University of Southern California Dissertations and Theses

The EITC, labor supply, and child development

(USC Thesis Other)

The EITC, labor supply, and child development

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THE EITC, LABOR SUPPLY, AND CHILD
DEVELOPMENT
by
Jeehyun Ko
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
May 2021
Copyright 2021 Jeehyun Ko
Acknowledgements
I would like to express my sincere gratitude to my advisor, Professor John
Strauss, forsharinghisinvaluableexpertiseandguidance. Hisinsightandfeedback
have sharpened my thought process and elevated my research to a much higher
level. He has also supported all the aspects of my research and guided me to be
professional. I also want to thank him for his invitations to his home and for that
fruit from his tree, which I will miss.
I would like to thank Professor Gary Painter for his helpful contributions to
my research and sharing his vast knowledge of the EITC literature. I would also
like to thank Professor Geert Ridder for his invaluable expertise in econometrics
and for suggesting the estimation based on the Fuzzy DID model in my first
chapter. I am grateful to Professor Hyungsik Moon, for his practical suggestions
for my identification strategy and his thoughtful advice. I would like to extend
my gratitude to Professor Jeffrey Weaver for offering his support and providing
constructive advice. I very much appreciate Professor Chul-In Lee in Korea, who
introduced me to economic research and has always been a mentor to me.
Many thanks also go to our Program Manager, Young Miller, and Ph.D. Pro-
gramAdvisor, AlexanderKarnazes, forprovidingexcellentadministrativesupport.
I would like to thank Research Data Administrator Jillian Wallis, and System Ad-
ii
ministrator Anshu Verma, at the USC Schaeffer Center. They kindly shared their
knowledge and experience in using the NLSY geocode data. I extend my thanks
to my English teachers at the American Language Institute, Mary Ann Murphy
and Reka Clausen, for their help in developing my writing and oral skills. Their
expertise and patience cannot be underestimated and have helped my dissertation
and presentations evolve to a higher level.
I would also like to thank my friends and cohort members at USC. With them,
I am stronger and even the hardest times of the journey were bearable and even
enjoyable. Special thanks to Hae Yeun Park, who has freely shared her wisdom
and who is always on my side. I am deeply indebted to the LA Sarang Church
Community, Gong Hang Bethel Church Community, and my friends in Korea. I
appreciate all their support and their prayers on my behalf.
All my love and thanks to my parents, Young Moo Ko and Sung Hee Lim, and
my younger brother, Samuel Ko. Without their prayers, unwavering support, and
great love, I would not have finished this journey.
Last but not least, I would like to express my deepest gratitude to my God,
who is and will always be with me, for His amazing love and grace throughout this
journey and my life.
iii
Contents
Acknowledgements ii
List of Tables vii
List of Figures viii
Abstract ix
Introduction 1
1 An Unintended Consequence of the Earned Income Tax Credit:
Maternal Labor Supply and Child Development 4
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 The Brief History of EITC . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 EITC and Child Development . . . . . . . . . . . . . . . . . . . . . 10
1.4 Identification Strategy . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5 Data and Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7 Common Trend Assumption and Robustness Checks . . . . . . . . . 22
1.7.1 Common Trend Assumption . . . . . . . . . . . . . . . . . . 22
1.7.2 Placebo Test . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.7.3 Other Welfare Reforms . . . . . . . . . . . . . . . . . . . . . 24
1.7.4 Child Care from Grandparents . . . . . . . . . . . . . . . . . 24
1.7.5 EITC and Marital Status . . . . . . . . . . . . . . . . . . . . 26
1.7.6 Unmarried Mother Sample . . . . . . . . . . . . . . . . . . . 28
1.8 Mechanisms of Impact . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.8.1 Maternal Labor Supply . . . . . . . . . . . . . . . . . . . . . 29
1.8.2 Other Family Incomes . . . . . . . . . . . . . . . . . . . . . 31
1.8.3 Investments in Children . . . . . . . . . . . . . . . . . . . . 32
1.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.10 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
iv
2 Revisiting the EITC and Child Development 53
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.2 EITC and Instrumental Variable . . . . . . . . . . . . . . . . . . . . 56
2.3 Data and Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.4 Panel Fixed Effect Estimation . . . . . . . . . . . . . . . . . . . . . 61
2.5 Instrumental-Variables (IV) Analysis . . . . . . . . . . . . . . . . . 62
2.6 Alternative Instrumental Variable . . . . . . . . . . . . . . . . . . . 65
2.7 Conclusion and Discussion . . . . . . . . . . . . . . . . . . . . . . . 67
2.8 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3 The EITC and the Labor Supply: A Case Study of Korea 88
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.2 Brief History of EITC in Korea . . . . . . . . . . . . . . . . . . . . 92
3.3 Theoretical Prediction . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.4 Data and Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . 95
3.5 Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.6 Conclusion and Future Works . . . . . . . . . . . . . . . . . . . . . 99
3.7 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Conclusion 114
Bibliography 117
A Appendix to Chapter 1 124
B Additional Tables to Chapter 1 129
v
List of Tables
1.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.2 Regression Results of Children’s Cognitive Achievement . . . . . . . 37
1.3 Regression Results of Children’s Noncognitive Achievement . . . . . 38
1.4 Tests for the Parallel Trends Assumption . . . . . . . . . . . . . . . 39
1.5 Placebo Tests with Children of Mothers with High-Income Levels . 40
1.6 Robustness of the Estimates . . . . . . . . . . . . . . . . . . . . . . 41
1.7 Impacts of Child Care (Presence of Grandparents in the Households) 42
1.8 Impact of the EITC on Change in Marital Status . . . . . . . . . . 43
1.9 Impact of the EITC on Probability to be Single Mothers . . . . . . 44
1.10 Changes in the Mothers’ Labor Supply . . . . . . . . . . . . . . . . 45
1.11 Changes in the Other Income Sources . . . . . . . . . . . . . . . . . 46
1.12 Changes in Home Environment . . . . . . . . . . . . . . . . . . . . 47
1.13 Changes in Cognitive Stimulation Related to Time Inputs (detailed) 48
1.14 Changes in Cognitive Stimulation Related to Goods Inputs (detailed) 48
2.1 Historical Changes in the Federal EITC Tax Credit Rates . . . . . . 69
2.2 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.3 Fixed Effect Model: Impacts of Mothers’ Labor Force Participation
on Child Development . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.4 Fixed Effect Model: Impacts of Mothers’ Annual Working Hours on
Child Development . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.5 Fixed Effect Model: Impacts of Mothers’ Number of Jobs on Child
Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.6 Instrumental-Variables Approach: First Stage Estimates with La-
bor Force Participation . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.7 Instrumental-Variables Approach: First Stage Estimates with An-
nual Working Hours . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.8 Instrumental-VariablesApproach: FirstStageEstimateswithNum-
ber of Jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.9 Instrumental-Variables Approach: Effect of Labor Force Participa-
tion on Child Development . . . . . . . . . . . . . . . . . . . . . . 77
vi
2.10 Instrumental-Variables Approach: Effect of Annual Working Hours
on Child Development . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.11 Instrumental-Variables Approach: Effect of Number of Jobs on
Child Development . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.12 Instrumental-Variables Approach: Effect of Labor Force Participa-
tion on Child Development (Additional Controls) . . . . . . . . . . 80
2.13 IV Estimates with PIAT Reading Scores by Mother’s Marital Status 81
2.14 ImpactsofMother’sLaborForceParticipationonthePIATReading
Tests by Income Levels . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.15 ImpactsofMother’sLaborForceParticipationonthePIATReading
Tests by Mother’s Education Level . . . . . . . . . . . . . . . . . . 83
2.16 ImpactsofMother’sLaborForceParticipationonthePIATReading
Tests by Number of Children . . . . . . . . . . . . . . . . . . . . . 84
2.17 Alternative Instrumental-Variable: Effect of Labor Force Participa-
tion on Child Development . . . . . . . . . . . . . . . . . . . . . . 85
3.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.2 Differences-in-DifferencesEstimatesof2011EITCExpansiononLa-
bor Force Participation . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.3 Differences-in-Differences Estimates of 2011 EITC Expansion on
Positive Earnings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.4 Differences-in-Differences Estimates of 2011 EITC Expansion on
Weekly Hours of Work . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.5 Tests for the Parallel Trends Assumption: Labor Force Participation 105
3.6 Tests for the Parallel Trends Assumption: Positive Earnings . . . . 106
3.7 Tests for the Parallel Trends Assumption: Weekly Hours of Work . 107
A1 Fuzzy DID Estimates for Combined PIAT Math and Reading Scores 128
B1 DID Regression Results of Children’s Cognitive Achievements (Ad-
ditional Controls) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
B2 Regression Results of Children’s Noncognitive Achievement (Addi-
tional Controls) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
B3 DID Regression Results of Children’s Cognitive Achievements (By
Age) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
B4 DID Regression Results of Children’s Cognitive Achievements (By
Ethnicity) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
B5 DID Regression Results of Children’s Non-cognitive Achievement
(By Ethnicity) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
vii
List of Figures
1.1 The EITC Benefit Schedule in 1994 for families with two children . 49
1.2 DistributionofFamilyIncomesandImputedEITCBenefits–Households
with More than Two Children . . . . . . . . . . . . . . . . . . . . . 50
1.3 Expansions of Earned Income Tax Credit for Families with More
than Two Children in the 1990s . . . . . . . . . . . . . . . . . . . . 51
1.4 Estimated Effect of the EITC for Years Before and After the 1993
Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.1 Federal and State EITC Exposure by Year and State for Families
with One Child . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.2 Federal and State EITC Exposure by Year and State for Families
with More than Two Children . . . . . . . . . . . . . . . . . . . . . 87
3.1 The EITC Schedule for Families with children in 2008 in Korea . . 108
3.2 The EITC Schedules for Families with children in 2011 . . . . . . . 109
3.3 TheoreticalPredictionoftheImpactsoftheEITConLaborSupply:
Change in the Budget Constraints . . . . . . . . . . . . . . . . . . 110
3.4 TheoreticalPredictionoftheImpactsoftheEITConLaborSupply:
Decisions of Non-workers . . . . . . . . . . . . . . . . . . . . . . . 111
3.5 TheoreticalPredictionoftheImpactsoftheEITConLaborSupply:
Decisions of Workers in the Phase-in Range . . . . . . . . . . . . . 112
3.6 TheoreticalPredictionoftheImpactsoftheEITConLaborSupply:
Decisions of Workers in the Plateau and Phase-out Ranges . . . . . 113
viii
Abstract
The dissertation consists of three essays on the Earned Income Tax Credit
(EITC), labor supply, and child development. The first chapter examines the
impact of the Earned Income Tax Credit (EITC) on the children of single mothers.
While the EITC is typically thought to benefit low-income children by increasing
family income, it may also decrease caregiving inputs as a result of increased
parental labor supply. Children of single mothers may be particularly sensitive to
such decreases due to the lack of other parental support. Using a difference-in-
differences (DID) approach to look at the impact of the 1993 EITC expansion, I
find that the EITC expansion reduces the combined math and reading test scores
of children of single mothers by 13.61 percent of a standard deviation. The most
important mechanism is reduced mother-child interactions due to the increased
maternal labor supply. These results suggest that for the EITC to be an effective
poverty reduction tool, it may need to be paired with other interventions such as
child care.
In the second chapter, I use an alternative approach, an instrumental-variables
(IV) approach, and address the same research question. Using changes in the
maximum credit rates of the EITC as an instrument for mothers’ labor supply, I
explore how the mothers’ labor supply affects children’s cognitive and noncognitive
ix
achievements. I find that labor force participation of single mothers results in a
decrease in the PIAT reading test scores by 1.3 standard deviation. The impacts
are larger for high impact groups, such as mothers with low household incomes.
In the third chapter, I explore the impact of the EITC on the labor supply
of married fathers in Korea. I utilize different benefit schemes of the EITC on
family size, which was due to the EITC expansion in 2011. Using a difference-in-
differences(DID)approachandtheKoreanLabor&IncomePanelStudy(KLIPS),
Ifindthattheexpansiondoesnotresultinasignificantincreaseinthelaborsupply
both at the extensive and intensive margins. A possible explanation is that the
married fathers in Korea have inelastic labor supply. In addition, the labor market
mightnotbeflexibleatleastintheshortrun, whichcannotabsorbthemiddle-aged
male workers, who are willing to work.
x
Introduction
The Earned Income Tax Credit (EITC) is a refundable tax credit to low- to
middle-income households and is designed to reward work. It became a part of the
U.S. tax code in 1975 and started as a small cash bonus program. However, since
the welfare reforms in the 1990’s, the tax credit has grown dramatically and is
now the largest cash transfer program in the U.S., distributing 62 billion dollars to
25 million recipients in 2020 (IRS, 2020). Many studies on the EITC evaluate its
impact on the labor supply and find that it increases the labor force participation
of single mothers by 10% (Meyer and Rosenbaum, 2001). Observing the success,
many other countries, such as United Kingdom, Canada, France, and Korea, have
started to implement their own programs.
Another issue often explored is how the EITC affects recipients’ children. As
the EITC was initiated as an anti-poverty program, it is important to evaluate its
inter-generational impacts. Previous studies report that the EITC affects children
positively by increasing household incomes.
However, there is a gap in the literature as previous studies mostly focus on
the impact of the EITC through changes in the income rather than parents’ labor
supply. Consideringthattimewithparentsisanimportantinputforchild’shuman
capital accumulation and may not be easily replaced by other inputs, ignoring this
1
impact renders our understanding of the EITC less complete.
Therefore, in the first chapter, I study the impact of the EITC on the chil-
dren of single mothers as they are more likely to suffer from the lack of their
mothers’ care giving inputs. The impacts on these children might differ those of
married couples. As single mothers are the primary earners and the childrens’
only caretakers, motivating mothers to work might result in a substantial decline
in caregiving inputs. In a two-parent household, however, the impact is mitigated
as, if one parent works more, there is still the other to take care of the child. Using
a difference-in-differences (DID) approach, I analyze the impact of the EITC ex-
pansion in 1993 and find that children of single mothers show reductions in their
combined math and reading, and Motor and Social Development (MSD) scores.
To explain the negative results, I explore possible mechanisms through which the
EITC affects these children. From the analysis, I find that reduced mother-child
interactions due to increased maternal labor supply mostly explains the negative
results.
In the second chapter, I address the same research question with an alternative
specification, the instrumental-variables (IV) approach. This approach, which
compensates for the weakness of the DID approach in the first chapter, shows
direct links between maternal work and child development, as well as incorporates
the multiple EITC expansions, including those at the state level. Using changes
in the policy parameters from the federal and state EITCs as an instrument of
the maternal labor supply, I find that labor force participation of single mothers
reduces the PIAT readings scores of their children.
In the third chapter, I further explore the impact of the EITC on the labor
supply, but change the context from the U.S. to Korea. Studying the case of
2
Korea is interesting as it will allow us to look at how the EITC works under the
different labor market conditions. Since the implementation in 2009, there have
been twelve payments and four expansions, the most recent one of which was in
2018. In the chapter, I evaluate the 2011 expansion and look at its impact on the
married fathers, relying on the different incentives dependent on family size. While
it has been widely studied in the US context, this paper looks at in Korea, case of
which has not been done so. Using a difference-in-differences (DID) approach, I
find that the expansion was not effective in changing the labor supply of married
fathers. If we compare the result to that of the U.S., Eissa and Hoynes (2006) find
married men slightly increase their labor supply by 0.2%. The difference can be
possibly explained by difference in the elasticity of labor supply and labor market
conditions.
3
Chapter 1
An Unintended Consequence of the
Earned Income Tax Credit:
Maternal Labor Supply and Child
Development
1.1 Introduction
Policymakers who design social safety nets must balance multiple competing
objectives. First, they need to provide adequate financial support to low-income
families. Second, at the same time, they should avoid disincentivizing work or
accumulation of human capital. The welfare reforms of the 1990s tried to find the
right balance between these two objectives and ended up putting more weight on
the second objective. For example, the Aid to Families with Dependent Children
(AFDC) was replaced by the more restrictive Temporary Aid to Needy Families
4
(TANF), which implemented work requirements and limited the duration of ben-
efits. The reform also included large expansions of the Earned Income Tax Credit
(EITC), which has encouraged people to work more by giving greater benefits
with greater earned income. As a result, the annual employment of single mothers
increased by about nine percentage points between 1986 and 1996 (Meyer and
Rosenbaum, 2001).
However, the impact of these pro-work welfare programs on children is am-
biguous. They may increase earnings and reduce the culture of dependency, which
affects the children positively. However, as they require parents to work for the
benefit or to work more for higher benefits, parents are not able to spend as much
time with their children. Requiring parents to work might cause a reduction in
both the quantity and quality of time with their children if the parents experience
stress or physical fatigue from working. This can have adverse effects on children.
This paper investigates how the EITC affects development outcomes of single
mothers’ children and looks into the mechanisms that can explain them. There
are two reasons that I focus on single mothers. First is that they are the primary
target of the EITC and the major beneficiaries. Single mothers made up 50%
of the overall recipients in 2008 and received 48% of the 2007 transfers (Meyer,
2010; Athreya et al., 2010). The second reason is that the effect on children of
single mothers will differ from that on children in two parent households. As single
mothers are both the primary earners and only caretakers for the children, taking
them away from their children and pushing them to work may cause a substantial
decline in both the amount and quality of childcare. On the other hand, in a two
parent household, if one of the parents works, there is still the other to take care
of the child, which makes it easier to find good quality, stable childcare. It is very
5
difficult and costly for single mothers to find in the market.
Many previous studies on the EITC analyze how it affects the labor supply.
They find that it successfully increases the labor supply of the recipients in the
large work incentive range (the phase-in range).
1
The most representative example
is the group of single mother recipients. As they are sole and primary earners in
the household and less educated, their income is low enough to put them in the
high work incentive range (the phase-in range). They increase their labor force
participation by 2.8% (Eissa and Liebman, 1996; Meyer and Rosenbaum, 2001;
Blundell et al., 2016; Kleven, 2019; Adireksombat, 2010). However, a compara-
ble group, married mothers are not responsive. It is because most of them are
secondary earners in the household, which makes their labor supply inelastic, and
they are in the plateau and phase-out ranges, where there are no labor supply
incentives (Eissa and Hoynes, 2004; Heim, 2010; Dickert et al., 1995).
2
Another main branch of the EITC literature studies how the increase in fam-
ily income from the EITC affects child development. So far, many studies have
focused on the income channel and have found positive impacts on birth weight,
academicscores, collegeenrollment, andfutureoutcomessuchasearningsinadult-
hood (Chetty et al., 2011; Bastian and Michelmore, 2018; Baker, 2008; Dahl and
Lochner, 2012; Manoli and Turner, 2018; Hoynes et al., 2015). However, these
papers don’t consider another important channel of the EITC: decrease in time
1
As shown in Figure 1.1, the EITC has three ranges: phase-in, plateau, and phase-out regions.
In the phase-in region, the benefit increases as the income increases. However, on the plateau or
phase-out region, the benefit amount stays the same or decreases. Thus, the phase-in range is
the only part where the EITC has the work incentive structure.
2
Some papers report significant impacts on subsamples of married mothers. Married mothers
with low-earning spouses increased their labor supply after the adoption of the EITC at the
extensivemargin(Bastian,2017), andself-employedmarriedmothersincreasedtheirlaborsupply
(Lim and Michelmore, 2018). Yang (2018) finds income seasonality caused by the EITC receipt
affects the labor supply of secondary earners.
6
with the mothers. This channel is particularly important for some recipients, such
as single mothers’ children, as the time effects are expected to be large.
The maternal labor supply channel is important but is rarely explored. To the
best of my knowledge, there is no paper that links the EITC and child outcomes
using the maternal labor supply mechanism except Agostinelli and Sorrenti (2018).
Their study explores the income and time effects of maternal work from the EITC,
building on Dahl and Lochner (2012) and using the IV strategy. They find that an
increase in mother’s labor supply influences children negatively, and this negative
impact dominates the positive income effect from the EITC for the low-income
households, which implies negative net impacts of the EITC.
My paper builds on the findings of Agostinelli and Sorrenti (2018). For ex-
ample, we both consider the mothers’ labor supply channel and explore potential
mechanisms through which maternal work negatively affects their children. When
exploring the mechanisms, however, Agostinelli and Sorrenti (2018) rely on cross-
sectional comparisons between children of mothers who are employed and not
employed across income groups, which is subject to selection bias. In addition,
they switch data sets when they analyze the mechanism, which might cause a
problem as the children in the new data set are not the same as in the other and
might have had different experience related to maternal work. In this paper, I
analyze more diverse mechanisms related to the EITC and mothers’ labor supply
in a more rigorous way. Using the same sample of children in a data set, and a
DID approach, I explore changes in income, home environment, goods investment
in children, and various forms of mothers’ labor supply. This will be helpful in
explaining why and how mothers’ work affects children, as well as in designing
policies to help the recipients more effectively.
7
This paper also contributes to an array of papers on maternal work and child
development. Even though empirical findings are mixed (James-Burdumy, 2005;
Schildberg-Hoerisch, 2011; Fort et al., 2016; Bernal, 2008; Bernal and Keane, 2011;
Felfe and Hsin, 2012; Brooks-Gunn et al., 2002; Paxson and Waldfogel, 2003),
studies on single mothers’ children suggest negative relationships. For example,
Bernal and Keane (2011) finds that welfare recipient single mothers’ work and
related child care leads to a 2.1% decrease in children’s test scores. Also, Felfe and
Hsin (2012) report that work-related stress or physical hazards, which low-income
mothers often face, decrease children’s cognitive and behavioral development. My
paper contributes to the literature by studying EITC-recipient single mothers, who
have been heavily targeted and impacted by welfare programs.
Using the difference-in-differences (DID) method, this paper finds that EITC
recipientsinglemothers’children, experiencedreductionsinthecombinedPeabody
IndividualAchievementTest(PIAT)mathandreadingscoresasmuchas13.61%of
a standard deviation compared to their counterparts, children of married mothers
who are likely to receive the EITC but less likely to have work incentives. The
reductions are attributed to reduced mother-child interactions due to increased
labor supply of mothers.
The remainder of the paper is divided into nine sections. In Section 2, I review
brief history and structure of the EITC. In Section 3, I review potential channels
through which the EITC may affect children. In Section 4, I presents the identi-
fication strategy. Section 5 explains data, the matched mother-child data of the
National Longitudinal Survey of Youth (NLSY), and provides summary statistics.
In Section 6, I present regression results with the regression model (Two-way fixed
effect model). Section 7 includes robustness checks including common trend as-
8
sumptions. In Section 8, I explore possible mechanisms through which the EITC
expansion could affect child development such as maternal labor supply, family
income, and child investment. In Section 9, I conclude with a summary. In the
appendix, I include the Fuzzy DID approach, which compensates for the weakness
of the DID approach.
1.2 The Brief History of EITC
It has been almost 45 years since the Earned Income Tax Credit became a
part of the 1975 U.S. tax code. Since then, the EITC has emerged as a popular
alternative method for transferring income to low-income families with children.
This popularity of the EITC stems from the structure of the tax credit which
allows people to believe that the EITC can redistribute income with much less
distortion of labor supply compared to other means-tested cash transfer such as
the former Aid to Families with Dependent Children (AFDC) and new Temporary
Aid to Needed Family (TANF) program.
The EITC encourages work by giving higher tax credits with higher earned
income. Figure 1.1 shows how it works. There are three regions in the EITC
benefit schedule: phase-in, plateau, and phase-out ranges. In the phase-in range,
the EITC benefit increases with an increase in earned income until it reaches the
maximum benefit, where the EITC has the largest work incentives by lowering
marginal tax rates. After that, there are plateau and phase-out ranges where the
benefit stays at the maximum and then phases out as earned income increases
(marginal tax rate stays and increases). The EITC has its largest work incentives
in the phase-in range.
9
Major expansions of the EITC started in the 1990s and the largest expansion
was in 1993 (Figure 1.3). As the EITC expanded, it gave larger EITC benefits
and covered households with higher income levels. The 1993 EITC expansion
increased the maximum tax credit amount from $6,810 in 1990 to $8,425 in 1994.
The expansion also provided larger work incentives to the recipients, particularly
those who are in the phase-in range by increasing the amount of tax credit per
dollar of earned income, which determines the phase-in rates. It increased the
phase-in rates from 14% in 1990 to 30% in 1994 to families with more than two
children. This means that the family received 19.5 cents per dollar of earned
income in 1990, and it rose to 30 cents in 1994.
The popularity in the EITC also made each state have its own state EITC as a
percentage of the federal EITC. Since Rhode Island enacted the first state EITC
in 1986, the EITC has become widespread at the state level with 28 states and
the District of Columbia. Each state decides when to adopt the state EITC, and
the rate in proportion to the federal EITC. This variation will be used in the third
chapter of this dissertation.
1.3 EITC and Child Development
Thereareseveralchannels, throughwhichtheEITCmaycausechangesinchild
development. Examples are changes in family income, maternal labor supply, and
family composition. In this section, I explore primary channels and discuss their
potential impacts on children.
First, EITC expansions increase household incomes (Neumark and Wascher,
2011, 2000; Scholz, 1994; Hoynes and Patel, 2018). Family income is an impor-
10
tant predictor of a child’s success and future opportunities. The impact of family
income on child development has been widely discussed. Previous studies have
reported a positive relationship between family economic conditions during child-
hood and various child outcomes (Cunha and Heckman, 2007; Chetty et al., 2011;
Blau, 1999; Duncan et al., 1998; Levy and Duncan, 2000; Ludwig and Miller, 2007;
Deming, 2009; Markowitz et al., 2017). Increases in income from the EITC expan-
sion also bring positive impacts on child development. Works such as Dahl and
Lochner (2012) and Bastian and Michelmore (2018) employ instrumental variable
techniques to confirm this positive effect both in the short-run and long-run.
A second possible channel is mothers’ employment. The EITC is designed
to encourage work, and it gives clear incentives to non-workers to start working.
Empirical studies provide evidence that the EITC encourages work among single
mothers, but little evidence that working women already in the labor market in-
crease their working hours (Hotz et al., 2001; Eissa and Liebman, 1996; Meyer and
Rosenbaum, 2001; Eissa and Hoynes, 2006). Secondary earners, such as some mar-
ried women, face incentives to reduce work, and empirical studies give consistent
results (Eissa and Hoynes, 2004). This discussion implies that the EITC has dif-
ferential impacts on mothers’ labor supply and as a result, it may have differential
impacts on their children.
Empirical findings on maternal work and child development are mixed (James-
Burdumy, 2005; Schildberg-Hoerisch, 2011; Fort et al., 2016; Bernal, 2008; Bernal
and Keane, 2011; Felfe and Hsin, 2012; Paxson and Waldfogel, 2003). The impacts
of maternal work differ depending on choices in child care, and both quantity
and quality changes in mothers’ care at home. Fort et al. (2016) find a negative
impact of the child care usages in affluent families. The negative impacts are
11
due to fewer interactions with adults, which are usually of high quality at home.
However, children of low-educated single mothers, whose care is not considered of
high quality, also experience similar negative impacts. Bernal and Keane (2011)
reports that their children experience a 2.1% decrease in test scores after their
mother’s employment and child care usages. This implies that single mothers’
choices for child care are limited, which is possible because of insufficient increases
in earnings.
Qualitative and quantitative changes in mothers’ care at home also affect chil-
dren. FelfeandHsin(2012)andBrooks-Gunnetal.(2002)reportthatwork-related
stress or physical hazards decrease children’s cognitive and behavioral develop-
ment. Bastian and Lochner (2020) show that the EITC recipients, particularly
unmarried mothers, reduce the time spent with their children, even though those
are mostly related to non-investment activities more than active learning and de-
velopment activities with their children. In this paper, I will explore how the
resources at home related to child development change and see how mothers’ em-
ployments change the home environment and affect their children.
Finally, the EITC may affect children by changing the household composition
such as co-residence, marriage, or fertility decisions. Theoretical predictions on
marital status and fertility are mixed. Consistently, empirical findings suggest
that the EITC does not induce mothers’ singlehood nor fertility (Baughman and
Dickert-Conlin, 2009; Dickert-Conlin and Houser, 2002; Eissa and Hoynes, 1999;
Ellwood, 2000). In the robustness check of the paper, I will explore how the
mother’s marriage decision and family composition are affected by the EITC.
12
1.4 Identification Strategy
The identification strategy is to compare the academic achievements of sin-
gle mothers’ children before and after the EITC expansion in 1993, using the
difference-in-differences (DID) approach. The treatment group in my paper is
children of single mothers who are likely to receive the EITC. To ensure the sam-
pleincludeEITCrecipients, Irestrictthesampletochildrenofmotherswithahigh
school education or less, and with (expected) household earnings below $30,000.
3
The low education level has been often used in previous studies such as Eissa and
Liebman (1996) and Meyer and Rosenbaum (2001) to make the sample a high-
impact group.
4
However, it is a loose restriction, and allows many high-income
familiestobeincluded, whoarenoteligiblefortheEITC.Iaddanotherrestriction,
low expected household earnings in 1992, below $30,000. $30,000 is the maximum
level of household earnings for the eligibility during the reference period.
As this project focuses on the impacts of mothers’ labor supply from the EITC
on children, an ideal control group would be a group: (1) which is similar to the
treatment group and has the same pre-trends in child outcomes, (2) which does
not experience an increase in the mothers’ labor supply after the expansion, and
(3) at the same time, whose changes in income related to mothers’ labor supply
does not differ from those of the treatment group. I explore the candidates for the
control group, and I find that the children of married mothers with low education
3
I estimated an earning’s equation using the sample of earners before 1993. For the regression,
I used exogenous variables, such as age (squared and cubed), an education level (squared and
cubed), and one race dummy variable (nonwhite) following Eissa and Liebman (1996). Using the
estimated coefficients and individual characteristics, I predict an earned income for each mother
in the sample.
4
Previous literature used the restriction based on their findings that over 60% of married
couples with less than a high school education were eligible for the EITC.
13
and low household incomes is a satisfactory control group. I will provide evidence
of the above in the following sections on robustness checks and mechanisms.
5
My final treatment and control groups are children of single mothers and mar-
ried mothers with low education levels (12) and low (expected) household earn-
ings ( $30;000). The two groups are similar in that they are all low-income
and likely EITC recipients. However, they are different in that only the treatment
group, children of single mothers experience changes in mothers’ labor supply. The
logic behind this is that the single mother recipients are usually in the phase-in
range and have large labor supply incentives while the married mothers are usually
in the plateau or phase-out ranges, and do not have the same incentives. The 1993
EITC expansion provided recipients in the phase-in range large tax cuts, 116%
decrease in the marginal tax rate (from -16% to -34%). At the same time, the
tax rates for the recipients in the plateau range did not change. However, for the
people in the phase-out range faced increase in the tax rates by 70%. Figure 1.2
shows the location of single and married mothers on the 1994 EITC benefit sched-
ule,
6
where 58% of single mothers were in the phase-in range, while only 15% of
married mothers were. As a result, the EITC gives large work incentives primarily
to single mothers.
Asthemaritalstatusisnotaperfectdeterminantofbeinginthephase-inrange,
the estimate of the DID will be smaller than the actual impact. To compensate for
this, I use the Fuzzy DID approach as an alternative approach by using the income
5
Plausible alternative control groups will be children of single mothers with high education
levels or single mothers who have a larger number of children. However, the mothers in those
groups either increase labor supply sharply or have large changes in incomes.
6
I use the EITC schedule disclosed in 1994, which was formed by the enactment of the 1993
expansions. The household incomes in 1993 are used. I calculate the EITC benefits based on
the information on the EITC schedules, incomes, and the number of children.
14
level for the treatment status. I use the income threshold $8,000 (in 1996 dollars),
where the phase-in range ends according to the 1994 EITC schedule. For this
alternative specification, the treatment group is children of single mothers with a
household income less than or equal to $8,000 in 1993, and the control group is
children of single mothers with a household income between $8,000 and $24,000.
However, as income level is endogenously determined and so is the treatment
status, I use the expected household income as an instrumental variable (IV),
which makes this approach different from the Sharp DID method. I found single
mothers’ children in the phase-in range show a reduction in their academic scores,
which is even larger than that of the DID approach using mothers’ marital status.
For more details on theses, see the appendix.
1.5 Data and Statistics
Toaddresstheresearchquestion,IbeginwiththeNationalLongitudinalSurvey
of Youth data 1979 (NLSY79), a nationally representative, longitudinal study.
Started in 1979, it includes schooling, labor market activity, marriage, and family
background information on 12,686 males and females, between the ages of 15 and
22 at that time. From this group, I only look at the female respondents, who
comprise the sample of mothers in my paper.
I match these mothers to their children using a related study, the National
Longitudinal Survey of Youth 1979 Child and Young Adults (NLSY79 Child and
Young Adults). Begun in 1985, the dataset interviews the children of the female
NLSY79 respondents and includes child-specific information on various child out-
comes in cognitive, socioemotional, and physiological assessments. Linking the
15
two datasets allows me to investigate the impact of the EITC both on the mothers
and their children.
Among the various child outcome measures provided by the NLSY, I use the
scores on the Peabody Individual Achievement Tests (PIAT) to evaluate cognitive
achievement. The PIAT measures academic achievement for children aged five
and over, and the NLSY79 provides raw, percentile, and standard scores on the
Mathematics, Reading Recognition, and Reading Comprehension assessments. I
utilize the standard scores of Mathematics and Reading Recognition and combine
the two scores, which are commonly used by the previous researchers (Agostinelli
andSorrenti,2018;BastianandMichelmore,2018;DahlandLochner,2012;Ruhm,
2004). The standard scores are derived on an age-specific basis from the child’s
raw score, and the norming sample has a mean 100 and a standard deviation of
15. To make the scores more interpretable, I create normalized test scores with a
mean of zero and a standard deviation of one based on the random sample of test
takers.
In addition to the measures of cognitive achievement, I use the Behavioral
Problem Index (BPI) and Motor and Social Development (MSD) as measures of
noncognitive achievement. The BPI measures the frequency of childhood behavior
problems from children age four and over (Zill and Peterson, 1986). The questions
ask mothers about specific behaviors that their children may have exhibited in
the previous three months. For the BPI, higher scores present higher frequency
or greater levels of behavior problems. The MSD measures the motor, social,
and cognitive development of young children from birth to three years by asking
mothers to answer questions on the age-appropriate motor and social development
items. Exploring the non-cognitive achievements is useful as it helps us to pre-
16
dict future earnings and educational attainment (Heckman and Rubinstein, 2001).
Especially, by exploring the MSD, we can see how much the early investment mat-
ters for the development of younger children, which has been explored by many
studies (Ruhm, 2004; James-Burdumy, 2005). For these measures, I also create
normalized scores with a mean of zero and a standard deviation of one.
Finally, I use information on maternal labor supply and family income from the
NLSY79 for the analysis of the mechanisms. The NLSY79 provides information on
weekly labor hours and the number of jobs. Based on the information, I created a
variable for annual labor hours and dummy variables for holding any jobs, positive
working hours, and working full-time. I also utilize information on family income
including earnings, government welfare, and imputed EITC benefits. The imputed
EITC amounts are calculated based on information on the number of children in
the household and household income using the TAXSIM program (version 27)
maintained by Daniel Feenberg and the National Bureau of Economic Research.
7
Out of the many EITC expansions in the 1990s and early 2000s, I use the
1993 expansion, which is the largest (see Figure 1.3) and is known to have caused
the biggest changes to low-income households including maternal labor supply.
8
Due to possible bias arising from the later policy changes, I limited my timeframe
to the period 1990 to 1996, because there was another EITC expansion, which
became effective in 1987 and possibly affected child outcomes in 1988 due to the
Tax Reform Act of 1986,
9
and 1996 was the year when there were another large
7
The TAXSIM program calculates federal and state income tax liabilities from typical survey
data. For more details, see http://www.nber.org/taxsim.
8
Kleven (2019) find the only the EITC reform in 1993 is associated with clear employment
increases. For the other expansions, he reports that the labor market conditions in the 1990s
contributed more to the increase in the labor supply.
9
Including 1986 and 1988, however, does not change the main findings.
17
policy changes for single mothers, replacing the AFDC with the TANF. As the
cognitive achievement variables are released biannually, the actual years of data
available are 1990, 1992, 1994 and 1996.
Thesampleincludeschildrenofmarriedandsinglemothers(widowed,divorced,
and never married), where single mothers comprise 40% of the sample. The moth-
ers are between 25 and 39 years old. I exclude the children with mothers who are
in school, who are not able to work, or who have a spouse who cannot work. I
also exclude the children in the household with extremely high values in net worth
(more than $50,000).
Table 1.1 presents the summary statistics of the treatment and control groups.
Column 1 presents the characteristics of children of single mothers with less than
or equal to high school education and with (predicted) household earnings below
$30,000 (treatment group); Column 3 presents characteristics of children of mar-
ried mothers with the same levels of education and household earnings (control
group). There are some differences between the two groups. A group of children of
single mothers, the treatment group, tends to be non-white and to be in families
with lower income, higher total welfare, and fewer adults as we can expect. Single
mothers’ children are more likely to live with their grandparents. They have the
lower PIAT math, reading, the Motor and Social Development (MSD) scores and
have more behavioral problems on average. The table also indicates that the mar-
ried and single mothers are almost at the same age, have similar education level
and have similar labor market activities on average. The employment rates are
60% for single mothers and 75% for married mothers.
10
10
Theyarehigherthanthoseinpreviousstudies, whichare49%and52%forsingleandmarried
mothers with less than high school education (Eissa and Liebman, 1996; Eissa and Hoynes, 2004).
The difference seems to be due to the different ethnicity composition.
18
1.6 Results
To analyze the impact of the 1993 EITC expansion on child development, I use
a difference-in-differences (DID) approach with the child fixed effect. I choose the
expansion as it is the largest expansion of the EITC and it gave large labor supply
incentives to single mothers. I estimate the following child development model:
y
it
=+Single
i
Expansion1993
t
+X
0
it

