the intestinal microbiome, providing long-term benefits
to breastfed infants.
Determinants of Breastfeeding
The many benefits of breastfeeding outlined above highlight the importance
of understanding the determinants of breastfeeding choices. Rollins et al.
(2016) provide an up-to-date review of the literature analyzing what determines
breastfeeding practices. They find that hospital practices can significantly affect
breastfeeding rates with decreased breastfeeding associated with early mother-
infant separation (Righard & Alade, 1990), prelacteal supplementation, and the
distribution of free infant formula samples (Wright et al., 1996). Further, Rollins et
al. suggest that health system interventions aimed at promoting breastfeeding may
have a significant positive impact on initiation and duration of breastfeeding.
Pérez-Escamilla et al. (2016) provide a review of one of the most popular
interventions aimed at promoting breastfeeding, the Baby-friendly Hospital
Initiative (BFHI). The BFHI was initially launched in 1991-1992 as part of the
WHO/UNICEF Global Strategy for Infant and Young Child Feeding. There
are currently 156 countries participating in the initiative with an estimate of
7
over 20,000 maternity hospitals and birthing centers ever designated as “Baby-
friendly” worldwide1 (World Health Organization, 2009). The BFHI consists of
ten-steps that health care facilities can implement to encourage breastfeeding
such as training medical staff in supportive breastfeeding practices and organizing
community support groups to encourage mothers in continued breastfeeding after
they leave the hospital. Pérez-Escamilla et al. find that implementing the ten steps
increases breastfeeding in both the short-term and the long-term. Additionally,
they find a dose-effect in the response of breastfeeding outcomes to the number of
steps a hospital has instituted.
Rollins and co-authors also suggest that individual characteristics play an
important role in breastfeeding decisions. Women who receive poor advice or
limited support that undermines their self-efficacy or increases the anticipation
of difficulties breastfeeding have lower rates of breastfeeding (Avery et al., 2009;
Brown et al., 2014). Further, maternal health factors including smoking (Leung
et al., 2002; Liu et al., 2006), overweight/obesity (Turcksin et al., 2014), and
depression (Dennis & McQueen, 2009) are associated with reduced breastfeeding.
Included in the extensive literature examining breastfeeding determinants
are a set of papers examining the effects of prenatal care. The wider literature
on prenatal care focuses mainly on birth outcomes, such as birth weight, with
mixed results (Conway & Deb, 2005; Reichman et al., 2009); however, two papers
have analyzed the effects on maternal postpartum health behaviors including
breastfeeding. Reichman et al. (2010) examine how the timing of prenatal care
affects maternal health behaviors postpartum finding that first trimester prenatal
care is associated with a decreased likelihood of postpartum smoking and an
1Estimates as of 2009
8
increased number of well-baby visits compared to prenatal care initiated later in
the pregnancy. They also find some suggestive evidence that early prenatal care
may improve breastfeeding outcomes, but their findings are sensitive to model
specification. Yan (2017) extends this analysis by examining the effects of prenatal
care utilization on both maternal health and maternal health behaviors. Yan finds
an increase in no breastfeeding associated with a low frequency of prenatal visits
and the late initiation of prenatal care, in the second or third trimesters. This
research expands on these studies by examining how a broad increase in access to
health care for all women affects breastfeeding outcomes with antithetical findings
possibly due to differences in health care facility practices.
Breastfeeding and Gender
In addition to the literature on the overall determinants of breastfeeding,
there have also been some studies that explore the determinants of heterogeneous
breastfeeding decisions along gender lines. Jayachandran and Kuziemko (2011)
provide a model whereby son preference can lead to decreased breastfeeding
duration for female children not because of inequitable valuing of health between
male and female children but because of the negative relationship between
breastfeeding and fertility. Specifically, they base their model on two important
facts regarding breastfeeding. First, women who are breastfeeding experience
reduced fertility. Second, women who become pregnant often stop breastfeeding
current children because of the physical demands of pregnancy. Their model
predicts that this negative relationship between breastfeeding and fertility coupled
with a “try until you have a son” fertility pattern may contribute to asymmetric
breastfeeding practices between male and female children. They test their model
9
empirically with data from India where there exists notably strong son preferences.
They find evidence supporting their theory that the decision to breastfeed female
children less is likely in part a “passive” outcome of wanting to have a son and
therefore choosing to reduce the period of decreased fecundity during breastfeeding
or ceasing breastfeeding current children if the mother becomes pregnant as a result
of trying for a son. Chakravarty (2015) and Hafeez and Quintana-Domeque (2018)
confirm Jayachandran and Kuziemko’s findings in Egypt and Pakistan, respectively.
Background
FIGURE 1. Map of Armenia
Source: CIA World Factbook Armenia
Armenia is located in the South Caucasus region of Asia neighboring Georgia,
Azerbaijan, Turkey, and Iran as seen in Figure 1. The population of Armenia is
10
2,972,700. Roughly one third of the population lives in the capital city of Yerevan2
(Statisitcal Committee RA, 2018). The official language of Armenia is Armenian
and a majority of the population is ethnically Armenian. A large majority of the
population is Christian with more than 92% of the population belonging to the
Armenian Apostolic Church as of 2011. Armenia is approximately 29,743 square
km and is divided into 11 regions (marz), one of which is the capital city. GDP per
capita (PPP) is $9,500 with 18.9% unemployment according to 2017 estimates.
The Armenia Demographic and Health Surveys (ADHS) conducted in 2000,
2005, 2010, and 2015-16 provide insight into fertility and family planning practices
in Armenia. According to these surveys, the average age at first marriage for
women in Armenia is roughly 21 with the first birth occurring on average about
one year later. Throughout the period considered in this study, the total fertility
rate in Armenia is 1.7. The percent of women who received some prenatal care
increased over the sample period from 92% in 2000 to more than 99% in 2015.
During the Soviet period, abortion was used as the primary means of controlling
fertility (National Statistical Service [Armenia], 2001). Abortion is still legal
in Armenia though the percentage of pregnancies ending in induced abortion
decreased drastically during the time period studied from over 50% in 2000 to 23%
in 2015. As noted in the introduction, the sex ratio at birth is currently among the
highest in the world with 113 males born for every 100 females. This figure has
been decreasing steadily from a high of 117 males for every 100 females born in
2002 (The World Bank, 2017).
2Estimates as of 2018
11
Breastfeeding in Armenia
Culturally, there is no stigma associated with breastfeeding in public and
family members have great respect for mothers who choose to breastfeed. Prior to
the onset of Soviet rule in Armenia, breastfeeding practices such as early initiation
and exclusive breastfeeding were common. During Soviet times, these practices
were abandoned and instead it was common to initiate breastfeeding later and to
provide prelacteal supplementation. In 1988, a devastating earthquake hit Armenia,
which in conjunction with the transition to independence in 1991, caused many
families to suffer economic difficulties and to further decrease their likelihood
of breastfeeding. Over this time period, rates of fully breastfeeding infants at 4
months of age dropped from 64% in 1988 to 23% in 1993 (Harutyunyan, 2004).
Contributing to this massive decline in breastfeeding was the widespread
availability of free infant formula provided by humanitarian aid after the
earthquake along with unfavorable health facility practices as well as a lack of
knowledge among medical professionals and the general population regarding
the benefits of breastfeeding. In response to the decline in breastfeeding, the
Ministry of Health (MOH) and aid organizations stopped their distribution of free
infant formula in 1995. Additionally, in 1993, the MOH implemented a national
breastfeeding promotional program which later included encouraging hospitals and
clinics to adopt “Baby-friendly” practices (Amirkhanyan, 2008). By 2001, the rate
of fully breastfed 4-month-old children was back up to 69%. In 2008, however,
the Baby-friendly Hospital Initiative and the National Breastfeeding Promotion
Committee were discontinued.
The WHO recommends that breastfeeding be initiated early, within the
first hour after delivery, and that children be exclusively breastfed until they
12
reach 6 months of age. More recent Demographic and Health Survey trends have
shown steady increases in these commonly used breastfeeding indicators. Early
initiation of breastfeeding increased from 24% in 2000 to 40% in 2015. Exclusive
breastfeeding under 6 months also increased over this time period from 29% in 2000
to almost 45% in 2015. Rates of ever breastfed children are consistent across the
2005, 2010, and 2015 waves of the survey at 96-97%, increasing slightly from 90% in
2000.
In 2014, the Armenian National Assembly passed a law limiting the marketing
of breast milk substitutes in line with international standards (Harutyunyan,
2015). A previous law adopted in 1999 had attempted to curb marketing of
breast milk substitutes but lacked enforcement mechanisms. Thus, prior to
2014, Armenia had significant reports of serious violations of the World Health
Organization’s International Code of Marketing of Breast-milk Substitutes such
as brand endorsements in health care facilities and infant formula sponsorships of
medical conferences (“Confidence” Health NGO, 2011).
“For You, Women” Program
The “For You, Women” campaign is a program based in Armenia’s capital
city of Yerevan which provides women with free health care services for one month
every year. The campaign began in 2009 at a single clinic, “Sourb Astvatsamayr
(St. Mary)” Medical Centre (SAMC). In the first year, 221 women applied for
medical services as part of the program (Yerevan City Hall, 2010). By 2013,
participation had increased twenty-fold, and the program expanded to include
multiple clinics in the Yerevan region (Yerevan City Hall, 2013). The campaign
consists of providing free medical services and if needed surgical interventions.
13
Some examples of the services provided include gynecological services, opthalmic
services, ultrasounds of the thyroid glands, and CT scans.
The campaign begins each year on March 8th, International Women’s Day,
and runs through April 7th, the Armenian holiday dedicated to motherhood and
beauty. If all the women who applied for consultations and treatment are unable
to be seen within the thirty-one day period, the participating clinics are instructed
to continue to provide services for free until everyone who applied can be seen.
Though the program is based in Yerevan, women from outside the region are able
to participate; however, participation rates for those outside the capital are low
at around 25%. Thus, for the proceeding analysis women living in Yerevan are
considered the treated group and women living in other regions of Armenia are
considered the control group.
Data
Integrated Living Conditions Survey
The baseline analysis utilizes data from the Integrated Living Conditions
Survey (ILCS) between the years of 2001 and 2016. The ILCS is an annual survey
conducted by the National Statistical Service of the Republic of Armenia (NSS)
with aid from the World Bank and USAID as well as additional assistance from
the Millennium Challenge Account - Armenia between the years 2007 and 2011.
Between 4,000 and 8,000 households are surveyed each year depending on the
funding.
The survey consists of two main components, namely, a questionnaire with
information about the household’s standard of living and a diary of consumption
and income for the past thirty days. Included in the questionnaire is a section
14
pertaining to health and health care utilization for children between the ages of
0 and 5 at the time of the survey. This section includes questions about whether
or not the child was breastfed and if so for how many months, which are used as
measures of the two main breastfeeding outcomes. Additionally, the survey includes
questions on current region of residence and birth year, which are used to create
the treatment dummy variables for the differences-in-differences analysis. The ILCS
also has information on the education and employment of individual’s over the age
of 15 as well as the number of members in the household and the number of rooms
in the household, all of which are used as controls.
For the main analysis, I limit the sample to households that include exactly
one woman of child-bearing age in order to include controls for education and
employment of the presumed mother, dropping roughly a quarter of the original
sample. Analysis containing the full sample can be found in the appendix. The
coefficients and significance levels for the full sample are consistent with the
findings from the limited sample used in the main analysis. Further, similarly
constricted sampling of the Demographic and Household Survey in Armenia,3 which
contains an identifying marker for mothers of children in the sample, suggests that
in households with one child-bearing aged female, that female is the mother of all
children under five in 99% of cases.
3This verification holds true across all years for which DHS data is available.
15
TABLE 1. Summary Statistics
Statistic Male Female Total
Ever Breastfed 0.927 0.929 0.928
(0.261) (0.257) (0.259)
Months Breastfed 8.981 8.854 8.923
(6.350) (6.120) (6.246)
Months Breastfed 9.692* 9.530* 9.618
If Ever Breastfed (6.052) (5.820) (5.947)
Number of 5.393 5.372 5.384
HH Members (1.283) (1.303) (1.293)
Number of 3.151 3.137 3.145
Rooms in HH (1.192) (1.191) (1.192)
Mother’s 2.229 2.233 2.231
Education (0.429) (0.429) (0.429)
Mother’s 0.277 0.272 0.275
Work Status (0.448) (0.445) (0.447)
Age 2.072 2.106 2.088
(1.420) (1.424) (1.422)
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Table 1 provides summary statistics for the variables of interest from the
ILCS for the subsample of households with one female of child-bearing age. The
table depicts averages for each variable of interest for males and females separately
as well as the total sample means. The number of months breastfed for children
who were ever breastfed is the only variable whose mean is statistically different
between the male and female subsamples. Standard deviations are in parentheses
below the mean. 54% of the children in the sample are male. This number is
higher than the natural rate but fairly consistent with the population of Armenia
where the sex ratio at birth is 53-54% over the sample period (The World Bank,
16
2017). Almost 93% of the children in the full sample are ever breastfed, and for
those children who are breastfed, the average duration of breastfeeding is about
9.6 months. Households on average have between 5 and 6 members with three
generations typically sharing a house, and about 28% of the presumed mothers
are employed at the time of the survey.
FIGURE 2. Armenia Breastfeeding Trends By Treatment Status
2000-2014
0.94
Birth Place
0.90 Yerevan
Other Regions
0.86
0.82
2000 2005 2010
Year of birth
Notes: Trend lines for the rate of children ever breastfed by treatment status calculated using the
ILCS subsample of household with only one child-bearing age female. Includes only years with
more than 500 observations. The vertical dashed line signifies the treatment year, 2009.
Sources: Author’s calculations from Armenia ILCS
Trends in the rates of children ever breastfed are depicted in Figure 2.
Separate trend lines are shown for Yerevan, the treatment group, and the other
regions of Armenia, the control group. Prior to 2009, both trend lines show
increasing rates of ever breastfeeding. After the intervention in 2009, however,
the trend line for Yerevan takes a downward turn while the trend line for other
17
Percent of children who were ever breastfed
regions continues to increase steadily. Hence, Figure 2 provides suggestive evidence
of parallel trends prior to treatment with trends diverging between the treatment
and control groups following the introduction of the “For You, Women” program in
2009.
Empirical Strategy
The baseline regression takes the following form. For child i, born in year t,
living in region r, and surveyed in ILCS wave w:
BFitrw = α + β1Postt + β2Y erevanr + β3Postt × Y erevanr + Xitrw
+γr + δt + ωw + εitrw
Here BFitrw refers to one of two main outcome variables–the number of
months spent breastfeeding or a dummy variable for whether or not the child was
ever breastfed. In the first context, BFitrw is a continuous variable running from 0
to 60 months for the baseline results and 0.1 to 60 months when considering only
the subsample of children who were ever breastfed. In the second context, BFitrw
is a dummy variable which takes on a value of 1 if the child was ever breastfed and
zero otherwise.
Two different definitions of Postt are used based on the outcome variable.
For the outcome variable of months breastfed among those children who were
ever breastfed, Postt is equal to 1 if a child is born during or after June 2008
and 0 otherwise. For this outcome variable, it is possible for children born prior
to the beginning of the intervention to be affected as the choice of when to stop
breastfeeding a child, given breastfeeding was initiated, can be made essentially
18
any time after the child is born. June 2008 is specifically chosen because it
corresponds to the average length of breastfeeding duration for the full sample.
Therefore, June 2008 seems a natural point at which to consider individuals
treated as this is the point when the average mother in the sample is choosing to
discontinue breastfeeding. When the outcome variable is whether or not the child
was breastfed, Postt is defined as any child born in or after March 2009 as this
is the month in which the program began. This definition of Postt is natural for
the outcome variable ever breastfed as breastfeeding initiation occurs only very
near to the time of the birth. Therefore, the decision of whether or not to initiate
breastfeeding children born in February 2009 or earlier should not have been
affected by a program that started in March 2009 and, therefore, these children are
considered untreated while children born after the program began are considered
treated.
Y erevanr is a dummy variable for whether or not the child currently resides
in the capital region where the program is based. Y erevanr is equal to 1 if the
child lives in Yerevan and 0 otherwise. The variable Y erevanr is used as a proxy
for whether or not the child was born in Yerevan because actual place of birth
is not recorded in early versions of the survey. Specifications that use a direct
measure of place of birth by excluding the early surveys show similar results.
Xitrw is a set of additional controls. Xitrw includes the number of rooms
in the house and the number of members living in the household as well as the
education level and employment status of the woman of child-bearing age. In
the regression tables to follow the two latter controls are referred to as mother’s
education and mother’s work status for ease, but as there is no direct variable
linking the child to a mother, the relationship is assumed rather than established
19
in the data. γr, δt, and ωw are fixed effects for region of current residence, birth
year, and wave of the survey, respectively. All standard errors are clustered using
the wild cluster bootstrap method as suggested by Cameron and Miller (2015) for
data with few clusters.
The coefficient of interest is β3, which represents the effect of a child being
born in Yerevan during the time of the program on the given breastfeeding
outcome. The identifying assumption for this to be considered the effect of the “For
You, Women” program on the breastfeeding outcome is that absent the “For You,
Women” program, trends in breastfeeding practices in Yerevan would not differ
from those in the other regions of Armenia. The effect captured by β3 should be
interpreted as an intent to treat because there is no direct measure of who is using
the free medical care. In other words, β3 captures the effect of the program which
increases access to maternal health care rather than the effect of utilization of the
free health care services.
