Three Essays in Health Economics

(1)

Three Essays in Health Economics: Evidence from U.S. Policies

By

Katherine G. Yewell

Dissertation

Submitted to the Faculty of the Graduate School of Vanderbilt University

in partial fulfillment of the requirements for the degree of

DOCTOR OF PHILOSOPHY in

Economics August 31, 2020 Nashville, Tennessee

Approved:

Michelle Marcus, Ph.D.

Analisa Packham, Ph.D Andrew Dustan, Ph.D.

Paige Skiba, Ph.D.

(2)

(3)

To my parents, who have always believed in me.

(4)

ACKNOWLEDGMENTS

I am tremendously grateful for everyone who has supported me through this experience and has been there through the many ups and downs. I know that I would not be who I am or where I am today without the love and support of my friends and family, and the many others that have helped me along the way.

First, I am incredibly blessed to have my wonderful parents and my brother. My mom gifted me with her creative spirit and intellectual curiosity, and has always been my biggest cheerleader.

She taught me to take pride in my accomplishments, to embrace my many quirks, and to give myself permission to recharge when needed. My dad is my rock and has always made himself available to help me in any way he can. He taught me how to find humor and lessons in the trials that life brings, and the importance of having people you can rely on when those trials arise. Most importantly, he taught me to “endeavor to persevere!” To my brother, who is incredibly smart and funny: thank you for sharing your dog with me when I needed snuggles and emotional support, for helping me move all of my stuff across states multiple times, and for all of the fun memories that we share. I’m a proud big sister and I know you can do anything you set your mind to.

I’m also thankful for my new family. My fianc´e, Scott Kriebel, has been a source of encouragement and support through the last few years and reminded me that I should be enjoying the prime years of my life. He has opened up a whole new world of experiences and brightened my existence with his enthusiasm and curiosity for life. I’m inspired every day by his work ethic and generosity and look forward to growing and building a life with him. I’m also incredibly grateful to Scott’s parents for bringing me into their family and always making me feel at home. Their love and support has meant a lot to me these past few years.

Next, I want to thank my many friends at Vanderbilt, particularly Emily Lawler, James Harri- son, Sebastian Tello-Trillo, Daniel Mangrum, Ben Ward, Aaron Gamino, Paul Niekamp, Salama Freed, Evan Elmore, Danielle Drago Drory, Jonah Yuen, Alper Arslan, Nick Ma¨eder, Jason Campbell, Michael Mathes, Katie Fritzdixon, and Chris Cotter. I could not have survived grad-

(5)

uate school without them! In addition to friendship and many fond memories, my peers have offered me support and guidance throughout my academic journey that is invaluable. I must especially acknowledge Emily Lawler for being the best of best friends and for always being there for me personally as well as academically. I have learned a lot from working on research with her and greatly appreciate her encouragement and feedback throughout the process of writing my dissertation.

This work would not have been possible without the guidance and support I received from my professors at Vanderbilt University. In particular I would like to thank my committee members Michelle Marcus, Analisa Packham, Andrew Dustan, and Paige Skiba for their time, feedback, and encouragement throughout the process of writing my dissertation. I would additionally like to thank Kitt Carpenter, Jennifer Reinganum, Andrew Daughety, Alejandro Molnar, and John Weymark who have also been generous with their time, advice, and support through the past eight years, for which I am very grateful. I am especially indebted to my advisor, Michelle Marcus, who has been particularly patient and generous with her time as I developed the work in this dissertation. In addition to helping me become a better researcher and economist, she has also been a mentor and a friend, providing constant support and advice as I have pursued both my career and life goals.

In addition to my Vanderbilt professors, I have been blessed throughout my life to have many other wonderful teachers that nurtured my love of learning and appreciation for quantitative rea- soning. My journey in economics began at Rhodes College where my professors fostered an early interest in both economics and advanced mathematics. I’m grateful to my professors there, especially Marshall Gramm, Teresa Gramm, Art Carden, and Jeff Hamrick for their enthusiasm for teaching and supporting students. My experience in their classrooms is one of the reasons I pursued an academic career, and I model my own commitment to teaching after their example.

I’m grateful to others that have recognized or appreciated my passion for teaching, which has given me purpose throughout graduate school. In particular, thank you to the many professors I worked with at Vanderbilt as a teaching assistant who have written recommendation letters for

(6)

various opportunities throughout the years. I’m grateful to Joel Rodrigue for giving me the chance to teach my own courses at Vanderbilt, which offered financial support, purpose, experience, and opportunities that would otherwise not have been possible. I’m also especially appreciative of Jamin Speer, who has been a generous friend, mentor, and confidant, and who has encouraged me to be more confident in my professional life. I’m thankful for my other colleagues at the University of Memphis, and especially my department chair Bill Smith, for their encouragement and support in my young professional career.

Finally, I want to thank a few other members of the Vanderbilt community that have been particularly supportive during my tenure there. Samantha York at the Center for Student Wellbeing was a breath of fresh air when I needed it, providing a non-judgmental safe place to talk about any personal or academic struggles as well as successes. Everyone needs a Samantha in their life.

I’m also grateful to David Heustess at the Sarratt Art Studios for providing a creative space that introduced me to pottery, which helped me find my center.

(7)

TABLE OF CONTENTS

Page

DEDICATION . . . iii

ACKNOWLEDGMENTS . . . iv

LIST OF TABLES . . . x

LIST OF FIGURES . . . xii

INTRODUCTION . . . 1

Chapter 1 THE EFFECT OF HOSPITAL BREASTFEEDING LAWS ON MATERNAL BEHAVIORS ANDHOUSEHOLD TIMEALLOCATION . . . 5

1.1 Introduction . . . 5

1.2 Background . . . 10

1.2.1 Breastfeeding Benefits . . . 10

1.2.2 Breastfeeding policy . . . 11

1.3 Data Description . . . 15

1.4 Empirical Strategy . . . 19

1.5 Main Results . . . 21

1.5.1 Descriptive Statistics . . . 21

1.5.2 Effects on Breastfeeding . . . 22

1.5.3 Effects on Maternal Smoking . . . 24

1.5.4 Effects on Household Time Use . . . 27

1.6 Additional Results . . . 28

1.6.1 Heterogeneous Policy Effects by Subpopulation . . . 28

1.6.2 Mechanisms . . . 29

1.6.3 Maternal Selection . . . 30

(8)

1.7 Conclusion . . . 31

1.8 References . . . 34

1.9 Figures . . . 38

1.10 Tables . . . 40

1.11 Appendix . . . 46

2 T^HE E^{FFECT OF} F^REE S^CHOOL M^{EALS ON} HÔUSEHOLD FÔOD PÛRCHASES: E^VI- DENCE FROM THE CÔMMUNITYE^LIGIBILITYP^ROVISION . . . 66