1
+Z
i
+
t
+"
it
; (1.1)
where y
it
is a cognitive or noncognitive development of child i at time t such as
the PIAT mathematics and reading scores, the Behavioral Problem Index (BPI),
and the Motor and Social Development (MSD) score. The control variables in
X
it
include child characteristics (ages of children), mother characteristics (ages of
mothers), nonwage income,
11
and region dummies.
12
Z
i
controls for child fixed
effects such as children’s characteristics (sex, race and time-invariant unobsereved
11
The NLSY provides information on income from a variety of sources such as income from
working, transfer from the government, transfer from nongovernment sources such as child sup-
port, alimony, and parental payments, and income from other sources such as scholarships. I
calculated the unearned income by subtracting the mother’s wages and salary and government
transfer from the total net family income variable. Total net family income is a composite income
figure from the income sources for household members related to the respondent by blood and
marriage. Therefore, for married mothers, unearned income includes a spouse’s earnings. All
income variables are in the past calendar year.
12
The NLSY provides rich information on the household composition such as ages of the
youngest child in the household, number of family members, number of adults and children in
the household, number of members in certain age groups, number of adults working, a dummy
variable for living with grandparents. However, in my paper, to avoid an argument regarding
the EITC’s impact on the household composition, I stick to the parsimonious model excluding
those variables. I will discuss this issue more detailed in the next chapter, the robustness check.
Including the control variables does not change the main findings.
19
ability) as well as maternal characteristics (education level, race and marital sta-
tus), and
t
controls for time effect. The remaining variables are all dummy
variables, where Single
i
equals one if the child’s mother is single in 1992, and
Expansion1993
t
equals one for any year after 1993.
This paper tests the size and direction of , the coefficient of the interaction
term between Single
i
and Expansion1993
t
. If the estimate for is a negative
number (positive for the BPI), this implies that children of single mothers obtain
lower test scores, develop more slowly, and exhibit more behavioral problems. In
other words, the children of single mothers have a negative net effect after the
expansion possibly because single mothers work so much more but end up with an
insufficient increase in income.
I start by analyzing the impacts on children’s cognitive development measures,
the PIAT math and reading scores. The first column in Table 1.2 shows the
estimation result of the combined PIAT math and reading scores, and the second
and third columns show the estimation results of the individual math and reading
scores, respectively. Results suggest that children of single mothers have lower
scores on the test. They have 13.61 percent reduction of a standard deviation
in the combined scores. The negative effects in the test scores are larger for the
reading test; the reading scores decrease by 14.36 percent of a standard deviation,
whilethemathscoresdecreaseby10.16percentofstandarddeviation. Thisimplies
that the major environment changes which the EITC brings to the children, are
more related to language development such as the decrease in interactions with
their mothers. The size of the effects is comparable to previous findings. Bernal
and Keane (2011) relate a 2.1% decrease in the test scores of children of single
mothers as a response to one year of child care instead of mother care. Agostinelli
20
and Sorrenti (2018) find a larger impact, a one-hundred-hour per year increase
in maternal work is related to a 6% standard deviation decrease in the children’s
math and reading test scores. As their findings can be converted to 6.3 and 18
percent standard deviation decrease for 3 years, my finding is between these two.
Table 1.3 shows the analysis of noncognitive development measured by the
Behavioral Problem Index (BPI) and the Motor and Social Development (MSD)
scores. I only find strong negative impacts on children, younger than 3 years old.
The MSD scores decrease by 68 percent of a standard deviation. On the other
hand, the BPI does not change significantly. Although many studies recognize
that time with mothers is especially important for early childhood development,
the impacts of mothers’ working on the MSD has not been widely shown before.
This result indicates that single mothers’ early employment have more adverse
impacts on younger children.
13
SubgroupanalysesforthePIATmathandreadingscores, andMotorandSocial
Development (MSD) are also available in the appendix table section (Table B3 to
Table B5). Overall, the negative impacts are concentrated among younger children
who are between 5 and 12 years, and children with Non-white mothers. The PIAT
scores decrease by 14% for children who are between 5 and 12 years old while
children who are older than 12 do not show the similar reductions. Children with
Non-white mothers show larger reduction in the PIAT and MSD scores by 14%
and 70%, respectively.
13
There are a number of studies on the impact of maternal work on early childhood develop-
ment. However, they report mixed conclusions. Some authors such as Sherlock et al. (2008),
Belsky (1988) and Baydar and Brooks-Gunn (1991) conclude that maternal employment in the
first year of child causes increased behavioral problems. In contrast, several studies did not find
significant impacts (Vandell and Ramanan, 1992; Desai et al., 1989).
21
1.7 Common Trend Assumption and Robustness
Checks
1.7.1 Common Trend Assumption
Since child and mother characteristics, child fixed effects and year fixed ef-
fects are included in the regressions, it is essentially a generalized difference-in-
differences (DID) model. The underlying assumption for an unbiased estimate of
is that the trends in the child outcomes for both control and treatment groups
before the expansion are parallel. I examine the parallel pretreatment trends as-
sumption for the outcome variables, particularly for the PIAT math and reading
scores, and the Motor and Social Development (MSD) scores. To test the assump-
tion, I estimate the following equation:
y
it
=+
X
k

k
Single
i
D
k
t
+X
0
it

1
+Z
i
+
t
+"
it
; (1.2)
where k = 1992;1994;1996. y
it
represents the child i’s outcome in year t. D
k
t
are dummy variables equal to one if it is in year k. Note that the dummy for
k = 1990, D
1990
t
is omitted so that the treatment effects are relative to 1990.
Thus, coefficients of Single
i
D
k
t
,
k
captures the average differences between
child outcomes of single and married mothers in year k compared to 1990. I
include the child fixed effects, the time effects, and the control variables as in
equation (1.1).
Table 1.4 reports the regression results. A test of the parallel trend assumption
shows that
1992
= 0 for the PIAT math and reading scores and Motor and So-
22
cial Development (MSD) scores, which means the differences in the outcomes are
constant overtime before the 1993 expansion. Based on this, I conclude that the
pretreatment trends in the child outcomes of single and married mothers are simi-
lar, and married mothers’ children can serve as a control group for single mothers’
children in the treatment period.
14
The table also confirms that the negative impacts come after 1993, in 1994 and
1996, and the impact on the combined PIAT math and reading scores is larger in
1996. Figure 1.4 provides a graphical display of the same information in the table.
1.7.2 Placebo Test
With this placebo test, I check if there are still similar negative impacts among
the less likely EITC recipients, children in the household with high-income levels.
It will make sure if the EITC has its largest effects among children most likely to
be eligible for the credit. For the test, I use the sample with children of single and
married mothers with the expected household income higher than $30,000.
As shown in Table 1.5, the impacts on both combined PIAT math and reading
and Motor and Social Development (MSD) scores become smaller and insignificant
with the less likely EITC recipient sample. In other words, the large reductions
in the scores among children of single mothers compared to those of married are
only for EITC recipients.
14
It might be ideal to test the assumption with data before the major EITC expansions started
such as before 1986. However, as the data is only available after 1986, I test the assumption
using 1990 and 1992 for pretreatment periods.
23
1.7.3 Other Welfare Reforms
Around 1993 and 1996, there were many welfare reforms other than the EITC
expansions. Even though the Aid to Families with Dependent Children (AFDC)
was replaced with the Temporary Asistance for Needy Families (TANF) in 1997 at
the national level, the federal government allowed states to experiment with their
welfareprograms, undertheheadingofwelfarewaivers. Thestatewaiversincluded
all the key elements that would later be implemented on national scale through
TANF reform, including time limits, work requirements, and financial incentives
to work.
To check if those state-level reforms were primary causes of the negative im-
pacts, I include additional state dummies and state dummies interacted with year
dummies. As the welfare reforms were mostly state-wide and the dates of the
implementation were determined by states, the state dummies and state by year
dummies capture the welfare changes well. Specification B in Table 1.6 shows
that the impacts on the combined PIAT math and reading scores, and Motor and
Social Development (MSD) scores remain negative and significant as they are in
Table 1.2 and 1.3, which are repeated in Specification A. The results confirm that
the EITC accounts for the most of the negative impacts.
1.7.4 Child Care from Grandparents
As mothers start working, they need child care and, depending on its quality,
theimpactsofmothers’workonchildrencandiffer. Inthispaper, ifsinglemothers’
alternative care is of lower quality than that of married mothers’, the estimates in
equation (1.1) will be downward biased.
24
To check this possibility, I use the information on the presence of grandparents
in the household. After the parents, grandparents provide the most childcare (Fort
et al., 2016; Blau and Currie, 2006). While the presence of grandparents within
a household does not necessarily mean they provide child care, if grandparents
decide to live with their grandchildren more often after the EITC expansion to
help the mothers, it can be used to see the impact.
15
I add a dummy variable for
the presence of grandparents and a triple interaction term among three dummy
variables for the presence of grandparents, single mothers, and the 1993 expansion
in addition to the other control variables in equation (1.1) as follows:
y
it
=+Single
i
Expansion1993
t
+
Single
i
Expansion1993
t
Grand
it
+X
0
it