Results
The results for the baseline regression, which looks at the total number of
months the child was breastfed, can be found in Table 2. The main specification
captures both the extensive and intensive margins with changes in duration and
prevalence estimated separately in subsequent sections. Columns (1) - (3) use the
first definition of Postt with treatment beginning in June 2008, and columns (4) -
(6) use the secondary definition of Postt with individuals only considered treated
if they were born after the program began in March 2009. Columns (1) and (4)
display the results for the simple OLS specification. Columns (2) and (5) include
the full set of fixed effects for year of birth, region of residence, and survey wave.
20
Household and mother controls are added in columns (3) and (6). Columns (3) and
(6) with the full set of fixed effects as well as the household and mother controls are
the preferred specifications.
The results of the baseline specification, though statistically insignificant,
provide the first evidence of a possible decrease in breastfeeding in response to the
program. The baseline definition of breastfeeding, however, may be too broad to
pick up significant changes in breastfeeding practices. Specifically, the baseline
definition of breastfeeding, which simply measures the total duration in months
a mother spent breastfeeding, does not distinguish between changes in initiation
of breastfeeding versus changes in the duration of breastfeeding. Therefore, to
better understand the true effect of the program on breastfeeding, it is necessary to
examine each of these aspects individually by exploring the extensive and intensive
margins separately.
Extensive Versus Intensive Margin
The extensive margin is measured by a dummy variable for whether or not
the child was ever breastfed. The results, which can be found in columns (1) - (3)
of Table 3, show how the prevalence of breastfeeding responds to the “For You,
Women” campaign. Additional controls are again included across the columns from
column (1) to column (3) with column (3) including the full set of fixed effects,
household controls, and mother controls.
The results presented in the first three columns of Table 3 suggest that
the program does have an effect on the prevalence of breastfeeding. For those
children exposed to the “For You, Women” program, there is a 4.9 percentage point
decrease in the likelihood of ever being breastfed. This estimate is robust across
21
TABLE 2. Effect of Access to Free Maternal Healthcare on the Number of Months
Spent Breastfeeding
Dependent variable:
Months Breastfed
(1) (2) (3) (4) (5) (6)
Post June08 −0.169 −0.489∗ −0.485∗
(0.210) (0.325) (0.307)
Post March09 −0.188 −0.865 −0.851
(0.204) (0.598) (0.602)
Yerevan −0.685 −0.717
(1.438) (1.412)
Number of −0.017 −0.017
HH Members (0.056) (0.057)
Number of −0.065 −0.065
Rooms in HH (0.074) (0.074)
Mother’s −0.292 −0.295
Education (0.299) (0.299)
Mother’s 0.021 0.025
Work Status (0.289) (0.290)
Post June08 −0.469 −0.832 −0.826
× Yerevan (0.797) (0.995) (0.994)
Post March09 −0.444 −0.744 −0.741
× Yerevan (0.308) (0.914) (0.921)
Time FE No Yes Yes No Yes Yes
Region FE No Yes Yes No Yes Yes
Wave FE No Yes Yes No Yes Yes
Household Controls No No Yes No No Yes
Mother Controls No No Yes No No Yes
Observations 17,358 17,358 17,349 17,358 17,358 17,349
R2 0.004 0.081 0.081 0.004 0.081 0.081
Adjusted R2 0.004 0.078 0.079 0.004 0.078 0.078
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 All standard errors are wild bootstrap standard errors.
22
TABLE 3. Extensive and Intensive Margins
Dependent variable:
Extensive Margin Intensive Margin
(1) (2) (3) (4) (5) (6)
Post March09 0.034∗∗∗ 0.002 0.002
(0.006) (0.016) (0.016)
Post June08 −0.526∗∗∗ −0.686∗∗ −0.689∗∗
(0.195) (0.307) (0.291)
Yerevan 0.002 −0.748
(0.031) (1.271)
Number of 0.004 −0.058
HH Members (0.003) (0.053)
Number of 0.002 −0.095
Rooms in HH (0.004) (0.068)
Mother’s −0.009 −0.243
Education (0.009) (0.275)
Mother’s 0.010∗ −0.084
Works Status (0.006) (0.266)
Post March09 −0.041∗∗ −0.049∗∗ −0.049∗∗
× Yerevan (0.021) (0.025) (0.025)
Post June08 −0.200 −0.483 −0.477
× Yerevan (0.8) (0.972) (0.960)
Time FE No Yes Yes No Yes Yes
Region FE No Yes Yes No Yes Yes
Wave FE No Yes Yes No Yes Yes
Household Controls No No Yes No No Yes
Mother Controls No No Yes No No Yes
Observations 17,358 17,358 17,349 16,104 16,104 16,095
R2 0.003 0.024 0.025 0.005 0.096 0.097
Adjusted R2 0.003 0.022 0.023 0.005 0.093 0.094
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 All standard errors are wild bootstrap standard errors.
23
all specifications and significant at the 5% level, though the naive OLS estimate is
slightly smaller. To better understand the magnitude of this change, the effects are
a little over half the size of the decrease in the likelihood of breastfeeding associated
with a mother returning to work within the first 12 weeks after delivery (Berger et
al., 2005).
To examine the intensive margin, the sample is limited to only those children
who were ever breastfed. The breastfeeding outcome measure is the number
of months the child was breastfed. Thus, the outcome variable is a measure of
breastfeeding duration for the subset of children who were ever breastfed. Columns
(4) - (6) of Table 3 contain the intensive margin results with controls again added
successively across columns. The results of the intensive margin analysis suggest
that on average breastfeeding duration does not respond to the increased access to
maternal health care provided by the “For You, Women” program.
Given these findings, we can conclude that changes in breastfeeding in
response to the program are concentrated among decreased prevalence rather
than decreased duration. In other words, exposure to the “For You, Women”
campaign is associated with an increase in the likelihood that a woman will choose
to never breastfeed but does not seem to affect the average length of time spent
breastfeeding for women who still decide to breastfeed.
Additional Analysis of the Extensive Margin
Event Study
As an additional test of the robustness of the extensive margin results, I
perform an event study where the dummy variable for ever breastfed is regressed on
a set of interaction terms between the indicator for living in the treatment region
24
of Yerevan and the indicators for the year of birth. The full set of fixed effects is
included. Figure 3 shows the results of the event study with the coefficients on the
interaction terms plotted across birth years for the years 1998-2015. Prior to the
treatment in 2009, the coefficients are fairly stable and all close to zero. Following
the intervention in 2009, however, there is evidence of a downward trend in the
coefficient, consistent with the findings in Table 3.
FIGURE 3. Event Study: Likelihood Ever Breastfed
0.2
0.0
−0.2
2000 2003 2006 2009 2012 2015
Year of birth
Notes: Estimates for the coefficients on the indicator for Yerevan as the region of residence
interacted with the child’s year of birth controlling for year, region, and wave fixed effects. The
ribbon reflects the 95% confidence intervals, and the dashed vertical line represents the start of
the program in 2009. Estimates are normalized to 2008, the year just prior to treatment.
Month of Birth Triple Difference
In addition to the event study, I also examine how the main effect differs by
month of birth using a triple difference specification. We would expect that if the
25
Change in likelihood of ever breastfeeding
decreased prevalence of breastfeeding is associated with the “For You, Women”
program that the largest effects should be seen in March and April, the months in
which the additional health care access is available. Figure 4 depicts the coefficient
of the main variable of interest, Postt × Y erevani interacted with each month of
birth. The dashed vertical line indicates the starting month of the program, March.
While none of the interaction coefficients are statistically significant, the largest
negative coefficient is in April which is consistent with the main findings.
FIGURE 4. Triple Difference Estimates by Month
0.2
0.0
−0.2
−0.4
1 2 3 4 5 6 7 8 9 10 11 12
Birth month
Notes: Estimates for the coefficients of the triple difference: the main variable of interest,
Postt × Y erevani, interacted with each month of birth. Regression includes year, region, and
wave fixed effects. The ribbon reflects the 95% confidence intervals, and the vertical dashed
line represents the beginning month of the intervention each year. Estimates are normalized to
January.
26
Change in likelihood of ever breastfeeding
Gender Heterogeneity
As noted previously, Armenia has a strong preference for sons as evidenced
by the high sex ratio at birth of 113 males for every 100 females (The World Bank,
2017). Thus, it is possible that breastfeeding practices differ across genders as is the
case in other countries with strong son preferences such as India and Pakistan.
To determine if there are any heterogeneous effects of the “For You, Women”
program by gender, I examine separate regressions for the subset of male and
female children. These results are found in Table 4 and Table 5. All specifications
include the full set of controls including all fixed effects, household controls, and
mother controls.
TABLE 4. Heterogeneity by Gender on Months Breastfed
Dependent variable:
Months Breastfed
Male Female Male Female
(1) (2) (3) (4)
Post June08 −0.388 −1.293
× Yerevan (1.335) (0.964)
Post March09 −0.265 −1.249
× Yerevan (1.138) (0.910)
Time FE Yes Yes Yes Yes
Region FE Yes Yes Yes Yes
Wave FE Yes Yes Yes Yes
Household Controls Yes Yes Yes Yes
Mother Controls Yes Yes Yes Yes
Observations 9,388 7,961 9,388 7,961
R2 0.078 0.092 0.079 0.092
Adjusted R2 0.073 0.086 0.073 0.086
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 All standard errors are wild bootstrap standard errors.
Table 4 presents the results of the overall change in the number of months
spent breastfeeding with no breastfeeding coded as zero for male and female
27
children. Here, we see that there is no statistically significant difference in
breastfeeding in response to the program among female versus male children
though the negative point estimate for females is larger than the point estimate
for males. Therefore, it is again possible that true differences are being masked by
the inclusion of both extensive and intensive margin effects in the main outcome
variable in Table 4. Thus, Table 5 looks at the extensive and intensive margins
separately for male and female children.
TABLE 5. Heterogeneity by Gender Extensive and Intensive Margins
Dependent variable:
Extensive Margin Intensive Margin
Male Female Male Female
(1) (2) (3) (4)
Post June08 0.054 −1.051
× Yerevan (1.329) (0.867)
Post March09 −0.053∗∗ −0.045∗∗
× Yerevan (0.030) (0.027)
Time FE Yes Yes Yes Yes
Region FE Yes Yes Yes Yes
Wave FE Yes Yes Yes Yes
Household Controls Yes Yes Yes Yes
Mother Controls Yes Yes Yes Yes
Observations 9,388 7,961 8,699 7,396
R2 0.033 0.025 0.095 0.108
Adjusted R2 0.028 0.019 0.090 0.101
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 All standard errors are wild bootstrap standard errors.
Columns (1) and (2) of Table 5 depict the extensive margin for the male and
female subsamples respectively, while columns (3) and (4) examine the intensive
margin. On the extensive margin, the coefficients of interest for the male and
female subsamples are both negative, significant at the 5% level, and similar in
magnitude to the results from the full sample. Additionally, though the coefficient
28
for the male sample found in column (1) is actually larger in magnitude than that
of the female sample found in column (2), the two are not statistically different.4
These results suggest that, on the extensive margin, there is no differential response
to the program by gender. Exposure to increased maternal health care access is
associated with a decrease in the likelihood of ever breastfeeding for both male
and female children. On the intensive margin, found in columns (3) and (4), the
coefficients on the interaction term PostJune08×Y erevan are insignificant for both
the female and male subsamples, and again the two are not significantly different.5
This indicates that there is also not strong evidence suggesting differences in the
duration of breastfeeding in response to the program based on the gender of the
child.
These results are surprising given the significant son preference in Armenia
and the findings from prior studies suggesting a connection between son preference
and breastfeeding asymmetries between male and female children. There are,
however, various possibilities for why these results might occur. First, it may
just be that the estimates here are not precise enough. These are intent to treat
estimates, and more precise measures of who exactly is taking up the program
may find different results. It is also possible that while this program does seem
to negatively impact breastfeeding, it may positively impact women’s attitudes
toward female children particularly given that the program is focused on women
and women’s health. Finally, the fertility rate in Armenia is low compared to the
4The t-statistic from the regression interacting a male dummy variable with the differences-in-
differences estimator is -1.1899 with a p-value of 0.430.
5The t-statistic from the regression interacting a male dummy variable with the differences-in-
differences estimator is 4.2258 with a p-value of 0.230.
29
countries previously studied in this context.6 This may cause a relaxing of the
biological time constraint for meeting the dual goals of ideal family size and son
preferences proposed by Jayachandran and Kuziemko (2011). This could mean that
women may feel less pressure to cease breastfeeding early to meet these goals.
Robustness
Placebo
I perform a set of placebo regressions to test the robustness of the main
findings. I consider placebo treatments in 2003 and 2005.7 The placebo treatment
dummies replace the variable Postt in the main regression equation and the same
differences-in-differences equation is estimated using the subset of pre-treatment
data. The results of the placebo regressions are found in Table 6. The placebo
regressions look only at the effect on the extensive margin with the outcome
variable ever breastfed since the extensive margin is where the main results are
focused. Hence, I use the second definition of Postt, defining the placebo treatment
group as children born during or after March of the placebo treatment year. The
results of the placebo regressions are similar to the corresponding coefficients from
the event study though smaller in magnitude. None of the interaction terms are
significant, supporting the main findings.
6Over the period studied the average fertility rate in Armenia is 1.66. At the time of the other
studies the fertility rates in India, Pakistan, and Egypt were all at least double this.
72003 and 2005 were chosen to maximize the balance in the number of pre- and post-treatment
observations given the size of the restricted dataset.
30
TABLE 6. Placebo Regressions
Dependent variable:
Ever Breastfed
(1) (2)
Post March03 −0.020
× Yerevan (0.074)
Post March05 −0.005
× Yerevan (0.051)
Full Set FE Yes Yes
Household Controls Yes Yes
Mother Controls Yes Yes
Observations 11,549 11,549
R2 0.029 0.028
Adjusted R2 0.025 0.025
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 All standard errors are wild bootstrap standard errors.
Female Employment
Throughout the study period the overall female employment in Armenia was
increasing. Because female employment has been shown to have a negative effect on
breastfeeding rates, the increase in female employment could be driving the main
results if female employment is growing faster in the capital city than the other
regions of Armenia. Therefore, I estimate a differences-in-differences model to test
the extent to which female employment changes differentially in the capital city at
the time of the program. The regression takes the following form. For woman j, in
survey wave w, living in region r:
EMPjrw = α + β1Postw + β2Y erevanr + β3Postw × Y erevanr + Xjrw
+γr + ωw + εjrw
31
Here EMPjrw is an indicator variable for whether or not the woman is
employed. The time subscript is dropped as the survey asks about current
employment so the year of employment perfectly correlates with the year of the
survey. Xjrw is a set of household and woman control variables that includes the
number of members in the household, the number of rooms in the household, the
woman’s marital status, and the woman’s education level. γr and ωw are fixed
effects for region of current residence and wave of the survey, respectively. If
there is a significant change in female employment in Yerevan coinciding with the
implementation of the program, we would expect β3 to be positive.
TABLE 7. Women’s Employment
Dependent variable:
Work Status
All Women Women in Subsample
(1) (2)
Post Jan09 −0.005 −0.016
× Yerevan (0.046) (0.071)
Full Set FE Yes Yes
Household Controls Yes Yes
Individual Controls Yes Yes
Observations 84,478 13,829
R2 0.093 0.071
Adjusted R2 0.093 0.069
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 All standard errors are wild bootstrap standard errors.
Table 7 presents the results of the above regression. The coefficient on the
interaction term suggests that there is no statistically significant change in female
employment status coinciding with the start of the “For You, Women” program.
Thus, differential changes in female employment between the treatment and control
regions over the course of the study is likely not driving the results found in the
baseline specification.
32
Mechanism
One possible mechanism that could explain these somewhat counterintuitive
results is the documented lack of support for breastfeeding and the promotion
of infant formula in Yerevan medical community. In a recent paper examining
the contributors to low levels of exclusive breastfeeding in Armenia, Demirchyan
and Melkomian (2020) find that a lack of counseling and lactation support,
misconceptions surrounding infant nutrition, and low motivation by doctors
to provide breastfeeding support all contribute to the low levels of exclusive
breastfeeding. Demirchyan and Melkomian also note that in Yerevan specifically,
there is evidence of “hospitals cooperating with pharmaceutical companies to
promote infant formula” as well as an increased likelihood of pre-lacteal feedings.
The infant formula promotion findings are further verified in work by Boban and
Zakarija-Grkovic (2016) and Harutyunyan et al. (2011).
Using data from the Demographic and Health Surveys from 2000, 2005, 2010,
and 2015, I examine hospital practices related to early breastfeeding and infant
formula use among children under six months to substantiate previous findings and
determine how these attitudes and practices could serve as contributors to the main
results. Figures 5 and 6 show that, compared to the other regions of Armenia, in
Yerevan a lower percentage of children were breastfed within the first hour after
delivery and a higher percentage of children were given something other than
breastmilk within the first three days. While neither of these gaps are especially
large, they are both consistent with lower support for breastfeeding in Yerevan
hospitals.