2.1 Introduction . . . 66

2.2.1 Policy Environment . . . 70

2.2.2 CEP Roll Out and School Adoption . . . 71

2.3 Data & Methods . . . 73

2.3.1 School-Level CEP Participation . . . 73

2.3.2 First Stage School Meal Data . . . 73

2.3.3 Household Purchase Data . . . 74

2.3.4 Food Insecurity Data . . . 77

2.3.5 Defining Treatment . . . 78

2.5 Results . . . 83

2.5.2 First Stage Effect of CEP Adoption on School Meals Served . . . 83

2.5.3 Effect of CEP Exposure on Grocery Purchases . . . 85

2.5.4 Heterogeneity of the Effect on Grocery Purchases . . . 88

2.5.5 Effects of CEP Exposure on Food Insecurity . . . 89

(9)

2.8 Figures . . . 95

2.9 Tables . . . 102

2.10 Appendix . . . 109

3 T^HEE^{FFECT OF} U^NIVERSALA^{CCESS TO}F^REES^CHOOL M^{EALS ON} T^EENH^EALTH . 116 3.1 Introduction . . . 116

3.3 Literature Review . . . 123

3.4 Data Description . . . 127

3.4.1 State CEP Participation and Controls . . . 127

3.4.2 Teen Health Outcomes . . . 129

3.4.3 Teen Doctor Visits . . . 130

3.4.4 Time Use . . . 130

3.6 Main Results . . . 134

3.6.2 Effects of CEP on Teen Health . . . 135

3.6.3 Robustness of the Effects of CEP on Teen Health . . . 137

3.6.4 Heterogeneity in the Effects of CEP on Teen Health . . . 137

3.6.5 Supplemental Results: Effects of CEP on Parent Time Use . . . 138

3.9 Figures . . . 145

3.10 Tables . . . 150

3.11 Appendix . . . 154

(10)

LIST OF TABLES

Table Page

1.1 Effects of Hospital Breastfeeding Support Laws on Breastfeeding Initiation and Duration, NIS-Child (2003-2017) . . . 40 1.2 Effect of Hospital Breastfeeding Support Laws on Maternal Smoking . . . 41 1.3 Effect of Hospital Breastfeeding Support Laws on Smoking, by age of infant . . . 42 1.4 Effects of Hospital Breastfeeding Support Laws on Parent Time Use, ATUS

(2003-2018) . . . 43 1.5 Effects of Hospital Breastfeeding Support Laws on Care During Postpartum Hos-

pital Stay, PRAMS (2000-2018) . . . 45 1.6 Descriptive Statistics, NIS-Child 2003-2017 . . . 55 1.7 Robustness of Breastfeeding Effects to Specification Choices, NIS-Child (2003-

2017) . . . 56 1.8 Heterogeneity in the Effects of Hospital Breastfeeding Support Laws, NIS-Child

2003-2017 . . . 57 1.9 Effects of Hospital Breastfeeding Laws on Time Use for RespondentsWithouta

Baby, ATUS (2003-2018) . . . 59 1.10 Effects of Hospital Breastfeeding Support Laws on Breastfeeding Initiation and

Duration, PRAMS (2000-2018) . . . 60 1.11 Effects of Lactation Consultant Law Component on Breastfeeding Initiation and

Duration, NIS-Child (2003-2017) . . . 61 1.12 Effect of Implementation of Baby Friendly Hospital Laws on ln(State Lactation

Consultants), 2006-2016, 2018 . . . 62 1.13 Maternal Characteristics Following the Implementation of Hospital Breastfeed-

ing Support Laws, NIS-Child (2003-2017) . . . 63 1.14 WHO/UNICEF “Ten Steps to Successful Breastfeeding” . . . 64

(11)

1.15 Time Use Categories . . . 65

2.1 Summary Statistics for Adopting versus Non-Adopting Zip Codes . . . 102

2.2 CEP Exposure and Household Characteristics . . . 103

2.3 First Stage Effect of CEP on NSLP Meals Served . . . 104

2.4 Effect of Overall CEP Exposure on Grocery Purchases . . . 105

2.5 Effect of School-Level CEP Exposure on Grocery Purchases . . . 106

2.6 Effect of Overall CEP Exposure on Health Categories of Food Spending . . . 107

2.7 Effect of Overall CEP Exposure on Food Spending by Lunch Eligibility . . . 107

2.8 Effect of CEP on Food Insecurity for Households with Kids (CPS December Supplement) . . . 108

2.9 Categories in Authors’ Defined Lunch Group . . . 111

2.11 USDA Health Categories . . . 111

2.12 Effect of School-Level CEP Exposure on Health Categories of Food Spending . . 114

2.13 Robustness of Main Results . . . 115

3.1 Descriptive Statistics . . . 150

3.2 The Effect of CEP on Teen Health (YRBSS) . . . 151

3.3 Effect of CEP on Teen Health (NIS-Teen) . . . 152

3.4 Effect of CEP on Parent Time Use (ATUS) . . . 153

3.5 Timing of CEP Exposure and Teen/State Characteristics . . . 156

3.6 The Effect of CEP on Teen Health (YRBSS), with alternate treatment definitions 157 3.7 The Effect of CEP on Teen Health (YRBSS), robustness to alternate specifications 158 3.8 The Effect of CEP on Teen Health (YRBSS), decomposed by gender . . . 159

3.9 The Effect of CEP on Teen Health (YRBSS), decomposed by race/ethnicity . . . 160

3.10 Time Use Categories . . . 161

(12)

LIST OF FIGURES

Figure Page

1.1 Timing of Adoption of Baby Friendly Hospital Laws Across States . . . 38

1.2 Event Study Estimates of the Effect of Hospital Breastfeeding Support Laws on Breastfeeding Outcomes, NIS-Child, 2003-2017 . . . 39

1.3 Timing of State Law Adoption and Sample Periods of Primary Data Sources . . . 46

1.4 Strength of State Breastfeeding Policies . . . 47

1.5 Components of State Breastfeeding Policies . . . 48

1.6 PRAMS Data Availability . . . 49

1.7 Event Study Estimates of the Effect of the Strength of Hospital Breastfeeding Support Laws on Breastfeeding Outcomes, NIS-Child, 2003-2017 . . . 50

1.8 Event Study Estimates of the Effect on Exclusive Breastfeeding Outcomes, NIS- Child, 2003-2017 . . . 51

1.9 Event Study Estimates of the Effect on Breastfeeding Outcomes PRAMS 2000- 2018 . . . 52

1.10 Event Study Estimates of the Effect on Maternal Smoking . . . 53

1.11 Event Study Estimates of the Effect on Time Use, ATUS, 2003-2018 . . . 54

2.1 School Adoption Timing Relative to State CEP Roll Out . . . 95

2.2 Percent of Zip Code Areas with CEP . . . 96

2.3 CEP Zip Code Exposure by School Year . . . 97

2.4 First Stage Effect of CEP on NSLP Meals Served . . . 98

2.5 Effect of CEP on Grocery Purchases for Households with Kids . . . 99

2.6 Effect of CEP on Grocery Purchases for Households with Kids, by Lunch Eligibility100 2.7 Food Insecurity for Households with Kids . . . 101

2.8 State Rollout of CEP . . . 109

(13)

2.9 Effect of CEP on Healthy/Unhealthy Food Spending for Households with Kids . 110

3.1 Timing of School CEP Adoption Relative to State CEP Availability . . . 145

3.2 Percent of All State Schools Adopting CEP by Year . . . 146

3.3 Effect of CEP on Teen Health Outcomes (YRBSS) . . . 147

3.4 Effect of CEP on Teen Doctor Visits (NIS-Teen) . . . 148

3.5 Effect of CEP on Parent Time Use (ATUS) . . . 149

3.6 YRBSS Observations by State and Year . . . 154

3.7 Percent of State High Schools Adopting CEP by Year . . . 155

(14)

INTRODUCTION

In the United States, food security and nutrition have become public health priorities for both children and adults, as many families lack adequate nutrition or reliable access to food.