1
+Z
i
+
t
+"
it
:
(1.3)
With the new regression, I first check if the treatment effects () change with
the additional variables. Then, I check the coefficient of the triple interaction (
),
which shows if the grandparents’ cares are different from the other alternative
cares for single mothers’ children.
Specification E in Table 1.6 shows that controlling for the presence of grand-
parents variables doesn’t change the estimates of the treatment effects much for
both combined PIAT math and reading scores and Motor and Social Development
(MSD) scores compared to the baseline estimates in Specification A in Table 1.6.
Table 1.7 shows more detailed results including estimates for the triple interaction.
15
An ideal way to explore the impacts of alternative care is to use the information on child care
usage while mothers work. However, unfortunately, the NLSY provides child care information
for very limited periods, only for 3 months after childbirth.
25
I find that both for the combined PIAT math and reading scores and the MSD
scores, the grandparents’ cares are not different from the other alternative cares
for single mothers’ children. Thus, the quality of the alternative child care, the
presence of grandparents, does not affect the estimates.
1.7.5 EITC and Marital Status
Whether the expansion in 1993 affected marriage decisions or not is important
in this study. If mothers’ marriage decisions were directly affected, and many of
the mothers changed their marital decisions, it would have violated the group-
composition condition for the difference-in-differences (DID), and the estimates
would have been biased. Even though available evidence suggests that the EITC
does not encourage the existence of female-headed families or family formation
(Dickert-Conlin and Houser, 2002; Eissa and Hoynes, 1999; Ellwood, 2000),
16
I
explorethispossibilityinmysample. Inthissubsection, Icheckhowmanymothers
change their marital status by looking at the statistics, and if the expansion causes
changesintheirmaritalstatusbyusingadifference-in-differences(DID)regression.
I first check the frequency of marital status change. I check the percentage of
mothers who changed their marital status around 1993, between 1992 and 1994,
and between 1992 and 1996. I first check the marital status change pattern of
mothers with all education levels. Between 1992 and 1994, 3% of married mothers
changed their marital status to single, and 12% of single mothers changed their
marital status to married. Between 1992 and 1996, the rates were 7% and 19%
16
Some recent studies find significant impacts. However, they do not give a solid answer as
they report conflicting results. Herbst (2011) finds that a $1,000 increase in the EITC benefit
resulted in a 4.9% decrease in the probability that single mothers would marry. However, Bastian
(2017) finds the opposite. He found that a 10% point increase in the state EITC rates led to a
1.5% point increase (or 2.9%) in the probability of being married the following year.
26
for the mothers. Similar patterns are observed among the likely EITC recipients,
mothers with low education levels. I find a 4% change for married mothers and
11% change for single mothers between 1992 and 1994. These become 9% and
18% respectively between 1992 and 1996. From the statistics, I observe that EITC
recipients were equally likely to change their marital status before and after the
expansioncomparedtotheallmothersamples. Thesimilarpatternsinthechanges
in the marital status among non-EITC recipients and likely EITC recipients imply
that the EITC would not affect mothers’ marital decision.
I then address this issue more rigorously by using the difference-in-differences
(DID) approach around the 1993 EITC expansion. The treatment group includes
likely EITC recipients, mothers with a high school education or less. The con-
trol group is nonrecipients, mothers with more than 12 years of education. The
empirical model is as follows:
y
it
=+LOW
i
Expansion1993
t
+X
0
it
+Z
i
+
t
+"
it
; (1.4)
where y
it
indicates whether the mother i’s marital status changes or whether the
mother i is single during the year t. LOW
i
equals one if the mother i is a likely
EITC recipient, or has a high school education or less in 1992.
Tables 1.8 and 1.9 show that the 1993 EITC expansion does not affect the
marriage decision of the EITC recipient mothers in my sample. I find no evidence
that the EITC increases the probability of marital status changes or being single
more compared to non-EITC recipients for all, single and married mothers.
Based on the analyses above, I conclude that it is less likely that the EITC
influences the children by changing the mothers’ marriage decisions. However, as
27
still some mothers change their marital status around 1993, 19% for single mothers
and 7% for married mothers between 1992 and 1996, I check if the results change
when I exclude the mothers who change their marital status. Specification C in
Table 1.6 includes only mothers who have constant marital status from 1990 to
1996. The estimates do not change much for the combined PIAT math and reading
scores, even larger. However, the impact on Motor and Social Development (MSD)
scores decreases and becomes insignificant.
1.7.6 Unmarried Mother Sample
By controlling for child and mother characteristics and child fixed effect, most
heterogeneities between the treatment and control groups are handled, and it is
partly proven as the parallel trends assumption is satisfied.
However, a concern that is hard to be dealt with is that single mothers’ chil-
dren experienced changes in family composition at least once when their parents
divorcedorlostaparentinthepast. Iftheeffectsofpasteventslastforalongtime,
and if the effects on children cannot be systematically controlled in the regression,
it would bias the estimates.
I check if this scenario happens by using a subsample of the treatment group,
children of unmarried mothers. As unmarried mothers did not experience divorce
or losing a partner, the concern on the family events can be moderated. Specifi-
cation D in Table 1.6 reports that there are similar patterns and actually larger
impacts for both the combined PIAT math and reading scores and Motor and
Social Development (MSD) scores. This provides an evidence that the decrease
in child outcomes are not driven by negative family events, which only the single
28
mothers’ children experience.
17
1.8 Mechanisms of Impact
In Sections 2 and 3, I have shown that the 1993 expansion had adverse impacts
on the children of single mothers. To explain these surprising findings, I explore
several mechanisms through which the EITC possibly affects the children. As the
EITC is designed to reward work, the primary channels are a change in the labor
supplyandachangeinhouseholdincomesrelatedtothework. Ifirstcheckifsingle
mothers work more than married mothers, and then if the changes accompany
related household income changes.
In addition, I explore a more direct mechanism, changes in investments in chil-
dren to show implications why the changes in mothers’ labor supply and incomes
influence children negatively.
1.8.1 Maternal Labor Supply
In this subsection, I test if there is a relative increase in the labor supply of
single mothers compared to that of married mothers. I use the mothers of the
estimating sample children. I use the similar DID approach in equation (1.1) as
follows:
mls
it
=+Single
i
Expansion1993
t
+X
0
it