33
FIGURE 5. Percent of Children Breastfed within the First Hour
60%
40% Yerevan
Other Regions
20%
1990 2000 2010
Year of birth
Sources: Author’s calculations from the Armenia DHS 2000, 2005, 2010, & 2015
FIGURE 6. Percent of Children Fed Something Other Than Breastmilk in First 3
Days
20.0%
15.0%
Yerevan
10.0% Other Regions
5.0%
0.0%
1995 2000 2005 2010 2015
Year of birth
Sources: Author’s calculations from the Armenia DHS 2000, 2005, 2010, & 2015
34
Percent of children fed something other Percent of children breastfed within 
       than breast milk in first 3 days after delivery        the first hour after delivery
FIGURE 7. Type of Food Given to Children in First 3 Days
60%
Formula
40%
Milk
Plain Water
Sugar Water
Tea
20% Other
0%
Yerevan Other Regions
Notes: Only children who were fed something other than breastmilk in the first 3 days are
included. The categories are non-exclusionary, so the same child may be included in more than
one bar. 2015 data is excluded because the question was dropped from that wave of the survey.
Sources: Author’s calculations from the Armenia DHS 2000, 2005, & 2010
Further, Figure 7 shows that while infant formula was not the type of food
or drink most often given to children during the first three days after delivery,
hospitals in Yerevan were more than twice as likely to feed children with infant
formula in the first three days compared to other regions of Armenia. Finally,
children under six months old living in Yerevan were slightly more likely than
children in other regions to have been fed infant formula in the past 24 hours
as seen in Figure 8. These data in Figures 7 and 8 are not only consistent
with the previous literature’s findings of greater infant formula promotion in
Yerevan hospitals, but they also provide some evidence for greater infant formula
usage more generally. Overall, this is consistent with a story in which increased
contact with the medical community in Yerevan, which has a lack of support for
35
Percent of children fed each type of liquid 
       during the first three days after delivery
breastfeeding and a documented prevalence of infant formula promotion, reduced
the likelihood of ever breastfeeding for women exposed to the “For You, Women”
program.
FIGURE 8. Percent of Children Fed Infant Formula in Last 24 Hours
20.0%
15.0%
Pre−Period Control
Post−Period Control
10.0% Pre−Period Treated
Post−Period Treated
5.0%
0.0%
Other Regions Yerevan
Sources: Author’s calculations from the Armenia DHS 2000, 2005, 2010, & 2015
Conclusion
Overall, the “For You, Women” program, which aimed to increase medical
access for women in Armenia, is associated with a significant decline in the
likelihood of breastfeeding by almost 5 percentage points. There is no significant
effect on the number of months spent breastfeeding among mothers who chose
to breastfeed. Taken together, this study finds that increased access to maternal
health care has a negative affect on the prevalence of breastfeeding while having no
significant impact on the duration.
Surprisingly, there is no strong evidence of a differential affect of the program
on breastfeeding practices based on the gender of the child as the previous
36
Percent of children under six months fed 
       infant formula in the past 24 hours
literature may have suggested. One potential reason for this discrepancy may be
the low fertility rate in Armenia. It is also possible, however, that my analysis
using the ILCS sample is simply not precise enough to capture gender specific
differences, and further research with more precise data would contribute to our
understanding of heterogeneity in these results.
The decline in breastfeeding that is associated with the “For You, Women”
program is likely driven by low levels of support for breastfeeding in medical clinics
in Yerevan. Therefore, updating practices at clinics providing services for the
women seeking treatment through the “For You, Women” program in line with
WHO recommendations may serve to curb these inadvertent outcomes.
37
CHAPTER III
IMPACTS OF THE RELOCATION PROGRAM ON NATIVE AMERICAN
MIGRATION AND FERTILITY
Introduction
Throughout US history, the United States government has regularly sought to
relocate Native Americans1 both through force and through policy. While some
of these policies are infamous–the Indian Removal Act of 1830 led to between
4,000 and 8,000 Cherokee deaths2–there is relatively little known about the long-
term economic and demographic consequences of more recent relocation policies.
In addition, the economic literature has largely ignored these programs and their
effects.
My paper examines the migratory impacts of one such program known simply
as the Relocation Program.3 The Relocation Program, which ran from 1952 to
1973, offered financial assistance and job counseling to Native Americans who left
tribal land and moved to one of the program’s target cities.4 As a crucial first
step in understanding the lasting impacts of these types of programs, this paper
1Because of the broad scope of the Relocation Program, I use the term Native America
throughout this paper to refer to the original inhabitants of North America generally, but I
acknowledge that the term is imprecise and disputed. In cases where the data or policy use
different terminology, I use the term that is consistent with the data and policy language.
2The 4,000 to 8,000 death estimate refers solely to deaths on the Trail of Tears. Historical
estimates place total population losses at greater than 10,000 (Thornton, 1984).
3The program was renamed the Employment Assistance Program when Congress expanded the
program through the Indian Relocation Act of 1956, but for consistency will be referred to as the
the Relocation Program or Relocation throughout this paper.
4More information about the history and details of the Relocation Program are provided in the
Background section.
38
analyzes the implications of the Relocation Program on the migratory outcomes of
the Native American population in the second half of the 20th century.
As far as I am aware, this is the first paper to estimate the causal effect of the
Relocation Program on Native American migration. Previous work in anthropology
and sociology has examined factors associated with migration through the program
(Ablon, 1965; Chadwick & White, 1973) and the demographic characteristics of
participants as well as economic and assimilation outcomes for those who were
part of the program (Ablon, 1964; Gundlach, Reid, & Roberts, 1977; Gundlach
& Roberts, 1978; Price, 1968; Sorkin, 1969, 1971, 1978). This paper extends the
literature by establishing a causal link between the program and Native American
migratory outcomes, providing valuable insights into the effects of such relocation
programs on migration and setting the stage for examining the causal impacts of
the Relocation Program on other outcomes.
Additionally, this paper establishes a basis for analyzing the long-run effects
of migration programs with the stated aim of expanding economic opportunities
for migrants, similar to programs such as the Moving to Opportunity program
in the US or the Transmigration Program in Indonesia. Thus, this paper adds to
the growing literature on the impacts of these employment targeting migration
programs (Chetty & Hendren, 2018a, 2018b; Chetty et al., 2016; Elmhirst, 2002;
Katz et al., 2001; Kling et al., 2007; Pollack et al., 2019, and more). Of particular
importance is the added value of the Relocation Program’s timing. The Moving
to Opportunity literature has specifically found that the main impacts of these
types of programs are realized among the young children of movers. The Moving to
Opportunity program, however, is still relatively recent and some of the longer term
or second-generation outcomes are yet to be determined. Because the Relocation
39
Program began almost 70 years ago, the long-term outcomes for movers, as well
as the movers’ children, have been realized and are able to be studied. Therefore,
establishing the effects of the program on migration provides a first step for further
research on long-term outcomes that are unable to be examined in more recent
programs. My analysis of the fertility rates for the second-generation of program
participants provides some preliminary evidence of long-run outcomes of the
program.
Finally, this paper also adds to the discussion of Indigenous population
migration, particularly regarding migration off reservations. There have been a
number of papers that have examined the unique factors that contribute specifically
to the migration outcomes of Indigenous populations (Cebula & Belton, 1994;
Davis et al., 2017; Gardner, 1994; Geib, 2001; Huskey et al., 2004). My research
contributes to this literature by analyzing how this type of jobs-related relocation
program affects the migration patterns of Native Americans.
I employ a generalized difference-in-differences approach to answer the
question of how the Relocation Program affected Native American migration
patterns. For my main specification, I use decennial census data to compare the
American Indian and Alaska Native population in the relocation target cities to
populations in all other metropolitan statistical areas (MSAs). I find that there
was a large and statistically significant increase in the Native American population
in the program’s target cities relative to the other MSAs. The baseline estimates
suggest that there was an increase in the Native American population of almost
12,000 individuals on average in the target MSAs relative to the non-target MSAs
between 1960 and 1990. The magnitude of this estimate is particularly stark given
that the median MSA had a Native American population of less than 100 (with a
40
mean population of about 1,000) in 1950, just prior to the start of the program.
The effect of the program is consistent in magnitude and significance across various
subsets of the main control group. Further, the results remain significant when
I apply a doubly robust difference-in-differences specification using propensity-
score methods though the magnitude is somewhat attenuated. To provide some
insight into the magnitude of this impact on tribal land areas, I also calculate
counterfactual population estimates for tribal land areas absent the program.
In addition to the main migration results, I also examine the impacts of
the Relocation Program on second-generation fertility. I analyze fertility for a
number of reasons. Firstly, fertility is a fundamentally important question in
this setting where the government has historically played a role in reducing the
Native American population and where there is a concurrent program of forcible
sterilization at government run clinics on reservations.5 Further, second-generation
fertility provides key insights into the long-run impacts of the program on the
primary goal of relocating the Native American population. Differential fertility
rates across different locations can alter the spatial distribution of populations
over time and is therefore vital to understanding the lasting impacts of relocation.
Lastly, fertility can also provide some noisy indication about the long-run impacts
of relocation on other outcomes, such as economic well-being and assimilation, both
of which have known associations with changes in fertility in certain cases.
I find that there is a gap in the general fertility rate among Native American
women living in urban areas in the 1990s relative to Native American women who
live in tribal land areas during this period. Specifically, the data show that the
5The forced sterilization of Native American women in the 1970s is discussed further in the
fertility results section as it provides important context for the study of fertility in this time
period.
41
fertility rate for Native American women living in cities was almost half that of
those living in the rural parts of states with large tribal land areas. While this
finding does not imply causality, it suggests changes in fertility outcomes among
the children of those who were affected by the Relocation Program. This result
could have important implications for the future of the Native American population
distribution across the US and for other programs that target or induce migration.
Background
The Relocation Program began in 1952 during what is known as the
termination era of US tribal policy. The joint policies of termination and relocation
aimed to assimilate the Native American population and move them off tribal lands
by terminating tribal trust relationships with the US federal government and urging
Native Americans to relocate to urban centers, away from reservations. While this
paper focuses on the effects of the latter policy, I also include a brief outline of the
termination policy as the two were inextricably linked.
Relocation
The Relocation Program originally began as part of an $88 million ten-
year rehabilitation act passed by Congress in April of 1950 which aimed to aid
the Navajo and Hopi Reservations after an extreme summer drought and severe
winter at the end of the 1940s devastated the area. In addition to providing
for a number of reservation development programs, the bill also allocated funds
to provide financial aid to individuals for finding off-reservation employment.
Specifically, the bill provided direct employment services for individuals from the
Navajo and Hopi Reservations to Denver, Salt Lake City, or Los Angeles. The
42
program was almost immediately expanded by the Bureau of Indian Affairs6 (BIA)
to allow for individuals from any US tribe to participate. During this period,
the BIA was run by Dillon S. Myer. It is important to note, that prior to being
named Commissioner on Indian Affairs in 1950, Myer served as the director of the
War Relocation Authority, the agency charged with the internment of Japanese
Americans during WWII. He brought many of his colleagues from the War
Relocation Services with him to work at the BIA in implementing the Relocation
Program on a national scale. The first relocatees of what became the Relocation
Program arrived at the newly opened relocation office of Chicago in February of
1952 (Fixico, 1980).
The Relocation Program grew throughout the early 1950s. In addition to
the relocation offices in Denver, Salt Lake City, Los Angeles, and Chicago, offices
were added in Cleveland, Cincinnati, Dallas, St. Louis, and San Francisco. The
BIA chose these cities for relocation offices because of their “diversified, abundant
industrial employment” (Madigan, 1956). The first step in the relocation process
was to apply at a local BIA area offices. There were no demographic restrictions,
such as age or gender, on who could apply, although Madigan (1956) suggests that
acceptance was based on the relocation officer’s expectations of an applicant’s
success in the program. Records show that drunkenness, physical or mental
health issues, marital problems, or arrests were all grounds for disqualification of
applicants. There were also complaints of prejudice among the relocation officers
who were mostly non-Native American (Fixico, 1980).
6The Bureau of Indian Affairs was established in 1824 under the Department of War but
became part of the Department of the Interior after its establishment in 1849. The BIA acts as
a liaison between the federal government and sovereign tribes. The president and Congress are
responsible for creating federal tribal policy, and the BIA is responsible for administering the
policies.
43
Upon acceptance to the program, a relocatee and her family received one-way
tickets to one of the relocation cities in the relocatee’s top choices as well as money
for the journey. In their relocation city, the relocatee met the relocation officer
who would assist relocatees in finding temporary housing and provide financial
assistance for the first month, including money for rent and some money for clothes
and food. Additionally, the relocation officer would provide the relocatee with
“intensive employment counseling” (Madigan, 1956).
Criticisms of the program, including complaints of slum housing, temporary
and unstable job placements, and inadequate preparation for cultural differences,
pushed Congress to reassess the program in 1955 and 1956. Congress responded
with the passage of Public Law 959–also known as the Indian Relocation Act of
1956–which consisted of two additional components: adult vocational training and
on-the-job training. The adult vocational training program trained participants
in industrial skills such as plumbing or steel work and targeted individuals
between the ages of 18 and 35. The on-the-job training program focused on finding
employment opportunities near reservations where individuals could be trained in a
specific industry while working. The BIA would subsidize wages of trainees during
their training period with the possibility of the trainees gaining full employment
upon completion of the training. These training programs were meant to serve as
a precursor to relocation in an attempt to better prepare relocatees for life in the
relocation cities (Fixico, 1980).
Termination
Termination policy consisted of ending the trust relationship between
tribes and the US federal government. This meant that if a tribe was officially
44
terminated, then any land that belonged to the tribe would be released from its
trust status, which generally resulted in the land being appraised and sold to
the highest bidder with the proceeds from the sale distributed to the members
of the terminated tribe. Additionally, Native Americans who were part of the
terminated tribe would be subject to state and federal laws and taxation from
which they were previously exempt, and all special federal programs at the tribal
and individual level were discontinued. Essentially, termination meant the end
of tribal sovereignty as without land, tribal governments had nowhere to exert
their jurisdiction (Wilkinson & Biggs, 1977). The stated aim of termination was
to “make Indians within the territorial limits of the United States subject to the
same laws and entitled to the same privileges and responsibilities as are applicable
to other citizens of the United States, to end their status as wards of the United
States and the grant them all of the rights and prerogatives pertaining to American
citizenship.7” Advocates of termination had two main motivators. First, they
wanted to “free” Native Americans from the federal trusteeship which proponents
of termination believed was paternalistic and kept Native Americans dependent
on government aid. Secondly, terminationists wanted to reduce the amount of
federal funds allocated to tribal programs. Those opposed to termination cited
the desire of many Native Americans to keep reservations intact to preserve tribal
homelands and to avoid the loss of sovereignty that comes with loss of communal
property. Additionally, termination opponents also condemned the meager payment
on assets in the forced sale of tribal land and suggested that high poverty rates
on reservations were evidence of continued need for federal education, health, and
welfare services (Sorkin, 1971). One of the most critical opponents of termination,
7House Concurrent Resolution no. 108
45
former Commissioner of Indian Affairs, John Collier, asserted that termination
was truly motivated by greed for land and natural resources owned by the tribes
(Fixico, 1980).
In 1953, House Concurrent Resolution no. 108 was passed. It called for the
termination of roughly 100 tribes throughout California, Florida, New York, Texas,
Wisconsin, Oregon, Utah, Oklahoma, South Carolina, and Nebraska, affecting over
10,000 Native Americans and over 1 million acres of tribal land. Tribal consent was
not deemed necessary, but tribes were nonetheless often coerced into providing
consent. At the same time, Congress passed another law aimed at termination
which granted states jurisdiction over civil and criminal offenses on tribal land
in five states, while allowing for all other states to adopt similar practices if they
chose. Strong opposition to these termination policies and poor outcomes among
early terminated tribes led to a reversal of policy in the 1960s, though many of the
terminated tribes did not regain tribal status until much later (Wilkinson & Biggs,
1977).
Data
Migration
The main data I use to measure migration comes from the U.S. Decennial
Census. Specifically, I use population estimates for the American Indian and Alaska
Native (AIAN) population by county and by metropolitan statistical area (MSA).
For the years 1910-1940, I calculate county-level population by race, by aggregating
the Integrated Public Use Microdata Series (IPUMS) complete count U.S. census
microdata. For the years 1950 onward, county-level population estimates for the
American Indian and Alaska Native population and the total population come from
46
U.S. Decennial Census Publications which have, in part, been digitized by Gardner
and Cohen (1992) and Haines (2010). All of the county-level population estimates
are corrected for changing county borders following Eckert et al. (2020). The
final county population estimates follow 1990 county borders. I then match these
counties to the 1990 U.S. Census MSAs to create MSA-level population counts.
The final dataset contains population estimates for the American Indian and Alaska
Native and total population for each MSA in the United States for every decennial
year from 1910-1990.
In addition to the county-level population data, I also use the IPUMS full
count data from 1930-1940 and the 1-5% samples from 1950-1980 to determine
migration patterns for a small number of Native American individuals. Specifically,
I use the place of birth and current place of residence variables to document where
individuals who identified as American Indian or Alaska Native migrated during
this period.
Fertility
To examine fertility outcomes, I use the National Center for Health Statistic’s
(NCHS) Vital Statistics Natality Birth Data along with the Survey of Epidemiology
and End Results (SEER) U.S. state and county population data. The natality data
contains all births in the United States from 1985-2002 with information on the
year of the birth as well as the mother’s race, current place of residence, and age.
The SEER population data gives population estimates at the county level by race,
sex, and single year of age from 1990-2018. I combine these data from 1990-2002 to
create general fertility rates by place of residence, year, and race.