Additionally, for infants it is argued by medical professionals that breastfeeding is the best source of nutrition, and yet breastfeeding rates in the U.S. remain low. In this dissertation I use rigorous empirical methodologies to examine the health and behavioral consequences of two policies that have been widely implemented in recent years in response to these public health concerns. My results point to spillover effects and previously unexplored costs and benefits that have important policy implications.

In the first chapter, The Effect of Hospital Breastfeeding Laws on Maternal Behaviors and Household Time Allocation(joint work with Emily C. Lawler), we provide novel evidence on the effects of state policies that require immediate post-partum in-hospital interventions to promote breastfeeding. The medical literature suggests that breastfeeding is associated with both short- and long-run health benefits for the mother and infant, yet breastfeeding rates in the U.S. remain low and many mothers do not meet recommendations set forth by the World Health Organiza- tion. The state laws we study here are an understudied subset of the policies targeting potential barriers to breastfeeding, and include a range of requirements for hospitals such as informing all new mothers of the benefits of breastfeeding, regularly training staff on initiation and support of lactation, and having a lactation consultant on staff. While seventeen states have adopted one of these laws, very little is known about their effects. We use a difference-in-differences strategy that allows us to take advantage of plausibly exogenous variation across states in the timing of policy adoption, and therefore avoid concerns that expecting mothers may positively select into individual hospitals that voluntarily adopt breastfeeding support policies. Using self-reported breastfeeding outcomes from the National Immunization Survey (NIS), we show that these laws were successful in achieving their intended goal: following law adoption, mothers are significantly more likely to initiate and sustain breastfeeding. In a heterogeneity analysis, we find the

(15)

largest effects for non-Hispanic Black mothers. Since these mothers also have lower rates of breastfeeding initially, the hospital breastfeeding support laws may serve to reduce disparities along this dimension.

Using data from the American Time Use Survey (ATUS) and the CDC’s Pregnancy Risk Assessment Monitoring System (PRAMS), our results also suggest that these laws had substantial spillover effects to maternal smoking and to household time allocation. Specifically, we find that adoption of hospital breastfeeding laws resulted in significant reductions in several measures of maternal smoking, with the reduction primarily occurring among mothers of relatively younger infants, which is the same group that experiences the largest increases in breastfeeding. We also find that these laws resulted in significant reallocation of maternal time: after implementation, mothers spend more time on child care and unpaid domestic work, and less time on formal work.

On the other hand, we find no significant changes in time use for fathers, nor for individuals in households without an infant. These behavioral complementarities have important empirical implications for research attempting to quantify the benefits of breastfeeding.

In the remaining chapters, I study a novel food and nutrition program that targets school- aged children and analyze spillover effects and previously unexplored benefits. This program, known as the Community Eligibility Provision (CEP), provides children with universal access to free meals at school. CEP is a part of the National School Lunch Program (NSLP), which provides free or reduced-price school meals to children in low-income families and is one of the largest federal safety net programs in the United States. Despite potential access to free and reduced-price meals, many low-income students do not participate in the traditional NSLP due to application costs (time and effort on the part of the parent or guardian) and the stigma associated with receiving free school meals. CEP addresses this concern by eliminating individual applica- tions and granting universal free school meals to all students in a participating school or district.

In this way, CEP reduces the cost of school meals for all students in the school: those that would have previously been eligible and may or may not have been partaking in free meals, as well as those that were previously ineligible for free meals based on their household income. Prior work

(16)

on the effect of CEP has focused on academic gains including improved test scores and modest reductions in school suspensions. However, this program also allows us a unique opportunity to examine additional spillover effects on household decision-making as well as health benefits associated with increasing access to food.

The second chapter, The Effect of Free School Meals on Household Food Purchases: Ev- idence from the Community Eligibility Provision (joint work with Michelle M. Marcus), focuses on changes in spending for households with children when nearby schools adopt CEP.

Our difference-in-differences strategy takes advantage of plausibly exogenous changes in household access to free school meals based on their zip code of residence and the timing of CEP adoption across schools. Using the Nielsen Consumer Panel, 2004-2016, we find that household food spending decreases by about $22 per month after children gain access to free school meals, which represents almost 11% of mean food spending at grocery stores. Of this $22, about $9 is due to a decrease in spending on typical lunch foods. We also find some evidence that the composition of food purchases is changing, with households buying less unhealthy food such as commercially prepared pre-packaged products, but also fewer healthy options such as vegetables.

The overall diet healthiness of at-home food spending decreases after CEP by about 13% from the mean health score of food purchases. These changes suggest that CEP may be altering the food consumption of other household members in addition to altering the food available to children in CEP participating schools. We also find that the spending effects are largest for families that would have been already eligible for free school lunch prior to CEP, suggesting that eliminating the application and stigma costs has a real effect on their household budget and behavior.

Finally, we provide novel evidence from the December supplement to the Current Population Survey that households with school-aged children experienced reduced food insecurity following CEP availability. Food insecurity is defined as a lack of dependable access to food due to money or other resources, and is associated with a lack of nutrition, psychological, and physiological stress that can result in adverse health outcomes. This suggests that CEP can have other health benefits by improving food security for households with school-aged children.

(17)

In the final chapter, The Effect of Universal Access to Free School Meals on Teen Health, I explore this possibility and consider the effects of CEP on a variety of new health outcomes for high school students. Using a similar empirical strategy that leverages changes in CEP availability across states over time, I show evidence that access to CEP decreased food insecurity and resulted in improved health for teens. In particular, I find that high school students go to the doctor less, have fewer incidences of asthma, are more likely to get eight or more hours of sleep, and report improved mental health following the increased access to meals at school.

Overall, these three chapters provide novel empirical evidence on the health and behavioral consequences of two recent policies rolled out to address food security and nutrition. My results point to spillover effects and previously unexplored costs and benefits of these policies that have important implications for policymakers looking to address other key health and well-being outcomes for families. In particular, the unique design features of both sets of policies can be leveraged in future law initiatives targeting similar outcomes.