1
+Z
i
+
t
+"
it
; (1.5)
17
There can be some arguments about solo motherhood of unmarried mothers. However, what
the psychology literature reports is that single motherhood itself does not result in psychological
problems for children. For more detail, please refer to Golombok et al. (2016).
29
wheremls
it
is a measure of motheri’s labor supply during the yeart. The control
variables inX
it
are similar to those in equation (1.1), and include mothers’ charac-
teristics (age, age squared, age cubed), nonwage income, and region dummies.
t
controls for time effect and Z
i
controls for mother fixed effects (race, and marital
status). The remaining variables are all dummy variables whereSingle
i
equals one
if the mother is single in 1992, and Expansion1993
t
equals one for any year after
1993. I test whether, the coefficient of the interaction term betweenSingle
i
and
Expansion1993
t
is positive, which indicates that single mothers work more after
the expansion.
I use four measures of the maternal labor supplymls
it
: annual labor hours, an
indicator for whether a mother reports positive hours worked during past calendar
year, an indicator for whether a mother has at least one job, and an indicator
for whether the mother works full time (i.e.,1650 + hours in a year). I analyze
the impacts for the all mothers sample and then the working mothers sample to
separate and see the intensive margin effect.
Fromtheanalysis, Iobservethatsinglemothersincreasetheirlaborsupplyonly
at the extensive margin. As in column 1 and 2 in Table 1.10, for the all mothers
sample, mothers are more likely to work (4.96%) and more likely to have a job
(5.13%) after the expansion. The size of the extensive margin effect is larger than
the previous findings, 2.8% (Eissa and Liebman, 1996). The effects on working
mothers arein the column 3and 4. As theprevious literature finds, single mothers,
who are already working, do not increase their annual labor hours or are not more
likely to work full time.
These regression results suggest that single mothers devote more time to work.
Theextensivemarginincreasesolelycanaccompanyalargereductionincaregiving
30
inputs as searching for jobs and starting to work is costly in terms of time and
energy. Those mothers need to spend time visiting multiple workplaces during job
search, to adjust themselves to new environments and tasks, or to commute to
the workplaces, which reduce both quantity and quality of childcare at home, and
potentially affect child outcomes.
1.8.2 Other Family Incomes
In this subsection, I explore changes in the family income sources related to
the mothers’ labor supply choices such as mother’s earned incomes, government
welfare amounts, and EITC benefits.
18
In equation (1.1) for the child outcome
analyses, I do not include these income sources as control variables since the EITC
also affects those income sources by changing mothers’ labor supply decisions. I
test if there are income source changes, which are plausibly attributable to the
child outcome by using the same two-way fixed effect model as equation (1.5) with
the same mother sample.
The regression results are in Table 1.11. I first check the sum of the three
income sources in column 1, and I find that it decreases but the decrease is not
significant. I look into subcategories of the sum and I find that maternal earn-
ings increase but the increase is statistically insignificant. However, the imputed
EITC and government welfare change significantly; the imputed EITC increases
by $351.78, and the government welfare decreases by $504.95. One explanation
of why the mothers’ earnings do not increase much is that single mothers are not
18
Among three income sources, the EITC benefits are calculated using the TAXSIM program
(version 27) maintained by Daniel Feenberg and the National Bureau of Economic Research. The
TAXSIM program calculates federal and state income tax liabilities from typical survey data.
For more details, see http://www.nber.org/taxsim.
31
rewarded much for their entry into the labor market as they are mostly low-skilled
workers, or they are not ready to work full time because of their child care bur-
dens. However, as the EITC gives the largest marginal benefits to those who start
working or are in the phase-in rage, single mothers still receive the greater benefits.
From the analysis, I found that the sharp increase in the maternal labor supply
does not result in a significant increase in the sum of other related incomes. The
insignificant income changes are because of an insignificant decrease in mothers’
earned income, a significant but small increase in the EITC benefit, and a signifi-
cant decrease in government transfers. As the change in the sum of other income
sources is insignificant, assuming that parents are indifferent about income sources
— do not use a specific source of income for children, I conclude that the changes
in income do not play an important role in explaining the changes in the child
outcomes.
1.8.3 Investments in Children
The EITC mostly affects children by increasing the maternal labor supply and
household incomes. However, those changes will be effective for children when they
are attributable to changes in their home environments. I explore the changes in
the quality of children’s home environments using the Home Observation Measure-
ment of the Environment-Short Form (HOME-SF) scores in the NLSY 79 Child
and Adults.
The HOME-SF measures the quality of a child’s home environment, such as
the quality of cognitive stimulation and emotional support provided by a child’s
family. It has been used in previous studies as an input, which helps to explain
32
other child characteristics or behaviors. It includes information on time and goods
investments in children such as the number of books and instruments children
have, whether to discuss TV programs, to eat dinner together or to do outdoor
activities. Questions are mostly asked to mothers but some questions are recorded
by interviewers. Answers are in binary or nominal variables, and the HOME-SF
index is constructed based on the answers.
For the analysis, I use the same DID model and control variables as in equation
(1.1). The sample is the estimating sample children. The dependent variable y
it
is the HOME-SF score and two subscores, cognitive stimulation and emotional
support scores. Column 1 in Table 1.12 shows that the HOME-SF scores do
not change significantly, implying that overall home environment does not change
much. However, a cognitive stimulation subscores decrease by 29.03, or by 2.4%
from the sample mean. This shows that single mothers’ children have less cognitive
stimulation at home after the expansion compared to married mothers’ children.
Emotional support subscores do not change.
I look into the items of the cognitive stimulation scores to see the details of the
decrease. As the questions are asked based on the child’s age, the analysis is also
age-specific. Among the 13 to 14 questions, I choose several questions which are
related to time and goods inputs for children at home.
Table 1.13 shows some signs that mothers spend less time on children’s cogni-
tive developments. In column 1, mothers have fewer discussions over TV programs
with their children (6-9 years old) by 15.96%. In addition, 12.72% fewer mothers
take their children (10-14 years old) to musical or theatrical performances.
19
On
19
The study of Bastian and Lochner (2020) finds different results with the American Time
Use Surveys data. Mothers decrease time in home production. However, they spend the same
amount of time attending museums or events with children.
33
the other hand, there are no signs of fewer or more goods inputs. As Table 1.14
shows, for both age groups, there are not significant changes in whether to have
enough number of books (ten or twenty) or to have a musical instrument.
The analysis on the home environment shows that single mothers’ children
have less cognitive stimulation at home and it is mostly due to less time invested.
The results are consistent with the findings that single mothers start working but
their incomes do not change much. Combining the analyses on mothers’ labor
supply, incomes and home environment, I conclude that reduced mother-child
interaction due to the increased mother’s work is attributable to reductions in the
child development.
1.9 Conclusion
Using the mother-child matched data (NLSY) from 1990 to 1996, I find that
the biggest EITC expansion in the 1990s has negative impacts on single mothers’
children. There is a large reduction in the combined PIAT math and reading scores
by 14 percent of a standard deviation and in the Motor and Social Development
by 68 percent of a standard deviation. This is surprising as the previous literature
has reported mostly positive impacts among the EITC recipients.
The analyses of the mechanisms of the EITC show that reduced mother-child
interaction due to an increase in the mother’s labor supply mostly explains the
negative impacts. After the expansion, 5% more single mothers start to work, and
5% have jobs. Their entry to the labor market leads to a increase in the EITC
benefits and a decrease in welfare. However, it only increases their earned income
slightly, and they end up with no significant changes in income. Plus, children
34
at home experience less cognitive stimulation, which is mostly attributable to less
time investment from mothers such as discussing TV programs or going to musical
or theatrical performances. There were no changes in goods investments on the
children.
These findings suggest that in case of the low-educated and low-income single
mothers, a more disadvantaged group, the time channel effect (less time with
mothers) can dominate income channel effect (increase in household income). This
dominance is also observed in Agostinelli and Sorrenti (2018), where children in
the low-income households with low hourly wages were less benefited from the
EITC, and their test scores decreased after mothers’ employment.
These results suggest a different but constructive perspective on the pro-work
welfare programs, which have been popular since they were initiated. For those
programs to be an effective poverty reduction tool, they need to be paired with
other interventions such as child care.
20
Otherwise, the financial benefit of the
program will be outweighed by the negative impact of decreased mother-child
interactions. It is particularly important for single mothers as they are mostly
low-skilled workers and they cannot afford a child care of good quality in the
market even though they start working.
21
20
There can be some arguments that larger cash transfers might work better to solve this prob-
lem without government’s intervention in the child care market. However, larger cash transfers
might cause another moral hazard problems. In addition, as Bergmann (1996) argues, that high-
quality child care provided by the government has more benefits to children than cash transfer
programs. One caveat on the cash transfer program is that parents may use up the grant and
purchase low-quality child care.
21
In that sense, recent child care subsidies from the federal and state governments such as
CalWORKs are expected to be helpful to mitigate these negative impacts.
35
1.10 Tables and Figures
Table 1.1
Summary Statistics
Sample: Children of Single Mothers Children of Married Mothers
Mean S.t.Dev Mean S.t.Dev
(1) (2) (3) (4)
1. Child Characteristics
Combined PIAT Math and Reading Scores (average) 95.57 11.82 99.69 11.71
PIAT Math Scores 94.33 12.76 97.90 12.71
PIAT Reading Scores 96.72 14.23 101.36 14.10
Behavioral Problem Index 108.23 16.09 106.25 13.97
Motor and Social Development 99.29 14.84 99.80 14.55
Age of child (in months) 100.69 49.71 91.94 48.29
Male 0.51 0.50 0.50 0.50
Black 0.60 0.49 0.23 0.42
Hispanic 0.24 0.43 0.37 0.48
2. Household Charactersitcs
Total net family income (truncated) 16058.27 13722.57 37018.94 19542.51
Nonwage income (truncated) 4599.55 10057.38 27226.07 17716.26
Total welfare amount (truncated) 4178.35 4221.77 549.42 1765.09
Number of family members in the housheold 4.07 1.60 4.73 1.33
Number of adults in the household 1.55 0.77 2.04 0.46
Number of children of mother in the household 2.68 1.26 2.70 1.19
Number of preschooler 1.53 1.07 1.51 1.02
Grandparents in the household 0.08 0.27 0.04 0.20
3. Mother Characteristics
Age of mother 31.10 2.94 31.15 2.72
Education level 11.15 1.48 11.22 1.73
Positive annual working hours 0.60 0.49 0.76 0.43
Number of jobs held 1.11 1.06 1.36 0.97
Annual working hours 953.12 994.78 1203.34 949.30
Annual working hours (conditional on working) 1593.01 796.46 1579.47 767.23
Working fulltime (conditional on working) 0.58 0.49 0.56 0.50
Mother’s earned income 13204.09 10033.05 13181.30 9063.12
Mother’s earned income (conditional on working) 7280.38 9931.18 9243.45 9695.57
Observations 2258 3267
Notes: This table shows the summary statistics of estimating sample. Data is from
the children of the NLSY79 Child and Young Adults and linked to their mothers in the
main NLSY79, which ranges from 1990 to 1996. The sample contains married and single
mothers with children between age 25 and 39. All dollar amounts are in 1996 dollars.
Thetotalnetfamilyincomeandtotalwelfareamountaretruncatedatthetop2%level.
36
Table 1.2
Regression Results of Children’s Cognitive Achievement
Dependent variables: PIAT Math+Reading PIAT Math PIAT Reading
(1) (2) (3)
Expansion in 1993 Single -0.1361*** -0.1016* -0.1436**
(0.0470) (0.0528) (0.0566)
Age of child -0.0323*** -0.0325** -0.0260***
(0.0105) (0.0129) (0.0082)
Age of mother 1.2087 1.6195 0.4931
(0.9478) (1.0818) (1.1281)
Age of mother squared -0.0381 -0.0517 -0.0144
(0.0300) (0.0341) (0.0359)
Age of mother cubed 0.0004 0.0005 0.0002
(0.0003) (0.0004) (0.0004)
Nonwage income (last calendar year) 0.0010 -0.0011 0.0030
(0.0037) (0.0044) (0.0043)
Constant -10.3112 -14.4511 -3.5748
(10.0078) (11.5043) (11.7834)
Child fixed effect Yes Yes Yes
Year and region dummies Yes Yes Yes
Number of children 1,829 1,832 1,849
Observations 3,753 3,773 3,812
R-squared 0.0201 0.0200 0.0154
Notes: Sample includes children of mothers with a high school education or less, and
with expected household earnings below $30,000 in 1992. Data are from the children of
theNLSY79ChildandYoungAdultslinkedtotheirmothersinthemainNLSY79,which
ranges from 1990 to 1996. Expansion1993 equals one for 1994, and 1996. Single equals
one if mother of child is single in 1992. I use nonwage income in terms of quadruple
root instead of log to include samples with zero amounts. Robust standard errors in
parentheses and clustered at the child level.***p<0.01, ** p<0.05,p<0.1
37
Table 1.3
Regression Results of Children’s Noncognitive Achievement
Dependent variables: Behavioral Problem Index Motor and Social Development
(1) (2)
Expansion in 1993 Single -0.0332 -0.6821**
(0.0571) (0.2780)
Age of child 0.0031 -0.0738*
(0.0096) (0.0391)
Age of mother -0.2506 3.3181
(1.0126) (4.7355)
Age of mother squared 0.0075 -0.0858
(0.0319) (0.1532)
Age of mother cubed -0.0001 0.0008
(0.0003) (0.0016)
Nonwage income (last calendar year) 0.0062 0.0042
(0.0044) (0.0157)
Constant 2.3945 -42.4780
(10.7052) (48.6067)
Child fixed effect Yes Yes
Year and region dummies Yes Yes
Number of children 1,995 798
Observations 4,233 1,081
R-squared 0.0151 0.0767
Notes: Sample includes children of mothers with a high school education or less, and
with expected household earnings below $30,000 in 1992. Data are from the children
of the NLSY79 Child and Young Adults linked to their mothers in the main NLSY79,
which ranges from 1990 to 1996. Expansion1993 equals one for 1994, and 1996. Single
equals one if mother of child is single in 1992. I use nonwage income in terms of
quadruple root instead of log to include samples with zero amounts. Robust standard
errors in parentheses and clustered at the child level.***p<0.01, ** p<0.05,p<0.1
38
Table 1.4
Tests for the Parallel Trends Assumption
Dependent Variables: PIAT Math+Reading Motor and Social Development
(1) (2)
Year 1992 Single 0.0501 -0.2631
(0.0497) (0.2055)
Year 1994 Single -0.0758 -0.9509***
(0.0583) (0.3522)
Year 1996 Single -0.1564** -0.8859*
(0.0718) (0.4729)
Child fixed effects Yes Yes
Year and region dummies Yes Yes
Controls Yes Yes
Number of children 1,829 798
Observations 3,753 1,081
R-squared 0.0215 0.0818
Notes: Sample includes children of mothers with a high school education or less,
and with expected household earnings below $30,000 in 1992. Data are from
the children of the NLSY79 Child and Young Adults linked to their mothers in
the main NLSY79, which ranges from 1990 to 1996. Expansion1993 equals one
for 1994, and 1996. Single equals one if mother of child is single in 1992. I use
nonwage income in terms of quadruple root instead of log to include samples with
zero amounts. Robust standard errors in parentheses and clustered at the child
level.***p<0.01, ** p<0.05,p<0.1
39
Table 1.5
Placebo Tests with Children of Mothers with High-Income Levels
Dependent variables: PIAT Math + Reading Motor and Social Development
(1) (2)
Expansion in 1993 Single -0.0570 -0.2975
(0.0513) (0.2402)
Age of child -0.0155 -0.0576
(0.0141) (0.0391)
Age of mother -0.5794 -6.1077*
(0.8468) (3.3454)
Age of mother squared 0.0135 0.1997*
(0.0261) (0.1049)
Age of mother cubed -0.0001 -0.0022**
(0.0003) (0.0011)
Nonwage income (last calendar year) 0.0020 0.0172
(0.0045) (0.0182)
Constant 9.7211 61.0221*
(9.1649) (35.5721)
Child fixed effect Yes Yes
Year and region dummies Yes Yes
Number of children 2,127 1,341
Observations 4,361 1,895
R-squared 0.0119 0.0387
Notes: Sample includes children with expected household earned income larger
than $30,000 in 1992. Data is from the children of the NLSY79 Child and
Young Adults and linked to their mothers in the main NLSY79, which ranges
from 1990 to 1996. Single equals one if mother of child is single in 1992. I use
nonwage income in terms of quadruple root instead of log to include samples with
zero amounts. Robust standard errors in parenthesis and clustered at the child
level.***p<0.01, ** p<0.05,p<0.1.
40
Table 1.6
Robustness of the Estimates
Dependent variable: PIAT Math +Reading Motor and Social Development
(1) (2)
A. Baseline sample -0.1361*** -0.6821**
(0.0470) (0.2780)
B. State dummies & State Year dummies -0.1218** -0.7580**
(0.0517) (0.2780)
C. Mothers with constant marital status -0.2080*** -0.4318
(0.0597) (0.3441)
D. Unmarried single mothers -0.2298*** -0.7664***
(0.0525) (0.2894)
E. Presence of Grand Parents -0.1320*** -0.6647**
(0.0478) (0.2876)
Child fixed effect Yes Yes
Year and region dummies Yes Yes
Notes: Specifications identical to those in Table 2 and 3. Estimates of treatment
effect, Expnasion in 1993 Single. Specification A repeats the results in Tables
2 and 3. Specification B includes state dummies and state dummies interacted by
year dummies to capture the effects of state-level welfare waivers. Specification C
is a regression with children of mothers with constant marital status. Specification
D is a regression with children of single mothers who have never married for the
treatment group. Control group is identical. Specification E includes a dummy
variable for the presence of grandparents and interactions among dummy variables
of the presence of grandparents, children of single mothers and years after 1993.
Table 9 shows the results in detail. Standard errors (in parentheses) are robust
for heteroskedasticity and clustered at the child level. *** p<0.01, ** p<0.05, *
p<0.1.
41
Table 1.7
Impacts of Child Care (Presence of Grandparents in the Households)
Dependent variables: PIAT Math + Reading Motor and Social Development
(1) (2)
Expansion in 1993 Single -0.1320*** -0.6647**
(0.0478) (0.2876)
Expansion in 1993 Single Grandparents -0.0676 -0.5825
(0.1824) (0.4394)
Grandparents in the household -0.0021 -0.0120
(0.0958) (0.4801)
Age of child -0.0323*** -0.0746*
(0.0105) (0.0389)
Age of mother 1.2046 3.3221
(0.9471) (4.7698)
Age of mother squared -0.0380 -0.0862
(0.0300) (0.1544)
Age of mother cubed 0.0004 0.0008
(0.0003) (0.0017)
Nonwage income (last calendar year) 0.0010 0.0040
(0.0037) (0.0162)
Constant -10.2645 -42.4796
(10.0025) (48.9479)
Child fixed effect Yes Yes
Year and region dummies Yes Yes
Number of children 1,829 798
Observations 3,753 1,081
R-squared 0.0202 0.0774
Notes: The sample is children of mothers with a high school education or less, and had
expected income less than $30,000 in 1992. Data from survey years 1990 to 1996 of
the matched data of NLSY79 and NLSY79 Child and Young Adults. Expansion1993
equals one for 1994, and 1996. Single in 1992 equals one if mother of child is single.
I use nonwage income in terms of quadruple root instead of log to include samples
with zero amounts. Robust standard errors in parentheses and clustered at the child
level.***p<0.01, ** p<0.05,p<0.1
42
Table 1.8
Impact of the EITC on Change in Marital Status
Dependent variable: Any Change in Marital Status
Samples: All Mothers Single Mothers Married Mothers
(1) (2) (3)
Expansion in 1993 Low education 0.0029 0.0067 0.0120
(0.0095) (0.0257) (0.0087)
Age of mother 0.1855 0.4143 -0.0008
(0.1943) (0.3928) (0.1991)
Age of mother squared -0.0052 -0.0129 0.0007
(0.0060) (0.0121) (0.0061)
Age of mother trupple 0.0001 0.0001 -0.0000
(0.0001) (0.0001) (0.0001)
Nonwage income (truncated) -0.0304*** -0.0278*** -0.0301***
(0.0016) (0.0021) (0.0024)
Constant -1.6886 -3.4395 0.0735
(2.1146) (4.2615) (2.1667)
Mother fixed effect Yes Yes Yes
Year and region dummies Yes Yes Yes
Number of mothers 2,763 871 1,781
Observations 9,231 2,763 6,265
R-squared 0.2286 0.2596 0.2682
Notes: Sample includes all mothers. Data comes from survey years 1990 to 1996
of NLSY79. Dependent variable is a dummy variable equal to one if the mother
is single. Expansion1993 equals one for 1994, and 1996. Low education equals
one if mother has a high school education or less. I use nonwage income in terms
of quadruple root instead of log to include samples with zero amounts. Robust
standard errors in parenthesis and clustered at the mother level.***p <0.01, **
p<0.05,p<0.1.
43
Table 1.9
Impact of the EITC on Probability to be Single Mothers
Dependent variable: Being Single Mothers
Samples: All Mothers Single Mothers Married Mothers
(1) (2) (3)
Expansion Low education 0.0024 0.0079 0.0103
(0.0093) (0.0256) (0.0085)
Age of mother 0.1689 0.4177 -0.0068
(0.1915) (0.3915) (0.1950)
Age of mother squared -0.0048 -0.0131 0.0008
(0.0059) (0.0121) (0.0060)
Age of mother cubed 0.0001 0.0001 -0.0000
(0.0001) (0.0001) (0.0001)
Nonwage income (truncated) -0.0305*** -0.0277*** -0.0304***
(0.0016) (0.0020) (0.0024)
Constant -1.4783 -3.4329 0.1838
(2.0823) (4.2463) (2.1200)
Mother fixed effect Yes Yes Yes
Year and region dummies Yes Yes Yes
Number of mothers 2,857 880 1,860
Observations 9,517 2,789 6,518
R-squared 0.2294 0.2604 0.2695
Notes: Sample includes all mothers. Data comes from survey years 1990 to 1996
of NLSY79. Dependent variable is a dummy variable equal to one if mothers
change the marital status. Expansion1993 equals one for 1994, and 1996. Low
education equals one if mother has a high school education or less. I use nonwage
income in terms of quadruple root instead of log to include samples with zero
amounts. Robust standard errors in parentheses and clustered at the mother
level.***p<0.01, ** p<0.05,p<0.1.
44
Table 1.10
Changes in the Mothers’ Labor Supply
Samples: All Mothers Working Mothers
Dependent Variables: Positive Working Hours Having a Job Annual Working Hours Working As Full Time
(1) (2) (3) (4)
Expansion in 1993 Single 0.0496* 0.0513** -25.2120 0.0137
(0.0265) (0.0261) (60.5966) (0.0371)
Mother fixed effect Yes Yes Yes Yes
Year and region dummies Yes Yes Yes Yes
Controls Yes Yes Yes Yes
Number of mothers 959 959 803 803
Observations 3,268 3,268 2,495 2,495
R-squared 0.0070 0.0087 0.0108 0.0159
Notes: Sample is the mothers of the estimating sample children for the combined
reading and math scores, the BPI and the MSD. Data from survey years 1990
to 1996 of the matched data of NLSY79 and NLSY79 Child and Young Adults.
Expansion1993 equals one for 1994, and 1996. Single in 1992 equals one if the
mother is single. I use nonwage income in terms of quadruple root instead of log
to include samples with zero amounts. Standard errors (in parenthesis) are robust
for heteroskedasticity and clustered at the mother level. ***p< 0:01, **p< 0:05,
*p<0:1.
45
Table 1.11
Changes in the Other Income Sources
Dependent Variables: Sum Mother’s Wages and Salaries Welfare Imputed EITC
(1)+(2)+(3) (1) (2) (3)
Expansion in 1993 Single -497.1784 144.9330 -504.9465*** 351.7840***
(623.4318) (556.5022) (110.3567) (68.3750)
Mother fixed effect Yes Yes Yes Yes
Year and region dummies Yes Yes Yes Yes
Controls Yes Yes Yes Yes
Number of mothers 947 959 959 947
Observations 2,489 3,268 3,268 2,489
R-squared 0.0397 0.0214 0.0386 0.1586
Notes: Sample is the mothers of the estimating sample children for the combined reading and math scores, the
BPI and the MSD. Data from survey years 1990 to 1996 of the matched data of NLSY79 and NLSY79 Child and
Young Adults. Expansion1993 equals one for 1994, and 1996. Single in 1992 equals one if the mother is single. I
calculated the imputed EITC benefits using the TAXSIM program (version 27) maintained by Daniel Freenberg
and the National Bureau of Economic Research. I use nonwage income in terms of quadruple root instead of log to
include samples with zero amounts. Standard errors (in parentheses) are robust for heteroskedasticity and clustered
at the mother level. ***p<0:01, **p<0:05, *p<0:1.
46
Table 1.12
Changes in Home Environment
Dependent Variables: HOME-SF Score Cognitive Stimulation Emotional Support
(1) (2) (3)
Expansion in 1993 Single -14.1150 -29.0268*** -1.9803
(8.8504) (9.5484) (10.2542)
Child fixed effects Yes Yes Yes
Year and region dummies Yes Yes Yes
Controls Yes Yes Yes
Number of children 2,301 2,267 2,176
Observation 5,222 5,037 4,647
R-squared 0.0176 0.0119 0.0205
Notes: Sample is the estimating sample children for the combined reading and math
scores, the BPI and the MSD. Dependent variables include the HOME-SF score and
two subscores, cognitive stimulation and emotional support. The HOME-SF represents
the Home Observation Measurement of the Environment-Short Form. The HOME-SF
is the primary measure of the quality of a child’s home environment included in the
NLSY79 child survey. Data from survey years 1990 to 1996 of the matched data of
NLSY79 and NLSY79 Child and Young Adults. Expansion1993 equals one for 1994,
and 1996. Single in 1992 equals one if mother of child is single. I use nonwage income in
terms of quadruple root instead of log to include samples with zero amounts. Standard
errors (in parentheses) are robust for heteroskedasticity and clustered at the child level.
***p< 0:01, **p< 0:05, *p< 0:1.
47
Table 1.13
Changes in Cognitive Stimulation Related to Time Inputs (detailed)
Dependent variables: Discuss TV Programs with Child Took Child to Musical/Theatrical Performance Last Year
Sample: 6-9 years Older than 10 years old
(1) (2)
Expansion in 1993 Single -0.1596** -0.1272*
(0.0716) (0.0673)
Child fixed effect Yes Yes
Year and region dummies Yes Yes
Controls Yes Yes
Number of children 1,188 1,227
Observations 1,684 1,934
R-squared 0.0257 0.1753
Notes: Samples are children who are 6-9 or 10-14 years old, and whose mothers have a
high school education or less, and have expected household earnings below $30,000 in
1992. Data from survey years 1990 to 1996 of the matched data of NLSY79 and NLSY79
Child and Young Adults. Expansion1993 equals one for 1994, and 1996. Single in 1992
equals one if mother of child is single. I use nonwage income in terms of quadruple
root instead of log to include samples with zero amounts. Robust standard errors in
parentheses and clustered at the child level.***p<0.01, ** p<0.05,p<0.1
Table 1.14
Changes in Cognitive Stimulation Related to Goods Inputs (detailed)
Dependent variables: Have More Than Ten Books Have More Than Twenty Books Musical Instrument for Child
Sample: 6-9 years old 10-14 years old 6-9 years old 10-14 years old
(1) (2) (3) (4)
Expansion in 1993 Single -0.0356 0.0868 0.0252 -0.0372
(0.0636) (0.0660) (0.0754) (0.0696)
Child fixed effect Yes Yes Yes Yes
Year and region dummies Yes Yes Yes Yes
Controls Yes Yes Yes Yes
Number of children 1,197 1,230 1,195 1,227
Observations 1,700 1,929 1,701 1,925
R-squared 0.0346 0.0349 0.0230 0.0173
Notes: Samples are children who are 6-9 years old or 10-14 years old. Data from survey
years 1990 to 1996 of the matched data of NLSY79 and NLSY79 Child and Young
Adults. Expansion1993 equals one for 1994, and 1996. Single in 1992 equals one if
mother of child is single. I use nonwage income in terms of quadruple root instead of log
to include samples with zero amounts. Standard errors (in parentheses) are robust for
heteroskedasticity and clustered at the child level. ***p< 0:01, **p< 0:05, *p< 0:1.
48
Figure 1.1
The EITC Benefit Schedule in 1994 for families with two children
Notes: The figure shows the EITC benefit schedule in 1994 for families with more
than two children. There are three regions in the EITC benefit schedule: the
phase-in, plateau and phase-out ranges. The benefit increases with earned income
upto a certain point (phase-in range), stays the same and decreases afterwards
(plateau and phase-out ranges). The phase-in rate was 30% in 1994, which means
people in the range received 30 cents per a dollar of earned income. The phase-out
tate was 17.68%, which means the EITC benefit decreased by 17.68 cents per a
dollar of earned income. The clear (extensive margin) work incentives exist only
within the phase-in range.
49
Figure 1.2
DistributionofFamilyIncomesandImputedEITCBenefits–HouseholdswithMore
than Two Children
Notes: Sample is the mothers with a high school education or less, and had ex-
pected incomes less than $30,000 in 1992. Gray vertical dashed line indicates the
end of the phase-in range of the EITC benefit schedule in 1994, $8,425. The single
mothers are located themselves within the phase-in range more frequently com-
pared to married mothers (68% compared to 15%). It suggests that single mothers
have more incentives for participating in the labor force.
50
Figure 1.3
Expansions of Earned Income Tax Credit for Families with More than Two Chil-
dren in the 1990s
Notes: The figure shows the changes in the federal EITC schedule for families
with two children. I calculated the EITC benfits over time using the TAXSIM
program (version 27) maintained by Daniel Freenberg and the National Bureau
of Economic Research. This picture shows how EITC has expanded through a
series of expansions in terms of maximum benefit amounts and coverage. The
big jump from the EITC benefit schedule of 1993 to that of 1994 was due to the
EITC expansion in 1993.
51
Figure 1.4
Estimated Effect of the EITC for Years Before and After the 1993 Expansion
Notes: These figures show that children of married and single mothers shared
common trends in combined PIAT Math and Reading scores and the MSD before
1993. After 1993, there were relative reductions in the scores of single mother’s
children.
52
Chapter 2
Revisiting the EITC and Child
Development
1
2.1 Introduction
The impact of maternal work on children has become an important issue as
the percentage of mothers who work has risen in the past decades. In 1980, only
56% of mothers were in the labor force. However, the number became 70% in
2017 (WB, 2018).
2
Furthermore, while the recent welfare reforms have succeed
in encouraging single mothers to work more, some concerns regarding potential
negative impacts on children have risen (Waldfogel et al., 2002; Agostinelli and
Sorrenti, 2018). Therefore, policymakers are increasingly interested in the trade-
off between mother’s time and financial inputs and consider it as an important
1
This research was conducted with restricted access to Bureau of Labor Statistics (BLS) data.
The views expressed here do not necessarily reflect the views of the BLS.
2
WB is an abbreviation for Women’s Bureau.
53
research question.
3
To analyze the impact of maternal work on children is challenging. This is
because mothers who decide to work may have unobservable attributes that also
affect their child’s development. In addition, as working generally results in a
higher household income, it is hard to isolate the impacts of these two variables.
To overcome the difficulties, many economics studies use various econometric ap-
proaches such as fixed effect model, regression discontinuity, propensity scores, and
instrumental-variables (IV) approaches (James-Burdumy, 2005; Carneiro et al.,
2015; Baum II, 2003; Carneiro and Rodrigues, 2009; Blau and Grossberg, 1992).
However, these approaches still have empirical problems (Carneiro et al., 2015).
The findings of the previous studies are mixed, and depend on their sample
choices and empirical methodologies. Some articles suggest that maternal work
has little impact on child outcome (Leibowitz, 1977; Desai et al., 1989; Blau and
Grossberg, 1992; Parcel and Menaghan, 1994; Dustmann and Schönberg, 2012).
However, other papers find evidence suggesting that maternal work does impact
children, mostly in a negative way (Han et al., 2001; Baum II, 2003; Waldfogel
et al., 2002; James-Burdumy, 2005; Duncan et al., 1998; Carneiro and Rodrigues,
2009; Carneiro et al., 2015; Vandell and Ramanan, 1992). The inconsistency in
the findings shows that maternal work might have differential impacts depending
on the context, which remains an empirical question. At the same time, it implies
that it is difficult to find proper exogenous changes in the mothers’ labor supply.
To overcome the difficulty, I use an instrumental-variables (IV) approach to
address the research question, using policy changes in the EITC. The instrumental
3
This trade-off may have particularly significant impacts on single mothers, who are both
primary earners and caretakers for their children.
54
variableIuseformaternallaborsupplyisapolicyparameter, theEITC’staxcredit
rate, which is defined as the maximum potential federal and state phase-in rates
that a child’s family could face, given their state of residence, family size, and tax
year.
4
The rate changes over time for an individual, based on federal and state
policy changes to the EITC as well as changes to the individual’s family size or
movements across states.
The EITC policy parameters have been used in previous studies as a measure
of exposure to it (Michelmore, 2013; Bastian and Michelmore, 2018; Hoynes et al.,
2015; Chetty et al., 2011). However, not many studies use the parameters as in-
strumental variables, and they only look at the impacts of the exposure (Hoynes
et al., 2015; Chetty et al., 2011). In addition, the tax credit rate is rarely utilized
while the amounts are often used. This is because previous studies focus on in-
come effects from the EITC. Bastian and Michelmore (2018), for instance, use the
maximum credit amount as an IV for household income and find positive impacts
on the children’s education and adult earnings. A benefit of using the tax credit
instead of the amounts to look at how the EITC affects children through maternal
work is that the rate is more closely related to both the labor supply incentives
and the labor supply.
In this paper, using changes in the maximum credit rates of the EITC as
an instrument for mothers’ labor supply, I explore how the single mothers’ work
affects children’s cognitive and noncognitive achievements. For the analysis, I
use the children of single mothers in the National Longitudinal Survey of Youth
(NLSY). Firstly, I find that an increase in the maximum credit rates results in
a significant increase in the labor supply of single mothers. 1% increase in the
4
It does not depend on family income or mothers’ marital status.
55
credit rate is associated with up to 1,190 annual labor hours, 75% increase in the
labor force participation rate, and 1.2 more jobs. The relationships are strong
and the estimates are significant at 1% to 5% level. In addition, I find that
mothers’ labor supply negatively impacts children. For instance, mothers’ labor
force participation results in a decrease in the PIAT reading test scores by 1.3
of standard deviation. However, the PIAT math scores and Behavioral Problem
Index (BPI) are not affected. Other measures of mothers’ labor supply such as
annual labor hours and the number of jobs do not have any significant impacts.
The paper proceeds as follows: In Section 2, I review a brief history of the
EITC and introduce the instrumental variable. Section 3 discusses the data and
summary statistics. Section 4 provides estimation results with a panel fixed-effect
model. Section 5 provides regression results with instrumental-variables analysis,
and Section 6 concludes.
2.2 EITC and Instrumental Variable
TheEarnedIncomeTaxCredit(EITC)isarefundabletaxcredittolowtomid-
dle income households with dependent children. When it first started and became
a part of the 1975 U.S. tax code, it was a small cash bonus program. However,
since the welfare reforms in the 1990s, which emphasized work and responsibil-
ity of the welfare recipients, the EITC has been a popular alternative for income
redistribution and has helped low-income families with children.
The popularity of the EITC is twofold. One, it encourages people to work.
Two, it increases families’ incomes both by increasing the amount of work and by
giving them the tax credits. Thus, it is structured in such a way to redistribute
56
income with little distortion of the labor supply compared to the other means-
tested cash transfer programs such as the former Aid to Families with Dependent
Children (AFDC).
The EITC encourages work by increasing the amount of credit as the number
of work hours increases. Figure 1.1 shows how it works. There are three ranges in
the EITC: phase-in, plateau, and phase-out ranges. The phase-in range appears
first, where the benefit increases with additional earned income until it reaches the
maximum benefit. After that, there are plateau and phase-out ranges where the
benefit stays at the maximum and then gradually phases out. Of the three, the
EITC has its largest work incentives in the phase-in range.
Primary expansions of the EITC started in the early 1990s, which includes the
largest expansion in 1993 due to the Omnibus Reconciliation Act of 1993. As the
EITC has expanded, more people have received larger benefits (Figure 1.3). In
1975, for instance, 6.2 million households received 0.1 billion EITC transfers. The
number grew dramatically, and in 2018, 64 billion dollars were distributed to 26
million households.
5
A noticeable point of these expansions is that the EITC gives larger work
incentives to the recipients, particularly to those who are in the phase-in range.
As the EITC has expanded, it has rewarded work more by increasing the phase-in
rates, which determine the amount of tax credit as a percentage of earned income.
For instance, in 1986, the rate for families with two children was 11%, which means
the tax credit increases by 11 cents per dollar income in the range. The rate rose to
40% in 1996, when the households received additional 40 cents to a dollar income.
Table 2.1 shows how the rate has increased throughout the expansions.
5
https://www.taxpolicycenter.org/statistics/eitc-recipients
57
In addition to the federal EITC, state governments also have started their own
state EITCs. In 1986, Rhode Island became the first state to enact the state
EITC. Since then, the state EITCs have become popular and in 2020, 28 states
and the District of Columbia implemented them. The state governments provide
the EITCs in a proportion of the federal EITC. California, for instance, has given
the tax credit since 2015, which is 85% of the federal EITC.
6
As each state decides
on the year to start the program and the rates, residents face different labor supply
incentives depending on which states they reside in.
By using the federal and state EITCs, I define and use the maximum EITC
rates as a measure of labor supply incentives from the EITC. The rate comes from
the two parts: federal EITC and the state EITCs. Thus, the rate changes as
the phase-in rates change with the federal government’s expansions and as state
governments implement or make changes in their own state EITCs. I incorporate
these two and define the maximum EITC rate, maxEITC
ist
, as follows:
maxEITC
ist
=Rfederal
t
(1+Rstate
ist
)100; (2.1)
where maxEITC
ist
denotes the maximum credit rate in percent. It is the maxi-
mum potential federal and state phase-in rates a child i’s family could face, given
their state of residence, family size, and tax year. The maximum rate is deter-
mined by Rfederal
t
, a maximum federal phase-in rate, and Rstate
ist
, a state
EITC rate. For instance, mothers with two children in California in 2015, faced
74%(= 0:4 (1 + 0:85) 100) of maxEITC
ist
, where Rfederal
t
was 40% and
Rstate
t
was 85%.
6
Refundibilities of the state EITCs also differ.
58
Figures 2.1 and 2.2 show how the maximum rate, maxEITC
ist
, have changed
by years and across states given family sizes. Most changes were in the late 1990s,
which was mainly attributable to the federal expansions. After 2000, however,
there were only smaller changes across states. These changes in the maxEITC
ist
generate variations in the labor supply incentives, which might result in increase
in mothers’ labor supply and as result affect their children.
2.3 Data and Statistics
I use data from the National Longitudinal Survey of Youth (NLSY) as in Chap-
ter 1. The NLSY provides two sets of longitudinal data for mothers and their
children. For mothers, the National Longitudinal Survey of Youth 1979 (NLSY79)
containsdetailedinformation ondemographic characteristics, education, and labor
market activities. For children, the National Longitudinal Survey of Youth 1979
Child and Young Adults (NLSY79 Child and Young Adults) provides a biannual
measure of family background and various child outcomes from 1986 to 2000. I
match these mothers from the NLSY 79 to their children, and use this matched
data for analysis.
Intheanalysis, Ifocusonchildrenofsinglemotherswhoarewidowed, divorced,
or never married. Single mothers compose almost half of the EITC recipients, and
they are usually in the phase-in range, where the labor supply incentive is the
largest. Therefore, to ensure the sample includes the most impacted group, it is
composed of children of single mothers.
For the measures of child outcomes, I use both academic and non-academic
achievements. For scholastic achievements, I use the Peabody Individual Achieve-
59
ment Test (PIAT) math and reading sections, which is administered to children
aged five and over. It is an age-appropriate test that evaluates math skills from ba-
sic to advanced, and the PIAT reading test does the same for reading achievements
such as word recognition and pronunciation ability. For non-academic achieve-
ments, I use the Behavioral Problem Index (BPI), which measures the frequency
of childhood behavior problems in children ages four and over. Mothers are asked
about specific behaviors that their children may have exhibited in the prior three
months. On the BPI, the higher the score, the greater the frequency of behavior
problems. These measures are all given in raw, percentile and standardized scores.
Among them, I use standardized scores and normalize them to have a mean of
zero and a standard deviation of one to facilitate the interpretation.
To measure the mothers’ labor supply, I look at their labor force participation,
annual working hours, and the number of jobs held. Their labor force participa-
tion shows an extensive margin effect and the other two measure intensive margin
effects. As a child can be affected differently by his or her mother’s type of em-
ployment, exploring different measures is important.
Table 2.2 shows the summary statistics of the estimating samples of the PIAT
math and reading scores, and Behavioral Problem Index (BPI). On average, chil-
dren of single mothers are 10 years old, and in households with more than 2 chil-
dren, and 2 adults. More than half of them are African Americans.
7
The mothers
are young, 33 years old on average, and they are mostly high-school graduates.
Almost 70% are in the labor force, they have at least one job, and they work 1,213
hours a year. Their annual family income is $28,156, one-third of which is from
government payments. Depending on their eligibility, they also receive tax credits
7
The larger proportion of African Americans is the result of oversampling of the NLSY.
60
up to approximately 30% of their earned income.
2.4 Panel Fixed Effect Estimation
Before the instrumental-variables (IV) analysis, I first explore the relationship
between maternal work and child development by using an individual fixed-effect
model. By doing so, I expect to check the differences in estimates, which will show
how much endogeneity can be adjusted by using the IV approach.
I estimate the following equation for a child’s human capital formation with
individual fixed effect:
Y
its
=+MLS
ist
+X
0
it