47
I break up the fertility data into three types of locations: target MSAs, other
MSAs, and the non-MSA parts of tribal states. I designate a state as “tribal” if
in all years prior to 1952, at least 1% of the population identified as American
Indian or Alaska Native in the decennial census.8 Because the place of residence
variable is censored in the publicly available natality data for counties with less
than 100,000 people, I am unable to examine the fertility rates specifically for tribal
lands. Instead, I use the tribal states as a proxy for people living on tribal lands as
most of the Native American population lived on tribal land prior to the relocation
program.
The general fertility rate equals the total number of births for a given location
category in a given year divided by the total female population aged 15-45 for that
location category and year multiplied by 1000. Therefore, the general fertility rate
represents the total births per 1000 women aged 15-45 for the given location type
in a given year.
Methodology
Migration
My main approach for answering the question of how the Relocation Program
affected migration patterns to urban areas is a generalized difference-in-differences
specification. Here, the two dimensions of the difference-in-differences approach
are (1) geography: whether or not the MSA was targeted by the program and (2)
time: when the MSA was targeted (if it was ever targeted). The main estimating
equation takes the following form.
8These states are Arizona, Montana, Nevada, New Mexico, Oklahoma, and South Dakota
48
For MSA, m, in decennial year, y:
AIANPopulationmy = β(TargetMSAm × Posty) + γm + δy + εmy
The dependent variable is the total Native American population in MSA,
m, in a decennial year, y. For the main analysis, I include all MSAs. Thus, MSAs
not targeted by the Relocation Program act as the control group. Note that
non-MSA areas are dropped. In the map in Figure 9, the target cities appear in
orange, while the main control group includes both the green areas (MSAs near
reservations) and purple areas (MSAs that are not near reservations). The tribal
lands and Oklahoma tribal statistical areas9 are shown in dark blue and light blue,
respectively, for reference.
The independent variable of interest is the interaction between TargetMSAm
and Posty. TargetMSAm is a dummy variable indicating that the MSA, m, is
one of the cities targeted by the Relocation Program.10 Posty is a dummy variable
indicating that the decennial year, y, is after the start of the program in 1952. This
means that for each of the decennial years from 1960 onwards, Posty takes on a
value of 1, while for each of the decennial years from 1910-1950, Posty is 0. The
interaction of these dummies gives us the program effect. γm and δy represent fixed
effects for the MSA and decennial year, respectively. εmy is the error term. The
errors throughout the primary analysis are clustered at the MSA.
9Oklahoma Tribal Statistical Areas represent former reservation lands that existed prior to
Oklahoma’s statehood in 1907. The borders were determined by the U.S. Census Bureau in
consultation with the American Indian tribes residing in Oklahoma.
10The target cities used for this analysis are Chicago, Cincinnati, Cleveland, Dallas, Denver, Los
Angeles, San Francisco, San Jose, Salt Lake City, and St. Louis. An official list of target cities for
the Relocation Program is not fully identified among historians. For this paper, I utilize the most
consistent list of cities I was able to find in the available historical literature and administrative
records.
49
FIGURE 9. Map of Relocation Program Target MSAs
Target MSA Near MSA Far MSA Tribal Land Oklahoma Tribal Statistical Area
Notes: MSA borders are based on the 1990 census and tribal land borders are based on 2018
borders.
Sources: MSA: NHGIS; Tribal Lands: Bureau of Indian Affairs (BIA) Branch of Geospatial
Support (BOGS) New Land Area Representation GIS dataset 2018
Identification for determining a causal effect of the program on migration
relies on the assumption that absent the treatment, the control and treatment
MSAs would have similar trends in Native American population counts. One
way that this assumption is often supported is by showing parallel trends of
the treatment and control units in the pre-treatment period. Figure 10 shows
the trends in the Native American population for the three types of MSAs: the
target MSAs, near tribal land MSAs, and all other MSAs. The colors in Figure 10
correspond to the map above as a reference for which MSAs fall into which of the
three MSA types. Figure 10 suggests that prior to the introduction of the program,
50
all three types of MSAs had similar average levels (near zero) and trends (no
growth) in the Native American population. After the introduction of the program
in 1952, we see that the average Native American population in target MSAs
quickly grows substantially larger than in the other two types of MSAs, providing a
first indication that the program did have an effect on Native American migration
patterns. To more carefully examine whether the parallel trends assumption likely
holds, I also conduct an event study which supports the assumption that prior to
the implementation of the Relocation Program, target MSAs and all other MSAs
had similar trends in Native American population counts.
FIGURE 10. Native American Population Trends by MSA Type
30,000
MSA Type
20,000
Target MSAs
Near MSAs
Far MSAs
10,000
0
1920 1940 1960 1980
Notes: Trends for the target and non-target MSAs from 1910-1990. Non-target MSAs are broken
into two groups based on proximity to tribal land. The points represent average MSA Native
American population. Best fit lines are estimated by MSA type for the pre and post periods.
The dotted lines correspond to the start and end of the Relocation Program in 1952 and 1973,
respectively. The standard errors are clustered at the MSA.
Sources: Author’s calculations from Decennial Census Data
51
Average Native American Population
Using all other MSAs as the control group, however, has its drawbacks as
the cities targeted by the program were mainly located in the West and Midwest,
where there is more tribal land, while the majority the MSAs in the US are located
in the South and on the East Coast. The geographic concentration of the target
MSAs relative to the set of control MSAs could be a problem given that the start
of the program coincided with the end of WWII which also may have been related
to Native American migration to urban centers. If there was greater migration after
the end of the war to cities near tribal lands, then this migration to target cities
could in part be driven by post-WWII migration patterns rather than program
effects. Therefore, I employ several more restrictive control groups in additional
specifications. The first alternative control group includes only MSAs located near
tribal land (MSAs colored green in Figure 9).11 Near tribal land MSAs are chosen
as they offer a plausible lowest cost alternative to migration through the Relocation
Program as they were likely easier and cheaper to get to than other MSAs. One
drawback of using this alternative control group is that with the expansion of
the program in 1956 to include vocational and on-the-job training in areas closer
to tribal land, it’s possible that people chose to stay in these near tribal land
cities where the job trainings took place rather than participating in the direct
employment program after their training period ended. Thus, these MSAs may not
be unaffected by the program. Therefore, the second alternative control group I
consider is far from tribal land MSAs. This control groups includes only the MSAs
that are not included in the near tribal land MSA group (shown in purple in Figure
9). This control group is considered because it is the least likely control group to
be at all affected by the actual treatment from the program. The third alternative
11An MSA is considered a near tribal land MSA if it is the closest MSA to at least one tribal
land area measuring from centroid to centroid.
52
control group includes only the MSAs that are in the same state as a target MSA.
This sample includes all MSAs in California, Colorado, Illinois, Missouri, Ohio,
Texas, and Utah. Finally, all but one of the target MSAs were among the top
fifty most populous cities in the US at the time of the program’s development.
The fact that the target cities were more populous on average than the group of
all other MSAs could have important consequences for my analysis since I am
specifically looking at population changes. Therefore, the final alternative control
group includes only the top fifty most populous MSAs in 1950.
Fertility
To study the impacts of the Relocation Program on fertility, I compare
second-generation fertility rates for women living in target cities to those of women
living in other urban areas and the rural parts of tribal states. I focus specifically
on second-generation outcomes for two reason. First, the previous migration
program literature has suggested that the economic and health outcomes of such
programs are greatest for those of the second-generation or the children of movers.
Secondly, the publicly available data on births in the US that is disaggregated by
geographic region and race is only available starting in 1990. Given the time of the
program, the sample of women who are of childbearing age from 1990 onward is
mainly second-generation women.
The main measure of fertility that I use is the generalized fertility rate which
gives the number of births per 1000 women of childbearing age (15-45). Specifically,
I calculate the generalized fertility rate for each year from 1990 to 2002 for three
racial groups, namely, white, black, and Native American, and by three types of
residence: target MSA, non-target MSA, and rural parts of tribal states. The rural
53
parts of tribal states area of residence includes all areas of tribal states that are not
included in an MSA in 1990. I compare the general fertility rate across these three
racial groups and the three types of residence to determine the unique fertility
patterns among Native American women in target MSAs. While the fertility
analysis does not provide causal estimates for the affect of the program on fertility
outcomes, it does provide descriptive evidence into how fertility rates differ between
these groups, providing valuable insight into how the program may impact long-
term health and demographic outcomes.
Results
Main Approach
The results of the main analysis are presented in Table 8. Column 1 shows
the results of the simple OLS specification without fixed effects, and column 2
adds year and MSA fixed effects. Both columns show a positive and statistically
significant increase in the Native American population associated with the
implementation of the Relocation Program. The estimates suggest that, on
average, between 1960 and 1990, the Relocation Program increased the Native
American population in target cities by almost 12,000 individuals relative to all
other MSAs. This constitutes a major shift in the Native American population of
these target cities considering that prior to the program the average MSA had a
Native American population of about 1,000 people and the median MSA, less than
100 people.
54
TABLE 8. Baseline Native American Population Response to the Relocation
Program
Dependent variable: Native American Population
(1) (2)
Post 1,446∗∗∗
(186)
Target MSA 432
(362)
Target MSA × 11,856∗∗ 11,866∗∗
Post (5,394) (5,392)
Constant 197∗∗∗
(40)
MSA FE No Yes
Year FE No Yes
Observations 2,526 2,526
R2 0.167 0.524
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01. The Native American population data is the total number
of individuals registered by the U.S. Census Bureau as American Indian or Alaska Native for each
decennial census year 1910-1990 by MSA. Standard errors are clustered at the MSA.
As a supplement to the main approach, I also run an event study analysis
to examine how this average effect was distributed throughout the course of the
program and to further establish the parallel trends assumption. The results of the
event study are shown in Figure 11, which plots the interaction of the TargetCitym
variable with each decennial year from 1910-1990. 1950 is the reference year. As
noted in the methodology section, it is clear that prior to the implementation of the
program, trends in Native American population for target and non-target MSAs are
very similar. After the program was implemented in 1952, we can see that there is
a sharp increase in the number of Native American individuals living in the target
55
cities relative to all other MSAs. This increase in the target MSA Native American
population relative to the other MSAs continues through 1970 and 1980. After
1980, the effects of the program level off. This is not surprising as the program
ended in 1973 (17 years prior to the 1990 census).
FIGURE 11. Estimates of the Relocation Program’s Migration Effects over Time
30,000
20,000
10,000
0
1920 1940 1960 1980
Notes: The figure shows the coefficent estimates for the interaction of the target city dummy and
each of the decennial years. The estimating equation includes MSA fixed effects, and the standard
errors are clustered at the MSA. The gray ribbon represents the 95% confidence interval. The
dotted lines represent the beginning and end of the program in 1952 and 1973, respectively
Finally, I also estimate the main regression with a number of different control
groups as described in the methodology section. Table 9 shows the results of these
regressions. For column 1, the control group is just the MSAs that are near tribal
lands. These are the MSAs shown in green in Figure 9. For column 2, the control
group is just the MSAs that are far away from tribal lands, as shown in purple in
Figure 9. The control group in column 3 includes only the MSAs that are located
in states with target MSAs. Column 4 uses a control group that consists of only
56
Coefficient Estimate
the 50 most populous MSAs around the time the target locations were being chosen
in 1950. All four columns produce estimates that are similar in significance and
magnitude to the baseline specification, further solidifying the main results of the
primary approach. Notably, column 2, which consists of the control group that is
least likely to have been partially treated, has the largest point estimate.
TABLE 9. Native American Population Response to the Relocation Program,
Alternative Control Groups
Dependent variable: Native American Population
Near MSAs Far MSAs Target State MSAs Top 50 MSAs
(1) (2) (3) (4)
Target MSA × 10,780∗∗ 12,394∗∗ 12,043∗∗ 10,206∗
Post (5,427) (5,3956) (5,437) (6,058)
MSA FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
Observations 882 1,725 675 448
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01. The Native American population data is the total number
of individuals registered by the U.S. Census Bureau as American Indian or Alaska Native for each
decennial census year 1910-1990 by MSA. Standard errors are clustered at the MSA.
Additional Robustness Analysis
One potential threat to identification in the main analysis would be if the
program specifically targeted cities that were growing differentially at the time
of the program’s implementation for reasons other than the program itself. To
examine this concern, I compare the change in Native American population to
that of the non-Native American population in the target cities over the period of
1910-1990. Figure 12 shows the population trends in the target cities for the Native
American population on the left and the non-Native American population on the
57
right. These graphs suggest that even with the large population growth in the
target cities over the time period, there does seem to be differential trends for the
Native American population following the start of the Relocation Program. This
suggests that the changes in the Native American population in the target cities
corresponding with the time of the Relocation Program seen in the main analysis
are likely indeed a result of the program and not other attributes specific to the
target cities from the 1950s onward.
FIGURE 12. Target City Population Trends by Race
Native American Non−Native American
75,000
10,000,000
50,000
5,000,000
25,000
0 0
1920 1940 1960 1980 1920 1940 1960 1980
Notes: The bold lines estimate linear trends in MSA population by race for the pre and post
periods. Background lines show individual MSA population totals by race for each of the target
cities. The dotted lines depict the start and end of the Relocation Program in 1952 and 1973,
respectively.
Sources: Author’s calculations from Decennial Census Data
Given that one of the program’s main aims was to move Native Americans
off of tribal lands to urban centers, I also compare the average Native American
population in target MSAs to the Native American population on tribal lands.
Because there is no data available on Native American population counts by
reservation throughout this time period, I examine Native American population
58
Average Population
FIGURE 13. Native American Population Trends for Target MSAs versus Tribal
Land
3000
2000
Target MSA Counties
Tribal Land Counties
1000
0
1920 1940 1960 1980
Notes: Trends in average county-level Native American population for the target MSA and tribal
land counties from 1910-1990. Best fit lines are estimated by county type for the pre and post
periods. The dotted lines correspond to the start and end of the Relocation Program in 1952 and
1973, respectively.
counts at the county level. Specifically, I compare average Native American
population counts in counties that are in target MSAs to those counties that
contain tribal land as shown in Figure 13. Figure 13 suggests that the increase in
the Native American population in the target MSAs is so large that the average
county-level target MSA Native American population may even exceed that of
tribal land counties by the end of the relocation program.
I also verify my main results by using a doubly robust difference-in-differences
specification. The second level of robustness relies upon the propensity-score
methods literature (Rosenbaum & Rubin, 1983)—i.e., inverse probability weighting.
Using logistic regression, I model the probability of an MSA being targeted by
the program as a function of total population, percent of the population that is
59
Average Native American Population
non-white, percent change in the total population since 1940, percent change in
the non-white population since 1940, the male and female employment rate, and
the minimum distance to a tribal land area. Using the logit regression estimates,
I calculate inverse probability of treatment weights using the formula: wi =
TargetMSAm + 1−TargetMSAm , and apply these weights to my baseline difference-in-
pm 1−pm
differences estimating equation to form a doubly robust estimator (Hirano et al.,
2003). The results of the doubly robust inverse probability weighted regression
are found in Table 10. While the estimate of the coefficient of interest on the
interaction of TargetMSA × Post is smaller than the baseline estimate, it is well
within the traditional confidence intervals and still suggests substantial movement
of the Native American population as a result of the Relocation Program.
TABLE 10. Doubly Robust DiD using Inverse Probability of Treatment Weighting
Dependent variable: Native American Population
Target MSA × 6,714∗∗∗
Post (2,273)
MSA FE Yes
Year FE Yes
Observations 954
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01. The Native American population data is the total number
of individuals registered by the U.S. Census Bureau as American Indian or Alaska Native for each
decennial census year 1910-1990 by MSA. Standard errors are clustered at the MSA.
Finally, up to this point, my analysis has focused on total population counts
rather than specific migration variables. This is mainly due to a lack of available
data on Native American migration over the period during which the program ran.
There is, however, a limited amount of data available in the IPUMS samples which
I use to examine migration in Figure 14. Here I explore the place of residence by
60
year for Native Americans born in one of the six states with the highest Native
American population prior to the start of the program in 1952.12 Figure 14 shows
the average percent of individuals who were born in a tribal state that are living
in a target MSA versus the average percent living in a non-target MSA at the
time of the census. While migration out of tribal states seems to be increasing for
both groups in 1960 and 1970, there is a much higher average percent of people
moving to the target MSAs than non-target MSAs. Overall, this supports the main
findings that Native Americans were migrating to target cities more than other
MSAs during the time of the Relocation Program.
FIGURE 14. Migration Trends for Native Americans Born in Tribal States
0.60%
0.50%
0.40%
0.30% Target MSAs
Other MSAs
0.20%
0.10%
0.00%
1940 1960 1980
Notes: Lines show the average percent of the Native American population that was born in one
of the six tribal states that now live in an MSA. The MSAs are broken into two types based on
whether or not the MSA was a target MSA.
Sources: Author’s calculations from IPUMS Data
12The tribal states here are the same those chosen for the fertility analysis. As a reminder these
states are those with at least 1% of the population identifying as American Indian or Alaska
Native in the decennial census for all years 1910-1950.
61
Average percent of Native American population 
       born in a tribal state that moved to an MSA
Fertility Differences
The fundamental aim of the Relocation Program was to alter the distribution
of Native Americans in the US and to promote assimilation. In the preceding
sections, I have documented the cross-sectional effects of the program, which
show a shift in the Native American population from rural reservations to urban
centers targeted by the Relocation Program. In this section, I examine the long-run
fertility differences among Native American women living in target MSAs versus
tribal states.
Before I present my fertility results, it is important to note that in the
1970s there was widespread forced sterilization of Native American women
living on reservations. Because the Indian Health Services had such singular
control over healthcare on tribal lands, Native American women were particularly
vulnerable to this practice during this period. Estimates suggest that one in four
Native American women of childbearing age were forcibly sterilized at this time
(Lawrence, 2000). Given that I see even lower rates of fertility for Native American
women living in urban areas compared to tribal land ares, this period of forced
sterilization does not change how I understand these fertility data as they relate
to the Relocation Program. It does, however, provide important context for this
discussion.
62
FIGURE 15. General Fertility Rate by Race and Current Residence
100
Line Color
White
Black
80 Native American
Line Type
Other MSA
60 Rural Tribal State
Target MSA
40
1990 1993 1996 1999 2002
Notes: The general fertility rate represents the number of births per 1,000 women age 15-45. The
lines above show the general fertility rate by race for each of three types of current residence:
target MSA, other MSA, and rural tribal state.
Sources: Author’s calculations from NVSS Natality and SEER population data
Figure 15 shows the general fertility rate for Native American women living
in target MSAs, non-target MSAs, and the rural parts of tribal states from 1990 to
2002. Figure 15 suggests that there is a large gap in the fertility rate between those
individuals living in an MSA versus those living in non-MSA areas of tribal states.
This provides some preliminary evidence of how the Relocation Program may have
contributed to shifting fertility outcomes for Native American women.
63
General Fertility Rate
FIGURE 16. General Fertility Rate by Age
15−20 21−25 26−30
200
150
100 Line Color
White
50
Black
0 Native American
31−35 36−40 41−45
200 Line Type
Other MSA
150 Rural Tribal State
Target MSA
100
50
0
1991 1996 2001 1991 1996 2001 1991 1996 2001
Notes: The general fertility rate by age represents the number of births to women within the
specified age range per 1,000 women in the age group.
Sources: Author’s calculations from NVSS Natality and SEER population data
Figure 16 disaggregates the fertility rate by age and shows that the gap is
largest for women ages 21-25, with the gap shrinking steadily in the older age
groups. This suggests that Native American women living in cities are likely
both delaying fertility as well as having less children overall. Given the timing of
the program, these women whose fertility I am examining here can most likely
be considered part of the second generation of movers (i.e., the children of the
original movers who were part of the program). Further research into the first
generation’s fertility outcomes and additional heterogeneity analysis may help
64
General Fertility Rate
explain the mechanisms driving the disparity in fertility rates seen here.13 The
fact that these differences in fertility exist in the generation after the Relocation
Program, however, suggests long-term, persistent demographic effects from the two-
decade long program.
Tribal Land Population Counterfactuals
To better understand the magnitude of these migration and fertility
effects on the tribal land population, I calculate a set of back-of-the-envelope
counterfactual population estimates for the tribal land areas. I consider four
possible counterfactual populations:
(1) Adjusted Migration, Unadjusted Birth Rate: the tribal land population
had those induced to migrate by the program remained on tribal land with no
adjustment to the birth rate for the period of forced sterilization
(2) Adjusted Migration, Adjusted Birth Rate: the tribal land population had
those induced to migrate by the program remained on tribal land and had
there been a 33% increase in the birth rate on tribal lands after 1970 to correct
for the estimated one quarter of women forcibly sterilized
(3) Unadjusted Migration, Adjusted Birth Rate: the tribal land population
with no migration adjustment but with a 33% increase in the birth rate on
tribal lands after 1970 to correct for the estimated one quarter of women
forcibly sterilized
(4) Unadjusted Migration, Unadjusted Birth Rate: the tribal land population
with no migration adjustment and with no adjustment made to the birth rate
13A lack of available Native American population data disaggregated by sex and age precludes
me from analyzing these questions here.
65
I create this fourth counterfactual population estimate with no adjustments
for migration or forced sterilization as a comparison group for the other
three population estimates rather than using the census population estimates
because there is a substantial increase in individuals identifying as “American
Indian/Alaksa Native” in the census during this period. There is no evidence
of a differential increase in identifying as Native American across my treatment
and control MSAs, so this should not cause any problems for the identification of
my main findings. It does, however, make it hard to compare my counterfactual
population estimates. Thus, I will instead be using the population described in (4)
above as a estimate of what the population growth on the tribal lands would have
looked like under the actual historic conditions.
To estimate the counterfactual Native American population living on tribal
land, I use the following population growth model:
For decennial year, y:
AIANPopulationy+10 = AIANPopulationy × (1 + r)10 +Migrationy+10
Here, AIANPopulationy is the Native American population living on tribal land
in decennial year y and AIANPopulationy+10 is the Native American population
living on tribal land in the following decennial year, 10 years after y. The variable r
is the population growth rate. r is equal to the per capita birth rate minus the per
capita death rate for Native Americans living in the tribal states described in the
data section. Finally, Migrationy+10 is the estimated total migration in decennial
year y + 10 based on the event study estimates in Figure 11. I calculate total
migration for each decennial year by multiplying the coefficient by the number of
66
treated units and subtracting the previous decennial year coefficient as the event
study coefficients are cumulative.
FIGURE 17. Counterfactual Native American Population on Tribal Land
1,500,000
Adjusted Migration
  and Unadjusted Birth Rate
Adjusted Migration
1,000,000   and Adjusted Birth Rate
Unadjusted Migration
  and Adjusted Birth Rate
Unadjusted Migration
  and Unadjusted Birth Rate
500,000
1950 1960 1970 1980 1990
Notes: The four models in the figure match the descriptions at the beginning of this section.
Birth and death rates are based on births and deaths in the tribal states as described in the data
section. Population estimates are based on the Native American population living on tribal land.
Sources: Author’s calculations from NVSS Natality data and author estimates of decennial
migration induced by the Relocation Program.
As you can see in Figure 17, there would have been a substantially larger
Native American population living on tribal land in 1990 if individuals had
remained on tribal lands rather than migrating as a result of the Relocation
Program. My population growth model suggests that absent the Relocation
Program, there would be over 250,000 additional Native Americans living on tribal
lands in 1990. Thus, in addition to the increased population from returning the
roughly 100,000 estimated migrants back to the tribal land area, there is a further
increase of 150,000 Native Americans living on tribal lands associated with the
tribal land population’s natural growth rate. Moreover, this additional population
67
Native American Population
would have been even more substantial if the forced sterilization in the 1970s had
not decreased the birth rate. In fact, model 2 suggests that absent these factors,
the Native American population living on tribal lands in 1990 would have been
almost double that of the unadjusted population estimates in model 4.
Conclusion
This paper shows how government-sponsored migration programs can
affect the migration patterns of targeted populations and potentially impact
long-run fertility. The Relocation Program had large and significant impacts
on the migration outcomes for Native Americans in the second half of the 20th
century. This policy led to a sorting of the Native American population into the
program’s target MSAs relative to all other MSAs, shaping the distribution of
the Native American population across the United States today. The magnitude
of these impacts for the tribal land area Native American population is made
clear in the counterfactual population estimates, which suggest that absent the
Relocation Program, there would have been an addition 250,000 Native American
individuals living on tribal lands in 1990. The gap in the fertility rate between
Native Americans living in MSAs versus traditionally high Native American
population states suggests that there are likely important long-run consequences
for those affected by the program.
The findings of this paper suggest some fulfillment of the program’s stated
aim–to relocate Native Americans away from tribal lands and into urban centers
with the overarching goal of assimilation. There is strong evidence of relocation of
the Native American population to the cities targeted by the program. Further,
68
the fertility findings provide some suggestive evidence of assimilation among those
living in the target cities.
While I am able to document one important dimension of the Relocation
Program by determining the effects on migratory outcomes, the full impacts of this
movement from reservations to major urban centers have yet to be examined. By
establishing the effect of the Relocation Program on migration, this paper sets up a
first stage for further research into the long-term economic and health outcomes for
those induced to move by the program.
69
CHAPTER IV
RIGHT TO CARRY LAWS AND INTIMATE PARTNER HOMICIDE
Introduction
Gun control is a fiercely debated topic in the United States. The regulation
of who and under what circumstances individuals can carry concealed handguns
is a topic that has grown increasingly prominent within this discussion and which
has seen substantial changes in recent years. At the state level alone, there have
been more than fifty changes made to concealed carry regulations in the past four
decades including a recent wave of eleven states adopting no permit1 concealed
carry in the past five years. In fact, the Supreme Court has just this year agreed
to review its first major gun control case in the past decade on the topic of states
rights to require permits for concealed carry weapons.2 Given the perennial and
divisive nature of this debate and the steady adoption of changes to concealed carry
laws in recent years, understanding the effects of concealed carry laws on violent
crime and homicide is vital.
While there is a vast literature examining these questions, there is still
uncertainty about the overall impact of concealed carry laws on violent crime
and homicide. One topic within this literature that is understudied is the effect of
concealed carry on homicides committed by an intimate partner. Access to guns is
known to be a major risk factor for intimate partner homicide (Bailey et al., 1997;
Campbell et al., 2003). Therefore, it is important to understand how laws that ease
1No permit style concealed carry regulations mean that an individual can carry a concealed
handgun in public without having to go through any type of permitting process.
2Scott Neuman and Nina Totenberg. “Supreme Court To Take Up 1st Major Gun Rights Case
In More Than A Decade.” NPR. April 26, 2021.
70
restrictions on gun access and increase the amount of time spent carrying a firearm
may impact intimate partner homicide rates.
In this paper, I analyze how right to carry laws3, which are the less restrictive
set of concealed carry regulations, affect rates of intimate partner homicide using
data from the Uniform Crime Reporting Program’s Supplementary Homicide
Reports. To answer this question, I employ three methods, namely: generalized
difference-in-differences, event study, and synthetic control methods (SCM). Using
these data and methodologies, I test a number of specifications with variation in
sample, fixed effects, and controls. Across specifications, I find no evidence for
an impact of right to carry laws on the rate of female victim intimate partner
homicide. I find only limited evidence of an impact on intimate partner firearm
homicide rates for male victims. Specifically, the adoption of right to carry laws
are associated with a decrease in the rate of male victim intimate partner firearm
homicides in some specifications using the generalized difference-in-differences
methodology. However, these findings are sensitive to reasonable variation in
specification and are not seen when using synthetic control methods. The event
study specifications suggest a possible violation of the parallel trends assumptions
which could be driving the difference-in-differences findings.
Overall, this paper shows that there is no strong evidence for an impact of
right to carry laws on intimate partner homicide rates, and any limited evidence of
a protective effect is solely realized for male victims. Given that 75% of male victim
intimate partner homicides are predated by male perpetrated intimate partner
violence, this suggests that the most likely mechanism driving these results is a
deterrent effect of right to carry laws (Campbell et al., 2003). Specifically, female
3Right to carry laws include both “shall issue” and “no permit” concealed carry laws.
71
partners who would have otherwise resorted to lethal means to escape a violent
relationship may be deterred from this extreme measure if they are concerned
their male partner may be more likely to be carrying a firearm. Additionally, these
findings suggest that when considering the impacts of right to carry laws, estimates
based on a single or only a select few specifications should be approached with
caution. Rather, researchers analyzing right to carry laws should provide readers
with specification sensitivity analysis so that it is clear how researcher choice
impacts treatment findings.
This paper primarily contributes to our understanding of the broad effects of
concealed carry laws and the impacts of gun laws on intimate partner homicide.
Because guns are known to be a significant risk factor for intimate partner
homicide, there have been a number of studies examining the impacts of gun
laws on rates of intimate partner homicides. Studies have found that certain
domestic violence restraining order laws (Vigdor & Mercy, 2006; Zeoli & Webster,
2010), the expansion of the gun control act (Raissian, 2016), and laws which
prohibit firearm possession for those convicted of violent misdemeanors (Zeoli
et al., 2018), all reduce rates of intimate partner homicide. Overall, this body
of work suggests that there are important protective effects against intimate
partner homicides of restricting access to firearms for certain groups of people
under specific circumstances. This paper contributes to this work by building up
our understanding of how right to carry laws specifically impact rates of intimate
partner homicide.
This paper also contributes to our understanding of how the loosening
of restrictions on concealed carry impacts different types of violent crimes and
homicide. There is still much debate within the concealed carry literature on the
72
impacts of concealed carry laws on violent crime. In their seminal paper on the
topic, Lott and Mustard (1997) find that less restrictive concealed carry laws lead
to lower rates of violent crime with some evidence of low levels of substitution
to property crime. Considerable work since Lott and Mustard’s initial findings
suggests that their results are quite sensitive to model specification and that there
is more work to be done to establish a clear understanding of the associations
between concealed carry and crime. In 2004, the National Research Council, part of
the National Academy of Sciences, investigated the early research and determined
that there was not enough evidence to support the findings that loosening
restrictions on concealed carry decreases violent crime (National Research Council,
2004). Since 2004, there have been a number of studies that have continued to shed
light on this question (La Valle & Glover, 2012; Aneja et al., 2014; Donohue et al.,
2019; Hamill et al., 2019, and many others) with many finding no significant or
strong association between less restrictive concealed carry laws and rates of violent
crime or homicide. There is still no consensus, however, with some papers finding
positive or negative associations among certain crimes or within certain changes
of the law. One recent paper that examines the impacts of concealed carry laws
on intimate partner homicide is Roberts (2009). Roberts (2009) finds that there
is a decrease in the overall rate of intimate partner homicide associated with less
restrictive concealed carry laws, but he finds no effects on intimate partner firearm
homicide rates. Given that concealed carry regulations are not the main focus of
Roberts’ paper, he ends his discussion of the findings by noting that he is unable to
explain these counterintuitive results.
This paper extends the small literature examining the relationship between
right to carry and intimate partner homicide to shed new light on the topic. I
73
contribute to the discussion in a number of ways. First, I apply new techniques
and consider additional specifications to test the robustness of treatment effects to
help further our understanding of inconsistencies in the concealed carry literature.
Also, I analyze about twenty years of additional data. This is of particular
importance given that the violent crime and concealed carry literature has shown
results are sensitive to the time period studied. Specifically, I add fifteen more
years of more recent data which are more relevant for new policy and suffer less
from contemporaneous issues like the violent crime spike in the 1990s. Finally, I
disaggregate my results by victim sex to better understand what might be driving
findings. Disaggregating by victim sex is vital within this setting because the
circumstances surrounding female versus male victim intimate partner homicides
are generally quite different. Therefore the mechanisms driving results for each
group are also likely to be different. For instance, roughly 70% of all intimate
partner homicides are predated by male perpetrated violence (Campbell et al.,
2003). Therefore, instances of female victim intimate partner homicide are typically
thought of as an escalation of male partner violence whereas male victim intimate
partner homicides are more often thought to be an extreme means by which female
partners attempt to escape violent relationships.
In addition to contributing to the concealed carry and intimate partner
homicide literatures, this paper also contributes to the literature analyzing
researcher choice. I find that my difference-in-differences results are highly sensitive
to defensible researcher choice of sample, fixed effects, and control variables. These
findings solidify what much of the research within this literature suggests–that
a more formal process for checking robustness to alternative specifications or
sensitivity analysis should be common practice to provide transparency about the
74
impacts of the research design process on the results. In this paper, I provide a
specification chart to formally illustrate specification sensitivity based on work by
Harrison and Rubinfeld (1978), Leamer (1983), and Simonsohn et al. (2020) – using
Ariel Ortiz Bobea’s design and R code.
Concealed Carry Laws and Intimate Partner Homicide
Concealed carry laws permit people to carry concealed handguns in public.
Intimate partner violence and homicides are often thought of as occurring inside
the home. Therefore, focusing on the impacts of concealed carry laws on intimate
partner homicides may initially seem counterintuitive. There are, however, a
number of ways by which the loosening of restrictions on concealed carry weapons
could impact intimate partner homicide. Firstly, less restrictive concealed carry
laws may increase the likelihood that a perpetrator has a gun on them when a
violent incident occurs, thereby increasing the probability that the violent incident
escalates to homicide. Secondly, I find some evidence that loosening restrictions
on concealed carry weapons increases overall gun sales4. Thus, it may be the
case that more permissive concealed carry laws could increase intimate partner
homicides by simply increasing the overall number of guns in possession. Thirdly,
it is possible that stricter concealed carry laws, which require more oversight
in the permitting process, serve as an additional check on the illegal possession
of firearms. Therefore, by loosening concealed carry laws, individuals who are
prohibited from owning firearms may be less likely to have any illegally possessed
firearms confiscated.
4The results for this analysis can be found in the Appendix in Table 1.
75
On the other hand, it is also possible that loosening restrictions on concealed
carry weapons may have a protective effect. It is possible that loosening concealed
carry laws could allow for a greater capacity for self-defense from intimate partners.
It is also possible that less restrictive concealed carry laws could have a deterrent
effect. For example, if people believe that it is more likely their partner will have a
gun on their person under a less restrictive concealed carry law, then they may be
less likely to engage in lethal violence.
Data
Supplementary Homicide Report
Data on intimate partner homicides come from the Uniform Crime Report’s
Supplementary Homicide Report (SHR). The SHR includes detailed information
on homicides reported to the FBI by local law enforcement agencies starting in
1976. Importantly, the SHR includes information on the relationship between the
victim and the offender which is used to determine instances of intimate partner
homicide. A homicide is considered an intimate partner homicide if the victim-
offender relationship is defined as spouse, ex-spouse, dating partner, or homosexual
relationship. Because of the marked differences in the circumstances surrounding
intimate partner homicides of males versus females, I also use the details on the sex
of the victim to separately analyze intimate partner homicides by victim sex. In my
main analysis, I include only homicides that were committed using a firearm given
that I am examining the effects of firearm restrictions. I also limit the observations
to only homicides that involved a single victim and single offender as is common
in the intimate partner homicide literature because the relationship variable is
only defined for the first victim. I combine the SHR data with the Surveillance,
76
Epidemiology, and End Results Program’s (SEER) U.S. Population Data 1969-2019
to create a panel dataset of annual, state-wide intimate partner homicide rates.
One drawback of using the SHR is that reporting is voluntary and fluctuates
over time. Further, there is no distinction made between instances of non-reporting
and instances of zero homicides. Therefore, I consider both the entire sample of
what is available from the SHR as well as a limited sample. The limited sample
includes only state-year observations where the number of total homicides reported
in the SHR is at least 50% of the number of homicides reported by vital statistics
data. The information on SHR and vital statistics matching is available from
1980-2019 by the National Center for Juvenile Justice. To allow for a more direct
comparisons between these two samples, I limit all analysis to 1980-2019. For state-
year observations with the 50% match, I impute zeros on any subsets of the data
which have no observations.
Figure 18 shows the trends in intimate partner homicide rates in the US
separately for male and female victims. Overall, intimate partner homicide rates
have decreased substantially in the past four decades, with the decline leveling off
in the 2000s. Intimate partner homicide rates decreased more dramatically for male
victims, falling by about 75% in the 1980s and 1990s.
77
FIGURE 18. Intimate Partner Homicides per 100,000 by Victim Sex
1.25
1.00
0.75
Female
Male
0.50
0.25
1980 1990 2000 2010 2020
Sources: Author’s calculations from the Supplementary Homicide Reports and SEER population
data
RAND State Firearm Law Database
Information on state-level concealed carry laws comes from the RAND State
Firearm Law Database. This database includes information on all state-level
concealed carry laws including when the law was passed and the level of the law’s
restrictiveness. The concealed carry laws are broken down into four categories,
namely: prohibited, may issue, shall issue, and permit not required. Prohibited
means that it is unlawful to carry a concealed handgun. May issue means that if
an individual meets certain criteria established by the law then a permit may or
may not be issued based on the discretion of the police. In this case, individuals
applying for permits usually must provide a reason for needing a concealed carry
permit. Shall issue means that if an individual meets the criteria established by
78
Intimate partner homicides per 100,000
the law then they will be granted a concealed carry permit. Finally, permit not
required means that it is lawful to carry a concealed handgun without obtaining a
permit. Prohibited is the most restrictive category while permit not required is the
most permissive. States with concealed carry laws that are shall issue or permit not
required are considered to have right to carry (RTC) laws. In this paper, I analyze
the impacts of these right to carry laws on intimate partner homicide rates.
FIGURE 19. Number of States that have Adopted Right to Carry Laws over Time
50
40
30
20
10
0
1980 1990 2000 2010 2020
Sources: Author’s calculations from the RAND state firearm database data
Throughout the time period analyzed, all the changes that states made to
their concealed carry laws served to make concealed carry less restrictive. Figure
19 shows how state adoption of right to carry laws has grown over time. The most
significant period of state adoption of right to carry laws occurred between 1994
and 1996 when fourteen states switched to right to carry. By the end of the period
43 of 51 states (counting the District of Columbia) had right to carry laws in place.
79
Number of states with RTC laws
Control Variables
I include a number of control variables across various specifications. My
baseline controls are the unemployment rate and the percent of the population that
lives in a metropolitan statistical area (MSA) as unemployment and urbanization
have been shown to impact a state’s likelihood of adoption of right to carry laws
as well as the intimate partner homicide rate. Data on unemployment comes from
the St. Louis Federal Reserve’s FRED database and data on the percent of the
state population living in an MSA is calculated using SEER population data and
US Census MSA delineation files. I also consider a number of other economic,
demographic, regulatory, and criminal justice system variables common in the
concealed carry and violent crime literature and the intimate partner homicide and
gun regulation literature. Information on what each measure means and where is
comes from is provided below, while details on the inclusion of each measure across
specifications is outlined in the methodology section.
Additional economic variables include information on real per capita personal
income, the poverty rate, maximum AFDC/TANF benefit levels for a family of
three, and the female to male employment ratio. Data on real per capita income is
from the St. Louis Federal Reserve’s FRED database. Information on the poverty
rate and the maximum AFDC/TANF benefit levels for a family of three comes
from the National Welfare Data that is made available through the University
of Kentucky’s Center for Poverty Research. The female to male employment
ratio is calculated using information from the Current Population Survey (CPS)
made available through IPUMS. I also use the CPS to calculate a number of the
demographic variables including the proportions of the population that are white
and black as well as the proportions of the population aged 15 and older who are
80
married, divorced, and widowed. The other demographic variables considered
are the set used in the Donohue et al. (2019) paper and are calculated using the
SEER population data. They consist of 6 age-sex-race groups that include white
males 15-19, black males 15-19, other males 15-19, white males 20-39, black males
20-39, and other males 20-39. Data on incarceration and police employment are
available through the FBI’s Uniform Crime Report and are used to calculate lagged
incarceration and police employment rates. I also control for the per capita level
of ethanol consumption from beer using data made available by Jacob Kaplan
from the Apparent per capita Alcohol Consumption: National, State, and Regional
Trends, 1977-2016 report from the National Institute of Health.
Lastly, I also control for several laws related to intimate partner homicides.
These include unilateral divorce laws, domestic violence misdemeanor firearm
confiscation laws, laws for police seizure of firearms at scenes of domestic violence,
and domestic violence restraining order firearm dispossession laws. Information
on unilateral divorce laws comes from Gruber (2004) with updates from Ciacci
(2018). Data on domestic violence misdemeanor firearm confiscation laws are from
Raissian (2016). Information on laws for police seizure of firearms at scenes of
domestic violence is available from Vigdor and Mercy (2006), which I updated with
information from the Giffords Law Center to Prevent Gun Violence. Finally, data
on domestic violence restraining order laws come from Zeoli et al. (2018).
Methodology
I take three broad approaches to determine how right to carry laws impact
intimate partner homicide rates. Firstly, I use a generalized difference-in-differences
approach. Within the difference-in-differences framework, I consider a number of
81
specifications with different samples, weighting, and sets of time-varying state-
level controls. I also employ an event study specification. This approach builds
on the difference-in-differences method by explicitly illustrating pre-trends and by
deconstructing the average treatment estimates from the difference-in-differences
specifications to determine how treatment effects vary over time. Finally, I use a
synthetic control methods approach to improve the pre-trend match between the
treated and control units. The details of each methodology and the particulars of
each specification considered are outlined below.
Method 1: Generalized DD Regression
For the difference-in-differences approach, the general estimating equation
takes the following form:
For state, s, in census region, r, in year, y:
IPHsy = α + βRightToCarrysy + γXsy + δy + λs + µry + εsy
Here, IPHsy is the rate of intimate partner homicide for a given state in a given
year per 100,000 population. In my main specifications, I include only intimate
partner homicides committed with a firearm in the calculation of the intimate
partner homicide rate and calculate separate rates for male and female victims.
As a robustness check, I also examine the rates of non-firearm intimate partner
homicides for male and female victims. RightToCarrysy is an indicator for whether
or not a given state has a right to carry law in place in year y. δy and λs are fixed
effects for year and state, respectively, and µry is a fairly strict region by year fixed
effect. εsy is the error term. For all difference-in-differences specifications, standard
82
errors are clustered at the state. Across many of the specifications, I use a weighted
least squares approach to adjust my estimates for the greater levels of noise in low
population states.
Xsy is a vector of state and time varying controls that differs across
specifications. The base controls are the state level unemployment rate and the
percent of the state population that lives in a metropolitan statistical area (MSA).
In additional specifications, I include sets of control variables that are common
in the concealed carry literature and the intimate partner homicide literature.
When considering the controls typical for the concealed carry literature, I use
the set of controls from Donohue et al. (2019). I refer to this set of controls as
the DAW controls. They include the one year lagged incarceration rate, the
one year lagged police employment rate per 100,000 population, the poverty
rate, the unemployment rate, per capita ethanol consumption from beer, the
percent of the population living in an MSA, real per capita personal income and
6 demographic age-sex-race groups which consist of white males 15-19, black
males 15-19, other males 15-19, white males 20-39, black males 20-39, and other
males 20-39. When considering the controls common to the relevant intimate
partner homicide literature, I include the following indicator variables for whether
or not the state has a unilateral divorce law, a domestic violence misdemeanor
firearm confiscation law, a law permitting or mandating police seizure of firearms
at scenes of domestic violence, or a domestic violence restraining order firearm
dispossession law that applies ex parte or has firearm relinquishment information. I
also include information on the ratio of male to female employment, AFDC/TANF
maximum benefit levels for a family of three, the poverty rate, the proportion of
the population that is white, the proportion of the population that is black, and
83
the proportions of the population 15 years and older that are married, divorced, or
widowed, individually.
Method 2: Event Study
The event study specifications look similar to the difference-in-differences
specifications but demonstrate the impacts of treatment over time.
For state, s, in year, y:
∑−2 ∑10
IPHsy = α + βiPreRTCisy + βiPostRTCisy + γXsy + δy + λs + µry + εsy
i=−10 i=0
IPHsy is still the rate of intimate partner homicide for a given state in a
given year per 100,000 population. δy and λs are again fixed effects for year and
state, µry is a region by year fixed effects, and Xsy is the same varying vector of
state-level time-varying controls as outlined above. In the event study equation,
though, instead of having a single indicator for whether or not a state has a right
to carry law in a given year, the event study equation considers a set of indicator
variables for how long it has been prior to or after the adoption of the right to
carry law for ten year in each direction. All other years before and after adoption
are included in the estimates of β−10 and β10, respectively. The reference year is the
year immediately prior to adoption. εsy is the error term, and standard errors are
clustered at the state.
Method 3: Synthetic Control Methods
Finally, I also use synthetic control methods (SCM; Abadie et al., 2010, 2015)
to answer the question of how right to carry laws impact intimate partner homicide
84
rates. The synthetic control method improves on the event study methodology
by creating synthetic control groups through the weighting of untreated units to
match pre-treatment outcomes for the treated units. SCM was originally designed
for a setting with a single treated unit and multiple control units that are weighted
to create the synthetic control. In this case, though, I am examining a number of
treated units with varying treatment times. One method for using SCM in this
setting is to simply find synthetic controls for each unit separately and then take
the average of the treatment effect for each individually estimated SCM regression.
However, this relies on finding good synthetic controls for each of the individual
units. Because several states do not have good matches in the pre-period when
estimated using the gsynth package, I use the partially pooled synthetic controls
method developed by Ben-Michael et al. (2021) for the main analysis. This method
balances the dual objectives of finding a good unit-level fit with finding a good
pooled fit. This methodology is ideal for settings like this where there are some
units that do not have good synthetic control fits at the individual level and where
we are interested in average treatment effects. In the augsynth package developed
by Eli Ben-Michael, which I use for this analysis, the hyperparameter ν determines
the relative importance that is placed on each goal and is set to equal the ratio of
the pooled fit to the average of the unit-level fit.
In this setting where treatment times are staggered, researchers have shown
that the generalized difference-in-differences approach with two-way fixed effects
can be misleading when treatment effects differ by treatment timing (Sun &
Abraham, 2020; Borusyak & Jaravel, 2017; Goodman-Bacon, 2018). The results of
the unit-level SCM found in the appendix suggest that treatment effects are likely
varying by treatment timing and maybe also by region. Therefore, the partially
85
pooled SCM provides a way to better estimate the treatment effects than the
generalized differences-in-differences method or the event study method described
above.
I generate the baseline SCM estimates by matching on the intimate partner
homicide rate for the ten years prior to adoption of the right to carry law. In
additional specifications, I also match on the 10 year lags of unemployment and the
percent of the population living in an MSA as well as the full set of DAW controls
described in the difference-in-differences methodology section.5 One synthetic
control is estimated for each time cohort–with a time cohort being the group of
states that adopt RTC laws in the same year. Currently, the augsynth package
only accepts balanced panels so all states with missing units are dropped. The
gsynth package is able to analyze unbalanced panels, so states that are dropped
in the main analysis are still included in the state-by-state estimates found in the
appendix. The SCM estimates all have jackknife standard errors.
Results
The main findings of the impacts of right to carry laws on intimate partner
homicides can be found in Figures 20 and 22. Figure 20 shows the impacts of RTC
laws on the rate of female victim intimate partner homicides where the weapon
used is a firearm. Across all of the specifications shown in the specification chart,
only the naive OLS estimates are significant at the 5% level. One SCM estimate
is significant at the 10% level, but no other estimates are significant at traditional
levels.
5I do not include in the results section a specification matching on the 10 year lags of the IPH
controls because so many of the IPH controls are binary variables and the non-binary variables are
not largely different from the DAW controls.
86
FIGURE 20. Specification Chart for the Impact of Right to Carry Laws on the
Rate of Intimate Partner Homicides with a Firearm: Female Victims
0.1
0.0
−0.1
−0.2
Method:
Difference−in−Differences
SCM
Samples:
Full
Matched
Balanced
Fixed Effects:
State
Year
Region by Year
Sets of Controls:
Base
DAW
IPH
Other:
WLS
State Linear Time Trends
Notes: This specification chart was created using the ‘spec chart’ function in R developed by
Ariel Oritz Bobea. Each coefficient on the chart is an estimate of the impact of right to carry
laws on the rate of female victim intimate partner homicides committed with a firearm from a
different specification that varies by methodology, sample, controls, and fixed effects as described
in the bottom half of the figure. The bars represent 95% and 99% confidence intervals. Standard
errors for the difference-in-differences estimates are clustered at the state. SCM specifications have
jackknife standard errors. The darkly shaded specification details correspond to the specifications
analyzed further in the event study graphs and Table 1. The baseline (1980) population weighted
mean for the rate female victim firearm intimate partner homicide is 0.828
Figure 21 shows the results of the event study specifications. Each line on the
graph is a different event study specification and corresponds to one of the darkly
87
Coefficient
shaded columns in the specification chart. Only seven specifications are included
here so that the results are still visible, but the results from other specifications
look similar. The event study estimates verify the results from the other methods
as they provide no evidence of an impact. None of the estimates are significant at
traditional levels. Overall, these results suggest that there is no effect of right to
carry laws on rates of female victim intimate partner firearm homicides.
FIGURE 21. Event Study Estimates for Female Victims, Weapon Type: Firearm
0.2
Model
0.1 Fixed Effects Full Sample
WLS Full Sample
WLS Matched Sample
0.0 Region by Year FE Matched Sample
Base Controls Matched Sample
DAW Controls Matched Sample
−0.1
IPH Controls Matched Sample
−0.2
−10 −5 0 5 10
Notes: Standard errors are represented by the ribbons around the estimates and are all clustered
at the state. Starting from the top of the event study graph legend the specifications included
in the event study graphs match up to the darkly shaded models in the bottom half of the
specification charts moving from left to right.
88
Event study coefficient estimate
FIGURE 22. Specification Chart for the Impact of Right to Carry Laws on the
Rate of Intimate Partner Homicides with a Firearm: Male Victims
0.2
0.1
0.0
−0.1
−0.2
−0.3
−0.4
Method:
Difference−in−Differences
SCM
Samples:
Full
Matched
Balanced
Fixed Effects:
State
Year
Region by Year
Sets of Controls:
Base
DAW
IPH
Other:
WLS
State Linear Time Trends
Notes: This specification chart was created using the ‘spec chart’ function in R developed by Ariel
Oritz Bobea. Each coefficient on the chart is an estimate of the impact of right to carry laws on
the rate of male victim intimate partner homicides committed with a firearm from a different
specification that varies by methodology, sample, controls, and fixed effects as described in the
bottom half of the figure. The bars represent 95% and 99% confidence intervals. Standard errors
for the difference-in-differences estimates are clustered at the state. SCM specifications have
jackknife standard errors. The darkly shaded specification details correspond to the specifications
analyzed further in the event study graphs and Table 11. The baseline (1980) population weighted
mean for the rate male victim firearm intimate partner homicide is 0.689
Figure 22 shows the impacts of right to carry laws on the rate of male
victim intimate partner homicides committed with a firearm. In Figure 22, we
89
Coefficient
see that there are a number of specifications with a negative coefficient that are
significant at traditional levels, but the estimates are not stable. Stricter difference-
in-differences specifications and synthetic controls method estimates suggest no
significant impacts of right to carry on male victim intimate partner firearm
homicide rates.
FIGURE 23. Event Study Estimates for Male Victims, Weapon Type: Firearm
Model
0.2
Fixed Effects Full Sample
WLS Full Sample
WLS Matched Sample
0.0 Region by Year FE Matched Sample
Base Controls Matched Sample
DAW Controls Matched Sample
IPH Controls Matched Sample
−0.2
−10 −5 0 5 10
Notes: Starting from the top of the event study graph legend, the specifications match up to the
darkly shaded models in the bottom half of the specification charts moving from left to right. The
ribbons represent 95% confidence intervals based on state-level clustered standard errors.
Figure 23 shows the event study results for a subset of the difference-in-
differences specifications in the specification chart with each line representing a
different specification. I include a number of the specifications where the difference-
in-differences coefficients are negative and significant. Here, the event study shows
an overall negative trend in the coefficients over time, including during the period
before treatment occurs. This trend is particularly pronounced among the less
90
Event study coefficient estimate
restrictive difference-in-differences specifications. The event study pre-trends
suggest a possible violation of the parallel trends assumption that may be one
reason for the significant and negative coefficient estimates in the less restrictive
specifications. Figure 24 shows the results of the partially pooled SCM estimates
over time for all but the naive OLS specification. Through the use of synthetic
controls, the SCM model is able to better match trends in the pre-period and
correct for the overall downward trend. All together, the evidence of an effect of
concealed carry on male victim intimate partner homicides is fairly weak.
FIGURE 24. Partially Pooled SCM for Male Victims, Weapon Type: Firearm
0.1
Matched On:
0.0 IPH Lags Only
IPH Lags and Base Controls
IPH Lags and DAW Controls
−0.1
−0.2
−10 −5 0 5
Notes: All models are matched on 10 lags and estimated for 10 time periods after treatment.
All models are estimated with two-way factors, and one synthetic control is estimated for each
group of states that adopted a right to carry law in the same year. The ribbons represent 95%
confidence intervals based on jackknife standard errors.
As an additional check on the main analysis, I examine the impact of right
to carry laws on rates of intimate partner homicides where the perpetrator used
91
Partially Pooled SCM coefficient estimate
TABLE 11. RTC Laws and Non-Firearm Intimate Partner Homicide Rates
Dependent variable: Rate of Intimate Partner Homicides per 100,000
Full SHR Sample SHR Sample with 50% Match to Vital Statistics
(1) (2) (3) (4) (5) (6) (7)
Panel A: Female Victims (1980 Population Weighted Mean: 0.380)
Right to Carry 0.035 0.039 0.038 0.051∗∗ 0.032 0.029 0.002
(0.023) (0.026) (0.028) (0.024) (0.026) (0.021) (0.019)
Panel B: Male Victims (1980 Population Weighted Mean: 0.306)
Right to Carry −0.015 −0.013 −0.017 0.025 −0.019 −0.013 −0.018
(0.019) (0.018) (0.019) (0.016) (0.019) (0.011) (0.018)
Two-Way FE Y Y Y Y Y Y Y
WLS N Y Y Y Y Y Y
Region x Year FE N N N Y N N N
Unemployment N N N N Y Y N
Percent MSA N N N N Y Y N
DAW Controls N N N N N Y N
IPH Controls N N N N N N Y
Observations 1,945 1,945 1,885 1,885 1,885 1,734 1,827
Notes: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01. Standard errors are clustered at the state. The weighted
least squares specifications are weighted by population. From left to right, these specifications
correspond to the darkly shaded models in the bottom half of the specification charts from left to
right.
a weapon other than a firearm. These results can be found in Table 11. This
analysis serves as an additional robustness check of my primary samples as we
would not expect right to carry laws to have an impact on non-firearm intimate
partner homicides. Table 11 includes only the subset of specifications shown in
the event study graphs that correspond to the darkly shaded specifications in the
specification charts. I show only this subset for brevity, but these are reflective
of the broader specification results. Here, we see no consistent evidence of an
impact of right to carry laws on non-firearm intimate partner homicides for male
92
or female victims. The single specification that is statistically significant is small
in magnitude relative to the baseline mean. Further, it is reasonable for one of
these to show statistical significance even if there is no impact given the number
of specifications and samples being tested.
Conclusion
Overall, I find little evidence of any impact of right to carry laws on intimate
partner homicide rates. Across three methodologies, multiple samples, and a variety
of included control variables, I find few specifications with statistically significant
results, almost all of which are found in specifications focusing on the male victim
sample. My findings indicate that the generalized difference-in-differences approach
may be inappropriate in this setting where the results are sensitive to model
specification and event study estimates call into question the validity of the parallel
trends assumption. This specification sensitivity suggests that previous estimates of
the relationship between right to carry laws and intimate partner homicide based
solely on difference-in-differences models should be considered carefully. These
results have important implications for advocates and policymakers working in
the areas of intimate partner violence and gun control as well as for researchers
studying concealed carry regulations.
Firstly, the limited evidence of an impact is completely concentrated among
male victims, while female victims do not see any protective effect of right to carry
laws. This provides some new insights into the possible mechanisms driving the
impacts of right to carry laws. Given that the majority of male victim intimate
partner homicides are predated by male offender intimate partner violence, it is
most likely that the right to carry laws serve as a deterrent. Specifically, women
93
may be deterred from resorting to lethal violence as a means of escaping a violent
relationship if they believe it is more likely their partner will have a firearm easily
accessible when a right to carry law is in place.
Additionally, the sensitivity of my results further the call for more formal
means of testing specification sensitivity particularly for researchers studying the
impacts of concealed carry. Results from the male victim sample shown in Figure
22 demonstrate this point well by showing just how sensitive the results are to
the particulars of the model specification. This sensitivity across defensible but
somewhat arbitrary researcher choices illustrates why more formal and extensive
sensitivity analysis is necessary and should be commonplace in the concealed
carry literature and in social science research more broadly. Further, these findings
provide one possibility for why there is so little consensus within the literature on
the effects of right to carry laws.
For policymakers and advocates seeking to reduce rates of intimate partner
homicide, these results suggest that the regulation of concealed carry laws may not
be a policy area with substantial impacts on this dimension. Therefore, advocates
and policymakers may want to spend their time focused on other areas of gun
control with larger documented protective effects such as restricting possession of
firearms for individuals with past violent misdemeanor offenses.
Although this paper does not show any strong evidence of an impact of
right to carry laws on intimate partner homicides, this does not mean that there
are no impacts for intimate partner violence more broadly. Intimate partner
homicides are just one extreme form of violence perpetrated by intimate partners.
Further research into how right to carry laws may impact other dimensions of
94
intimate partner violence is needed as right to carry laws could affect other forms of
aggression like controlling or threatening behavior.
While other research has shown right to carry laws have important impacts
for other outcomes, I find little evidence of an impact on intimate partner homicide.
These findings provide insight into possible mechanisms driving right to carry law
impacts and highlight the need for greater sensitivity analysis in the concealed
carry literature. Further, they suggest that policymakers and advocates seeking
to reduce rates of intimate partner homicides may have greater impacts by focusing
their efforts on other gun control legislation.
95
CHAPTER V
CONCLUSION
Across the three main chapters of this dissertation, I provide insight into
how a number of policies have impacted public health as well as, for one policy,
migration.
In Chapter II, “The Effect of Women’s Access to Free Health Care on
Breastfeeding Practices: Evidence from Armenia,” I estimate the impacts of
increased access to free health care for women on the prevalence and duration
of breastfeeding. I find that providing free health care to women in this setting
decreased the likelihood of ever breastfeeding among those exposed to the program.
I find no impact of the program on the duration of breastfeeding. Analyzing data
on hospital practices related to infant feeding suggests that a lack of support
for breastfeeding in the treated region’s medical community may be driving this
decrease. Well-documented infant formula promotion in health care facilities
throughout the treated region further supports this explanation. Overall, these
results suggest that policymakers may want to take additional steps to ensure
up-to-date practices at the clinics providing increased access to health care when
implementing programs like this. In particular, facilitators of programs targeting
female health care may want to pay special attention to the gender-specific
practices at the clinics to curb these types of unintended consequences.
Chapter III analyzes the effects of the Relocation Program on Native
American migration and fertility outcomes. Comparing the program’s target
cities to cities that were not targeted, I employ a difference-in-differences model
to estimate the effect of the Relocation Program on Native American migration.
96
I find that the Relocation Program led to a significant 12,000 person increase in
the Native American population in cities that were targeted by the Relocation
Program relative to the non-targeted cities. This constitutes a considerable shift
in the spatial distribution of the Native American population who prior to 1950
primarily resided on rural tribal lands. I also find that there is a substantial gap in
the rates of second-generation fertility for Native American women living in urban
areas compared to those living on tribal lands. This suggests that the program
likely had lasting impacts across generations for those exposed to the program.
In Chapter IV, I examine the impact of right to carry laws on rates of
intimate partner homicide. I analyze to what extend right to carry laws effect
intimate partner homicide rates separately for male and female victims using
three methodologies, namely: difference-in-differences, event study, and synthetic
control methods. I find no indication of an effect of right to carry on female
victim intimate partner homicide rates and only weak evidence of an effect for
male victims. These limited effects are highly sensitive to model specification and
vary considerably across the three methodologies. This specification sensitivity
suggests that policymakers should approach estimated effects of right to carry
laws with caution. Further, this highlights the need for more formal sensitivity
analysis within this literature where sensitivity has been demonstrated but also
more broadly as these results show that defensible but arbitrary researcher choices
can produce misleading conclusions.
97
APPENDIX
ADDITIONAL TABLES AND FIGURES
CHAPTER II: Additional Tables and Figures
TABLE A1. Effect of Access to Free Maternal Healthcare on Breastfeeding for the
Full Sample
Dependent variable:
Months Breastfed Extensive Intensive
Margin Margin
(1) (2) (3) (4)
Post June08 −0.424 −0.515
(0.369) (0.363)
Post March09 −0.552 0.006
(0.563) (0.016)
Number of 0.008 0.008 0.002 −0.010
HH Members (0.031) (0.031) (0.002) (0.027)
Number of −0.058 −0.058 0.004∗ −0.108∗
Rooms in HH (0.065) (0.066) (0.003) (0.065)
Post June08 −0.670 −0.312
× Yerevan (0.864) (0.857)
Post March09 −0.573 −0.045∗∗
× Yerevan (0.773) (0.023)
Full Set FE Yes Yes Yes Yes
Household Controls Yes Yes Yes Yes
Mother Controls No No No No
Observations 23,090 23,090 23,090 21,408
R2 0.083 0.083 0.021 0.102
Adjusted R2 0.081 0.081 0.019 0.100
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
98
FIGURE A1. Pre-Treatment Coefficient Estimates by Month of Birth
0.2
0.1
0.0
−0.1
−0.2
1 2 3 4 5 6 7 8 9 10 11 12
Birth month
Note: Estimates for the coefficients on the indicator for Yerevan as the region of residence
interacted with the child’s month of birth in the per-treatment period, namely, prior to 2009.
Regression includes year, region, and wave fixed effects. The ribbon reflects the 95% confidence
intervals, and the vertical dashed line represents the beginning month of the intervention each
year. Estimates are normalized to January.
99
Change in likelihood of ever breastfeeding
CHAPTER III: Additional Tables and Figures
FIGURE A2. Target City Native American Population Trends
Chicago Cincinnati Cleveland
100,000
75,000
50,000
25,000
0
Dallas Denver Los Angeles
100,000
75,000
50,000
25,000
0
Salt Lake City San Francisco St. Louis
100,000
75,000
50,000
25,000
0
1920 1940 1960 1980 1920 1940 1960 1980 1920 1940 1960 1980
Notes: Total Native American population for each of the target MSAs individually from 1910-
1990. Dotted lines represent the start and end of the Relocation Program in 1952 and 1973,
respectively.
100
Native American Population
FIGURE A3. Native American Population Trends by MSA Type, Log Scale
10000
1000 MSA Type
Target MSAs
Near MSAs
100 Far MSAs
10
1920 1940 1960 1980
Notes: Population Growth for the target and non-target MSAs from 1910-1990. Non-target MSAs
are broken into two groups based on proximity to tribal land. The points represent log average
MSA Native American population. Best fit lines are estimated by MSA type for the pre and post
periods. The dotted lines correspond to the start and end of the Relocation Program in 1952 and
1973, respectively.
101
Average Native American Population
CHAPTER VI: Additional Tables and Figures
TABLE A2. Impact of Right to Carry Laws on Rates of Handgun Sales
Dependent variable:
Rate of Handgun Sales per 100,000
Right to Carry 479.041∗∗
(179.115)
Observations 756
R2 0.870
Adjusted R2 0.859
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01. Standard errors are clustered at the state. Data from the
FBI’s NICS background checks is used as a proxy for gun sales. These data are made available on
GitHub by BuzzFeedNews.
FIGURE A4. Partially Pooled SCM for Female Victims, Weapon Type: Firearm
0.1
0.0 Matched On:
IPH Lags Only
IPH Lags and Base Controls
−0.1 IPH Lags and DAW Controls
−0.2
−10 −5 0 5
Notes: All models are matched on 10 lags and estimated for 10 time periods after treatment.
All models are estimated with two-way factors, and one synthetic control is estimated for each
group of states that adopted a right to carry law in the same year. The ribbons represent 95%
confidence intervals based on jackknife standard errors.
102
Partially Pooled SCM coefficient estimate
FIGURE A5. Adoption of Right to Carry Laws over Time
GA
PA
WV
ID
MS
OR
AK
AZ
WY
AR
NC
NV
UT
VA
OK
SC
TN
TX
MI
CO
MN
MO
OH
CA
HI
MA
MD
NJ
NY
RI
1980 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 2013 2016 2019
control states treated states (before RTC) treated states (after RTC)
Notes: Includes only states from the balanced panel used in the synthetic control methods.
103
FIGURE A6. Individual State SCM Estimates
Female Victims, Weapon Type: Firearm
Georgia (1989) Pennsylvania (1989) West Virginia (1989)
switch from may issue switch from may issue switch from may issue
1.0 1.5
1.0
0.5 1.0
0.5 0.5
0.0
0.0 0.0
−0.5 −0.5
−0.5
−1.0 −1.0
−5 0 5 10 −5 0 5 10 −5 0 5 10
Idaho (1990) Mississippi (1990) Oregon (1990)
switch from may issue switch from prohibited switch from may issue
1 1.0
1.0
0.5
0.5 0
0.0 0.0
−0.5 −1 −0.5
−1.0 −1.0
−5 0 5 10 −5 0 5 10 −5 0 5 10
Alaska (1994) Arizona (1994) Wyoming (1994)
switch from prohibited switch from prohibited switch from may issue
1.0
2 1.0
0.5
1 0.5
0.0
0.0
0
−0.5 −0.5
−1 −1.0 −1.0
−2 −1.5
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10
Arkansas (1995) North Carolina (1995) Nevada (1995)
switch from prohibited switch from prohibited switch from may issue
1.5 1.0
1.0 0.5 0.5
0.5 0.0
0.0
0.0
−0.5
−0.5 −0.5
−1.0 −1.0 −1.0
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10
104
ATT ATT ATT ATT
ATT ATT ATT ATT
ATT ATT ATT ATT
Utah (1995) Virginia (1995) Louisiana (1996)
switch from may issue switch from prohibited switch from may issue
1.0 1.0 1.0
0.5 0.5 0.5
0.0 0.0 0.0
−0.5 −0.5 −0.5
−1.0 −1.0
−1.0
−1.5 −1.5
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10
Texas (1996) Michigan (2001) Colorado (2003)
switch from prohibited switch from may issue switch from may issue
1.0 1.0
0.5
0.5 0.5
0.0
0.0 0.0
−0.5
−0.5 −0.5
−1.0
−1.0 −1.0
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10
Minnesota (2003) Missouri (2004) New Mexico (2004)
switch from may issue switch from prohibited switch from prohibited
1.0 1.01.0
0.5
0.5 0.5
0.0
0.0 0.0
−0.5
−0.5 −0.5
−1.0
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10
Ohio (2004) Iowa (2011) Wisconsin (2011)
switch from prohibited switch from may issue switch from prohibited
0.5 0.5 0.5
0.0 0.0 0.0
−0.5
−0.5 −0.5
−1.0
−10 −5 0 5 10 −10 −5 0 5 −10 −5 0 5
Notes: Each graph represents the estimate of the Average Treatment on the Treated from a SCM
specification with inclusion of factors determined by cross-validation. Matches are based on 10
years of lagged intimate partner homicide rates as well as lagged unemployment and the percent
of the state’s population living in an MSA.
105
ATT ATT ATT ATT
ATT ATT ATT ATT
ATT ATT ATT ATT
FIGURE A7. Individual State SCM Estimates
Male Victims, Weapon Type: Firearm
Georgia (1989) Pennsylvania (1989) West Virginia (1989)
switch from may issue switch from may issue switch from may issue
0.8
1
0.5
0.4
0
0.0
0.0 −1
−0.5
−0.4
−1.0 −2
−5 0 5 10 −5 0 5 10 −5 0 5 10
Idaho (1990) Mississippi (1990) Oregon (1990)
switch from may issue switch from prohibited switch from may issue
0.4
1.0
0.5
0.0
0.5
0.0
0.0 −0.4
−0.5
−0.5 −0.8
−1.0
−1.0 −1.2
−5 0 5 10 −5 0 5 10 −5 0 5 10
Alaska (1994) Arizona (1994) Wyoming (1994)
switch from prohibited switch from prohibited switch from may issue
2 0.5 1.5
1 1.00.0
0.5
0
−0.5 0.0
−1 −0.5
−1.0
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10
Arkansas (1995) North Carolina (1995) Nevada (1995)
switch from prohibited switch from prohibited switch from may issue
1.0 1.0
0.5
0.5 0.5
0.0 0.0 0.0
−0.5 −0.5 −0.5
−1.0 −1.0
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10
106
ATT ATT ATT ATT
ATT ATT ATT ATT
ATT ATT ATT ATT
Utah (1995) Virginia (1995) Louisiana (1996)
switch from may issue switch from prohibited switch from may issue
0.5 0.5
0.4
0.0 0.00.0
−0.4 −0.5
−0.5
−1.0
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10
Texas (1996) Michigan (2001) Colorado (2003)
switch from prohibited switch from may issue switch from may issue
1.0
0.5 0.250.4
0.0 0.00
0.0
−0.5 −0.25
−0.4
−1.0
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10
Minnesota (2003) Missouri (2004) New Mexico (2004)
switch from may issue switch from prohibited switch from prohibited
0.6
0.3
0.25 0.4
0.2 0.0
0.00
0.0
−0.3
−0.25 −0.2
−0.4 −0.6
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10
Ohio (2004) Iowa (2011) Wisconsin (2011)
switch from prohibited switch from may issue switch from prohibited
0.50
0.4 0.25
0.25
0.2
0.00
0.0 0.00
−0.25
−0.2 −0.25
−0.50
−0.4 −0.50
−10 −5 0 5 10 −10 −5 0 5 −10 −5 0 5
Notes: Each graph represents the estimate of the Average Treatment on the Treated from a SCM
specification with inclusion of factors determined by cross-validation. Matches are based on 10
years of lagged intimate partner homicide rates as well as lagged unemployment and the percent
of the state’s population living in an MSA.
107
ATT ATT ATT ATT
ATT ATT ATT ATT
ATT ATT ATT ATT
REFERENCES CITED
Abadie, A., Diamond, A., & Hainmueller, J. (2010). Synthetic control methods for
comparative case studies: Estimating the effect of California’s tobacco control
program. Journal of the American Statistical Association, 105(490), 493-505.
Abadie, A., Diamond, A., & Hainmueller, J. (2015). Comparative politics and the
synthetic control method. American Journal of Political Science, 59(2),
495-510.
Ablon, J. (1964). Relocated American Indians in the San Francisco Bay Area:
Social interaction and Indian identity. Human Organization, 23(4), 296-304.
Ablon, J. (1965). American Indian relocation: Problems of dependency and
management in the city. Phylon, 26(4), 362-371.
Amirkhanyan, Y. (2008). Knowledge, attitude and practice of mothers and health
care providers towards optimal breastfeeding in Yerevan, Armenia. The New
Armenian Medical Journal , 2(2), 41-47.
Aneja, A., Donohue, J. J., & Zhang, A. (2014). The impact of right-to-carry laws
and the NRC report: Lessons for the empirical evaluation of law and policy.
American Law and Economics Review , 13(2), 565–631.
Avery, A., Zimmermann, K., Underwood, P. W., & Magnus, J. H. (2009).
Confident commitment is a key factor for sustained breastfeeding.
Birth-Issues in Perinatal Care, 36(2), 141-148.
Bailey, J. E., Kellermann, A. L., Somes, G. W., Banton, J. G., Rivara, F. P., &
Rushforth, N. P. (1997). Risk factors for violent death of women in the home.
Archives of Internal Medicine, 157(7), 777-782.
Ben-Michael, E., Feller, A., & Rothstein, J. (2021). Synthetic controls with
staggered adoption.
Berger, L. M., Hill, J., & Waldfogel, J. (2005). Maternity leave, early maternal
employment and child health and development in the US. Economic Journal ,
115(501), F29-F47.
Boban, M., & Zakarija-Grkovic, I. (2016). In-hospital formula supplementation of
healthy newborns: Practices, reasons, and their medical justification.
Breastfeeding Medicine, 11(9), 448-454.
Borusyak, K., & Jaravel, X. (2017). Revisiting event study designs.
108
Brandtzaeg, P. (2010). The mucosal immune system and its integration with the
mammary glands. Journal of Pediatrics , 156(2), S8-S15.
Brown, C. R. L., Dodds, L., Legge, A., Bryanton, J., & Semenic, S. (2014). Factors
influencing the reasons why mothers stop breastfeeding. Canadian Journal of
Public Health-Revue Canadienne De Sante Publique, 105(3), E179-E185.
Cameron, A. C., & Miller, D. L. (2015). A practitioner’s guide to cluster-robust
inference. Journal of Human Resources , 50(2), 317-372.
Campbell, J. C., Webster, D., Koziol-McLain, J., Block, C., Campbell, D., Curry,
M. A., . . . Laughon, K. (2003). Risk factors for femicide in abusive
relationships: Results from a multisite case control study. American Journal
of Public Health, 93(7), 1089-1097.
Cebula, R. J., & Belton, W. J. (1994). Voting with ones feet: An empirical analysis
of public welfare and migration of the American Indian, 1985-1990. American
Journal of Economics and Sociology , 53(3), 273-280.
Chadwick, B. A., & White, L. C. (1973). Correlates of length of urban residence
among Spokane Indians. Human Organization, 35(1), 9-16.
Chakravarty, A. (2015). Gender-biased breastfeeding in Egypt: Examining the
fertility preference hypotheses of Jayachandran and Kuziemko (2011). Journal
of Applied Econometrics , 30(5), 848-855.
Chetty, R., & Hendren, N. (2018a). The impacts of neighborhoods on
intergenerational mobility I: Childhood exposure effects. Quarterly Journal of
Economics , 133(3), 1107-1162.
Chetty, R., & Hendren, N. (2018b). The impacts of neighborhoods on
intergenerational mobility II: County-level estimates. Quarterly Journal of
Economics , 133(3), 1163-1228.
Chetty, R., Hendren, N., & Katz, L. F. (2016). The effects of exposure to better
neighborhoods on children: New evidence from the moving to opportunity
experiment. American Economic Review , 106(4), 855-902.
Ciacci, R. (2018). The effect of unilateral divorce on prostitution: Evidence from
divorce laws in U.S. states.
“Confidence” Health NGO. (2011). Armenia: Code violations. The International
Baby Food Action Network.
Conway, K. S., & Deb, P. (2005). Is prenatal care really ineffective? Or, is the
’devil’ in the distribution? Journal of Health Economics , 24(3), 489-513.
109
Davis, J., Sellers, S., Gray, C., & Bilsborrow, R. (2017). Indigenous migration
dynamics in the Ecuadorian Amazon: A longitudinal and hierarchical
analysis. Journal of Development Studies , 53(11), 1849-1864.
Demirchyan, A., & Melkomian, D. M. (2020). Main barriers to optimal
breastfeeding practices in Armenia: A qualitative study. Journal of Human
Lactation, 36(2), 318-327.
Dennis, C. L., & McQueen, K. (2009). The relationship between infant-feeding
outcomes and postpartum depression: A qualitative systematic review.
Pediatrics , 123(4), E736-E751.
Donohue, J. J., Aneja, A., & Weber, K. D. (2019). Right-to-carry laws and violent
crime: A comprehensive assessment using panel data and a state-level
synthetic control analysis. Journal of Empirical Legal Studies , 16(2), 198–247.
Eckert, F., Gvirtz, A., Liang, J., & Peters, M. (2020). A method to construct
geographical crosswalks with an application to US counties since 1790.
National Bureau of Economic Research Working Paper Series , No. 26770.
Elmhirst, R. (2002). Daughters and displacement: Migration dynamics in an
Indonesian transmigration area. Journal of Development Studies , 38(5),
143-+.
Fixico, D. (1980). Termination and relocation: Federal indian policy in the 1950’s
(Unpublished doctoral dissertation). University of Oklahoma, Norman, OK.
Gardner, P. (1994). Aboriginal community incomes and migration in the NWT:
Policy issues and alternatives. Canadian Public Policy-Analyse De Politiques ,
20(3), 297-317.
Geib, E. Z. (2001). Do reservation Native Americans vote with their feet? A
re-examination of Native American migration, 1985-1990. American Journal
of Economics and Sociology , 60(4), 815-827.
Goodman-Bacon, A. (2018). Difference-in-differences with variation in treatment
timing. National Bureau of Economic Research Working Paper Series , No.
25018.
Gruber, J. (2004). Is making divorce easier bad for children? The long-run
implications of unilateral divorce. Journal of Labor Economics , 22(4),
799-833.
Gundlach, J. H., Reid, P. N., & Roberts, A. E. (1977). Migration, labor mobility,
and relocation assistance: The case of the American Indian. Social Service
Review , 51(3), 464-473.
110
Gundlach, J. H., & Roberts, A. E. (1978). Native American Indian migration and
relocation: Success or failure. Pacific Sociological Review , 21(1), 117-128.
Hafeez, N., & Quintana-Domeque, C. (2018). Son preference and gender-biased
breastfeeding in Pakistan. Economic Development and Cultural Change,
66(2), 179-215.
Hamill, M. E., Hernandez, M. C., Bailey, K. R., Zielinski, M. D., Matos, M. A., &
Schiller, H. J. (2019). State level firearm concealed-carry legislation and rates
of homicide and other violent crime. Journal of the American College of
Surgeons , 228(1), 1-8.
Harrison, D., & Rubinfeld, D. L. (1978). Hedonic housing prices and demand for
clean air. Journal of Environmental Economics and Management , 5(1),
81-102.
Harutyunyan, S. A. (2004). Armenian case study. Emergency Nutrition Network.
Harutyunyan, S. A. (2015). Armenia assessment report. The World Breastfeeding
Trends Initiative.
Harutyunyan, S. A., Poghosyan, K. P., Saribekyan, K. S., & Ghazaryan, A. E.
(2011). The effect of health care practices on infant feeding. The New
Armenian Medical Journal , 5(1), 36-39.
Hirano, K., Imbens, G. W., & Ridder, G. (2003). Efficient estimation of average
treatment effects using the estimated propensity score. Econometrica, 71(4),
1161-1189.
Huskey, L., Berman, M., & Hill, A. (2004). Leaving home, returning home:
Migration as a labor market choice for Alaska Natives. Annals of Regional
Science, 38(1), 75-92.
Jayachandran, S., & Kuziemko, I. (2011). Why do mothers breastfeed girls less
than boys? Evidence and implications for child health in India. Quarterly
Journal of Economics , 126(3), 1485-1538.
Katz, L. F., Kling, J. R., & Liebman, J. B. (2001). Moving to opportunity in
Boston: Early results of a randomized mobility experiment. Quarterly Journal
of Economics , 116(2), 607-654.
Kling, J. R., Liebman, J. B., & Katz, L. F. (2007). Experimental analysis of
neighborhood effects. Econometrica, 75(1), 83-119.
La Valle, J. M., & Glover, T. C. (2012). Revisiting licensed handgun carrying:
Personal protection or interpersonal liability? American Journal of Criminal
Justice, 37(4), 580–601.
111
Lawrence, J. (2000). The Indian Health Service and the sterilization of Native
American women. American Indian Quarterly , 24(3), 400-419.
Leamer, E. E. (1983). Let’s take the con out of econometrics. American Economics
Review , 73(1), 31-43.
Le Doare, K., Holder, B., Bassett, A., & Pannaraj, P. S. (2018). Mother’s milk: A
purposeful contribution to the development of the infant microbiota and
immunity. Frontiers in Immunology , 9.
Leung, G. M., Ho, L. M., & Lam, T. H. (2002). Maternal, paternal and
environmental tobacco smoking and breast feeding. Paediatric and Perinatal
Epidemiology , 16(3), 236-245.
Liu, J. H., Rosenberg, K. D., & Sandoval, A. P. (2006). Breastfeeding duration and
perinatal cigarette smoking in a population-based cohort. American Journal
of Public Health, 96(2), 309-314.
Lott, J. R., & Mustard, D. B. (1997). Crime, deterrence, and right-to-carry
concealed handguns. Journal of Legal Studies , 26(1), 1-68.
Madigan, L. V. (1956). The American Indian relocation program. New York, NY:
The Association of American Indian Affairs, Inc.
National Research Council. (2004). Firearms and violence: A critical review.
Washington, D.C.: The National Academies Press.
National Statistical Service [Armenia]. (2001). Armenia demographic and health
survey 2000. Calverton, Maryland: National Statistical Service, Ministry of
Health, and ORC Macro.
Pérez-Escamilla, R., Martinez, J. L., & Segura-Pérez, S. (2016). Impact of the
baby-friendly hospital initiative on breastfeeding and child health outcomes: a
systematic review. Maternal and Child Nutrition, 12(3), 402-417.
Pollack, C. E., Blackford, A. L., Du, S., Deluca, S., Thornton, R. L. J., & Herring,
B. (2019). Association of receipt of a housing voucher with subsequent
hospital utilization and spending. Jama-Journal of the American Medical
Association, 322(21), 2115-2124.
Price, J. A. (1968). Migration and adaptation of American Indians to Los Angeles.
Human Organization, 27(2), 168-175.
Raissian, K. M. (2016). Hold your fire: Did the 1996 federal gun control act
expansion reduce domestic homicides? Journal of Policy Analysis and
Management , 35(1), 67-+.
112
Reichman, N. E., Corman, H., Noonan, K., & Dave, D. (2009). Infant health
production functions: What a difference the data make. Health Economics ,
18(7), 761-782.
Reichman, N. E., Corman, H., Noonan, K., & Schwartz-Soicher, O. (2010). Effects
of prenatal care on maternal postpartum behaviors. Review of Economics of
the Household , 8(2), 171-197.
Righard, L., & Alade, M. O. (1990). Effect of delivery room routines on success of
1st breast-feed. Lancet , 336(8723), 1105-1107.
Roberts, D. (2009). Intimate partner homicide: Relationships to alcohol and
firearms. Journal of Contemporary Criminal Justice, 25(1), 67–88.
Rollins, N. C., Bhandari, N., Hajeebhoy, N., Horton, S., Lutter, C. K., Martines,
J. C., . . . Lancet Breastfeeding Series, G. (2016). Why invest, and what it
will take to improve breastfeeding practices? Lancet , 387(10017), 491-504.
Rosenbaum, P., & Rubin, D. (1983). The central role of the propensity score in
observational studies for causal effects. Biometrika, 70(1), 41-55.
Simonsohn, U., Simmons, J. P., & Nelson, L. D. (2020). Specification curve
analysis. Nature Human Behaviour , 4(11), 1208-1214.
Sorkin, A. L. (1969). Some aspects of American Indian migration. Social Forces ,
48(2), 243-250.
Sorkin, A. L. (1971). American Indian and federal aid. Washington, D.C.:
Brookings Institute.
Sorkin, A. L. (1978). The urban American Indian. Lexington, MA: Lexington
Books.
Statisitcal Committee RA. (2018). The demographic handbook of Armenia, 2018.
Sun, L., & Abraham, S. (2020). Estimating dynamic treatment effects in event
studies with heterogeneous treatment effects. Journal of Econometrics , In
Press.
The World Bank. (2017). Sex ratio at birth (male births per female births).
https://data.worldbank.org/indicator/SP.POP.BRTH.MF. (accessed: March
13, 2019)
Thornton, R. (1984). Cherokee population losses during the trail of tears: A new
perspective and a new estimate. Ethnohistory , 31(4), 289-300.
113
Turcksin, R., Bel, S., Galjaard, S., & Devlieger, R. (2014). Maternal obesity and
breastfeeding intention, initiation, intensity and duration: a systematic
review. Maternal and Child Nutrition, 10(2), 166-183.
Ursell, L. K., Metcalf, J. L., Parfrey, L. W., & Knight, R. (2012). Defining the
human microbiome. Nutrition Reviews , 70, S38-S44.
Victora, C. G., Bahl, R., Barros, A. J. D., Franca, G. V. A., Horton, S., Krasevec,
J., . . . Lancet Breastfeeding Series, G. (2016). Breastfeeding in the 21st
century: epidemiology, mechanisms, and lifelong effect. Lancet , 387(10017),
475-490.
Vigdor, E. R., & Mercy, J. A. (2006). Do laws restricting access to firearms by
domestic violence offenders prevent intimate partner homicide? Evaluation
Review , 30(3), 313-346.
Wilkinson, C. F., & Biggs, E. R. (1977). The evolution of the termination policy.
American Indian Law Review , 5(1), 139-184.
World Health Organization. (2009). Baby-friendly hospital initiative: Revised,
updated and expanded for integrated care. Geneva: World Health Organization
and UNICEF.
Wright, A., Rice, S., & Wells, S. (1996). Changing hospital practices to increase
the duration of breastfeeding. Pediatrics , 97(5), 669-675.
Yan, J. (2017). The effects of prenatal care utilization on maternal health and
health behaviors. Health Economics , 26(8), 1001-1018.
Yerevan City Hall. (2010). The results of a month of women’s free medical aid have
been summarized. https://www.yerevan.am/en/news/the-results-of-a-month-
of-women-s-free-medical-aid-have-been-summarized/. (accessed: March 21,
2019)
Yerevan City Hall. (2013). All the registered women will make use of free medical
aid. https://www.yerevan.am/en/news/all-the-registered-women-will-make-
use-of-free-medical-aid/. (accessed: March 21, 2019)
Zeoli, A. M., McCourt, A., Buggs, S., Frattaroli, S., Lilley, D., & Webster, D. W.
(2018). Analysis of the strength of legal firearms restrictions for perpetrators
of domestic violence and their associations with intimate partner homicide.
American Journal of Epidemiology , 187(11), 2365-2371.
Zeoli, A. M., & Webster, D. W. (2010). Effects of domestic violence policies,
alcohol taxes and police staffing levels on intimate partner homicide in large
US cities. Injury Prevention, 16(2), 90-95.
114