(18)

CHAPTER 1

T^HEE^{FFECT OF} H^OSPITALBREASTFEEDINGL^{AWS ON}M^ATERNAL BEHAVIORS AND

H^OUSEHOLDT^IME A^LLOCATION

With Emily C. Lawler

1.1 Introduction

Breastfeeding is widely considered to be an important parental investment in child health and development. A large body of research in the medical literature shows breastfeeding is associated with positive infant and maternal outcomes (Eidelman and Schanler, 2012; Ip et al., 2007);

in light of this research, both the American Academy of Pediatrics (AAP) and the World Health Organization (WHO) recommend that infants be breastfed for at least the first year of life (AAP, 2012; WHO, 2011). Despite this, in the United States only 84 percent of mothers ever initiate breastfeeding, and only 36 percent of infants are still breastfed at one year (CDC, 2019). In an effort to increase these persistently low rates of breastfeeding initiation and duration, policymakers have implemented a broad set of policies which target a range of potential barriers to breastfeeding, including provision of workplace accommodations, insurance coverage of lactation-related services, and information-based interventions.

We examine the effects of one such policy - state hospital breastfeeding support laws - which are intended to increase breastfeeding by requiring certain care standards for new mothers and their babies during their hospital stay. Over the past two decades, adoption of these laws has become increasingly widespread, and as of 2019, seventeen states have implemented a hospital breastfeeding support law. Although the specifics of the laws vary across states, frequent requirements include that all new mothers be informed of the benefits of breastfeeding, that hospital staff be regularly trained on initiation and support of lactation, and that there be a lactation consultant on staff. In spite of their widespread adoption, very little is known about the effects of these laws.

(19)

In this paper, we provide novel evidence on the effects of hospital breastfeeding support laws on a range of maternal behaviors. To identify the effects of the laws we estimate difference-in- differences models that leverage plausibly exogenous variation across states in the timing of law adoption. We first examine the effects on breastfeeding initiation and duration, using self-reported breastfeeding outcomes from the National Immunization Survey-Child (NIS-Child). Using data from the American Time Use Survey (ATUS) and the CDC’s Pregnancy Risk Assessment Mon- itoring System (PRAMS), we next consider potential spillover effects of the laws. Specifically, we estimate the impact of the laws on maternal smoking and on household time allocation across child care, formal work, unpaid domestic work, and leisure. Finally, we provide suggestive evidence on the mechanisms through which these laws affect outcomes.

Our results show that the hospital breastfeeding support laws were successful at increasing both initiation and duration of breastfeeding. We find that after the adoption of a law, the probability of breastfeeding initiation increases by 4.1 percentage points, and breastfeeding at 6 months and 1 year by 3.8 and 1.6 percentage points, respectively. Across these different outcomes, estimated effects consistently represent a 5.4 to 8.7 percent increase relative to the respective outcome mean. We also explore potential heterogeneity in these effects across different subgroups of mothers, and find that the laws had the largest effect among non-Hispanic Black mothers. No- tably, in our baseline year, non-Hispanic Black mothers are nearly 17 percentage points less likely to initiate breastfeeding than white mothers, and this gap persists for measures of breastfeeding duration. Our findings suggest that hospital breastfeeding support laws may reduce disparities in breastfeeding initiation and duration.

We also find strong evidence that these laws had unintended effects on other maternal behaviors. First, we show that the hospital breastfeeding laws resulted in significant reductions in maternal smoking during the postpartum period. While maternal smoking is not recommended regardless of infant feeding method, evidence shows that chemicals found in cigarettes, including nicotine, can be passed through the breast milk to the infant (CDC, 2019; AAP, 2013). Thus, maternal smoking may change in response to the adoption of a hospital breastfeeding law if the

(20)

law increases either the probability that the recommendation to not smoke is conveyed to mothers by hospital staff or if it increases the overall salience of the recommendation. Consistent with the idea that the observed reduction in maternal smoking is driven by the change in breastfeeding, we find that the reductions in smoking are largest for women with the youngest infants (i.e., the same group which experiences the largest increase in breastfeeding). We also find no evidence of changes in smoking for women without infants in their household.

Second, we find that the hospital breastfeeding laws substantially impacted maternal time allocation across child care, formal work, and unpaid domestic work. Changes in breastfeeding behavior may result in changes in maternal and household time use since it requires maternal time to be spent on infant feeding (either breastfeeding or pumping), while other types of infant feeding do not; we may also observe changes in household time allocation if breastfeeding is generally more (or less) time intensive than other methods of infant feeding.¹ We find that after a hospital breastfeeding law is implemented, the overall time households spend on basic childcare and on unpaid domestic work increases, suggesting breastfeeding is more time intensive relative to other forms of infant feeding. These increases primarily occur for females, and, perhaps most strikingly, our results show that women offset their increased childcare and domestic work time burden by significantly reducing their time spent on formal work. As with the smoking outcomes, we find that these changes in time use are driven by households with the youngest infants, and find no evidence of changes in time use for households without infants.

Finally, we provide evidence on the mechanisms of the laws, and show that they increased the probability that breastfeeding mothers report being helped to breastfeed by hospital staff, being allowed to breastfeed on demand during their hospital stay, or being connected with a breastfeeding support group prior to hospital discharge. We also provide suggestive evidence that laws which required hospitals to have a lactation consultant on staff resulted in an increase in the number of International Board Certified Lactation Consultants (IBCLCs) in the state. We supplement this with a set of analyses in which we characterize the hospital breastfeeding laws

1Throughout we use the term “breastfeeding” to refer to both breastfeeding directly, or to pumping breast milk

(21)

based on whether they do or do not require the hospital to have a lactation consultant on staff, and then re-estimate the effects of the laws on breastfeeding outcomes. These results suggest that provision of a lactation consultant during the postpartum period is independently important for sustained breastfeeding.²

This paper makes a number of contributions to the literature on the effects of breastfeeding policies, and on the determinants of parental investment more broadly. Our analysis of state-level hospital breastfeeding support laws provides the first causal evidence on the effects of immediate postpartum interventions on initiation and duration of breastfeeding. While a small set of papers in the medical and public health literature have examined the effects of similar hospital-level policies, they either rely on cross-sectional comparisons, or they fail to address endogenous selection of a delivery hospital (Kuan et al., 1999; DiGirolamo et al., 2001; Taddei et al., 2000; Kramer et al., 2001; Coutinho et al., 2005; Philipp et al., 2001; Hawkins et al., 2015).³ Fitzsimons and Vera-Hern´andez (2015, 2016) leverage variation in access to hospital lactation support induced by staff scheduling, and show that mothers that give birth on the weekends are less likely to receive lactation support in the hospital and also less likely to breastfeed. Beyond these papers, most of the existing literature on the determinants of breastfeeding has focused on policies that apply only after the mother and child leave the hospital, such as paid family leave (e.g., Baker and Milligan, 2008; Pac et al., 2019), laws that address breastfeeding in the workplace (Hawkins et al., 2013), or laws mandating insurance coverage of lactation support services and equipment (Kapinos et al., 2017; Gurley-Calvez et al., 2018). Given that breastfeeding is an extremely time sensitive parental investment,⁴ analyzing policies that target the immediate postpartum period is

2Ideally, we would fully characterize each law based on the specific set of components it contains, in order to identify which policy component is most important for affecting outcomes. Unfortunately, however, because states adopt these law components in bundles we are limited in our ability to separately identify the effects of individual components. As the requirement to provide a lactation consultant is relatively well-identified, we focus on this law component.

3The exception to this is Andersen (2019), which examines the effects of hospital-level policies, and overcomes the endogenous selection issue by assigning treatment exposure based on the treatment status of the hospital closest to the mother’s residence, as opposed to the treatment status at the hospital the mother delivers at. That paper, however, does not examine breastfeeding outcomes.

4In particular, milk removal from the breasts soon after birth is associated with increased efficiency of milk production; if milk is not removed then biological mechanisms cause the cells to stop producing milk. Thus, the

(22)

crucial for understanding the determinants of breastfeeding.

Second, as the first to credibly identify the behavioral complementarity between maternal breastfeeding and smoking, we contribute to the strand of the literature that focuses on the determinants of smoking during pregnancy and the postpartum period. The existing literature consistently shows that maternal smoking is responsive to a range of targeted smoking policies, such as as counseling interventions, indoor smoking bans, and cigarette excise taxes (Chamberlain et al., 2017; McGeary et al., 2019; Colman et al., 2003). Our results, however, show that maternal smoking is also responsive to breastfeeding policies, thus suggesting a complementarity between the two behaviors. Additionally, our finding that mothers reinforce their initial positive investment in their child (breastfeeding) through another positive investment (not smoking) has important empirical implications regarding the estimated benefits of breastfeeding. In particular, it suggests that the existing literature may have overestimated the infant health benefits of breastfeeding, as some of the observed changes may have been driven by complementary positive parental investment.⁵

We also contribute to the sparse literature that examines the determinants of household allocation of time across formal work and childcare. Given that breastfeeding is a uniquely gendered activity, it likely has different effects on household time allocation relative to most other shocks that have been examined, such as unemployment or changes in childcare prices (Gorsuch, 2016;

Amuedo-Dorantes and Sevilla, 2014).⁶ In particular, our finding that hospital breastfeeding laws decrease the time women spend on formal work suggests that policies aimed at increasing breastfeeding may have the unintended consequence of reducing maternal attachment to the labor force.

The rest of the paper proceeds as follows: section 1.2 gives some background information

(Neville and Morton, 2001; Hurst, 2007).

5Notably, there is some evidence in the medical literature suggesting that breastfeeding is associated with reductions in ear infections, asthma, respiratory infections, and sudden infant death syndrome (SIDS) (Ip et al., 2007;

Eidelman and Schanler, 2012), all of which are also associated with secondhand smoke exposure during infancy (US DHHS, 2014).

6Most similar to our context is work examining the impacts of California’s 2004 Paid Family Leave Act (PFLA), which, although available to fathers and mothers, primarily increased maternal leave-taking. Trajkovski (2019) finds that in response to PFLA mothers significantly increase their time spent on childcare, even after they return to work, while (Bailey et al., 2019) show that it significantly reduced employment for first time mothers in both the short and

(23)

on the research regarding the benefits of breastfeeding, as well as some existing policies that address breastfeeding. In sections 1.3 and 1.4, we describe our data sources and empirical strategy, respectively. Our main results on breastfeeding and maternal behavior are presented in section 1.5, and we supplement those with some additional results in section 1.6. Finally, section 1.7 concludes.

1.2 Background

1.2.1 Breastfeeding Benefits

There is a substantial body of research in the medical literature that shows breastfeeding is associated with a wide range of positive short- and long-run outcomes for the child (Eidelman and Schanler, 2012; Ip et al., 2007). Much of this literature, however, relies on cross-sectional variation in breastfeeding across children, and therefore is unable to address important unobserved confounders that may drive both breastfeeding behavior and other positive outcomes. The causal evidence on the benefits of breastfeeding is much sparser, and primarily comes from a large randomized breastfeeding support intervention conducted in Belarus in the 1990s. This study shows that increased breastfeeding significantly reduces gastrointestinal infections, eczema, and other skin rashes in the first year of life. However, the study also examines a broad array of other outcomes, including respiratory infections during the first year of life, allergies, asthma, height, body mass index, and blood pressure in early childhood, and measures of child behavior, cog- nitive development, and mother-child bonding, and finds no evidence of benefits across those dimensions (Kramer et al., 2001, 2007, 2008; Yang et al., 2018).

Although the causal evidence on the short- and long-run benefits to breastfeeding is limited, it is heavily promoted as the best method of infant feeding. For example, the Centers of Dis- ease Control and Prevention (CDC) refers to breast milk as “the clinical gold standard”⁷ and the AAP says it is “uniquely superior for infant feeding” (AAP, 1997). Moreover, the World Health

7See: https://www.cdc.gov/breastfeeding/about-breastfeeding/why-it-matters.html. Last Accessed: July 7,

(24)

Organization (2011) and the American Academy of Pediatrics (2012) recommend that, unless medically contraindicated, babies should be exclusively breastfed for the first 6 months of life with continued breastfeeding recommended through at least 1 year of age.⁸ Additionally, survey evidence suggests that the vast majority of mothers in the United States believe breastfeeding to be the best method of infant feeding. In the Infant Feeding Practices Survey II, a nationwide survey conducted in 2005, 79 percent of pregnant women reported believing that breastfeeding was the best way to feed an infant and nearly 47 percent of pregnant women reported that they believed infants should be breastfed exclusively for the first 6 months.⁹

1.2.2 Breastfeeding policy

Despite the WHO and AAP breastfeeding recommendations and the widespread belief regarding the benefits of breastfeeding, rates of initiation and duration in the United States are persistently low. For infants born in 2016, nearly 16 percent of mothers never initiated breastfeeding, only 25 percent were exclusively breastfeeding when the infant was 6 months old, and only 36 percent were breastfeeding at all when the infant reached one year of age (CDC, 2019).

There are also persistent disparities in breastfeeding rates; in particular, Black mothers are consistently much less likely to breastfeed than either white or Hispanic mothers.¹⁰

As a result, improving breastfeeding initiation and duration has long been a public health priority in the United States. Notably, in 2011 the U.S. Surgeon General issued a call to action to support breastfeeding (US DHHS, 2011), and improvements in breastfeeding outcomes have been specific objectives of the Department of Health and Human Services “Healthy People”

initiative for the past several decades (CDC, 2001; US DHHS, 2019). National-level policies that have explicitly aimed to improve breastfeeding outcomes include the Special Supplemental

8Medical contraindications to breastfeeding are rare, but can be due to conditions of either the infant (e.g.

glactosemia) or the mother (e.g. human T-cell leukemia virus type I) (AAP, 2012).

9See: https://www.cdc.gov/breastfeeding/data/ifps/results.htm. Last accessed: July 7, 2020.

10Evidence from the IFPS II does not demonstrate substantial differences across races/ethnicities in terms of beliefs about the benefits of breastfeeding. We note, however, that the IFPS II is not nationally representative, and

(25)

Nutrition Program for Women, Infants, and Children (WIC) program and the Affordable Care Act (ACA).¹¹ There is also a patchwork of state-level policies, the majority of which focus on breastfeeding rights in the workplace or mothers’ legal rights to breastfeed in a given location (Hawkins et al., 2013). Nearly all states currently allow breastfeeding in any public or private location; the majority also exempt breastfeeding mothers from public indecency laws.