+
i
+
s
+t+"
ist
; (2.2)
where Y
ist
is cognitive or noncognitive development of child i in state s at year
t such as the PIAT math and reading scores and the Behavioral Problem Index
(BPI). MLS
ist
measures maternal labor supply such as mothers’ labor force par-
ticipation, annual working hours, or the number of jobs held. X
it
is a vector of
child, mothers and household characteristics such as children’s age, age of mothers,
mother’s education level, the age of the youngest child in the household, the num-
ber of children in the household, and nonwage income. I also control for individual
fixed effect, state fixed effect, and year effect. Here the coefficient of interest is
beta, which shows the relationship between the mothers’ labor supply and child
developments.
The fixed effect estimation results are displayed in Table 2.3 to 2.5. The esti-
mation results show that there are no significant relationships between any forms
61
of maternal labor supply and child outcomes. This might imply that there are no
relationships. However, it is still possible that the endogeneity related to mothers’
labor supply choices is the cause. For instance, if mothers with high motivation
are more likely to work and if their children who inherit the high motivation out-
perform in the tests, the estimate for can be upward biased, which leads to
insignificant estimates even though there is a negative relationship. Furthermore,
if the characteristic varies by years, and thus cannot be fully controlled by the fixed
effect, the fixed-effect model is not sufficient to improve the endogeneity problem.
The IV estimation results in the next section will show that this is highly likely.
2.5 Instrumental-Variables (IV) Analysis
I next use an instrumental-variables (IV) approach to analyze how mothers’ la-
borsupplymotivatedbytheEITCaffectschildren. Usingchangesinthemaximum
credit rates as an instrument for maternal work, I model a first stage-equation as
follows:
MLS
ist
=
1
+
1
lmaxEITC
ist
+X
0
it

1
+
i
+
s
+year
t
+"
ist
; (2.3)
whereMLS
ist
is a measure of labor supply of child i’s mother in states at yeart.
For the measure of labor supply, I use the labor force participation, annual labor
hours, and the number of jobs. Here, the lmaxEITC
ist
is the log of maximum
credit rates, which will be used as an IV for maternal work.
8
X
it
is a vector of
child, mother and household characteristics such as children’s age, age of mothers,
8
It is the maximum phase-in rate the household of the child i could face in year t and state s.
62
mother’s education level, the age of the youngest child in the household, number of
children in the household, and nonwage income. I also control for individual fixed
effect, state fixed effect and year effect. In this equation, I check the estimates of

1
, and obtain the predicted maternal labor supply.
An increase in the credit rate,maxEITC
ist
, gives work incentives by increasing
effective wages. It will clearly attract non-workers to the labor market. However,
aneffectformotherswhoarealreadyinthemarketislessclear. Astherearesubsti-
tution and income effects in opposite directions, the net impact will be determined
by the relative sizes of those two effects, which can be revealed empirically.
Table 2.6 to 2.8 present estimates from the first-stage regressions for maternal
work on EITC exposure. There are sizable increases in mothers’ labor supply from
the positive tax incentives. 1% increase in the credit rate is associated with from
70% to 75% increase in the labor force participation rate (Table 2.6), from 787
to 1,190 annual labor hours (Table 2.7), and 1 to 1.2 more jobs (Table 2.8). The
relationships are strong and the estimates are significant at 1% to 5% level, while
the labor force participation shows the strongest relationship.
Using the predicted mothers’ labor supply from equation 2.3, I then estimate
the impact of increasing mothers’ labor supply on child outcomes in the following
second-stage equation:
y
ist
=
2
+
2
^
LFP
ist
+X
0
it

+
i
+
s
+year
t
+"
ist
; (2.4)
where Y
ist
is a cognitive or noncognitive achievement of child i in state s at year
t, which is measured by standard PIAT math, reading scores, and Behavioral
Problem Index (BPI) scores.
^
LFP
ist
is the fitted value of maternal labor supply,
63
whichisestimatedinequation(2.3). X
it
isavectorofchild, mother, andhousehold
characteristics such as children’s age, age of mothers, mother’s education level, the
age of the youngest child in the household, the number of children in the household
and nonwage income. I also control for individual fixed effect, state fixed effect,
and year effect. Here, the coefficient of interest is
2
, which shows the impact of
mother’s labors supply on child developments.
Results from the second-stage equation (2.4) are presented in Table 2.9 to
2.11. From the estimation, I find that the maternal work generally has negative
impacts on child outcomes but only on the PIAT reading test scores. In Table 2.9,
single mothers’ labor force participation decreases the PIAT reading scores of their
children by-1.3 of standard deviation. Similarly, an increase in the number of jobs
decreases the score by -1.01 standard deviation as in Table 2.11. However, changes
in the annual labor hours do not have a similar impact. On the PIAT math
scores and Behavioral Problem Index (BPI), none of the maternal labor supply
has significant impacts.
The results are robust when I include additional control variables related to
household composition. They include age of the youngest child in the household,
thenumberoffamilymembers, thenumberofadultsandchildreninthehousehold,
the number of members in certain age groups,the number of adults working and a
dummy variable for living with grandparents. The estimates do not change much,
and I find that the impact of the labor force participation on the PIAT reading
scores slightly rises to -1.4 of the standard deviation as in Table 2.12.
The sizes of the impacts differ across subgroups. First, I compare the impacts
among children of single and married mothers. As Table 2.13 shows, single moth-
ers are more likely to participate in the labor market after the EITC expansions,
64
and as a result, the reading scores of their children decreased by -1.4 standard
deviation, while married mothers and their children are not affected.
In addition, I find that the negative impacts are larger for the single mothers,
who are more disadvantaged or more likely to be EITC recipients. Single mothers
with low household income level increase their labor supply when the credit rates
increases, and mothers’ labor force participation is associated with a decrease in
the PIAT reading scores by -1.3 of standard deviation (Table 2.14). In contrast,
mothers with high household income level do not respond to changes in the rate,
and there is no association between the mothers’ labor supply and child develop-
ment.
A similar contrast is found among children of single mothers with low and high
education levels. Mothers with low-education levels increase their labor supply in
response to the changes in the rate, and there is a weak but negative association
between mothers’ participation and the PIAT reading tests, -0.9 of standard de-
viation as in Table 2.15. In contrast, single mothers with high education levels
do not change their labor supply much, and there are no significant relationships.
Analysis on subgroups with different family sizes is available in Table 2.16 where
I find no differences across the subgroups.
2.6 Alternative Instrumental Variable
For an instrumental-variables (IV) approach to be valid, there are two condi-
tions to be satisfied. The first assumption is that the instrumental variable should
be strongly associated with the endogenous variable, here labor supply of mothers.
This assumption can be easily tested by using F-statistics. According to the rule
65
of thumb, if the F-statistics is larger than 10, the first assumption is considered
satisfied. The second assumption is that an instrument variable is exogenous. It
means that maxEITC
ist
is not related to error terms of child outcomes. There
is no structured way to test this assumption. However, we can indirectly assure
that the assumption is satisfied by testing and ruling out potential threats to the
exogeneity.
A possible threat to the exogeneity is that maxEITC
ist
depends on family
sizes, or the number of children. Even though many previous studies find no
evidence that the EITC affects fertility decisions, it is still possible that mothers’
characteristics, which affect fertility decisions also impact child outcomes. To
compensate for the weakness, I make and use an alternative instrument variable
maxEITC2
ist
, which is equivalent to maxEITC
ist
, but does not depend on the
number of children in the households. I run the same regression of equation (2.4)
by using a new IV, maxEITC2
ist
.
From the analysis, I find that a similar but weaker impact of mothers’ labor
supply on the reading scores, by -1.1 standard deviation as in Table 2.17. It
is partly due to the weak association between alternative IV and mothers’ labor
supply. With the Behavioral Problem Index (BPI) estimating sample, there is also
a weaker association between the IV and the mother’s participation. However,
I find there are fewer behavioral problems, by a -1.62 standard deviation after
mothers’ employment.
Alternative IV is more exogenous as it does not depend on family sizes. How-
ever, it has a less strong relationship with the mother’s labor supply compared to
the previous IV. It is possibly due to the fact that the IV is more related to family
income instead of maternal labor supply, which results in positive impacts on the
66
BPI.
9
2.7 Conclusion and Discussion
A limitation with the DID analysis in Chapter 1 is that it only shows indirect
links between maternal labor supply, and child development, and therefore, it is
hard to quantify the impact. In addition, it is hard to incorporate multiple EITC
expansions in the 1990s and early 2000s. The instrumental-variables (IV) method,
in this chapter, compensates for the weakness.
Using changes in the maximum credit rates of the EITC as an instrumental
variable for mothers’ labor supply, this paper shows that single mothers’ labor
force participation results in negative impacts on child outcomes. Mothers’ labor
force participation leads to a decrease in the PIAT reading test scores by 1.3 to 1.4
standard deviation decrease. The size of the impact is comparable to that of the
DID estimation. As the IV estimation gives the Local Average Treatment Effect
(LATE), it may not be best to compare the IV estimate to that of a DID approach.
However, to have a sense of how accurately they were estimated, I converted the IV
estimate, assuming that there are 5% increases in the mother’s labor supply and
children are exposed to the change for 3 years, and the IV estimate became 19%
percent standard deviation reduction. Under the same condition, the decrease in
the scores was 13% in the previous DID approach.
On the PIAT math scores, however, these two approaches do not arrive at
a consensus. While the IV estimation reports insignificant impacts of mothers’
9
Another concern related to the exogeneity is if the EITC affects the residence choices. This
issue has been least explored in the previous literature. Bastian and Michelmore (2018) finds that
only 1-2% of the sample moves across the states in a given year in the NLSY sample. However,
this should be explored further.
67
labor supply, previous DID approach finds negative impacts. The differences in
the results can be partly explained by the fact that the impacts on math scores are
weaker and smaller compared to those on the reading scores in the DID approach,
and the contrast is shown more clearly in the IV estimation.
A possible future work related to the research question and the IV approach is
to identify the impact of earned income, and isolate it from the impact of mothers’
labor supply. In this paper, only the maternal work is instrumented, and another
endogenous variable earned income is not. Thus, predicted maternal work, to a
certain extent is also affected by changes in the earned income, and it can results in
a bias as we estimate the impacts on child outcomes. Fortunately, as the income
has positive impacts on children, the bias underestimates the negative impacts
and is a less serious problem than overestimation. However, to accurately measure
the impact of maternal work, finding proper identification strategy or empirical
model is necessary, which will be a large contribution to the literature. Candidates
include to find another IV for the earned income as Agostinelli and Sorrenti (2018)
do or to find groups which experience relative changes in labor hours but not in
income.
68
2.8 Tables and Figures
Table 2.1
Historical Changes in the Federal EITC Tax Credit Rates
Calendar Year Credit Rate (percent)
One Child More than Two Children
1986 11 11
1988 14 14
1989 14 14
1990 14 14
1991 16.7 17.3
1992 17.6 18.4
1993 18.5 19.5
1994 23.6 30
1995 34 36
After 1996 34 40
Source: https://www.taxpolicycenter.org/statistics/eitc-parameters
69
Table 2.2
Summary Statistics
Mean St.Dev
(1) (2)
1. Child Characteristics
Math 96.31 13.43
Reading recognition 100.04 14.58
Behavioral Problem Index (BPI) 108.77 15.60
Hispanic 0.18 0.38
Black 0.52 0.50
Age of Children (months) 112.01 36.06
2. Mother’s Characteristics
Mother’s age 32.65 4.21
Mother’s education 12.09 1.92
Annual working hours 1213.45 1012.20
Labor force particiatpion 0.68 0.47
Number of jobs 1.21 1.03
3. Household Characteristics
Number of family members in the huosehold 3.89 1.56
Number of children in the household 2.63 1.35
Number of adults in the household 1.52 0.77
Age of the youngest child in the household 6.37 3.58
4. Income Variables
Total net family income 28156.23 36906.64
Wages and salaries 16166.74 19166.5
Non-wage income 11989.49 31043.4
Government payments 10958.44 10117.42
Maximum EITC rates 0.28 0.12
Observations 5,502
Notes: This table shows the summary statistics of the estimating samples, children
whose mothers are single, who are widowed, divorced or never married. Data from
survey years 1988 to 2000 of the matched data of NLSY79 and NLSY79 Child and
Young Adults. Standard errors (in parentheses) are robust for heteroskedasticity and
clustered at the child level. ***p< 0:01, **p< 0:05, *p< 0:1.
70
Table 2.3
Fixed Effect Model: Impacts of Mothers’ Labor Force Participation on Child De-
velopment
PIAT Math PIAT Reading BPI
Labor Force Participation 0.5620 0.4963 -0.2614
(0.5370) (0.5075) (0.6136)
Child fixed effect Yes Yes Yes
Year and state dummies Yes Yes Yes
Control Yes Yes Yes
Number of children 2,549 2,550 3,047
Observations 4,859 4,848 5,734
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
insteadoflogtoincludesampleswithzeroamounts. Standarderrors(inparentheses)are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
71
Table 2.4
Fixed Effect Model: Impacts of Mothers’ Annual Working Hours on Child Devel-
opment
PIAT Math PIAT Reading BPI
Annual Working Hours -0.0000 -0.0000 -0.0001
(0.0003) (0.0003) (0.0003)
Child fixed effect Yes Yes Yes
Year and state dummies Yes Yes Yes
Control Yes Yes Yes
Number of children 2,682 2,684 2,891
Observations 5,366 5,348 5,871
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
insteadoflogtoincludesampleswithzeroamounts. Standarderrors(inparentheses)are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
72
Table 2.5
Fixed Effect Model: Impacts of Mothers’ Number of Jobs on Child Development
PIAT Math PIAT Reading BPI
Number of Jobs -0.0210 0.0185 -0.2528
(0.2317) (0.1973) (0.2427)
Child fixed effect Yes Yes Yes
Year and state dummies Yes Yes Yes
Control Yes Yes Yes
Number of children 2,682 2,684 2,891
Observations 5,366 5,348 5,871
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
insteadoflogtoincludesampleswithzeroamounts. Standarderrors(inparentheses)are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
73
Table 2.6
Instrumental-Variables Approach: First Stage Estimates with Labor Force Partic-
ipation
Dependent Variable: Labor Force Participation
Child Outcomes: Math Reading BPI
(1) (2) (3)
Maximum Phase-in Rate 0.736*** 0.7078*** 0.7449***
(0.2045) (0.2063) (0.1811)
Kleibergen-Paap rk LM statistic 12.2 11.21 15.469
Kleibergen-Paap rk Wald F-statistic 12.94 11.78 16.911
Number of children 1288 1279 1369
Observations 3175 3148 3441
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
insteadoflogtoincludesampleswithzeroamounts. Standarderrors(inparentheses)are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
74
Table 2.7
Instrumental-Variables Approach: First Stage Estimates with Annual Working
Hours
Dependent Variable: Annual Working Hours
Child Outcomes: Math Reading BPI
(1) (2) (3)
Maximum Phase-in Rate 787.19*** 769.72*** 1190.04***
(472.06) (471.72) (413.18)
Kleibergen-Paap rk LM statistic 2.817 2.697 8.215
Kleibergen-Paap rk Wald F-statistic 2.781 2.662 8.295
Number of children 1450 1440 1507
Observations 3711 3685 3965
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
insteadoflogtoincludesampleswithzeroamounts. Standarderrors(inparentheses)are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
75
Table 2.8
Instrumental-Variables Approach: First Stage Estimates with Number of Jobs
Dependent Variable: Number of Jobs
Child Outcomes: Math Reading BPI
(1) (2) (3)
Maximum Phase-in Rate 1.0094** 0.9583** 1.2383***
(0.4442) (0.4433) (0.3963)
Kleibergen-Paap rk LM statistic 5.087 4.611 9.373
Kleibergen-Paap rk Wald F-statistic 5.165 4.674 9.746
Number of children 1450 1440 1507
Observations 3711 3685 2965
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
insteadoflogtoincludesampleswithzeroamounts. Standarderrors(inparentheses)are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
76
Table 2.9
Instrumental-Variables Approach: Effect of Labor Force Participation on Child
Development
Dependent variable: PIAT Math PIAT Reading BPI
(1) (2) (3)
Labor Force Participation 0.5753 -1.2959** -0.4080
(0.4724) (0.6602) (0.4545)
Child fixed effect Yes Yes Yes
Year and state dummies Yes Yes Yes
Control Base Base Base
Kelibergen-Paap rk LM statistic 12.204 11.209 15.469
Kleibergen-Paap rk Wald F-statistic 12.938 10.512 16.911
Number of children 1,288 1,279 1,507
Observaiotions 3,175 3,148 3,964
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
insteadoflogtoincludesampleswithzeroamounts. Standarderrors(inparentheses)are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
77
Table 2.10
Instrumental-Variables Approach: Effect of Annual Working Hours on Child De-
velopment
Dependent variable: PIAT Math PIAT Reading BPI
(1) (2) (3)
Annual Working Hours 0.0005 -0.0013 -0.0002
(0.0005) (0.0008) (0.0003)
Child fixed effect Yes Yes Yes
Year and state dummies Yes Yes Yes
Control Base Base Base
Kelibergen-Paap rk LM statistic 2.817 2.697 8.215
Kleibergen-Paap rk Wald F-statistic 2.781 2.662 18.295
Number of children 1,440 1,450 1,507
Observaiotions 3,685 3,711 3,965
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
insteadoflogtoincludesampleswithzeroamounts. Standarderrors(inparentheses)are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
78
Table 2.11
Instrumental-VariablesApproach: EffectofNumberofJobsonChildDevelopment
Dependent variable: PIAT Math PIAT Reading BPI
(1) (2) (3)
Number of Jobs 0.4211 -1.0191* -0.1451
(0.3619) (0.5925) (0.2507)
Child fixed effect Yes Yes Yes
Year and state dummies Yes Yes Yes
Control Base Base Base
Kelibergen-Paap rk LM statistic 5.087 4.611 9.373
Kleibergen-Paap rk Wald F-statistic 5.165 4.674 9.746
Number of children 1,450 1,440 1,507
Observaiotions 3,711 3,685 3,965
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
insteadoflogtoincludesampleswithzeroamounts. Standarderrors(inparentheses)are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
79
Table 2.12
Instrumental-Variables Approach: Effect of Labor Force Participation on Child
Development (Additional Controls)
Dependent variable: PIAT Math PIAT Reading BPI
(1) (2) (3)
Labor Force Participation 0.6772 -1.3997** -0.2528
(0.5288) (0.7121) (0.5141)
Child fixed effect Yes Yes Yes
Year and state dummies Yes Yes Yes
Control Additional Additional Additional
Kelibergen-Paap rk LM statistic 11.038 10.118 12.711
Kleibergen-Paap rk Wald F-statistic 11.510 10.512 13.451
Number of children 1,245 1,237 1,327
Observaiotions 3,054 3,030 3,319
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple
root instead of log to include samples with zero amounts. Additional control variables
includenumberoffamilymembersinthehousehold. Standarderrors(inparentheses)are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
80
Table 2.13
IV Estimates with PIAT Reading Scores by Mother’s Marital Status
Dependent variable: PIAT Reading Scores
Sample: All Single Married
(1) (2)
Labor Force Participation -2.6813* -1.3997** -30.9536
(1.4325) (0.7121) (204.6778)
Child fixed effect Yes Yes Yes
Year and state dummies Yes Yes Yes
Control Additional Additional Additional
Kleibergen-Paap rk LM statistics 5.638 10.118 0.023
Kleibergen-Paap rk Wald F-statistic 5.660 10.521 0.023
Number of children 4,435 1,237 3,035
Observaiotions 11,638 3,030 7,808
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
instead of log to include samples with zero amounts. Additional control variables include
number of family members in the household. Standard errors (in parentheses) are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
81
Table 2.14
Impacts of Mother’s Labor Force Participation on the PIAT Reading Tests by
Income Levels
Dependent variable: PIAT Reading Test
Sample: Low Income High Income
(1) (2)
Labor Force Participation -1.3480* -1.8644
(0.7582) (2.2618)
Child fixed effect Yes Yes
Year and state dummies Yes Yes
Control Additional Additional
Kleibergen-Paap rk LM statistics 8.805 1.227
Kleibergen-Paap rk Wald F-statistic 9.088 1.779
Number of children 1,131 60
Observaiotions 2,733 139
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data fromsurvey years1988 to2000of thematched dataofNLSY79 and
NLSY79 Child and Young Adults. Low education means 12 or less years of education.
High education implies more than 12 years of education. I use nonwage income in terms
of quadruple root instead of log to include samples with zero amounts. Additional
control variables include number of family members in the household. Standard errors
(in parentheses) are robust for heteroskedasticity and clustered at the child level.
***p< 0:01, **p< 0:05, *p< 0:1.
82
Table 2.15
Impacts of Mother’s Labor Force Participation on the PIAT Reading Tests by
Mother’s Education Level
Dependent variable: PIAT Reading Test
Sample: High School More than High School
(1) (2)
Labor Force Participation -0.8687 -4.2533
(0.5684) (6.8044)
Child fixed effect Yes Yes
Year and state dummies Yes Yes
Control Additional Additional
Kleibergen-Paap rk LM statistics 10.059 0.483
Kleibergen-Paap rk Wald F-statistic 10.941 0.475
Number of children 893 330
Observaiotions 2,186 790
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
instead of log to include samples with zero amounts. Additional control variables include
number of family members in the household. Standard errors (in parentheses) are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
83
Table 2.16
Impacts of Mother’s Labor Force Participation on the PIAT Reading Tests by
Number of Children
Dependent variable: PIAT Reading Test
Sample: One Child Two Children More than Two
(1) (2) (3)
Labor Force Participation -3.6339 -1.1332 -5.2765
(8.406) (1.0900) (22.1276)
Child fixed effect Yes Yes Yes
Year and state dummies Yes Yes Yes
Control Additional Additional Additional
Kleibergen-Paap rk LM statistic 0.263 3.222 0.075
Kleibergen-Paap rk Wald F-statistic 0.252 2.550 0.072
Number of children 182 409 561
Observaiotions 437 971 1,349
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
instead of log to include samples with zero amounts. Additional control variables include
number of family members in the household. Standard errors (in parentheses) are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
84
Table 2.17
Alternative Instrumental-Variable: Effect of Labor Force Participation on Child
Development
Dependent variable: PIAT Math PIAT Reading BPI
(1) (2) (3)
Labor Force Participation 0.7544 -1.0860 -1.6174**
(0.6318) (0.8089) (0.8978)
Child fixed effect Yes Yes Yes
Year and state dummies Yes Yes Yes
Control Base Base Base
Kelibergen-Paap rk LM statistic 6.777 6.585 7.870
Kleibergen-Paap rk Wald F-statistic 7.274 7.062 9.581
Number of children 1,167 1,157 1,205
Observaiotions 2,895 2,866 3,060
Notes: Samples are children whose mothers are single, who are widowed, divorced or
never married. Data from survey years 1988 to 2000 of the matched data of NLSY79
and NLSY79 Child and Young Adults. I use nonwage income in terms of quadruple root
instead of log to include samples with zero amounts. Additional control variables include
number of family members in the household. Standard errors (in parentheses) are
robust for heteroskedasticity and clustered at the child level. ***p < 0:01, **p < 0:05,
*p< 0:1.
85
Figure 2.1
Federal and State EITC Exposure by Year and State for Families with One Child
Notes: The figure shows changes in the maximum federal and state EITC tax
credit rates across states and by year. The maximum tax credit rate is defined as
the maximum potential credit rates, which an individual in one-child household
could face in a given year and state. Each colored line denotes individual states.
Source: author’s calculations.
86
Figure 2.2
Federal and State EITC Exposure by Year and State for Families with More than
Two Children
Notes: The figure shows changes in the maximum federal and state EITC tax
credit rates across states and by year. The maximum tax credit rate is defined
as the maximum potential credit rates, which an individual in a household with
more than two children could face in a given year and state. Each colored line
denotes individual states. Source: author’s calculations.
87
Chapter 3
The EITC and the Labor Supply: A
Case Study of Korea
3.1 Introduction
The Earned Income Tax Credit (EITC) is a refundable tax credit, which is
given to low to middle-income families with dependent children. In the United
States, it was introduced and became a part of the tax code in 1975 as a negative
income tax. Since its implementation, the EITC has been a popular method for
transferring income to low-income families with children, growing from $3.9 billion
(in 1999 dollars) in 1975 to $31 billion in 2000 through a series of expansions.
Its popularity stems from the structure of the tax credit, which allows people
to believe that it can redistribute income with much less distortion in the labor
supply decisions compared to other means-tested cash transfers such as the former
Aid to Families with Dependent Children (AFDC). The tax credit increases with
additional earned income until it reaches the maximum benefit, and gradually
88
phases out, creating a well-known trapezoid shape of the benefit schedule (see
Figure1.1).
The EITC has succeeded in encouraging work but only to a certain extent.
How the EITC affects the labor supply depends on whether individuals are in the
labor market, and in which ranges of the EITC they are. For the non-workers,
economic theory predicts that it encourage them to participate in the labor market
as the EITC increases their effective wages. Consistently, empirical findings such
as Eissa and Liebman (1996), Meyer (2002), and Meyer and Rosenbaum (2001)
report that it increases the labor force participation .
However, for people who are already working, the impact is ambiguous. Theo-
retically, the recipients either increase or decrease their working hours depending
on the relative sizes and directions of substitution and income effects that the
EITC generates. Most of the empirical evidence, however, suggests that the EITC
has no impact on the working hours of the recipients who are in the labor mar-
ket (Eissa and Hoynes, 2004). Furthermore, some studies report that EITC even
reduces the labor hours (Hoffman and Seidman, 2003; Dickert et al., 1995).
1
Whether the individual is a primary earner or a secondary earner of the house-
hold also matters. Secondary earners are usually considered less responsive to the
EITC because of the dominance of income effects. Studies of Eissa and Hoynes
(2006), and Yang (2018), for instance, find that married mothers usually stay at
home or even decrease their labor supply after the EITC expansions. On the
contrary, primary earners are responsive. Many studies report that single mothers
whoare primaryearners inthe householdusually increasetheir laborsupply(Eissa
and Liebman, 1996; Meyer and Rosenbaum, 2001). Meyer and Rosenbaum (2001),
1
Keane and Moffitt (1998), however, finds a positive impact.
89
for instance, reports that single mothers increase their labor force participation by
10% due to the EITC expansions. To summarize, the previous studies find that
the EITC has extensive margin effects on the primary earners in the household,
mostly single mothers. However, its impacts differ depending on the contexts and
still remain an empirical question.
Observing the popularity and effectiveness of the EITC in the United States,
many other countries such as Canada, United Kingdom, and France start to adopt
similar programs. In the early 2000s, the Korean government also joined the
queue. In 2005, the government started to discuss the implementation, and in the
same year, the implementation was finalized. As a result, there have been twelve
payments to the beneficiaries since the first payment was made in September 2009.
Korea’s EITC is very similar to that of the United States. It has a similar
incentive structure including phase-in, plateau, and phase-out ranges, and targets
low-income households with dependent children. However, specific details differ
such as household income thresholds, size of benefits and differential schemes.
Mostly, Korea’sEITCprogramissmaller, asithasalowerthresholdsforhousehold
earnings, a lower phase-in rate and a smaller maximum benefits.
With expectations on the EITC, even before the first payment was made, many
researchers start their research to predict the effectiveness. Most studies predict
the plausible beneficiaries or its impact on the labor supply by using structural
models with simulation methods. The common prediction of those papers is that
the EITC will not change the labor supply as the incentives from the EITC are
not large enough.
By using a policy simulation, Chun (2009), for instance, predicts that the
EITC in 2009 does not provide large incentives to the low-income households, and
90
he predicts that it will not result in increases in the labor supply. He explains
that the amounts of tax credit given to the recipients are small, considering the
existing cash transfer program, Minimum Living Standards Security (MLSS).
2
After the implementation, there are several empirical analyses. However, the
results are mixed. Yeom and Jun (2014) and Song and Chun (2011) find that the
EITC significantly increases the labor supply of likely EITC recipients. Park and
Kim (2011) also find that the 2008 EITC expansion has positive but insignificant
impacts. On the contrary, Ki et al. (2015) report that the EITC reduces the earned
income.
The main focus of the previous literature is to look at the post-treatment
outcomes of the EITC recipients rather than to explore the causal relationship
between the EITC and the labor supply. In other words, the studies usually rely
on comparisons between the actual EITC recipients and the non-recipients, and
find differences in their labor supply decisions. However, as the recipients decide
on their labor supply and as a result decide whether to participate in the EITC
program or not, the estimation based on their receipt history may be biased. In
addition, the studies rarely explore the EITC’s incentive structure and its impact
on the labor supply. My paper contributes to the literature by utilizing the EITC’s
incentive structure from the policy changes and focusing primarily on the causal
relationship, which provides more useful information to the policymakers to design
the program.
In this paper, I evaluate the impact of the 2011 EITC expansion on the labor
supply of married fathers in Korea. I utilize the differential EITC benefit schedules
2
Cho (1998), on the other hand, uses a simulation model based on the earnings equation and
predicts that the reduction in the tax rate caused by the EITC results in increasing labor force
participation of married mothers while there are no impacts on the working hours.
91
dependent on family sizes, which was first introduced by the 2011 EITC expansion.
The expansion gives larger EITC benefits to households with more children, which
is the largest for the household with more than three children. Using a Difference-
in-Differences (DID) estimation with the Korean Labor and Income Panel Study
(KLIPS), I find that the expansion does not have significant impacts on the labor
supply of married fathers both at the extensive and intensive margins.
Thepaperproceedsasfollows. InSection2, IreviewabriefhistoryoftheEITC
in Korea. Section 3 reviews the theoretical prediction of the impact of the EITC
on labor supply. Section 4 discusses the data and empirical strategy, difference in
differences (DID). Section 5 provides estimation results, and Section 6 concludes.
3.2 Brief History of EITC in Korea
In 2008, the Korean government provide the first plan for the EITC benefit
schedule. The EITC was supposed to be given to the households with annual
earnings lower than $17,000 and had at least one dependent child younger than
18 years old. The phase-in rate was 10%, the maximum credit was $800 and the
phase-our rate was 16% (Choi, 2013).
Afterward, the government has expanded the EITC program, and the first
expansion was in 2008 right before the first payments of the EITC were made.
The government decided to expand the EITC program earlier than expected to
support the low-income households suffering from the 2008 economic crisis (Choi,
2013). The government increased the phase-in rate from 10% to 15%, the phase-
out rate from 16% to 24%, and the maximum benefits from $800 to $1200, which
becomes the first EITC benefit schedule (see Figure 3.7).
92
In 2011, the government made another expansion. In this second expansion,
the government included married couples without dependent children and applied
different schemes (phase-in rate, phase-out rate, maximum benefit, and income
range) for households depending on the numbers of children as in Figure 3.2. It
was to support small-sized low-income households and to increase the birth rate.
The third change was in 2013. In the third expansion, the different benefit
schemes for the different number of children were removed, and Child Tax Credit
(CTC) took the role. Instead, different schemes for the differential marital status
(single parents or married couples) and labor force participation (single earner
or multiple earners) were added.
3
Through a series of EITC expansions, the
number of recipients increases from 0.6 million in 2008 to 1.4 million in 2015. The
government spending also increased from $453 million in 2008 to $1 billion in 2015.
Compared to the EITC program in the United States, Korea’s EITC covered
fewer households and provides fewer work incentives and smaller tax credits. In
2011, forinstance, themarriedcoupleswithtwochildrenwereeligiblefortheEITC
if their household earnings were below $21,000. The maximum credit was $1,700
and the phase-in rate was 18.9%. In contrast, the United States gave the EITC to
families with higher incomes. Married couples with two children with household
earnings below $46,471 were eligible for the EITC. The maximum credit was also
greater, $5,160, and the phase-in rate was higher, 40%. Thus, as the previous
studies also noted, Korea’s EITC seems to have smaller work incentives, which
potentially leads to smaller impacts on labor supply.
3
In 2014, the EITC also became available to the self-employed households
93
3.3 Theoretical Prediction
How the EITC affects labor supply can be analyzed with a leisure-labor supply
decision model. Adoption or expansions of the EITC affects individuals’ decisions
by changing the budget constraints from the linear line to the kinked line as Fig-
ure 3.3 shows. However, the impact differs depending on the previous working
status of individuals and which ranges the individual is in.
First, when an individual is a primary earner in a household and is not in
the labor market, the EITC gives positive work incentives and motivates labor
force participation by increasing the effective wages and as a result increasing the
opportunity cost of not participating (extensive margin) (see Figure 3.4).
However, when an individual is a primary earner in a household and is already
working, the impact might differ depending on the cases. If the individual is in the
phase-in range, two effects, which off-set each other arise simultaneously. There
is a substitution effect that increases the labor hours and an income effect that
decreases the labor supply given that leisure is a normal good. Thus, depending
on the relative sizes of the substitution and income effects, the individual would
increase or decrease the labor supply. If the income effect dominates the substitu-
tion effect, the individual will reduce labor hours (see Figure 3.5). Otherwise, the
individual will increase the working hours. In the plateau and phase-out ranges,
individuals decrease their labor supply. In the plateau range, there is only an in-
come effect, which decreases working hours, and in the phase-out range, both the
substitution and income effects make the individual decrease the labor hours (see
Figure3.6).
Thus, the economic theory gives clear expectations on the labor supply for
94
the non-workers, and the recipients in the phase-out ranges. The recipients in
the phase-in and plateau ranges might or might not increase their labor hours
depending on the size and directions of the two effects. Thus, how the EITC
affects the labor supply of the low-income households depends on which interval
he or she is in, in other words, how the government designs the program.
3.4 Data and Empirical Strategy
For data, I use the Korean Labor and Income Panel Study (KLIPS). The
KLIPS is a longitudinal study of households and individuals living in urban areas.
Started in 1998, it has been conducted annually on a sample of 50,000 households
and members of those households.
4
The latest Wave 18 was conducted in 2015
and completed in 2016.
The KLIPS is useful to investigate the impact of the EITC on the labor supply.
The data provides information on income, labor supply, and family background
bothatthehouseholdandindividuallevels, whichisusefultopredicttheeligibility
of the EITC. In addition, KLIPS provides information on the actual application
and amounts of the EITC that the household receives since the 13th Wave (con-
ducted in 2010). Even though the numbers of observation of those variables are
small at the current level, it is useful to explore if the EITC helps improve poverty
problems.
Here, I use a sample of married fathers with low household income (less than
$25,000 in 2009) and do not hold additional ownership of houses for investment
purposes. This makes the sample likely to be EITC recipients according to the
4
It has tracked their characteristics of the economic activities, labor market activities, income,
expenditures, education, job training, and social activities.
95
2011 EITC rule.
5
Among the three expansions of the EITC in Korea, I use the 2011 EITC ex-
pansion for two reasons. First, the expansion in 2011 was the first expansion
that moved forward from the restrictive design at the initial level, which people
expected would have a significant impact on the labor supply. Second, the expan-
sion gave differential labor supply incentives to the households depending on the
number of children, which allow the quasi-experiment opportunity. To be more
specific, before 2011, the maximum benefit for a household with dependent chil-
dren was $1,200 regardless of the number of the dependent children. However, in
2011, the maximum benefit increased, and the increases were larger for families
with more children. The households without child received up to $700. For house-
holds with one child, the maximum credit was $1,400. For the households with
two children, the maximum benefit rose to $1,700. The households with more than
three children received the largest amount $2,000.
Utilizing the differential incentive structures among families of different sizes,
I address the research questions by using the difference-in-differences (DID) ap-
proach. To be more specific, I compare changes in the labor supply between
married fathers who are affected more by the 2011 EITC expansion (treatment
group) and those who are affected less (control group).
I use three pairs of treatment and control groups. The first pair is married fa-
thers with children and without children (Primary treatment and control groups).
The second pair is married fathers with one child and without child (First alter-
5
According to the EITC schedule in 2011, households with more than $25,000 and/or owning
another house for investment purposes are not eligible for the EITC. Furthermore, the report on
EITC recipients in 2011 also confirms the fact that by showing that 81.8% of EITC recipients
did not hold ownership for a house.
96
native treatment and control groups). The third pair is married fathers with two
children and with one child (Second alternative treatment and control groups).
The summary statistics for those groups are presented in Table 3.1. Overall,
fathers with children are younger, have higher education levels, have more labor
market activities, and have higher earned income. For instance, married fathers
with one child are on average 43 years old while the married fathers without a child
are 60 years old. 42% of married fathers with one child graduate university/college
or have higher education while only 20% of the married fathers without child do.
Also, married fathers with one child have more labor market activities. Almost
93% of the married fathers with one child are in the labor market and have higher
earned incomes. On the contrary, married fathers with two children, compared
to married fathers with one child, are relatively older and have lower education
levels. In addition, they have fewer labor market activities, which results in lower
household earnings.
3.5 Regression Results
Using the treatment and control groups introduced above, I estimate the fol-
lowing equation, which is a typical DID model with individual fixed effect:
y
it
=+Treat
i
EXP2011
t
+X
0
it