The state laws we study, hospital breastfeeding support laws, are unique relative to the other state-level breastfeeding interventions in that they focus on the immediate postpartum period and serve to set standards for the care that hospitals provide to new mothers. Although these laws vary across states, the most frequent requirements include the following: (1) hospitals must have a lactation consultant on staff, (2) patients must be informed about the benefits of breastfeeding, (3) obstetric staff must receive regular lactation training, (4) hospitals must develop a written policy promoting breastfeeding, and (5) that patients must be permitted to have their baby stay with them 24 hours a day (“rooming in”). We provide more detail on the specific provisions of each of the state laws in Appendix Figure 1.5.

To the best of our knowledge, we are the first to examine the effects of state-level hospital breastfeeding support laws. There are, however, a number of papers in the medical and public health literature that have examined the breastfeeding effects of hospital-level implementation of a similar intervention: the international Baby-Friendly Hospital Initiative’s (BFHI) “Ten Steps to Successful Breastfeeding” program. The Ten Steps program, which was launched in 1991 by the World Health Organization (WHO) and UNICEF (UNICEF, 2005), outlines a set of ten recommended policies for hospitals to implement that are intended to increase breastfeeding initiation and duration. These ten policies correspond closely to the components of the laws we study,¹²and include adoption of a written breastfeeding policy, training staff on lactation support,

11WIC primarily provides health and nutrition assistance for low-income pregnant or postpartum women and their children, but since the 1980s it has also provided breastfeeding education and support (Oliveira and Fraz˜ao, 2015). Notably, WIC also provides free infant formula (via vouchers) for infants whose mothers choose not to exclusively breastfeed. Several components of the Affordable Care Act explicitly pertain to breastfeeding, including the requirements that employers provide adequate break time and space for employees to express milk, and that all new insurance policies in the individual and group market, and new Medicaid coverage provided under the Medicaid expansion, cover lactation support and equipment rental with no cost sharing (Hawkins et al., 2015).

(26)

and informing all pregnant women on the benefits of breastfeeding (see Appendix Table 1.14 for all ten steps). If hospitals implement all ten steps, as well as follow the International Code of Marketing of Breast-Milk Substitutes regarding the advertising and provision of breast-milk substitutes to families, they are then designated as a Baby-Friendly^® hospital.¹³

The primary evidence on the effects of hospital adoption of the BFHI Ten Steps program is from two randomized control trials (RCTs) implemented in Brazil and one implemented in Belarus. The results of the RCTs show that in those contexts the implementation of the BFHI Ten Steps significantly increased the duration of breastfeeding by approximately 20% (Taddei et al., 2000; Kramer et al., 2001; Coutinho et al., 2005). Notably, however, although these studies randomized the implementation of the BFHI policy across hospitals and are therefore able to overcome concerns regarding endogenous adoption of the policy at the hospital level, the designs of the RCTs do not enable them to address concerns about mothers differentially sorting across hospitals in response to policy adoption. That is, since women do not randomly choose their delivery hospital, changes in hospital policy may cause a change in the composition of women choosing to give birth there. As we expect women to positively select into giving birth at a Baby Friendly^® Hospital, this selection problem could cause studies that rely on hospital –level variation to overestimate the effects of the policy on breastfeeding outcomes.¹⁴

Much of the other work on the effects of BFHI policies has relied on cross-sectional or before- and-after comparisons of hospital-level breastfeeding outcomes (see, for example, Philipp et al., 2001; Kuan et al., 1999; DiGirolamo et al., 2001); in the U.S. only two studies have used a

For example, California’s regulation states that hospitals must have an infant-feeding policy that promotes breastfeeding, and this policy should “follow guidance provided by the Baby-Friendly Hospital Initiative or the State Department of Public Health Model Hospital Policy Recommendations” (Cal. Health & Safety Code 123366(c)).

13See https://www.babyfriendlyusa.org/about/

14Additionally, there are a number of reasons why we may expect different results from the implementation of Baby Friendly Hospital policies in the United States, as compared to Belarus and Brazil. Notably, the maternity leave policies in both Belarus and Brazil are substantially more generous than the policy in the U.S. At the time of each of the respective studies, mothers in Brazil were entitled to 120 days of paid leave with full salary (Machado and Neto, 2016), while in Belarus they are entitled to 126 days of fully paid leave, and up to 3 years partially paid (Kasmach, 2016). Differences in the relative cost of formula feeding across countries may also be substantial, as, for example, during the Belarus study, the cost of formula was nearly 20 percent of an average monthly salary (Kramer et al., 2001). There may also be differences in the baseline standard of maternity care and general knowledge about

(27)

quasi-experimental approach to study their effects. Hawkins et al. (2015) estimates difference- in-differences models using data from 32 hospitals across five states, and finds that BFHI accred- itation of a hospital is associated with increases in breastfeeding initiation and duration among lower-educated mothers who give birth in that hospital. As with the RCTs, however, these estimates will be biased if there is differential sorting of mothers across hospitals in response to policy adoption.¹⁵ Andersen (2019) studies the effects of California hospitals being designated as Baby Friendly^®using a difference-in-differences framework in which exposure to the policy is based on the designation of the hospital closest to a woman’s residence (as opposed to her delivery hospital). Although the author does not consider effects on breastfeeding, he does show that the Baby Friendly^® designation results in a significant reduction in the probability of a woman having cesarean deliveries and a significant reduction in infant mortality.

By studying the effects of state-level hospital breastfeeding support laws on breastfeeding, as well as on maternal smoking and household time allocation, this paper makes several important contributions to the existing literature. First, the adoption of state-level laws represents a more plausibly exogenous shock to the immediate postpartum care that an individual mother receives compared to hospital-level adoption of these policies. Second, while the state laws we study are similar to the BFHI’s Ten Steps program, they do differ in several key ways. Specifically, the majority of the hospital breastfeeding laws require only a relatively small subset of the ten steps to be implemented, and many of the laws include the requirement that hospitals have a full time lactation consultant on staff, which is a provision not addressed by the BFHI. Providing evidence on the effectiveness of these widely adopted laws is independently important. Finally, we are the first to consider the spillover effects of breastfeeding policy on maternal smoking and household time allocation. These outcomes have not been previously considered in this context and represent potentially important unintended costs and benefits of breastfeeding policy.

15Notably, in the US the information about which hospitals are designated as Baby Friendly is easily accessible

(28)

1.3 Data Description

We use several data sets to identify the effects of hospital breastfeeding support laws on breastfeeding, maternal smoking, and household time allocation.

Data on breastfeeding are from the National Immunization Survey–Child (NIS-Child), 2003- 2017. The NIS-Child is an annual state-representative survey conducted by the CDC that targets children aged 19-35 months. Breastfeeding outcomes are self-reported, and include information on both initiation and duration of breastfeeding. Since the hospital breastfeeding laws apply to care received during the immediate postpartum period, we assign policy exposure based on year of birth and, as we only observe state of residence at time of survey, we restrict our sample to the set of children still residing in their state of birth. An additional limitation of the NIS- Child is that child age at time of survey is provided only in bins (19-23 months, 24-29 months, and 30-35 months) and month of survey is not included in the public-use files. Given this, we approximate child’s year of birth as (year of survey-2) for infants that were 19-23 months at the time of survey, and as (year of survey-3) for infants that were 24-29 months or 30-35 months at the time of survey.¹⁶

Our data on household time allocation is from the American Time Use Survey (ATUS), 2003- 2018. The ATUS is a nationally representative sample of adults drawn from the population of households that complete the Current Population Survey (CPS). Respondents are asked to record a detailed time diary of all activities over a given 24 hour period, including location of the activity and who else was present. Surveys are distributed approximately 2 to 5 months after completion of the CPS,¹⁷ and are equally split across weekends and weekdays. For our main estimates we restrict our sample to households that report having an infant under one year of age, although we

16Based on the calendar months that the NIS-Child is fielded, we calculate that infants that were between 19-23 months of age when surveyed in year t, should have been born between February of year t-2 and May of year t-1;

infants that were between 24-29 months of age when surveyed in year t, should have been born between July of year t-3 and December of year t-2; infants that were between 30-35 months of age when surveyed in year t, should have been born between February of year t-3 and June of year t-2. Measurement error in the birth year should bias our estimates towards zero.

17The average gap is 3 months, with 70% of households responding to the ATUS 3 months after their last CPS

(29)

also examine outcomes for households without infants as a falsification test. In some specifications we further allow the effect of the policy to vary across relatively younger versus relatively older infants. Although information about infant age in the ATUS and CPS is limited to 1 year age bins, we combine information about the number of people and the presence of an infant in the household at the time of each survey to impute a range of possible ages (in months) for the infant.¹⁸ Infants that were present in both the CPS and the ATUS will on average be between 3 and 11 months old at the time of the ATUS (relatively older), infants that were born between the CPS and the ATUS will on average be between 0 and 3 months old at the time of the ATUS (relatively younger).

For our time use analyses we group reported activities into four primary categories: childcare, formal work, unpaid domestic work, and leisure; for some analyses we further decompose childcare into two sub-categories: time spent on basic physical care and time spent on educa- tional/recreational care. We note here that breastfeeding falls under the basic physical child care category, and unfortunately is unable to be disaggregated from other infant care activities, including giving child a bottle, and feeding a child. We present in Appendix Table 3.10 more detail on the types of activities that are included in each of the time use categories. To examine maternal smoking, we focus on reported minutes of tobacco or drug use (a component of leisure time).¹⁹

We additionally use data from the CDC’s Pregnancy Risk Assessment Monitoring System (PRAMS), 2000-2018, to provide evidence on the effects and mechanisms of the hospital breastfeeding support laws. The PRAMS surveys women who had a live birth in the past 2 to 4 months, drawn from a sample of state birth certificate records, and contains self-reported information on breastfeeding, maternal smoking, and information regarding the types of breastfeeding-related care the mother received during her immediate postpartum hospital stay. This data set has several notable limitations, however. First, the set of states with available PRAMS data varies substan-

18Specifically, consider an infant that is present and less than 1 year of age in both the ATUS and the CPS. If those two surveys were fieldedmmonths apart, the youngest the child could be at the ATUS ismmonths (if they had just been born at the time of the CPS), the oldest is 11 months. If an infant was not present at the time of the CPS and is present at the time of the ATUS, they must be between 0 andmmonths of age. For the vast majority of the sample (70 percent) the number of months between the two surveys is three.

(30)

tially across years, with between 19 and 36 states reporting in a given year.²⁰ Second, due to a survey revision that took place in 2009, the questions pertaining to maternal smoking are not comparable across the entire sample period. Given this survey change, and the fact that only one state that adopted a hospital breastfeeding law during 2009-2018 is consistently in the PRAMS data (as opposed to 3 states that adopted during the 2000-2008 period), we examine maternal smoking using this data set only for the period 2000-2008. Finally, the survey items that pertain to breastfeeding-related care received at the hospital are part of an optional module for states, and thus the set of states and years during which these questions are asked is further restricted. These survey questions also have the substantial limitation of only being asked to mothers who initiated breastfeeding. Appendix Figure 1.6 provides information on the set of state-years the PRAMS data are available for, as well as how that coincides with state law implementation.

To provide further evidence on the mechanisms through which the hospital breastfeeding laws impact breastfeeding outcomes, we use data on the number of International Board Certified Lactation Consultants (IBCLCs) in a given state, collected from the CDC’s annual Breastfeeding Report Card²¹ and from archived versions of the International Board of Lactation Consultant Examiners website,²² for 2006-2016, and 2018. While IBCLC represents the only professional certification for lactation consultants, we note that there are other certifications available that would satisfy the law requirements (such as Certified Lactation Specialists and Certified Lactation Counselors). We focus on IBCLCs both due to data availability and because this is the key measure that the CDC includes in their annual Breastfeeding Report Card.

Information on the state hospital breastfeeding support laws were obtained from the LawAt-

20This variation is due both to states choosing not to participate in the survey in a given year, and because data are not released for a given state-year if response rates did not meet a pre-specified threshold. The number of states choosing to participate has increased over time, from 20 states in 2000 to 48 states in 2018. The response rate threshold that must be met in order for the data to be publicly released has also changed over time, decreasing from 70 percent for 2000-2006, to 65 percent for 2007-2011, to 60 percent for 2012-2014, to 55 percent from 2015 to present.

21Current and historic CDC Breastfeeding Report Cards are available here: https://www.cdc.gov/breastfeeding/

data/reportcard.htm.

22Current year state-level counts of IBCLCs are available here: https://iblce.org/about-iblce/

current-statistics-on-worldwide-ibclcs/. Historic counts were retrieved from archived versions of the website,

(31)

las Policy Surveillance Program database;²³ adoption dates were identified through independent review of state statutes and state administrative codes. We graphically present the timing of law adoption across states in Figure 1.1; in Appendix Figure 1.3 we show how that timing coincides with the available sample periods of our primary data sets. While there is generally substantial variation across space and time in the adoption of these laws, there is some clustering of adoption in the Northeast and South, and notably only one state in the western census region (California) ever adopts a hospital breastfeeding support law. In order to address the potential concern that unobserved region-level shocks are driving both the adoption of the laws and the observed changes in outcomes, we estimate robustness checks that include region-by-year fixed effects.

As previously discussed, across states there is substantial heterogeneity in the specific components of the laws. In order to capture this heterogeneity, we characterize the relative strength of each law as the fraction of eleven possible components (each of the ten items corresponding to the WHO/UNICEF “Ten Steps for Successful Breastfeeding,” plus the requirement for a lactation consultant) that a given law mandates.²⁴ We follow the law component categorization provided by the LawAtlas database. Appendix Figure 1.4 presents this measure of law strength for the set of states that adopt during our sample period; the specific components of each state’s law are provided in Appendix Figure 1.5. Only one state (New York) adopts a law that mandates all eleven possible categories; the median adopting state mandates two out of the eleven categories.

Ideally, we would also characterize the laws based on the specific set of components that they contain, in order to identify which policy component is most important for affecting outcomes.

Unfortunately, however, because states adopt these law components in bundles we are limited in our ability to separately identify the effects of individual components. As the requirement to provide a lactation consultant is relatively well-identified (adopted by 9 separate states, 3 of which

23http://lawatlas.org/datasets/baby-friendly-hospital-1525279705

24The eleven categories are as follows: required to communicate policy to staff; required to train healthcare staff;

required to inform patients about breastfeeding; required to make lactation consultant available; required to help initiate breastfeeding; required to provide mothers instruction on how to breastfeed and how to maintain lactation, even if they are separated from infant; requirements regarding provision of non-milk; required to allow breastfeeding on demand; no pacifiers/artificial nipples; required to permit rooming-in; required to provide information on/refer

(32)

mandate only a lactation consultant and no other components), we do provide some suggestive evidence about the importance of this law component.

1.4 Empirical Strategy

To identify the effects of hospital breastfeeding support laws, we estimate dynamic difference- in-differences models that rely on plausibly exogenous variation in the timing of law adoption across states. Specifically, we estimate:

Y_ist =b₀+

Â

k2K

b₁^kHospitalLaw^k_st+b₂X_ist+Z_st+t_t+d_s+e_ist (1.1)

whereYist is the outcome of interest for motheriresiding in stateswho had an infant born in year t. X_ist is a vector of individual characteristics, as available in a given dataset, including: child’s gender, child’s race/ethnicity (Hispanic, white, Black, with other as the excluded category), an indicator variable for receiving WIC benefits, number of other children under 18 years old living in the home (only 1 child, 2 to 3 children, with 4 or more children as the excluded category), an indicator variable for being the mother’s first born child, maternal education (less than high school, high school, some college, with college or above as the excluded category), an indicator variable for if the mother is over the age of 29,²⁵ and an indicator variable for whether the mother is married.²⁶ In regressions using NIS-Child data, we additionally control for fixed effects for child’s age at time of survey (19- 23 months, 24-29 months, with 30-35 months as the excluded category).

HospitalLaw^k_st is a vector of indicator variables equal to one if state s in year t has had a hospital breastfeeding support law in effect forkyears,K={ 6, 5, ..., 2,0, ...,5, 6}, and is zero otherwise (year -1 is the omitted category), thusb₁^k represents our vector of coefficients of interest and captures the effects of the laws. For robustness we also estimate specifications in

25This is the most detailed maternal age information that is consistently available across NIS-Child survey waves.

26In the NIS-Child, all household level variables are measured at the time of survey (when the child is 19-35

(33)

which the binary variable that captures whether the state has any law is replaced with a continuous variable (between zero and one) that captures the relative strength of the law. By estimating event study specifications we are able to test for dynamic policy effects, as well as examine the extent to which outcomes were trending similarly in treatment and control states during the periods prior to law adoption. As the key identifying assumption in this difference-in-differences model is that outcomes would have evolved similarly in states that did and did not adopt a hospital breastfeeding law, evidence of parallel trends during the pre-treatment period provides significant support for this assumption.

Z_st is a vector of other state policies, as well as state demographic and economic characteristics, which may potentially affect maternal behaviors and breastfeeding. Specifically, we control for the following state policies: laws that encourage or require employers to provide break time and/or private space for breastfeeding or expressing milk; laws prohibiting employer discrimination against breastfeeding employees; laws that allow breastfeeding in any public or private location; laws that exempt breastfeeding mothers from public indecency laws; laws that exempt breastfeeding mothers from jury duty; laws that require states to provide paid maternity leave; and an indicator variable for whether or not a state has expanded Medicaid.²⁷ Information on workplace breastfeeding laws was obtained from Nguyen and Hawkins (2013), the National Council of State Legislators (2018),²⁸ and the United States Department of Labor Women’s Bu- reau (2019).²⁹ Information on the implementation of Medicaid expansion is from the Kaiser Family Foundation (2015). Annual state-level demographic measures (fraction female; fraction Black, Hispanic, and other non-white races; fraction of individuals with high school degrees and

27While the ACA was a national-level policy, the effects may not be absorbed by year fixed effects since the requirement that all new insurance plans cover breastfeeding equipment and supplies, as well as lactation support and counseling without cost-sharing differentially affected households with private insurance. We do not control for this in our baseline specification because in the NIS-Child insurance status is not observed at time of birth (only at time of survey), and it is only observed for approximately 50 percent of our sample. As a robustness check we verify that all main results are not sensitive to controlling for whether the child is currently on Medicaid and including an interaction between post-ACA and Medicaid status. We also estimate equations where we include the interaction between post-ACA and either maternal education fixed effects or WIC receipt, as proxies for Medicaid status at time of birth. Our results are similarly robust to the inclusion of these controls.

28https://www.ncsl.org/research/health/breastfeeding-state-laws.aspx. Last accessed: July 16, 2020.

(34)

college or more; fraction of individuals under 21 and between 21-64; and fraction of individuals below the federal poverty line) are constructed from IPUMS-Current Population Survey (King, 2015); we obtain annual state unemployment rates from the Bureau of Labor Statistics. In specifications in which maternal smoking is the outcome of interest, we additionally control for the log of the state cigarette tax, and for whether there are any state-level indoor smoking bans in effect.³⁰ In order to best capture the state characteristics that would have feasibly been relevant to the postpartum behaviors considered here, all variables contained in theZst vector are measured in the year in which the child was born.

All models additionally control for a full set of state of residence and child birth year fixed effects. We use sample weights as provided by each data set, and cluster standard errors at the state level (Bertrand et al., 2004).³¹ We also provide standard difference-in-differences estimates for all models, in which the estimated effect of hospital breastfeeding laws is summarized as the single coefficient on the interaction termPost_t⇥HospitalLaw_s. For these specifications,Post is an indicator variable equal to one if the state had adopted a hospital breastfeeding law by the year prior to the infant’s birth yeart; HospitalLaw_s is an indicator variable capturing if states ever adopts a law.

1.5 Main Results

1.5.1 Descriptive Statistics

Descriptive statistics for the NIS-Child sample are presented in Appendix Table 1.6. We provide variable means for the full sample (column 1), and separately for individuals who lived

30State cigarette tax information is from the CDC National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health ”Tax Burden on Tobacco, 1970-2018” database, available at https:

//chronicdata.cdc.gov/Policy/The-Tax-Burden-on-Tobacco-1970-2018/7nwe-3aj9. Data on state indoor smoking bans is from the CDC National Center for Chronic Disease Prevention and Health Promotion, Office on Smok- ing and Health ”Tobacco Legislation - Smokefree Indoor Air” database, available at https://www.cdc.gov/oshdata/

Legislation.html

31In 2011 NIS-Child switched from single frame landline-only sampling to dual frame sampling that included landlines and cell phones, and in that year only both single and dual frame weights are provided. In all reported estimates we use dual frame weights starting in 2011. None of the main results are sensitive either to this decision