+
i
+Year
t
+"
it
; (3.1)
where y
it
refers to labor supply of individual i at time t. There are two dummy
variables, Treat
i
and EXP2011
t
, which indicate the treatment status and post-
treatment respectively. Treat
i
equals 1 if the individuali is in the treatment group
97
based on the number of children. EXP2011
t
equals 1 if the year is after 2011.
Here the coefficient of interest is , which captures changes in the labor supply
of the treatment group after the 2011 EITC expansion. X
it
, includes individual
characteristics(age, education), incomes (unearned income except for spouse’s in-
come and spouse’s income), and others (residing city).
6

i
controls for individual
fixed effects such as personality or ability, andYear
t
capture common time effects
such as business cycles.
Regression results of the DID estimations with the treatment and control
groups are in Table 3.2, 3.3, and 3.4. Overall, I cannot find any significant impacts
of the expansion on the labor supply. Analysis with labor force participation in
Table 3.2 shows that married fathers with children, with one child or more than
two children, are not more likely to participate in the labor force compared to their
counterparts, married fathers without a child, or with one child. I can see similar
results when I use whether there are positive earnings as a dependent variable. In
Table 3.3, there are no changes in the probability to have positive earnings after
the expansion for three different pairs of treatment and control groups. Thus, at
the extensive margin, there are no impacts. This is also true at the intensive mar-
gin. From analysis with weekly working hours in Table 3.4, I find no significant
impacts.
For these DID estimates to be unbiased, there is an important assumption
to be satisfied, common trend assumption. This assumption means that the dif-
ferences in the labor supply between treatment and control groups are constant
overtime before the policy intervention. I check the assumption by using an event
6
Other family backgrounds (number of households, number of dependent children under age
18)areavailable. However,consideringthepossibilitythattheEITCaffectsthefertilitydecisions,
I do not include family composition variables.
98
study approach. For the test, I include interaction terms between the treatment
status and each year’s dummies, excluding the Year2009Treatment to use it
as a base.
From the test, I check if the coefficient of those interaction terms which are
related to the pretreatment are similar to each other. Here, this can be checked by
testing that the coefficient of Year2010Treatment term equals zero or not. As
shown in Tables 3.7, 3.6 and 3.5, I find that the assumption is satisfied except
when I use weekly working hours as dependent variables and second alternative
groups (parity 2 + versus 1).
3.6 Conclusion and Future Works
In this paper, I evaluate the impact of the 2011 EITC expansion on the labor
supply of married fathers in Korea. In 2011, the Korean government introduced
differential benefit schedules for families depending on the number of children. By
utilizing this quasi-experimental opportunity and Korea Labor and Income Panel
Study (KLIPS), I find that the expansion was not effective in encouraging married
fathers to work.
These results are comparable to the finding in Eissa and Hoynes (2006). In
their study, they evaluate the impact of 1993 EITC expansion on married fathers.
Theyfindmarriedmenslightlyincreasetheirlaborsupplyby0.2%. Itisimpressive
considering that the labor force participation rates of those fathers are almost 97%,
which is even higher than the rate in my sample, 94%. In addition, considering
that the phase-in rates are lower than those in Korea, the difference is surprising.
A possible explanation for the difference in effectiveness is elasticities of the
99
labor supply. The married fathers in the sample of Eissa and Hoynes (2006) are
on average 40 years old while the fathers are in my sample are around 50 years
old. Thus, it is likely that the elasticities of the fathers in my sample might be
lower. The low variances in the dependent variables support this possibility.
In addition, less flexible labor markets in Korea might have been accounted for
the results. Even though the fathers are willing to start working, the market does
not have capacities for them at least in the short-run. In this case, the government
has to think about demand-side labor market policies to increase the employment
of the low-income households.
Thismightchangeifwelookatdifferentrecipientgroupssuchassinglemothers
or married mothers, who might have higher labor supply elasticities. However, to
do so, it should be also explored if the differential benefit structure affects fertility
decisionsofmothers. Asmothers’laborsupplywillbealsoaffectedbytheirfertility
decisions, how the EITC affect the family size and labor supply is important. So
far, many previous studies on the EITC in the United States find no impact of
EITC on the decision, which might be the same in Korea as the benefits are not
large. The future research on EITC, fertility and labor supply of mothers will be
helpful in deciding on how to design the Child Tax Credit (CTC) and the EITC.
Another important topic related to the EITC expansions is how the other re-
lated outcome variables of interest are affected. Examples are household earnings,
EITC benefits, or welfare payments from the government.
100
3.7 Tables and Figures
Table 3.1
Summary Statistics
No Child One Child More than Two Children
Mean Standard Deviation Mean Standard Deviation Mean Standard Deviation
Individual Characteristics:
Age 60.13 12.46 43.94 9.64 53.32 14.05
High school draduate 0.32 0.47 0.38 0.49 0.34 0.47
University/College graduate 0.18 0.38 0.38 0.49 0.28 0.45
Master/Doctoral degree 0.02 0.15 0.04 0.20 0.04 0.19
Labor Force Particiation 0.68 0.47 0.93 0.25 0.78 0.42
Average weekly working hours 42.07 16.02 44.36 13.73 44.12 14.83
Positive Earnings 0.71 0.21 0.94 0.05 0.80 0.16
Family Characteristics:
Number of dependent children (under age 18) 1.15 0.40 1.11 0.33 1.97 0.56
Number of household members 2.93 1.01 3.42 0.86 3.37 1.08
Reside in Seoul 0.21 0.40 0.18 0.39 0.19 0.39
Household earned income $2,6110.05 8245.45 $40,090.18 9611.87 $31,630.59 8721.06
Unearned income $6,650.64 5545.10 $3,320.06 4313.91 $5,480.71 6228.39
Observations 7,296 2,290 11,690
Notes: Samples are married fathers with low household income (less than $25,000
in 2009) and who do not hold any additional ownership of houses for investment
purposes. Data from survey years 2009 to 2012 of the Korea Labor and Income Panel
Study (KLIPS). Standard errors (in parentheses) are robust for heteroskedasticity and
clustered at the child level. ***p< 0:01, **p< 0:05, *p< 0:1.
101
Table 3.2
Differences-in-Differences Estimates of 2011 EITC Expansion on Labor Force Par-
ticipation
Dependent Variable: Labor Force Participation
Model: Parity 1+ versus 0 Parity 1 versus 0 Parity 2+ versus 1
Expansion in 2011 Treatment -0.00807 -0.00707 -0.00729
(0.00915) (0.0123) (0.00972)
Individual Fixed Effect Yes Yes Yes
Year Effect Yes Yes Yes
Regeion Effect Yes Yes Yes
Number of Individuals 3,752 2,709 4,050
Observations 13,502 9,586 13,980
R-square 0.008 0.007 0.023
Notes: Samples are married fathers with low household income (less than $25,000
in 2009) and who do not hold any additional ownership of houses for investment
purposes. Data from survey years 2009 to 2012 of the Korea Labor and Income Panel
Study (KLIPS). Standard errors (in parentheses) are robust for heteroskedasticity and
clustered at the child level. ***p< 0:01, **p< 0:05, *p< 0:1.
102
Table 3.3
Differences-in-Differences Estimates of 2011 EITC Expansion on Positive Earnings
Dependent Variable: Positive Earnings
Model: Parity 1+ versus 0 Parity 1 versus 0 Parity 2+ versus 1
Expansion in 2011Treatment 0.00499 0.00170 -0.00605
(0.00799) (0.0110) (0.00846)
Individual Fixed Effect Yes Yes Yes
Year Effect Yes Yes Yes
Regeion Effect Yes Yes Yes
Number of Individuals 3,752 2,709 4,050
Observations 13,501 9,585 13,979
R-square 0.006 0.008 0.006
Notes: Samples are married fathers with low household income (less than $25,000
in 2009) and who do not hold any additional ownership of houses for investment
purposes. Data from survey years 2009 to 2012 of the Korea Labor and Income Panel
Study (KLIPS). Standard errors (in parentheses) are robust for heteroskedasticity and
clustered at the child level. ***p< 0:01, **p< 0:05, *p< 0:1.
103
Table 3.4
Differences-in-Differences Estimates of 2011 EITC Expansion on Weekly Hours of
Work
Dependent Variable: Weekly Hours of Work
Model: Parity 1+ versus 0 Parity 1 versus 0 Parity 2+ versus 1
Expansion in 2011 Treatment -0.0524 1.156 -0.651
(2.076) (2.590) (2.136)
Individual Fixed Effect Yes Yes Yes
Year Effect Yes Yes Yes
Regeion Effect Yes Yes Yes
Number of Individuals 685 465 730
Observations 685 819 1,240
R-square 0.023 0.026 0.024
Notes: Samples are married fathers with low household income (less than $25,000
in 2009) and who do not hold any additional ownership of houses for investment
purposes. Data from survey years 2009 to 2012 of the Korea Labor and Income Panel
Study (KLIPS). Standard errors (in parentheses) are robust for heteroskedasticity and
clustered at the child level. ***p< 0:01, **p< 0:05, *p< 0:1.
104
Table 3.5
Tests for the Parallel Trends Assumption: Labor Force Participation
Dependent Variable: Labor Force Participation
Model: Parity 1+ versus 0 Parity 1 versus 0 Parity 2+ versus 1
Year 2010 Treatment 0.000410 0.00574 -0.00985
(0.0106) (0.0153) (0.0133)
Year 2011 Treatment -0.00628 -0.000456 -0.0141
(0.0120) (0.0164) (0.0135)
Year 2012 Treatment -0.0108 -0.00826 -0.0107
(0.0139) (0.0181) (0.0139)
Individual Fixed Effect Yes Yes Yes
Year Effect Yes Yes Yes
Regeion Effect Yes Yes Yes
Number of Individuals 3,752 2,709 4,050
Observations 13,502 9,586 13,980
R-square 0.008 0.007 0.023
Notes: Samples are married fathers with low household income (less than $25,000
in 2009) and who do not hold any additional ownership of houses for investment
purposes. Data from survey years 2009 to 2012 of the Korea Labor and Income Panel
Study (KLIPS). Standard errors (in parentheses) are robust for heteroskedasticity and
clustered at the child level. ***p< 0:01, **p< 0:05, *p< 0:1.
105
Table 3.6
Tests for the Parallel Trends Assumption: Positive Earnings
Dependent Variable: Positive Earnings
Model: Parity 1+ versus 0 Parity 1 versus 0 Parity 2+ versus 1
Year 2010 Treatment -0.00814 -0.00903 0.00210
(0.00923) (0.0137) (0.0116)
Year 2011 Treatment -0.00156 -0.00753 0.000949
(0.0105) (0.0146) (0.0117)
Year 2012 Treatment 0.00144 0.00182 -0.0115
(0.0121) (0.0161) (0.0121)
Individual Fixed Effect Yes Yes Yes
Year Effect Yes Yes Yes
Regeion Effect Yes Yes Yes
Number of Individuals 3,752 2,709 4,050
Observations 13,501 9,585 13,979
R-square 0.006 0.009 0.007
Notes: Samples are married fathers with low household income (less than $25,000
in 2009) and who do not hold any additional ownership of houses for investment
purposes. Data from survey years 2009 to 2012 of the Korea Labor and Income Panel
Study (KLIPS). Standard errors (in parentheses) are robust for heteroskedasticity and
clustered at the child level. ***p< 0:01, **p< 0:05, *p< 0:1.
106
Table 3.7
Tests for the Parallel Trends Assumption: Weekly Hours of Work
Dependent Variable: Weekly Hours of Work
Model: Parity 1+ versus 0 Parity 1 versus 0 Parity 2+ versus 1
Year 2010 Treatment -1.720 -4.800 5.782**
(2.382) (3.076) (2.738)
Year 2011 Treatment -1.219 -2.329 3.398
(2.731) (3.487) (2.968)
Year 2012 Treatment -1.512 -1.812 2.194
(3.260) (3.949) (3.076)
Individual Fixed Effect Yes Yes Yes
Year Effect Yes Yes Yes
Regeion Effect Yes Yes Yes
Number of Individuals 685 465 730
Observations 1,182 819 1,240
R-square 0.024 0.033 0.033
Notes: Samples are married fathers with low household income (less than $25,000
in 2009) and who do not hold any additional ownership of houses for investment
purposes. Data from survey years 2009 to 2012 of the Korea Labor and Income Panel
Study (KLIPS). Standard errors (in parentheses) are robust for heteroskedasticity and
clustered at the child level. ***p< 0:01, **p< 0:05, *p< 0:1.
107
Figure 3.1
The EITC Schedule for Families with children in 2008 in Korea
Notes: This figure shows the EITC schedule for families with children in 2008.
The same schedule was applied regardless of the number of children. The phase-in
range was between 0 and $8,000 with a phase-in rate of 15%. The plateau range
was between $8,000 and $12,000, and the phase-out range was between $12,000
and $17,000 with a phase-out rate of 24%. The maximum tax credit was $800.
108
Figure 3.2
The EITC Schedules for Families with children in 2011
Notes: This figure shows the EITC schedules for families with children in 2011.
The different benefit schedules were applied depending on the different family
sizes. A family with more than three children faced 22.2% of a phase-in rate if
the family’s earned income was less than $9,000. They received $2,000 for the tax
credit if their income was between $9,000 and $12,000. Finally, they faced 15.4%
of a phase-out rate if their income was between $12,000 and $25,000.
109
Figure 3.3
Theoretical Prediction of the Impacts of the EITC on Labor Supply: Change in
the Budget Constraints
Notes: This figure shows how the adoption of the EITC changes the budget con-
straint from the straight line which is colored blue to the kinked one, which is
colored green.
110
Figure 3.4
Theoretical Prediction of the Impacts of the EITC on Labor Supply: Decisions of
Non-workers
Notes: This figure shows labor supply decisions of the non-workers when the EITC
is introduced. There is a clear incentives for non-workers to start working.
111
Figure 3.5
Theoretical Prediction of the Impacts of the EITC on Labor Supply: Decisions of
Workers in the Phase-in Range
Notes: This figure shows labor supply decisions of the recipients who are already
in the labor market. Particularly, recipients in the phase-in range, the decision
depends on the relative sizes of the substitution and income effects. Usually, the
income effect is larger than the substitution effect, and as a result they reduce
their labor supply as the figure displays.
112
Figure 3.6
Theoretical Prediction of the Impacts of the EITC on Labor Supply: Decisions of
Workers in the Plateau and Phase-out Ranges
Notes: This figure shows labor supply decisions of the recipients, who are already
in the labor market. Particularly, recipients in the plateau and phase-out ranges
decrease their working hours, given the leisure is a normal good.
113
Conclusion
A feature of recent welfare programs is that they emphasize welfare recipients’
work and responsibility. This addition comes in response to criticisms of the
previous means-tested cash transfer programs for their moral hazard problem,
disincentivizing work.
To mitigate the problem, the EITC has been a good alternative and now is
a large cash transfer program in the U.S. as well as in other countries. In this
dissertation, I analyze the EITC’s impact on the labor supply decisions and its
impact on their children.
In the first chapter, I explore how the EITC affects the labor supply of single
mothers and development outcomes of their children. From the analysis, I find
that the 1993 EITC expansion increases single mothers’ labor force participation
even though it does not result in a significant increase in household income. The
mothers’ participation increases their EITC benefits and decreases their govern-
ment welfare payments. However, it does not lead to a significant increase in their
earned income. A possible explanation is that, as single mothers with low educa-
tion levels have low attachment to the labor market, it would be hard for them
to find stable and well-paid jobs. When I look at the experiences of children at
home, I see similar changes. The children of single mothers interact less with their
114
mothers, but there are no significant changes in financial resources available to
them. These results have a policy implication, which is, for the EITC to be an
effective poverty reduction tool, it may need to be paired with other policy inter-
ventions such as child care. In that sense, the recent adoption of federal and state
government subsidies for child care services may offset or prevent these potential
negative impacts.
In the second chapter, I explore the impact of the EITC on children using a
different approach, an instrumental-variables (IV) approach. As an instrumental
variable (IV) for mothers’ work, I use changes in policy parameters of the EITC
both at the federal and state government levels, which are closely related to the
labor supply incentives. From the analysis, I first find that the changes in the pol-
icy parameters are strongly associated with single mothers’ labor supply such as
labor force participation, the number of jobs, and annual working hours. However,
married mothers’ labor supply is not affected by the changes. Using the predicted
labor supply of mothers, I estimate the impact of maternal work on child devel-
opment. What I find is that single mothers’ labor force participation reduces the
PIAT reading scores of their children, whereas the children of married mothers are
not affected.
In the third chapter, I study the EITC in Korea, which was introduced in 2009.
Since the adoption, there have been twelve payments and four EITC expansions.
I evaluate the 2011 EITC expansion, which allows for different benefit schedules
dependingonfamilysize. Fromtheanalysis, Ifindthatthelaborsupplyofmarried
fathersdoesnotincrease, whichispartlyattributabletotheirinelasticlaborsupply
and labor market conditions.
The findings in this dissertation are not restricted to the EITC. In particular,
115
the analyses of the EITC and child development have useful implications for other
welfare programs that are conditional on work. Considering that the labor itself
is an important part of life and means more than earning income, recent welfare
reformswhichencouragepeopletoworkbygivingfinancialincentivesaredesirable.
However, as the recipients are most likely to have low education levels and low
labor market attachments, their entry to the labor market might not be rewarded
much in the short run. This can result in unintentional results, one of which is
shown in this dissertation as a decrease in child development. Thus, research on
the program from diverse perspectives is necessary to improve the programs as
effective poverty reduction tools to achieve efficient income redistribution.
116
Bibliography
Adireksombat, K. (2010). The effects of the 1993 earned income tax credit ex-
pansion on the labor supply of unmarried women. Public Finance Review,
38(1):11–40.
Agostinelli, F. and Sorrenti, G. (2018). Money vs. time: family income, maternal
labor supply, and child development. University of Zurich, Department of
Economics, Working Paper, (273).
Andrews, I., Stock, J. H., and Sun, L. (2019). Weak instruments in instrumen-
tal variables regression: Theory and practice. Annual Review of Economics,
11:727–753.
Athreya, K., Reilly, D., and Simpson, N. (2010). Earned income tax credit recipi-
ents: Income, marginal tax rates, wealth, and credit constraints.
Baker, K. (2008). Do cash transfer programs improve infant health: Evidence from
the 1993 expansion of the earned income tax credit. manuscript, University
of Notre Dame.
Bastian, J. (2017). The rise of working mothers and the 1975 earned income tax
credit. Unpublished mimeo.
Bastian, J. and Lochner, L. (2020). The eitc and maternal time use: More time
working and less time with kids? Technical report, National Bureau of Eco-
nomic Research.
Bastian, J.andMichelmore, K.(2018). Thelong-termimpactoftheearnedincome
taxcreditonchildren’seducationandemploymentoutcomes. Journal of Labor
Economics, 36(4):1127–1163.
Baughman, R. and Dickert-Conlin, S. (2009). The earned income tax credit and
fertility. Journal of Population Economics, 22(3):537–563.
117
Baum II, C. L. (2003). Does early maternal employment harm child develop-
ment? an analysis of the potential benefits of leave taking. Journal of labor
Economics, 21(2):409–448.
Baydar, N. and Brooks-Gunn, J. (1991). Effects of maternal employment and
child-care arrangements on preschoolers’ cognitive and behavioral outcomes:
Evidence from the children of the national longitudinal survey of youth. De-
velopmental Psychology, 27(6):932.
Belsky, J. (1988). The “effects” of infant day care reconsidered. Early Childhood
Research Quarterly, 3(3):235–272.
Bergmann, B. R. (1996). Saving our children from poverty: What the United States
can learn from France. Russell Sage Foundation.
Bernal, R. (2008). The effect of maternal employment and child care on children’s
cognitive development. International Economic Review, 49(4):1173–1209.
Bernal, R. and Keane, M. P. (2011). Child care choices and children’s cogni-
tive achievement: The case of single mothers. Journal of Labor Economics,
29(3):459–512.
Blau, D. and Currie, J. (2006). Pre-school, day care, and after-school care: who’s
minding the kids? Handbook of the Economics of Education, 2:1163–1278.
Blau, D. M. (1999). The effect of income on child development. The Review of
Economics and Statistics, 81(2):261–276.
Blau, F. D. and Grossberg, A. J. (1992). Maternal labor supply and children’s
cognitive development. The Review of Economics and Statistics, pages 474–
481.
Blundell, R., Costa Dias, M., Meghir, C., and Shaw, J. (2016). Female labor
supply, human capital, and welfare reform. Econometrica, 84(5):1705–1753.
Brooks-Gunn, J., Han, W.-J., and Waldfogel, J. (2002). Maternal employment
and child cognitive outcomes in the first three years of life: The nichd study
of early child care. Child Development, 73(4):1052–1072.
Carneiro, P., Løken, K. V., and Salvanes, K. G. (2015). A flying start? mater-
nity leave benefits and long-run outcomes of children. Journal of Political
Economy, 123(2):365–412.
118
Carneiro, P. and Rodrigues, M. (2009). Evaluating the effect of maternal time
on child development using the generalized propensity score. Institute for the
Study of Labor, 12th IZA European Summer School in Labor Economics.
Chetty, R., Friedman, J. N., and Rockoff, J. E. (2011). New evidence on the
long-term impacts of tax credits. IRS Statistics of Income White Paper.
Cho, S.-J. (1998). Earned income tax credit and female labor supply. Quarterly
Journal of Labor Policy, 9(3):29–54.
Choi, H. S. (2013). Eitc expansions, ctc implementation, and their policy agenda.
Korea Institute for Health and Social Affairs Issue Focus, (0):1–8.
Chun, Y. J. (2009). The impact of earned income tax credit on the work incentive.
Monthly Review of Labor, pages 55–68.
Cunha, F. and Heckman, J. (2007). The technology of skill formation. American
Economic Review, 97(2):31–47.
Dahl, G. B. and Lochner, L. (2012). The impact of family income on child achieve-
ment: Evidence from the earned income tax credit. American Economic Re-
view, 102(5):1927–56.
De Chaisemartin, C. and d’Haultfoeuille, X. (2018). Fuzzy differences-in-
differences. The Review of Economic Studies, 85(2):999–1028.
de Chaisemartin, C., D’Haultfœuille, X., and Guyonvarch, Y. (2019). Fuzzy
differences-in-differences with stata. The Stata Journal, 19(2):435–458.
Deming, D. (2009). Early childhood intervention and life-cycle skill development:
Evidence from head start. American Economic Journal: Applied Economics,
1(3):111–34.
Desai, S., Chase-Lansdale, P. L., and Michael, R. T. (1989). Mother or market? ef-
fects of maternal employment on the intellectual ability of 4-year-old children.
Demography, 26(4):545–561.
Dickert, S., Houser, S., and Scholz, J. K. (1995). The earned income tax credit
and transfer programs: a study of labor market and program participation.
Tax Policy and the Economy, 9:1–50.
Dickert-Conlin, S.andHouser, S.(2002). Eitcandmarriage. National Tax Journal,
pages 25–40.
119
Duncan, G. J., Yeung, W. J., Brooks-Gunn, J., and Smith, J. R. (1998). How
much does childhood poverty affect the life chances of children? American
Sociological Review, pages 406–423.
Dustmann, C. and Schönberg, U. (2012). Expansions in maternity leave coverage
and children’s long-term outcomes. American Economic Journal: Applied
Economics, 4(3):190–224.
Eissa, N. and Hoynes, H. W. (1999). Good news for low-income families? tax-
transfer schemes and marriage. polycopié, University of California, Berkeley.
Eissa, N. and Hoynes, H. W. (2004). Taxes and the labor market participation of
married couples: the earned income tax credit. Journal of Public Economics,
88(9-10):1931–1958.
Eissa, N. and Hoynes, H. W. (2006). Behavioral responses to taxes: Lessons from
the eitc and labor supply. Tax Policy and the Economy, 20:73–110.
Eissa, N. and Liebman, J. B. (1996). Labor supply response to the earned income
tax credit. The Quarterly Journal of Economics, 111(2):605–637.
Ellwood, D. T. (2000). The impact of the earned income tax credit and social
policy reforms on work, marriage, and living arrangements. National Tax
Journal, pages 1063–1105.
Felfe, C. and Hsin, A. (2012). Maternal work conditions and child development.
Economics of Education Review, 31(6):1037–1057.
Fort, M., Ichino, A., and Zanella, G. (2016). Cognitive and non-cognitive costs of
daycare 0–2 for girls.
Golombok, S., Zadeh, S., Imrie, S., Smith, V., and Freeman, T. (2016). Single
mothers by choice: Mother–child relationships and children’s psychological
adjustment. Journal of Family Psychology, 30(4):409.
Han, W.-J., Waldfogel, J., and Brooks-Gunn, J. (2001). The effects of early ma-
ternal employment on later cognitive and behavioral outcomes. Journal of
Marriage and Family, 63(2):336–354.
Heckman, J. J. and Rubinstein, Y. (2001). The importance of noncognitive skills:
Lessonsfromthegedtestingprogram. American Economic Review,91(2):145–
149.
120
Heim, B. T. (2010). The impact of the earned income tax credit on the labor
supply of married couples: A structural estimation. Manuscript. School of
Public and Environmental Affairs, Indiana University, Bloomington, IN.
Herbst, C. M. (2011). The impact of the earned income tax credit on marriage
and divorce: Evidence from flow data. Population research and Policy review,
30(1):101–128.
Hoffman, S. D. and Seidman, L. S. (2003). Helping working families: The earned
income tax credit. WE Upjohn Institute.
Hotz, V. J., Mullin, C. H., Scholz, J. K., et al. (2001). The earned income tax
credit and labor market participation of families on welfare. Poverty Research
News, 5(3):13–15.
Hoynes, H., Miller, D., and Simon, D. (2015). Income, the earned income tax
credit, and infant health. American Economic Journal: Economic Policy,
7(1):172–211.
Hoynes, H. W. and Patel, A. J. (2018). Effective policy for reducing poverty
and inequality? the earned income tax credit and the distribution of income.
Journal of Human Resources, 53(4):859–890.
IRS (2020). Eitc fast facts. https://www.eitc.irs.gov/partner-toolkit/
basic-marketing-communication-materials/eitc-fast-facts/
eitc-fast-facts#:~:text=EITC%20is%20for%20workers%20whose,
jointly)%20with%20one%20qualifying%20child. Retrived on March 1st.
James-Burdumy, S. (2005). The effect of maternal labor force participation on
child development. Journal of Labor Economics, 23(1):177–211.
Keane, M. and Moffitt, R. (1998). A structural model of multiple welfare program
participation and labor supply. International Economic Review, pages 553–
589.
Ki, J. R., Kim, J. H., and Kim, J. H. (2015). The study of the effects of eitc on
hourly wage of recipients. Korean Journal of Policy Analysis and Evaluation,
25(3):298–312.
Kleven, H. (2019). The eitc and the extensive margin: A reappraisal. Technical
report, National Bureau of Economic Research.
Leibowitz, A. (1977). Parental inputs and children’s achievement. The Journal of
Human Resources, 12(2):242–251.
121
Levy, D. and Duncan, G. J. (2000). Using sibling samples to assess the effect of
childhood family income on completed schooling. jcpr working paper.
Lim, K. and Michelmore, K. (2018). The eitc and self-employment among married
mothers. Labour Economics, 55:98–115.
Ludwig, J. and Miller, D. L. (2007). Does head start improve children’s life
chances? evidence from a regression discontinuity design. The Quarterly
Journal of Economics, 122(1):159–208.
Manoli, D. and Turner, N. (2018). Cash-on-hand and college enrollment: Evi-
dence from population tax data and the earned income tax credit. American
Economic Journal: Economic Policy, 10(2):242–71.
Markowitz, S., Komro, K. A., Livingston, M. D., Lenhart, O., and Wagenaar,
A. C. (2017). Effects of state-level earned income tax credit laws in the us
on maternal health behaviors and infant health outcomes. Social Science &
Medicine, 194:67–75.
Meyer, B. D. (2002). Labor supply at the extensive and intensive margins: The
eitc, welfare, and hours worked. American Economic Review, 92(2):373–379.
Meyer, B.D.(2010). Theeffectsoftheearnedincometaxcreditandrecentreforms.
Tax Policy and the Economy, 24(1):153–180.
Meyer, B. D. and Rosenbaum, D. T. (2001). Welfare, the earned income tax credit,
and the labor supply of single mothers. The Quarterly Journal of Economics,
116(3):1063–1114.
Michelmore, K. (2013). The effect of income on educational attainment: Evidence
from state earned income tax credit expansions. Available at SSRN 2356444.
Neumark, D. and Wascher, W. (2000). Using the eitc to help poor families: New
evidence and a comparision with the minimum wage. Technical report, Na-
tional Bureau of Economic Research.
Neumark, D. and Wascher, W. (2011). Does a higher minimum wage enhance the
effectiveness of the earned income tax credit? ILR Review, 64(4):712–746.
Olea, J. L. M. and Pflueger, C. (2013). A robust test for weak instruments. Journal
of Business & Economic Statistics, 31(3):358–369.
Parcel, T. L. and Menaghan, E. G. (1994). Early parental work, family social capi-
tal, and early childhood outcomes. American Journal of Sociology, 99(4):972–
1009.
122
Park, S. H. and Kim, T. I. (2011). Analysis on the impact of the eitc on labor
supply. (0):1–23.
Paxson, C. and Waldfogel, J. (2003). Welfare reforms, family resources, and child
maltreatment. Journal of Policy Analysis and Management, 22(1):85–113.
Ruhm, C. J. (2004). Parental employment and child cognitive development. Jour-
nal of Human resources, 39(1):155–192.
Schildberg-Hoerisch, H. (2011). Does parental employment affect children’s edu-
cational attainment? Economics of Education Review, 30(6):1456–1467.
Scholz, J. K. (1994). The earned income tax credit: Participation, compliance,
and antipoverty effectiveness. National Tax Journal, pages 63–87.
Sherlock, R., Synnes, A., and Koehoorn, M. (2008). Working mothers and early
childhood outcomes: Lessons from the canadian national longitudinal study
on children and youth. Early Human Development, 84(4):237–242.
Song, H. J. and Chun, Y. J. (2011). Effect of earned income tax credit on labor
supply.
Vandell, D. L. and Ramanan, J. (1992). Effects of early and recent mater-
nal employment on children from low-income families. Child Development,
63(4):938–949.
Waldfogel, J., Han, W.-J., and Brooks-Gunn, J. (2002). The effects of early mater-
nal employment on child cognitive development. Demography, 39(2):369–392.
WB (2018). Mothers and families. https://www.dol.gov/agencies/wb/data/
mothers-and-families. Retrived on Feb 22nd.
Yang, T.-T. (2018). Family labor supply and the timing of cash transfers evidence
from the earned income tax credit. Journal of Human Resources, 53(2):445–
473.
Yeom, J.-Y. and Jun, H. W. (2014). Effect of earned income tax credit on labor
supply. Journal of the Korean Official Statistics, 19(2):73–98.
Zill, N. and Peterson, J. L. (1986). Behavior problems index. Washington, DC:
Child Trends.
123
Appendix A
Appendix to Chapter 1
In this appendix, I explain the alternative specification where I use an income
threshold to determine the treatment status with the Fuzzy DID method. The
idea behind this is that as single mothers in the phase-in range in 1994 have larger
work incentives due to the 1993 EITC expansion, their children have less time
with their mothers, which possibly affects child development. Children in plateau
or phase-out ranges do not have the same impacts as the expansion decreases the
labor supply incentives for their mothers.
For the treatment group, I use single mothers’ children on the phase-in range
in 1994 as their household earned incomes were less than $8,000 in 1993. The
control group is single mothers’ children in plateau or phase-out ranges in 1994 as
their household earned incomes were between $8,000 and $24,000 in 1993.
To determine the treatment status, I use expected household earned incomes
instead of actual incomes. As earned incomes depend on mothers’ labor supply
decisions, treatment based on it will be endogenous. I estimate the expected
household earned incomes based on mothers’ age, education, and ethnicity.
However, using the predicted incomes does have costs; it makes the DID design
fuzzy. As the predicted income is not the same as the actual earned income,
124
treatment status based on the predicted income might not be the same as the
actual treatment. For instance, if the actual income levels are less than $8,000
but the predicted incomes are $8,050, the treatment status is zero while they are
treated.
EmpiricalModel: FuzzyDID To deal with the possible fuzziness in the stan-
dard sharp DID approach, I use the fuzzy DID approach. The fuzzy DID approach
links an actual (observed) treatment with treatment status by using the following
IV strategy:
First Stage:
D
it
=
1
+
1
G
i
Expansion1993
t
+X
0
it

1
+Z
i
+
t
+u
it
(A.1)
Second Stage:
y
it
=
2
+
2
D
it
+X
0
it

1
+Z
i
+
t
+"
it
; (A.2)
where y
it
is child i’s academic score at time t. G
i
Expansion1993
t
is the treat-
ment status based on predicted incomes, and D
it
is observed treatment based on
actual incomes. G
i
equals one if predicted income is less than or equal to $8,000.
Expansion1993
t
equals one if the year is after 1993. D
it
equals one only when
the household income is less than or equal to $8,000, and the time is after 1993.
The control variables inX
it
include child characteristics (ages of children), mother
characteristics (ages of mothers), nonwage income, and area characteristics (SM-
SAs). Z
i
controls for child fixed effects such as children’s characteristics (sex, race,
125
and unobservable ability) and mothers’ characteristics (education level, race, and
marital status) and
t
controls for time effect. In this model, I fix the fuzziness in
GiExpansion1993
t
by utilizing the information on D
it
.
Results Table A1 shows the Fuzzy DID estimation results for the combined
PIAT Math and Reading scores. I analyze for children of all mothers, single
mothers, and married mothers. The first stages are strong for three samples and
have large Kleibergen-Papp F statistics.
1
For single mothers’ children in Column
2, I found that children of single mothers on the phase-in range have lower scores
compared to those on the plateau or phase-out ranges by -1.8 standard deviation.
It is larger than the DID specification based on the marital status in Section 3
(-0.1361). For children of married mothers, the result is not the same. Married
mothers’ children on the phase-in range increase by 0.47 standard deviation in
the scores.
2
One explanation is that as low-income married mothers are usually
secondaryearners, theirspousesworkmoreandhavehigherincomes, whilemarried
mothers stay at home.
For now, as I have not explored the related mechanisms fully such as changes
in mothers’ labor supply or household earned income, I should interpret the results
with caution. However, it seems apparent that the EITC has differential impacts
on children of single and married mothers. This is consistent with the results in
1
As Andrews et al. (2019) notes when there is only one instrument variable, an effective first-
stage F-statistics of Olea and Pflueger (2013) equals a robust first-stage F-statistics, Kleibergen-
Paap F-statistics. Thus, we should compare the Kleibergen-Paap F-statistics to the Stock-Yogo
critical values. The Kleibergen-Paap F-statistic for single mothers’ children is larger than the
critical value for 10% maximal IV size, and The Kleibergen-Paap F-statistic for all and married
mothers’ children are larger than the critical value for 15% maximal IV size.
2
The estimates are Wald-DID. To insist that the estimates are the Local Average Treatment
Effect (LATE), I need to show additional two conditions: common trends assumption and stable
treatment effect over time. For more details about the assumptions please refer to De Chaise-
martin and d’Haultfoeuille (2018) and de Chaisemartin et al. (2019).
126
the DID approach using mothers’ marital status for the treatment status, where
the EITC had negative impacts only on the children of single mothers.
127
Table A1
Fuzzy DID Estimates for Combined PIAT Math and Reading Scores
Dependent variables: PIAT Math +Reading
Samples: All mothers Single mothers Married mothers
(1) (2) (3)
Expansion in 1993 Phase-in Range (D) 0.2343 -1.8200*** 0.4701
(0.5071) (0.2850) (0.4652)
Cragg-Donald Wald F statistic 30.436 2.126 36.182
Kleibergen-Paap rk Wald F statistic 27.927 64.403 24.793
Child fixed effects Yes Yes Yes
Year dummies and region dummies Yes Yes Yes
Number of children 363 181 182
Observations 951 464 487
R-squared 0.0457 -0.8950 0.0658
Notes: Samples are children with expected household earnings below $24,000. Data
from survey years 1990 to 1996 of the matched data of NLSY79 and NLSY79 Child and
Young Adults. Expansion1993 equals one for 1994, and 1996. Phase-in Range equals
one if the children are in the household with the household earnings below $8,000, and
they are in the phase-in range in 1993. I use predicted value for Expansion1993
phase-in range by using exogenous treatment status variable (G) which is based on the
predicted earned income. I use nonwage income in terms of quadruple root instead of
log to include samples with zero amount. Standard errors (in parenthesis) are robust for
heteroskedasticity and clustered at the child level. ***p< 0:01, **p< 0:05, *p< 0:1.
128
Appendix B
Additional Tables to Chapter 1
Table B1
DID Regression Results of Children’s Cognitive Achievements
(Additional Controls)
Dependent variables: PIAT Math+Reading PIAT Math PIAT Reading
(1) (2) (3)
Expansion in 1993 Single -0.1317*** -0.1140** -0.1176**
(0.0481) (0.0544) (0.0575)
Child fixed effect Yes Yes Yes
Year and region dummies Yes Yes Yes
Controls Additional Additional Additional
Number of children 1,795 1,814 1,800
Observations 3,612 3,667 3,632
R-squared 0.0256 0.0203 0.0275
Notes: The sample is children of mothers with a high school education or less, and had
expected income less than $30,000 in 1992. Data from survey years 1990 to 1996 of the
matched data of NLSY79 and NLSY79 Child and Young Adults. Expansion1993 equals
onefor1994,and1996. Singlein1992equalsoneifmotherofchildissingle.Iusenonwage
income in terms of quadruple root instead of log to include samples with zero amounts.
Additional control variables are on household composition such as ages of the youngest
child in the household, number of family members (adults, children, children in certain
age groups, adults who worked), and a dummy variable for living with grandparents.
Robust standard errors in parentheses and clustered at the child level.***p <0.01, **
p<0.05,p<0.1
129
Table B2
Regression Results of Children’s Noncognitive Achievement (Additional Controls)
Dependent variables: Behavioral Problem Index Motor and Social Development
(1) (2)
Expansion in 1993 Single -0.0519 -0.7064**
(0.0588) (0.1764)
Child fixed effect Yes Yes
Year and region dummies Yes Yes
Controls Additional Additional
Number of children 1,995 778
Observations 4,061 1,040
R-squared 0.0303 0.1764
Notes: Sample includes children of mothers with a high school education or less, and
with expected household earnings below $30,000 in 1992. Data are from the children
of the NLSY79 Child and Young Adults linked to their mothers in the main NLSY79,
which ranges from 1990 to 1996. Expansion1993 equals one for 1994, and 1996. Single
equals one if mother of child is single in 1992. Additional control variables are on
household composition such as ages of the youngest child in the household, number of
family members (adults, children, children in certain age groups, adults who worked),
and a dummy variable for living with grandparents. I use nonwage income in terms of
quadruple root instead of log to include samples with zero amounts. Robust standard
errors in parentheses and clustered at the child level.***p<0.01, ** p<0.05,p<0.1
130
Table B3
DID Regression Results of Children’s Cognitive Achievements (By Age)
Dependent variables: PIAT Math+Reading
Samples: All 5-12 years Older than 12 years old
(1) (2)
Expansion in 1993 Single -0.1361*** -0.1429*** -0.0662
(0.0470) (0.0502) (0.1200)
Child fixed effect Yes Yes Yes
Year and region dummies Yes Yes Yes
Controls Yes Yes Yes
Number of children 1,829 1,353 477
Observations 3,753 2,981 773
R-squared 0.0201 0.023 0.0284
Notes: Sample includes children of mothers with a high school education or less, and
with expected household earnings below $30,000 in 1992. Data are from the children
of the NLSY79 Child and Young Adults linked to their mothers in the main NLSY79,
which ranges from 1990 to 1996. Expansion1993 equals one for 1994, and 1996. Single
equals one if mother of child is single in 1992. The control variables include child
characteristics (ages of children), mother characteristics (ages of mothers), nonwage
income, and region dummies. I use nonwage income in terms of quadruple root instead
of log to include samples with zero amounts. Robust standard errors in parentheses and
clustered at the child level.***p<0.01, ** p<0.05,p<0.1
131
Table B4
DID Regression Results of Children’s Cognitive Achievements (By Ethnicity)
Dependent variables: PIAT Math+Reading
Samples: All Non-White White
(2) (3)
Expansion in 1993 Single -0.1361*** -0.1356** 0.0306
(0.0470) (0.0550) (0.1028)
Child fixed effect Yes Yes Yes
Year and region dummies Yes Yes Yes
Controls Yes Yes Yes
Number of children 1,829 1,349 480
Observations 3,753 2,716 1,037
R-squared 0.0256 0.0280 0.0395
Notes: Sample includes children of mothers with a high school education or less, and
with expected household earnings below $30,000 in 1992. Data are from the children
of the NLSY79 Child and Young Adults linked to their mothers in the main NLSY79,
which ranges from 1990 to 1996. Expansion1993 equals one for 1994, and 1996. Single
equals one if mother of child is single in 1992. The control variables include child
characteristics (ages of children), mother characteristics (ages of mothers), nonwage
income, and region dummies. I use nonwage income in terms of quadruple root instead
of log to include samples with zero amounts. Robust standard errors in parentheses and
clustered at the child level.***p<0.01, ** p<0.05,p<0.1
132
Table B5
DID Regression Results of Children’s Non-cognitive Achievement (By Ethnicity)
Dependent variables: Motor Social Development
Samples: All Non-White White
(2) (3)
Expansion in 1993 Single -0.6820** -0.6898* -0.4828
(0.2780) (0.3928) (0.6053)
Child fixed effect Yes Yes Yes
Year and region dummies Yes Yes Yes
Controls Yes Yes Yes
Number of children 798 489 309
Observations 1,081 646 435
R-squared 0.0767 0.1129 0.1371
Notes: Sample includes children of mothers with a high school education or less, and
with expected household earnings below $30,000 in 1992. Data are from the children
of the NLSY79 Child and Young Adults linked to their mothers in the main NLSY79,
which ranges from 1990 to 1996. Expansion1993 equals one for 1994, and 1996. Single
equals one if mother of child is single in 1992. The control variables include child
characteristics (ages of children), mother characteristics (ages of mothers), nonwage
income, and region dummies. I use nonwage income in terms of quadruple root instead
of log to include samples with zero amounts. Robust standard errors in parentheses and
clustered at the child level.***p<0.01, ** p<0.05,p<0.1
133

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on family and labor economics

PDF

Essays on development and health economics: social media and education policy

PDF

Essays in labor economics: demographic determinants of labor supply

PDF

Measurement error in income and consumption data and economic mobility

PDF

Essays on labor and development economics

PDF

Essays on econometrics

PDF

Behavioral approaches to industrial organization

PDF

Essays in information economics and marketing

PDF

Essays in development economics

PDF

Essays on health insurance programs and policies

PDF

Three essays on human capital and family economics

PDF

Essays on nonparametric and finite-sample econometrics

PDF

Essays on development economics

PDF

Three essays in education and urban economics

PDF

Three essays on the evaluation of long-term care insurance policies

PDF

Essays on causal inference

PDF

Essays on health economics

PDF

Leveling the playing field: unbiased tests of the relative information content of book income and taxable income

PDF

Essays on price determinants in the Los Angeles housing market

PDF

Distributive equality and labor burdens

Asset Metadata

Creator Ko, Jeehyun (author)

Core Title The EITC, labor supply, and child development

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 04/19/2021

Defense Date 03/04/2021

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Child development,education,EITC,maternal labor supply,OAI-PMH Harvest

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Strauss, John (committee chair), Moon, Hyungsik Roger (committee member), Painter, Gary Dean (committee member), Ridder, Geert (committee member)

Creator Email kojeehyu@usc.edu,kojeehyun21@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-447341

Unique identifier UC11667812

Identifier etd-KoJeehyun-9499.pdf (filename),usctheses-c89-447341 (legacy record id)

Legacy Identifier etd-KoJeehyun-9499.pdf

Dmrecord 447341

Document Type Dissertation

Rights Ko, Jeehyun

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA