,P‘l“ ‘.

5’!"

”man

mg

I“?!

nmmmt

ﬂﬂiﬂl‘l"! “VIN"

I “

‘2‘!

rue (-r‘

N
u
h
u

-i

~hv1'0l dug“

 

at 3“.

 

 

(100?

   

LIBRARY
Michigan State
University

 

This is to certify that the
dissertation entitled

THE EFFECTS OF THE EARNED INCOME TAX CREDIT ON
LABOR MARKETS AND INDIVIDUAL BEHAVIOR

presented by

KAMPON ADIREKSOMBAT

has been accepted towards fulﬁllment
of the requirements for the

PhD. degree in Economics

 

ItawBWa/UL 16W

U0 Major Profésdor’ 5 Signature
Wil/Otﬁ

Date

MSU is an afﬁnnativecaction, equal-opportunity employer

-.. -.-.-.-.-._ It--.

‘;-J-....-.-I_-.-..--r-o-n-.->-J-._

..-.-.----g.--._.-

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K lProilAcc8.Pres/CIRCIDateDue indd

 

 

THE EFFECTS OF THE EARNED INCOME TAX CREDIT ON LABOR MARKETS
AND INDIVIDUAL BEHAVIOR
By

Kampon Adireksombat

A DISSERTATION
Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY

Department of Economics

2008

ABSTRACT

THE EFFECTS OF THE EARNED INCOME TAX CREDIT ON LABOR MARKETS
AND INDIVIDUAL BEHAVIOR

By

Kampon Adireksombat

With the ultimate goal to examine the incidence of the 1993 Earned Income Tax
Credit (EITC) expansion, my dissertation consists of three chapters that examine the
effects of the EITC on labor market and individual behavior.

In Chapter 1, I test whether differential increases‘in the maximum credits of the
EITC due to the 1993 expansion resulted in differential changes in labor supply. To avoid
the complex joint labor supply decisions of spouses, this study focuses on the labor
supply decisions of unmarried women. Using national survey data from the March
Current Population Survey (CPS) and differential increases in the maximum credits
available to unmarried women with two or more children compared to those with one
child and those with no children, I ﬁnd that the 1993 EITC expansion increased the
probability of labor force participation of the two or more children group by 3.6
percentage points relative to the no children group and 3.0 percentage points relative to
the one child group. Regarding the hours of work, I ﬁnd evidence that Ordinary Least
Squares (OLS) estimates may be biased towards zero, and therefore estimate the annual
hours of work by a Tobit model. Results suggest that following the EITC expansion,
unmarried women with two or more children increased their annual hours of work
compared to those with no children and those with one child by 130 and 91 hours,

respectively. However, when the sample is restricted to those who were already in the

labor force, I ﬁnd no evidence of statistically signiﬁcant changes in their annual hours of
work.

The EITC is more effective if its incidence is captured by the worker. Interest in
whether the EITC incidence is captured by the worker or the employer follows from the
EITC’s larger role in increasing low-income workers’ labor supply. Using cross-sectional
and cross—time variation in the federal and state EITCs with data from the CPS, in
Chapter 2, I examine the incidence of the 1993 EITC expansion on the wages Of
unmarried women. Accounting for the sample selection problem and the endogeneity of
the EITC, I ﬁnd no evidence of a statistically and economically signiﬁcant decrease in
wages received by unmarried women who face larger increases compared to those who
face smaller increases in their EITC.

Results in Chapters 1 and 2 rely on the assumption that the marriage decision is
exogenous to the value of the EITC. Therefore, in Chapter 3, I test this hypothesis. Using
a sample of cohabiting women with children from the 2001 panel of the Survey of
Income and Program Participation, I describe the distribution of the changes in the EITC
marriage incentives associated with the EITC expansion for married couples in 2001 and
examine whether the changes in the EITC marriage incentives are correlated with
marriage decisions. Accounting for individual ﬁxed effects and the endogeneity of the
EITC, I ﬁnd no relationship between changes in the EITC marriage incentives faced by
cohabiting women and their marriage decisions. I conclude that the marriage decision is

exogenous to the EITC.

To my family and the memory of Ngangsong Saekow

iv

ACKNOWLEDGEMENTS

I would like to thank my committee chair, Professor Stacy Dickert-Conlin. I am
very grateful to for her guidance, support, patience, and encouragement. I would also like
to give special thanks to Professor Charles Ballard. Without his special help, I would not
be able to focus on my ﬁeld of specialization, public economics. I would like to thank the
other members of the committee; Professor Leslie Papke and Professor Edmund Outslay
for providing me helpful comments and suggestions.

I would like to thank my parents, who did not have an opportunity to even ﬁnish
elementary school but strongly believe that education can give their kids a great
opportunity. Their love and sacriﬁce has been a solid base from which this child could
launch many pursuits. Pa and Ma, thank you. I need to thank my sister and brother, I e
Nueng and Tuang, who always believe in me and give me their support and
encouragement. I also need to thank Ae, my girlﬁ-iend, for her love, support, and
patience. You guys have always been there for me. I love you all.

I want to thank my friends from Suankularb Wittayalai and Chulalongkom
University. I absolutely cherish all of the years we have spent together, and your ability
to make me forget about all of the stress in my life, or at least for a short while. I also
want to thank my MSU friends, who give me a lot of comments and supports.

Finally, I would like to take this opportunity to thank people of Michigan and
Michigan State University for providing me the scholarship and fellowship. It is also my
pleasure to spend ﬁve years in the United States. Thank you for your generosity. It is

greatly appreciated.

TABLE OF CONTENTS

LIST OF TABLES .................................................................................. vii
LIST OF FIGURES ................................................................................. ix
CHAPTER 1
THE EFFECTS OF THE 1993 EARNED INCOME TAX CREDIT (EITC)
EXPANSION ON THE LABOR SUPPLY OF UNMARRIED WOMEN ................... 1
1.1 INTRODUCTION ...................................................................... 1
1.2 INSTITUTIONAL DETAILS AND THEORETICAL PREDICTION... .......3
1.3 LITERATURE REVIEW .............................................................. 6
1.4 IDENTIFICATION STRATEGY ................................................... 10
1.5 DATA AND EMPIRICAL APPROACH .......................................... 11
1.6 REGRESSION SPECIFICATION AND RESULTS ............................. 19
1.7 CONCLUSION ........................................................................ 33
CHAPTER 2
THE INCIDENCE OF THE EITC ............................................................... 34
2.1 INTRODUCTION ..................................................................... 34
2.2 THE EITC AND ECONOMIC INCIDENCE ..................................... 36
2.3 LITERATURE REVIEW ............................................................ 43
2.4 DATA AND EMPIRICAL METHOD ............................................. 48
2.5 RESULTS .............................................................................. 57
2.6 CONCLUSION ........................................................................ 62
CHAPTER 3
EVIDENCE OF THE EITC MARRIAGE-PENALTY-RELIEF ............................ 64
3.1 INTRODUCTION ..................................................................... 64
3.2 INSTITUTIONAL DETAILS ....................................................... 67
3.3 EXISTING LITERATURE .......................................................... 69
3.4 DATA AND DESCRIPTIVE EVIDENCE ........................................ 74
3.5 ARE EITC MARRIAGE INCENTIVES CORRELATED WITH MARRIAGE
DECISIONS? ........................................................................... 82
3.6 CONCLUSION ........................................................................ 89
APPENDIX .......................................................................................... 90
BIBLIOGRAPHY ................................................................................... 93

vi

Table 1.1

Table 1.2

Table 1.3

Table 1.4

Table 1.5

Table 1.6

Table 1.7

Table 1.8

Table 2.1

Table 2.2

Table 2.3

Table 2.4

Table 2.5

Table 2.6

Table 2.7

Table 2.8

Table 3.1

Table 3.2

Table 3.3

LIST OF TABLES

Summary Statistics ................................................................ 13

Labor Force Participation Rates of Unmarried Women Before and

After 1994 .......................................................................... 15
Annual Hours Worked by Unmarried Women Before and

After 1994 (Including Those with Zero Hour of Work) ..................... 17
Annual Hours Worked by Unmarried Women Before and

Aﬁer 1994 (Conditional on Hours of Work Exceeding Zero) ............... 18
Marginal Effects of Probit Model ............................................... 24
Marginal Effects of Probit Model by Levels of Education ................... 26
Results of OLS Estimation ....................................................... 29
Marginal Effects of Tobit Model ................................................ 31
Summary Statistics ................................................................ 49
Median Marginal Tax Rates of Unmarried Women Before and

After 1994 .......................................................................... 51
Labor force Participation Rates of Unmarried Women Before and

After 1994 .......................................................................... 52
Wages for Unmarried Women Before and After 1994 ....................... 53
Marginal Effects of Probit Model ............................................... 58
Test of Selection Bias ............................................................ 60
Results of Wage Equation ........................................................ 60
Results of Wage Equation (Including the Minimum Wage) ................ 62
Marriage Penalty Relief Provisions ............................................. 69
Descriptive Statistics .............................................................. 84

Linear Probability Model for All Women with
Children ............................................................................ 86

Vii

Table 3.4 Linear Probability Model for All Women with
Children by Education Level .................................................... 88

Table A] Federal Earned Income Tax Credit Parameters, 1990-2000 ................. 90

viii

Figure 1.1
Figure 2.1
Figure 2.2

Figure 3.1

Figure 3.2
Figure 3.3
Figure 3.4
Figure 3.5

Figure 3.6
Figure A.1

Figure A.2

Figure A.3

Figure A.4

LIST OF FIGURES

The EITC Maximum Credits during the 19905 ................................ 5
The EITC Maximum Credits during the 19903 .............................. 38
Labor Demand and Labor Supply Model ...................................... 42
Weighted Number of Cohabiting Women Facing an EITC

Marriage Subsidy/Penalty in 2001, 2002, and 2005 .......................... 76
Weighted Means of the EITC Marriage Subsidy .............................. 78
Weighted Means of the EITC Marriage Subsidy by Education ............ 79
Weighted Number of Women Facing an EITC Marriage Subsidy ......... 79
Weighted Means of the EITC Marriage Incentive by Race ................. 81

Weighted Means of the EITC Marriage Incentive by Race for
Women with Less than High School Education and Two or More

Children ............................................................................. 81
Labor Force Participation of Unmarried Women between 1991 and
2000 ................................................................................. 91
Labor Force Participation of Unmarried Women with Two or More
Children ............................................................................. 91
Labor Force Participation of Unmarn'ed Women with One Child ......... 92

Labor Force Participation of Unmarried Women with No Children ....... 92

ix

CHAPTER 1
THE EFFECTS OF THE 1993 EARNED INCOME TAX CREDIT EXPANSION

ON THE LABOR SUPPLY OF UNMARRIED WOMEN

1.1 INTRODUCTION

The Earned Income Tax Credit (EITC) is a refundable income tax credit targeted
at Iow- and middle- income working families in the United States. The tax credit is paid
as a lump sum with the annual tax return. Working as an earning subsidy, in theory, the
EITC expansion will encourage labor force participation but reduce hours worked by
EITC-eligible taxpayers who are already in the labor force. To avoid complicated joint
labor supply decisions, existing research mainly focuses on the labor supply of unmarried
women and suggests that EITC expansions increased the labor force participation, but
reduced hours worked by women who were already in the labor force. I

To examine the effect of the EITC on the intensive margin of labor supply,
previous studies use Ordinary Least Squares (OLS) to estimate the total annual hours of
work, which takes positive values and the value of zero for a nontrivial fraction. As a
result, OLS provides inconsistent estimators (Wooldridge 2002). This study uses a Tobit
model to estimate annual hours of work to correct the inconsistency problem. Moreover,
this study accounts for the major welfare reform in 1996, which is another important
government policy that signiﬁcantly encourages labor supply among low-income people.2

Failure to control for welfare reform may overestimate the effects of the EITC expansion.

 

' See l-lotz and Scholz (2003) for a summary of labor supply response to the EITC.
2 See O’Neill and Hill (2001), Schoeni and Blank (2000) Grogger(2003), Kaushal and Kaestner(2005).

To identify the effect of the EITC expansion in 1993 on labor supply response,
this study uses a differential increase in the maximum credits available to a two-or-more
child family relative to a one child family and a childless family. The Omnibus Budget
Reconciliation Act 1993 (OBRA-93) Signiﬁcantly increased the maximum credits
available to a family with two or more children relative to that with one child and that
with no children. In 2005 dollars the difference between the maximum credits available
to a family with two or more children and a family with one child rose from $105 in 1993
to $1,745 in 1996 when OBRA-93 became fully effective. Likewise, relative to a family
with no children, the difference rose from $2,042 in 1993 to $4,024 in 1996.

In this study, I test whether differential increases in the maximum values of the
EITC available to unmarried women with two or more children (the treatment group)
compared to those with no children (the comparison group (A)) and those with one child
(the comparison group (B)) resulted in differential changes in their labor supply.3

Using data from the March Current Population Survey (CPS) during the 19905,
results suggest that after the 1993 EITC expansion, the treatment group increased their
probability of labor force participation relative to the comparison groups (A) and (B) by
3.6 and 3.0 percentage points, respectively. However, between the comparison groups
(A) and (B), I do not ﬁnd any statistically Signiﬁcant change in their labor force
participation.4

For the effect on the intensive margin of labor supply, my ﬁndings suggest that

the previous work may underestimate the effect of the EITC on the total hours of work.

 

3 Meyer (2002) examines the effects of the EITC on the labor supply of a variety of demographic groups
and concludes that the EITC primarily affected unmarried women.

4 These ﬁndings are consistent with Hotz et al. (2005), who use Californian data to examine the EITC
effects on the labor market participation.

Using a Tobit model, my results Show that the treatment group increased their annual
hours worked relative to the comparison groups (A) and (B) by 130 and 91 hours,
respectively. However, when I restrict the sample to examine only the hours worked by
those who were already in the labor force, I do not ﬁnd any statistically signiﬁcant
change in annual hours worked by the treatment group compared with the comparison
groups.

Chapter 1 proceeds as follows: Section 1.2 discusses the structure of the EITC and
its effect on labor supply. Section 1.3 reviews the previous work on the labor supply
response to the EITC. Section 1.4 describes the identiﬁcation strategy. Section 1.5
describes the data and empirical approach used in this study. Section 1.6 describes the

regression speciﬁcation and provides results. And, section 1.7 concludes.

1.2 INSTITUTIONAL DETAILS AND THEORETICAL PREDICTION
A. The E] T C Structure and Its History

The credit equals a speciﬁed percentage of earned income up to a maximum
dollar amount over the “phase-in range.” Over a range of income termed the “ﬂat range,”
taxpayers receive the maximum credit. The credit then diminishes to zero over the
“phase-out range.” 5 The EITC is refundable and claimants are paid regardless of whether
the credit-qualiﬁed taxpayer has any federal income tax liability. The EITC payment is
typically made once a year as an adjustment to tax liabilities or refunds.6 For those who

have children and want to claim the EITC, their children need to pass an age,

 

5 Table A] in Appendix shows the EITC parameters between 1990 and 2000.

6 An “advance payment” option was added in 197 8 so that workers would be able, if they so chose, to
receive the credit incrementally throughout the year. Only 1.1 percent of EITC recipients with children
used the advance payment option in 1998 (Hotz and Scholz 2003).

relationship, and residence tests to qualify. For example, the age test in 2000 requires the
qualifying child to be under 19 years old or under 24 years old if she/he is a full-time
student, or any age if she/he is completely disabled. The relationship test requires the
claimant to be the parent or the grandparent of the qualifying child.

The EITC has provided tax reductions and earning subsidies for low-and middle-
income working families since 1975. The EITC payments were eroded by inﬂation until
the Tax Reform Act 1986 (TRA-86) increased the maximum credit in 1987 to have a real
value equal to that of the credit in 1975, and indexed the EITC value for inﬂation. The
Omnibus Budget Reconciliation Act of 1990 (OBRA-90) introduced differential credit
rates and maximum credits available to a family with one and a family with two or more
children. However, it was OBRA-93 that substantially increased the maximum credit
available for a family with two or more children, relative to that with no children and that
with one child. OBRA-93 also expanded the beginning and ending incomes of the
“phase-out range.” Figure 1.1 presents the maximum credits available to a family with
two or more children, a family with no children, and a family with one child in 2005
dollars for the period from 1991 to 2000. The maximum credit available to a family with
two or more children increased substantially between 1994 and 1996, becoming fairly
constant afterward, as the reform was ﬁilly phased in (from $3,230 to $4,400). For a
family with one child and a family with no children, the maximum credits increased after
OBRA-93 was implemented, and become constant after 1994. The difference between
maximum credits available to a family with two or more children and a family with no

children rose from $2,042 in 1993 to $3,233 in 1996. Relative to the one child group, the

difference rose from $104 to $1,404. In addition, OBRA—93 introduced the EITC to

childless taxpayers aged between 25 and 65 years.

«I -_ A, ,miLL A.;L bmm

Figure 1.1 The EITC Maximum Credits during the 19905
S (in 2005 dollars)

4500 *
4000 W
3500 T
3000
2500 ‘
2000 “
1500 "
1000
500 *

 

 

 

1991 I992 I993 1994 1995 1996 1997 1998 1999 2000

 

Ybar

 

EBB} Children E One C1616- TWo ChildreEI

 

B. Theoretical Predictions

From the static labor-leisure model, the EITC expansion affects the intensive and
the extensive margins of the labor supply of unmarried women.

For a non-worker who was out of the labor force before the expansion, the static
labor-leisure model predicts that the EITC expansion will expand her budget set when
she enters the labor force. With no earned income before the expansion, there will be
only a positive substitution effect but no income effect due to an increase in the effective
wage (marginal value of working). As a result, some will choose to participate in the
labor force.

For a worker who was already in the labor force, the effect of the expansion on

her hours of work is ambiguous, depending on the range of EITC in which her income

falls before and after the expansion. If her income falls in the “phase-in range,” in theory,
there will be a positive substitution effect and a negative income effect, assuming that
leisure is a “normal good.” Thus, the net effect is ambiguous. If her income falls in the
“ﬂat range,” there is only a negative income effect; consequently, the expansion leads to
a decrease in hours of work. If her income falls in the “phase-out range,” a diminishing
credit implies a lower effective wage relative to the absence of the EITC. This negative
substitution effect results in a reduction in hours of work, as does the negative income
effect. Finally, if her income was beyond the credit region, she may decide to reduce her
hours of work to be eligible for the credit.

With a substantial increase in the credit available to a family with two or more
children, in theory, there will be a relatively larger increase in the labor force
participation of the treatment group compared to the comparison groups. For the effect on
hours worked, if their incomes fall in the “ﬂat range” or the “phase-out range,” the
treatment group will reduce their hours of work more than the comparison groups. If their
incomes fall in the “phase-in range,” the effect is ambiguous due to a positive substitution

effect and a negative income effect.

1.3 LITERATURE REVIEW

Due to a substantial increase in the labor force participation of unmarried women
during the 19903 and to avoid the complex joint labor supply decisions of husband and
wife, most of the EITC studies focus on the labor supply decisions of unmarried women
(with the exception of Dickert, Houser, and Scholz 1995; Eissa and Hoynes 2004; Heim

2005). In general, these studies ﬁnd a positive effect of EITC expansion on the labor

force participation of unmarried women. On the intensive margin, focusing on those who
were already in the labor force, they do not ﬁnd statistically signiﬁcant effects on hours
worked. Categorized by econometric methods, there are three strands of the previous
work.

The structural model “explicitly parameterizes the preferences and constraints
facing individuals and then exploits the theory of optimal decision-making to characterize
the likelihood function that is used to reconcile observed labor supply and program
participation behaviors” (Hotz and Scholz 2003, p.181). However, the structural model is
more complicated to estimate relative to other methods because it requires greater
knowledge of the structure of individual and household preferences and their choice
processes. The examples Of studies using the structural model are Keane (1995) and
Keane and Mofﬁtt (1998). They examine the EITC effect on the labor supply of single
mothers. With the fourth wave of the 1984 Survey of Income and Program Participation
(SIPP), they ﬁnd that the EITC expansions between 1984 and 1996 led to an increase in
the labor force participation of single mothers by 10.7 percentage points, from a base of
64.7 percent. In addition, they ﬁnd that the expansions led to an increase in weekly hours
of work from 24.1 to 26.5 hours.

The second method is the quasi-structural model. To identify the effect on the
EITC, this method employs variation in effective wages or effective tax rates. Dickert et
al. (1995) use monthly data from the 1990 SIPP to examine the effect of the EITC on the
labor supply of single parents and couples. Results from their simulation model Show that
the EITC expansion in 1993 increased the labor force participation of single-parent

families by 3.3 percentage points. For the intensive margin, with the assumption that new

entrants work for 20 hours/week and 20 weeks/year, new entrants would increase hours
of work by 72.8 million hours, but those who were already in the labor force would
reduce hours of work by 26.4 millions hours. Using data from the 1984-1996 March and
Outgoing Rotation Group (ORG) CPS, Meyer and Rosenbaum (2001) ﬁnd that the EITC
and other tax changes increased annual employment of single mothers by 7.2 percentage
points, relative to single women with no children.

The last method is the reduced-form model, which is the method I use in this
study. The objective of this approach is to determine the overall effects of the policy
changes on a particular behavior. The reduced-form model is commonly referred to as the
natural experiment or the Difference-in-Difference (DID) approach. The advantage of
this approach is simplicity and transparency in the assumptions that allow the
identiﬁcation of key parameters. Unlike a true experiment, in which treatment and
comparison groups are randomly chosen, the treatment (affected) and comparison (not
affected) groups in natural experiments arise from the particular policy change.
Therefore, to control systematic differences between the treatment and comparison
groups, two periods of data (before and after the policy change) are needed. Other
exogenous variables, which are included in the regression equation, control for the fact
that the populations sampled may differ systematically over the two periods (Wooldridge
2006).

For example, Eissa and Liebman (1996) employ a DID approach with a treatment
group of single women with children and a control group of those with no children to
examine the effect Of the 1986 EITC expansion on the labor supply of single women aged

between 16 and 44 years. Using the March CPS data from 1985 to 1987 and from 1989 to

1991, they ﬁnd that the expansion resulted in a 2.8 percentage point increase (from a base
of 74.2 percent) in labor force participation of the treatment group compared with that of
the control group. However, they do not ﬁnd any statistically signiﬁcant change in armual
hours worked by those single women who were already in the labor force.

Using a longer time frame, Darragh (2002) replicates Eissa and Liebman’s 1996
study by using the March CPS data between 1982 and 1996 to focus on single women
aged between 25 and 50 years. He ﬁnds that the 1986 EITC expansion led to an increase
in labor force participation of single mothers by 1.7 percentage points relative to single
women without children, while the 1993 expansion led to a larger increase, ranging
between 3.5 and 4.9 percentage points.

To avoid bias due to changes in the composition of treatment and comparison
groups in repeated cross-sectional studies, Hotz et al. (2005) use longitudinal data
between 1991 and 2000 from California’s Medi-Cal Eligibility Data System (MEDS) and
the California Employment Development Department (EDD) Base Wage Files, as well as
data from federal tax returns, to examine the effect of the EITC on labor market
participation of single-parent families on welfare.7 Taking advantage of the longitudinal
data, their empirical approach controls for covariates and household-speciﬁc ﬁxed
effects. They ﬁnd that the 1993 EITC expansion resulted in an increase in employment
by as much as 3.4 percentage points for families with two or more children relative to

families with one child.

 

7 MEDS provides information on welfare program participation and demographic characteristics. EDD
contains employer-reported taxable wage payments, labor force participation, employment, and earning.
Finally, federal tax returns data provide tax ﬁling and EITC claiming information.

1.4 IDENTIFICATION STRATEGY

To identify the effect of the 1993 EITC expansion on the intensive and extensive
margins of the labor supply of unmarried women, as discussed earlier, I use a differential
increase in maximum credits available to those with two or more children (the treatment
group) relative to those with no children (the comparison group (A)) and those with one
child (the comparison group (B)). Using the DID approachs, the estimation strategy
compares changes in the labor supply of the treatment and comparison groups before and
after 1994. The comparison groups are also affected by the EITC expansion, and in
theory, they are expected to move in the same direction as the treatment group.
Therefore, the effect of the EITC on the labor supply in this study might be
underestimated. To mitigate this potential problem, I include a variety of treatment and
comparison groups.

By selecting a sample based on marital status and using the number of children as
a source of identiﬁcation, I need to assume that marriage and fertility decisions are
exogenous to the EITC expansion. Dickert-Conlin and Houser (1998) describe that the
tax and transfer systems are not marriage neutral. They calculate the changes of tax
liability and transfer beneﬁts that would occur for a sample of unmarried women with
children, if they were married and living with a spouse. They ﬁnd that poor single women
with children face a loss of transfer beneﬁts if they marry, but the tax system (including
the EITC) subsidizes marriage, and thus mitigates the loss in transfer beneﬁts. On the

other hand, near-poor single women with children not only faced a decline in transfer

 

8 Meyer and Rosenbaum (2000) suggest that after welfare reform, several welfare features, such as
sanctions, diversion, time limits, and welfare to work program, are complicated to characterize when one
uses the quasi-structural model. Thus, I choose to use the reduced-form method because the sample
period in this study includes the years after welfare reform.

10

beneﬁts but also an increase in tax liability if they marry. The empirical evidence on the
effect of the EITC and tax system on these decisions is mixed. Hoynes (1997) concludes
that marriage decisions are not sensitive to ﬁnancial incentives. Ellwood (2000) ﬁnds
little or no effect of the EITC and welfare on marriage. The empirical study of the EITC
effect on the fertility decision is very limited. Using state-level data on birth rates,
Baughman and Dickert-Conlin (forthcoming) ﬁnd that the effect of the EITC expansions
during the 19905 on fertility is signiﬁcant only for higher-order births among white

women. However, the effect is extremely small.

1.5 DATA AND EMPIRICAL APPROACH

I use data from the March CPS Annual Demographic File The CPS is the monthly
survey of unemployment and labor force participation of about 50,000 households,
including labor market and income information for the previous year. 9 The sample period
Of this study corresponds to 1991 to 1993 and 1995 to 2000 tax years. OBRA-93 became
effective in January 1994, thus I drop the 1995 CPS data to allow one year for those
women to adjust their labor supply response.

The sample includes unmarried women, who were aged 25 to 55 years, were not

self-employed, and ﬁled either head-of-household or single tax return, depending on

 

9 I choose the March CPS over the Merged Outgoing Rotation Group (MORG) ﬁles because the MORG
ﬁles do not contain data of the number of children between 1994 and 1998. Another potential dataset is
the SIPP. However, I do not use data from the SIPP because in 1996 the SIPP was redesigned to over-
sample of households from areas with high poverty concentrations. This will lead to a dramatic change in
the distribution of labor supply and wages. In addition, before 1996, SIPP covered less than 20,000
households.

11

whether or not she has dependents.10 I exclude those aged younger than 25 years because
of the concern that many of them were still in school, and because childless women must
be between the ages of 25 and 65 to be eligible for the credit. I exclude those aged over
55, because they are less likely to participate in the labor force. The sample size, after
pooling all nine years, is 82,064 observations.

In addition to the number of children, following ﬁndings from EITC eligibility
literature that women with lower education are more likely to eligible for the EITC, I also
categorize the sample into four groups by their levels of education: Less than High
School, High School, Some College, and College (see Dahl and Lochner 2005; Baughman
and Dickert-Conlin forthcoming). Therefore, I expect that the treatment group with low
education will increase their labor supply compared to the comparison groups.

Table 1.1 presents the demographic characteristics, intensive and extensive
margins of labor supply, and incomes of women in the sample. It is worth noticing that
the two-or-more child group, on average, tend to (a) attain lower education than the one
child and the no children groups (12.67, 13.24, and 13.94 years of education,
respectively), (b) be younger (34.92 versus 37.00 and 39.15), (c) be less likely to
participate in the labor force (0.74 versus 0.83 and 0.85 for the probability of
participating in the labor force), (d) work in smaller numbers of annual hours and weeks,
(e) have higher incomes from other sources, and (f) conditional on working, have less

earned income.

 

10 Those who claim of head of household ﬁlling status are those who are unmarried and pay more than 50
percent of the costs of keeping up a home for themselves and their dependent(s) or other qualifying
individuals (Form W-4, Internal Revenue Service)

12

NE ”SN 8% 3% mg
2.: a: 3% m2: Sm

”2mm 2:; 3.5.. $3” 3%...
8.32 $.§_ $.32 3.85 8.32

E. E. E. E as
a: 2 E 5 5
Na Na 3 E E.
3 EM a: 2.: a:
so 2; 2; E E
as a: as N2 32

BB 23 saw a:

03:8 ”sow .3: as m3 E.
28230 0.58 .5 oar—r

”$— Nﬂz 3.8 3.8 $2 3.: NE— aém 3.3 82 3.5

a: 3.: EM 3: 2.: as a: E a: a; 5:
5.53:2. 533
Ea E: E3 33 3:... ﬁg 2...; Es 3.32 a? Ba
as: as: as: is 32 2% BE 35 is as: s:
is, s 52. .255
5. E E as E i 5. g E E. 33
N2 3 as i A: i a: as i B a:
size: a:
E E E B E E 2.... § § 3 SE
was an EM 2: 2.: EM 3: as is Em a:
02
§ 3° E E E as so 2; 2: N: as
a: a: as so a: :2 a: as i a: a:
523
BE sew Ea? BB saw BEE
age can E: game .2 use as” a: as a: .2
:5 25 5:5 02

Beam Beam 2 2E

13

8%:

”EN
~3—

o;
mod

9%
”no
go
Go

=<

23 £3 a} $3
8.3 8.: $2 2.:
8.3 3: mm: 8.0
3.2 :6 MS. 86
5.0 2% 8w :6
as mad as 3%
3o :6 NE 93
:6 Rd 23 3o
NE So 3o :6
age sea sea as
09:8 05% :2: scams;
5:250 82: s a:

:3“

3.3
2.2

3.2
3m
88
one
98
8o

owo=oo 28m

$3

3.8
3.8
8.»
a?
$6
m3
:3
Rd

83 Exm

2.2
N2:

”2.
gm
2%
so
mg
m8

m2:

Ks

m3
m?
”to
m3
5%
mg

ﬁn:

3.8
3:
3.”
N:
5%
N8
Q3
Ed

35 .023 .2532
E: .253 __<

220 25

”3.8 RS:

3%
3%

3.:
3:
ad
“:5
cod
cod

omo=ou 088

S. a 2.8 2% _
$2 3.8 o3
as 2% 3m
23 SN 2a
NE NE 28
N3 m3 3o
86 cod cod
85 86 23

Bag BEE E: B 55 Be? a was:
3:8 28 a as am case see Ba 85 Bscoso 88 3&8 pa BEE mag. Baa: 25 was? Swansea see 3.3

saga 05 5 Ba was» a BE: Ema gﬁﬁo Bag: 05 e 52% BEBE .Eo a8» 3.8 Ema. as?» causes. a Baas 2:
$8.82 Ea maioa as mac» 3 an 633 $8.32 Ea 32-32 Eu 532 ”28

«8.2 mm?

08.2

QR
8.2

2.0
gm
2%
mg
cod
8.0

38 Ea saw as

sagas 32% Beam : Be

a: a: a:
58:5 02

=<

82:33

so .3

so:

Amoco: 3.8.: Beam
>5 .3

=82
AmooocsaegEas
>5 .3

as:

32252

>5 .3

:82

56:5 .8385

14

Table 1.2 Labor Force Participation Rates of Unmarried Women Before and After 1994

(l) (2) (3) (4)
Panel 1: All Before1994 Aﬁer1994 Difference (2)-(l) DID
Treatment group TwoPlus 0.647 (.007) 0.781 (.005) 0.134" (.009)
Comparison group (A) NoChildren 0.863 (.003) 0.852 (.002) -0.010*"‘ (.003) 0.144" (.010)
Comparison group (B) OneChild 0.805 (.006) 0.840 (.004) 0.035" (.007) 0.099” (.011)
Panel [1: Less than High School
Treatment group TwoPlus 0.390 (.015) 0.578 (.012) 0.189" (.020)
Comparison group (A) NoChildren 0.560 (.012) 0.557 (.009) -0.002 (.014) 0.191" (.024)
Comparison group (B) OneChild 0.545 (.019) 0.615 (.014) 0.070" (.024) 0.119” (.030)
Comparison group (C) TwoPlus w/ College 0.870 (.013) 0.900 (.008) 0030* (.016) 0.156" (.025)
Panel III: High School
Treatment group TwoPlus 0.662 (.012) 0.795 (.008) 0.134" (.014)
Comparison group (A) NoChildren 0.840 (.006) 0.817 (.004) -0.012“ (.007) 0.155” (.016)
Comparison group (B) OneChild 0.820 (.009) 0.840 (.007) 0.020“ (.012) 0.113" (.019)
Comparison group (C) TwoPlus w/ College 0.870 (.013) 0.900 (.008) 0.030“ (.016) 0.104“ (.021)
Panel IV: Some College
Treatment group TwoPlus 0.774 (.014) 0.853 (.009) 0.080" (.017)
Comparison group (A) NoChildren 0.900 (.005) 0.888 (.004) -0.012" (.007) 0.091" (.018)
Comparison group (B) OneChild 0.870 (.011) 0.880 (.007) 0.010 (.013) 0.070" (.021)
Comparison group (C) TwoPlus w/ College 0.870 (.013) 0.900 (.008) 0.030‘ (.012) 0.049" (.023)

The numbers showed are the estimates of mean. The numbers in parentheses are standard errors
” Statistically signiﬁcant at 5% level

‘ Statistically signiﬁcant at 10% level
Data: March CPS 1992-1994 and 1996-2001 which are tax years of 1991-1993 and 1995-2000,
The sample is unmarried women aged 25-55 years old

Means are weighted with CPS March supplement weights

Table 1.2 presents the labor force participation rates of unmarried women before

and after 1994, as well as the changes categorized by levels of education and numbers of

children.11 '2 Columns (1) and (2) show the average participation rates before and after

1994, and Column (3) shows the difference between them. Negative values mean that

they decreased their participation rate. In Column (4), the DID estimates compare

 

" Figure A.l to A.4 in the Appendix show the labor force participation of unmarried women between 1991
and 2000

‘2 For the deﬁnition of labor force participation, 1 follow Eissa and Liebman (1996) who deﬁne labor force
participation as working a positive number of hours during the year. Meyer and Rosenbaum (2001) also
support this labor force participation deﬁnition. They suggest that “if our goal is to provide a sharp test of
theoretical predictions, whether a woman worked last year is a better outcome measure."

15

changes of the labor force participation before and after 1994 between the treatment
group and the comparison group.

Panel 1 presents the participation rate categorized by the number of children and
levels of education. Column (3) shows that, after 1994, the treatment group signiﬁcantly
increased their participation rate (13.4 percentage points) compared with the comparison
groups (-1.0 and 3.5 percentage points). As a result, the DID estimates in Column (4)
suggest that the treatment group increased their labor force participation rate relative to
the comparison groups (A) and (B) by 14.4 and 9.9 percentage points.

As discussed earlier, the EITC eligibility literature suggests that women with
lower education tend to be more likely to be eligible for the credit. Therefore, in Panels
11, III, and IV, I focus on those who do not have a college degree as the treatment group
and those with a college degree as the comparison group. 13 The DID estimates from
Panels II, III, and IV suggest that the treatment group increased their labor force
participation relative to the comparison groups. For example, from Panel II among high
school dropouts, the participation rate of the treatment group increased by 19.1
percentage points relative to the comparison group (A), 11.9 relative to the comparison
group (B), and 15.6 relative to the comparison group (C).

The EITC expansion had the potential to not only affect the participation margin
but also affect the intensive margin of the labor supply of unmarried women. Table 1.3
presents annual hours worked by unmarried women (including those with zero hour of
work). In all panels, Column (3) shows that the treatment group and the comparison

group (B) statistically increased their annual hours worked after 1994 (with the exception

 

‘3 In these panels I also include the comparison group (C), those with two or more children who hold a
college degree, to compare the treatment group with more educated women who have the same number
of children.

16

of the comparison group (B) in Panel IV). An increase in annual hours worked by the

treatment group ranges from 185 in the Some College group to 319 hours in the Less than

High School group. Moreover, the DID estimates in Column (4) suggest that the

treatment group significantly increased their hours worked compared with the

comparison groups. For example, in panel II, the treatment group increased their hours

worked by 319 compared with the comparison group (A), 168 compared with the

comparison group (B), and 236 compared with the comparison group (C).

Table 1.3 Annual Hours Worked by Unmarried Women Before and After 1994 (Including those with zero hour of work)

Panel 1: All

Treatment group TwoPlus
Comparison group (A) NoChildren
Comparison group (B) OneChild

Panel II: Less than High School

Treatment group TwoPlus
Comparison group (A) NoChildren
Comparison group (B) OneChild
Comparison group (C) TwoPlus w/ College

Panel III: High School

Treatment group TwoPlus
Comparison group (A) NoChildren
Comparison group (B) OneChild
Comparison group (C) TwoPlus w/ College

Panel IV: Some College

Treatment group TwoPlus
Comparison group (A) NoChildren
Comparison group (B) OneChild
Comparison group (C) TwoPlus w/ College

(1)
Beforel994

1074.350 (15.543)
1673.936 (7.364)
1436.422 (14.270)

516.313 (24.416)
950.750 (24.773)
312.553 (35.373)
1637.363 (33.962)

1035.679 (24.704)
1533.433 (13.533)
1456.323 (22.630)
1637.363 (33.962)

1313.299 (33.493)
1734.773 (15.274)
1593.677 (27.466)
1637.363 (33.962)

(2)
Aﬁerl994

1339.433 (10.390)
1637.261 (5.359)
1562.500 (10.250)

335.237 (22.641)
950.633 (17.332)
963.233 (23.531)
1720.730 (21.200)

1336.759 (17.563)
1573.339 (10.195)
1556.664 (17.102)
1720.730 (21.200)

1503.514 (22.329
1736.573 (11.293
1629.533 (19.994
1720.730 (21.200

vvvv

The numbers showed are the estimates of mean. The numbers in parentheses are standard errors

" Statistically significant at 5% level
’ Statistically signiﬁcant at l0% level

Data: March CPS 1992-1994 and 1996-2001 which are tax years of 1991-I993 and 1995-2000,

The sample is unmarried women aged 25-55 years old

Means are weighted with CPS March supplement weights

17

(3)

Difference (2)-(l)

264.583"
8.274
126.077"

318.969"
-0.066
150.684"
82.862"

251.079"
-l4.598
9984]"
82.862"

l85.215"
1.800
35.91 I

82.862"

(13.933)
(9.103)
(17.570)

(33.293)
(30.557)
(45.445)
(40.036)

(30.314)
(16.643)
(23.405)
(40.036)

(40.254)
(13.996)
(33.973)
(40.036)

(4)
DlD

256.307"
B8505"

319.036”
l68.285“
236.107“

265.678”
lSl.238”
l68.2l7"

183.4l4“
149.303"
102.352‘

(21.055)
(25.366)

(45.194)
(56.339)
(52.074)

(34.727)
(41.542)
(50.213)

(44.511)
(52.764)
(56.774)

Table 1.4 Annual Hours Worked by Unmarried Women Before and After 1994 (Conditional on Hours of Work Exceeding Zero)

(1) (2) (3) (4)

Beforel994 Aﬁerl994 Difference (2)-(1) DID
Panel I:All
Treatment group TwoPlus 1659.791 (14.310) 1713.995 (9.226) 54.204" (17.026)
Comparison group (A) NoChidren 1946.358 (5.683) 1979.985 (4.052) 33.627" (6.980) 20.577 (18.401)
Comparison group (B) OneChild 1783.496 (11.689) 1859.418 (8.088) 75.922” (14.215) -21.718 (22.180)
Panel II: less than High School
Treatment group TwoPlus 1325.558 (39.205) 1444.87 (24.806) 119.312" (46.394)
Compan'son group (A) NoChidren 1699.407 (25.907) 1705.399 (17.547) 5.992 (31.288) 113.320**(55.959)
Comparison group (B) OneChild 1490.932 (39.596) 1567.064 (29.032) 76.132 (49.099) 43.180 (67.551)
Comparison group (C) TwoPlus w/ College 1882.170 (25.888) 1911.482 (16.582) 29.313 (30.744) 90.000 (55.656)
Panel III: High School
Treatment group TwoPlus 1641.080 (22.491) 1680.774 (14.946) 39.695 (27.004)
Comparison group (A) NoChidren 1895.978 (9.933) 1926.754 (7.353) 30.776” (12.358) 8.919 (29.698)
Comparison group (B) OneChild 1776.771 (18.924) 1852.299 (13.524) 75.528" (23.260) -35.833 (35.640)
Comparison group (C) TwoPlus w/ College 1882.170 (25.888) 1911.482 (16.582) 29.313 (30.744) 10.382 (40.919)
Panel IV: Some College
Treatment group TwoPlus 1703.499 (28.698) 1761.875 (18.859) 58.376“ (34.340)
Comparison group (A) NoChidren 1928.185 (12.158) 1955.591 (8.722) 27.405“ (14.963) 30.970 (37.458)
Comparison group (B) OneChild 1832.234 (21.807) 1852.707 (16.217) 20.473 (27.176) 37.902 (43.792)
Comparison group(C) TwoPlus w/ College 1882.170 (25.888) 1911.482 (16.582) 29.313 (30.744) 29.063 (46.091)

The numbers showed are the estimates of mean. The numbers in parentheses are standard errors

” Statistically signiﬁcant at 5% level
‘ Statistically signiﬁcant at 10% level

Data: March CPS 1992-1994 and 1996-2001 which are tax years of 1991-1993 and 1995-2000

The sample is unmarried women aged 25-55 years old who have annual hours worked exceeding zero
Means are weighted with CPS March supplement weights

In theory, the EITC expansion is predicted to have reduced hours worked by

unmarried women who were already in the labor force. Thus, in Table 1.4, I restrict the

sample to include only those with annual hours of work exceeding zero. Column (3) in all

panels still shows that annual hours worked increased after 1994. However, the DID

estimates in Column (4) suggest that the treatment group did not statistically signiﬁcantly

increase their hours worked compared with the comparison groups (except in Panel 11,

compared with the comparison group (A)).

1.6 REGRESSION SPECIFICATION AND RESULTS

In Tables 1.2 and 1.3, the DID estimates show that the increases in the
participation rates and annual hours of work of treatment groups are greater than those of
the comparison groups for all cases. This suggests that a substantial increase in the
maximum credits for the treatment group may be able to explain their increase in labor
supply after 1994 relative to the comparison groups. However, the treatment and
comparison groups are not observationally the same. For example, the treatment group
tends to attain lower education, as shown in Table 1. In addition, major policy changes,
such as welfare reform, may have had differential effects on comparison and treatment
groups. To account for other characteristics and policy changes that could result in
differential labor supply outcomes, I use a Probit model to analyze changes in the
extensive margin and use an OLS estimation and a Tobit model to analyze changes in the
intensive margin of labor supply.

In the regression format, I include a dummy variable for having one child
(OneChiId) and a dummy variable for having two or more children (TwoPlus) in the
model. I control for other factors that may differentially affect the labor supply of
unmarried women, including demographic and area characteristics, and other government
programs.

I estimate the following probit model:

P (prlt=1) = a + ﬂIOneChildt + ﬂszoPlus, + ﬂ3Aﬁer1994,+ ﬂaAﬁer1994,*0neChild,-,
+ ﬂ5AﬂerI994 ,*Tw0Plus,-, + ﬂaTANF), + ﬂy Waiver), + BgUnemprs,

+ ,B,Year, + Bqu + 81‘:

19

where i indexes individuals and t indexes years. 11%).; is a dummy variable that equals 1 if
a woman reported working at least one hour, and equals 0, otherwise.14 X), is a set of
demographic characteristic variables. OneChildi, is a dummy variable that equals 1 if an
unmarried woman had one child, and otherwise equals 0. TwoPlus), is a dummy variable
that equals 1 if an unmarried woman had two or more children, and otherwise equals 0. )6)
and ,6; measure the effect of the number of children on labor force participation of
unmarried women that is not due to the EITC expansion. The inclusion of these two
variables implies that the no children group is the base group. After1994, is equal to l for
years after 1994, and otherwise equals 0. Therefore, ,83, represents the average change in
labor force participation for both the treatment and comparison groups in the years after
1994, relative to those before 1994. From Panel I of Table 2, I expect ,83 to be positive if
unmarried women increased their labor force participation, even after controlling for
other factors.

The coefﬁcient of interest, ,65, measures the change in the probability of
participating in the labor force of the two or more children group relative to the change of
the no children group, all else equal. )64 measures the change of the one child group
relative to the change of the no children group, all else equal. Theoretically, I expect ,64
and ,65 to be positive, which implies that the two or more children group and the one child
group increased their probability of participating in the labor force after 1994, relative to
the no children group. In addition, to measure the difference between the effect of the

expansion on the probabilities of labor force participation of the two or more children

 

'4 For the deﬁnition of labor force participation, I follow Eissa and Liebman (1996) who deﬁne labor force
participation as “working a positive number of hours during the year.” As a deﬁnition check, I also use
other positive numbers and ﬁnd consistent results.

20

group and the one child group, I also estimate the same Probit model with the one child
group as the base group.15

Besides the EITC expansion in 1993, the major welfare reform in 1996, the
Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA), may
have increased the labor supply of low-income unmarried women during the 19905.16
Therefore, I also include Waiver”, a dummy variable that equals 1 for the year when the
state where an unmarried woman lived implemented the Aid to F arnilies with Dependent
Children (AFDC) waiver, and otherwise equals 0. These waivers were mostly designed to
allow states to more stringently enforce work requirements for welfare recipients.”
T ANF ,-, is a dummy variable that equals 1 for all years when the state had the Temporary
Assistance for Needy Families (TANF) in place, and otherwise equals 0. TANF sets a
maximum of 60 months of beneﬁts within one's lifetime and requiring clients to ﬁnd
employment.

To account for area characteristics, I include Unempru, the unemployment rate of
the metropolitan areas and for unmarried women outside of metropolitan areas I use the

unemployment rates at the state level. 18 In addition, to reﬂect the fact that the population

 

‘5 Because of nonlinearity of Probit model, I cannot use [35 - B4 to measure the difference.

‘6 The major provision of PRWORA included devolution of greater program authority to the state, ongoing
work requirement, and ﬁve year maximum time limit (Blank 2002). Several studies examine the effect of
welfare reform on labor supply. See, for example, Schoeni and Blank (2000), O’Neill and Hill (2001),
Grogger (2003), and Kaushal and Kaestner (2005).

I would like to thank Stacy Dickert-Conlin and Katie Fitzpatrick who kindly provided this dataset. Data
of the AFDC waiver and TANF implementation are from http://aspe.hhs.gov/hsp/Waiver-

Policies99/T able A.PDF and
http://www.acf.hhs.gmLprograms/ofa/anweporm/chapter12/chap12.htm#4 . Data of expiration of
AFDC waiver are from the HHS
(http://www.acf.hhs.gov/programs/ofa/annuilreport6/chapter12/chap12.htm#4).

‘8 Data of the unemployment rate are from the Local Area Unemployment Statistics, Bureau of Labor
Statistics.

21

may have different distributions in different time period, I include Yeart, a set of year
dummies.19

Demographic characteristics include other income (in 10005 in 2005 dollars), the
number of never married children in the family age between 7 and 18, the number of
children in the family under age 6, and race. In addition, I include dummy variables for
women’s ages and levels of education.20
Results

Table 1.5 presents the marginal effects of the Probit model estimating the labor
force participation of unmarried women between 1991 and 2000. All speciﬁcations
include year dummies. Panel I reports the marginal effects when the no children group is
the base group. In Column (1), when demographic characteristics are included, the
marginal effects suggest that the probability of participating in the labor force of the two
or more children group increased by 3.7 percentage points, compared with the no
children group at statistically signiﬁcant levels. When those demographic characteristics
are not included, the marginal effect increases from 3.7 to 3.9 percentage points as shown
in Column (2).

To check the robustness of my result, in Column (3), I duplicate Eissa and
Liebman (1996) by categorizing unmarried women into two groups; those with children

and those with no children. With the different period of time, I ﬁnd a very similar result

as in Eissa and Liebman (1996). After 1994 unmarried women with children increased

 

'9 Wooldridge (2006) suggests that in case of pooling independent cross sections across time such as CPS
data, we should allow the intercept to differ across periods by including dummy variables for all except
the base year (1991 is the omitted year in this regression).
I include three dummy variables for education levels, which represent high school, some college, and college

(less than high school is the omitted category); and ﬁve dummy variables for ages, which represent the ages of 3 1-
35, 36-40, 41-45, 46-50, and 51-55 (25-30 is the omitted category).

22

their probability of participating by 2.6 percentage points relative to those with no
children. Using the smaller EITC expansion in 1986, Eissa and Liebman (1996) ﬁnd a 2.8
percentage point increase.

Back to results in Column (1), after controlling for other factors affecting the
participation rate, dummy variables for the number of children and After1994, are not
statistically different from zero. For the effect of welfare programs, the AFDC waivers
have a signiﬁcant effect on the labor force participation. Unmarried women who lived in
states that were implementing a waiver have higher probability of participating in the
labor force by 0.5 percentage points. On the other hand, I do not ﬁnd a statistically
signiﬁcant effect of the TAN F implementation. A one percent increase in the
unemployment rate lowers the probability of participating in the labor force by 0.5
percentage points.

The estimate of Aﬁerl 994,*0neChild,-, is not statistically signiﬁcant. This
suggests that the probability of the one child group participating in the labor force is not
statistically different from the no children group. In Panel 11, when using the one child
group as the base group, the marginal effect shows that after the expansion the two or
more children group increased their probability of participating in the labor force by 3.0

percentage points, relative to the one child group.

23

Table 1.5 Marginal Effects of Probit model
Dependent Variable:Labor force participation

 

 

(1) (2) (3)
Demographic Without Covariates With Children VS
Characteristics With No Children
Panel I: Relative to the comparison group (A): (the no children group)
OneChild 0.007 (0.006) -0.014" (0.006) -0.006 (0.005)
TwoPlus -0.012 (0.010) 0076" (0.007) ----- -----
Afterl994 .0015 (0.009) 0016* (0.009) -0014 (0.009)
Aﬁer1994‘0neChild 0.009 (0.006) 0.009 (0.006) 0.026" (0.004)
Aﬁerl994‘TwoPlus 0.037" (0.004) 0.039M (0.004) - -...._..
TANF 0.009 (0.008) 0.010 (0.008) 0.009 (0.008)
Waiver 0.005" (0.003) 0.008" (0.003) 0.005“ (0.003)
Unemployment rate -0.005" (0.001) 0006" (0.001) 0005" (0.001)
Other income (10005) 0002" (0.001) ----- -0.002" (0.001)
Number of children aged 7-18 0008" (0.003) ------ ------ -0.007" (0.002)
Number of preschool children -0.026" (0.004) ------ ------ -0.024" (0.003)
Nonwhite 0032" (0.003) ----- ------ -0.032“ (0.003)
Aged 31-35 -0.002 (0.004) ----- ------ -0.002 (0.004)
Aged 36—40 -0.007" (0.004) ------- ------ -0.007‘ (0.004)
Aged 4145 0009" (0.004) mm .0009" (0.004)
Aged 46-50 0009" (0.004) ----- ------ 0.010" (0.004)
Aged 51-55 -0.014" (0.004) ----- ----- -0.014" (0.004)
High School 0.034" (0.003) ......... .-....... 0.034" (0.003)
Some College 0.053" (0.002) ------- --------- 0.053" (0.002)
College 0.068" (0.003) ----- -------- 0.068" (0.003)
No. of observations 82,064 82,064 82,064
Panel II: Relative to the comparison group (B): (the one child group)
NoChildren -0.007 (0.005)
TwoPlus -0.020" (0.009)
Aﬁer1994 0.001 (0.010)
After1994“NoChi1dren -0.009 (0.006)
Afterl994'TwoPlus 0.030" (0.006)

Data: March CPS 1992-1994 and 1996-2001. The dependent variable is labor force participation.
The sample is unmarried women aged 25-55 years old

The robust standard errors are in parentheses.

Regressions are weighted with March CPS weights. All speciﬁcations include year dummies.

” Statistically signiﬁcant at 5% level

‘ Statistically signiﬁcant at 10% level

Demographic characteristic variables have the expected signs. All else equal,
relative to childless women, women with a preschool child have 2.6 percentage points
lower probability in participating in the labor force, while those who had an older child
have 0.8 lower probability. Non-white unmarried women have a lower probability of
participating in the labor force by 3.2 percentage points compared with white women. All

else equal, older unmarried women have a lower participation rate. The level of education

24

has a positive effect on the participation rate.” Finally, an increase in other incomes by a
thousand dollars lowers the probability of participating by 0.2 percentage points

As discussed earlier, women with lower education are more likely to be eligible -
f0r the EITC. Therefore, I expect that the high school dropouts and high school graduates
would have a larger participation response than those with higher education. To test this
hypothesis, I estimate Probit models of labor participation by levels of education with the
same speciﬁcation as in Column (1) in Table 1.5, but excluding dummies for levels of

education.

 

2‘ I also estimate a Probit model by using the actual age and years of education (not dummy variables).
Results are very similar to those of the speciﬁcation above. But estimates of age and education variables
are not statistically signiﬁcant.

25

Table 1.6 Marginal Effects of Probit model by levels of education
Dependent Variablchabor force participation

Less than high school High school Some College College
(1) (2) (3) (4)

Panel 1: Relative to the comparison group (A): (the no children group)

OneChild 0.071" (0.033) 0.010 (0.014) 0.008 (0.010) 0.001 (0.006)
TwoPlus 0.016 (0.045) -0.032 (0.027) 0.012 (0.015) 0.001 (0.010)
Afterl994 0.048 (0.034) -0.007 (0.023) 0.001 (0.020) 0.009" (0.006)
Aﬁerl994*0neChild 0.014 (0.039) 0.017 (0.013) 0.002 (0.01 1) 0.006 (0.005)
Aﬁer1994‘TwoP1us 0.122” (0.034) 0.063” (0.010) 0.019" (0.009) 0.011 (0.008)
TANF 0.039 (0.045) 0.024 (0.020) 0.010 (0.014) 0.001 (0.006)
Waiver 0.009 (0.019) 0.011 (0.007) 0.001 (0.005) 0.002 (0.002)
Unemployment rate .0007" (0.004) 0.003" (0.002) 0.004" (0.001) 0.003” (0.001)
Other income(1000s) -0.027" (0.003) 0.005" (0.001) 0.002" (0.001) 0.001” (0.001)
Number of children aged 7-18 0008 (0.014) -0.006 (0.009) 0.013" (0.006) 0.008“ (0.004)
Numberof preschool children 0070" (0.016) 0.026" (0.007) 0.029” (0.005) 0.016“ (0.005)
Nonwhite 0.092“ (0.016) 0.039“ (0.007) 0.021" (0.005) 0.017" (0.003)
Aged31-35 0.008 (0.023) 0.001 (0.009) 0.008 (0.007) 0.003 (0.003)
Aged 3640 0.003 (0.025) 0.001 (0.009) 0.017” (0.008) 0.002 (0.003)
Aged4l45 0.008 (0.026) 0.001 (0.009) 0.018" (0.008) 0006‘ (0.003)
Aged 46-50 0.164 (0.025) -0.001 (0.009) 0.017" (0.009) 0.007“ (0.004)
Aged51-55 0.165 (0.025) 0.002 (0.009) 0.034" (0.010) 0.010" (0.004)
No. of observations 1 1,755 24,672 14,859 24,733

Panel H: Relative to the comparison group (B): (the one child group)

NoChildren 0.072" (0.033) -0.010 (0.014) 0.008 (0.010) -0.001 (0.001)
TwoPlus -0.057 (0.042) 0.044” (0.021) 0.004 (0.013) -0.001 (0.008)
Aﬁerl994 0.062 (0.042) 0.010 (0.026) 0.013 (0.020) 0.007 (0.005)
Aﬁerl994‘NoChildren -0.014 (0.039) -0.017 (0.014) -0.002 (0.01 1) 0.006 (0.007)
Aherl994‘TwoPlus 0.109" (0.043) 0.050" (0.131) 0.017 (0.011) 0.007 (0.005)

Data: March CPS 1992-1994 and 19962001. The dependent variable is labor force participation.
The sample is unmarried women aged 25-55 years old

The robust standard errors are in parentheses.

Regressions are weighted with March CPS weights. All speciﬁcations include year dummies.

” Statistically signiﬁcant at 5% level

‘ Statistically signiﬁcant at 10% level

Table 1.6 reports the marginal effects of Probit models by levels of education. In
Panel I Columns (1), (2), and (3) show that, among high school dropouts, high school
graduates, and college dropouts, the two or more children group increased their
participation rate relative to the no children group by 12.2, 6.3, and 1.9 percentage points,

respectively. However, I do not ﬁnd any statistically signiﬁcant change in the labor force

26

participation among those with a college degree. This might be because college women
are almost all working.

In Panel 11 of Table 1.6, the marginal effect shows that relative to the one child
group, the two or more children group who are high school dropouts and high school
graduates increased their participation rate by 10.9 and 5.0 percentage points,
respectively.

The effect of the EITC on hours of work

Results in the previous section show that the EITC expansion in 1993 increases
the labor force participation of unmarried women. This suggests that the EITC could
result in an increase in total hours of work. On the other hand, theory predicts that the
EITC expansion reduces hours worked by women already in the labor force. Previous
studies use Ordinary Least Squares (OLS) to estimate the total annual hours and hours
conditional on working. However, the total annual hours takes the value of zero for a
nontrivial fraction and continuously distributes over positive values. As a result, OLS
provides inconsistent estimators (Wooldridge 2002). To correct the inconsistency
problem, I use a Tobit model to estimate total hours of work and use OLS to estimate
hours worked by those who are already in the labor force. Following is the empirical
model:

Annual Hours), = y + 610neChild, + (52TwoPlus, + 53After1994,
+ 54After1994 ,*OneChild, + 65After1994 ,*TwoPlus, + 56TANFH
+ (5; Waiver), + (38Unemprs, + 6,Year, + 6kX,-, + 6;,

where Annual Hours“ is the annual hours worked by unmarried women.

27

Table 1.7 presents results of the OLS estimation of annual hours of work. In
Column (1), when conditional on annual hours exceeding zero, the coefﬁcient of interest,
55, is not statistically signiﬁcant.22 Consistent with Eissa and Liebman (1996), this
suggests that among those who were already in the labor force, the two or more children
group did not reduce their hours worked relative to the no children group.

In Column (2), when the sample includes women who worked zero hours, the
coefﬁcient of interest suggests that after 1994 the two or more children group
signiﬁcantly increased their annual hours of work by 61 hours, relative to the no children
group. While this methodology is consistent with previous works, for example Eissa and
Liebman (1996), the estimate is inconsistent due to the fact that the variable of total hours
of work takes a value of zero for a non-trivial fraction. Of the 86,051 unmarried women

in my sample, 14,949 observations have zero hours worked.

 

22 The number of observation decreases by 14,949 to 71,102 observations.

28

Table 1.7 Results of OLS Estimation
Dependent Variable: Annual Hours of Work

Annual Hours Annual Hours
Conditinal on Hours>0 Unconditional
(1) (2)

OneChild -35.229 ** (15.645) -44.034 " (13.519)
TwoPlus -32.843 (27.781) -80.504 " (21.894)
After1994 18.009 (11.738) 19.082 " (11.331)
Afterl994‘OneChi1d 26.812 * (14.225) 31.874 "”" (13.866)
Afterl994‘TwoPlus 23.888 (15.909) 60.725 " (14.396)
TANF -6.622 (17.859) 6.489 (17.457)
Waiver 4.273 (6.537) 9.856 (6.294)
Unemployment rate -13.328 ** (1.557) -l6.569 " (1.491)
Other income (10005) -9.426 ** (0.615) -10.558 " (0.536)
Number of children aged 7-18 -38.628 ** (10.578) -23.505 " (7.725)
Number of preschool children -99.989 ** (12.364) -80.967 ” (8.778)
Nonwhite -27.972 ** (6.427) -60.363 " (6.099)
Aged 31-35 71.995 ** (8.044) 60.652 " (7.944)
Aged 36-40 94.314 ** (8.092) 78.391 " (7.977)
Aged 41-45 99.741 ** (8.166) 82.868 " (7.988)
Aged 46-50 109.259 ** (8.768) 91.726 *"‘ (8.459)
Aged 51-55 101.982 ** (9.631) 79.070 " (8.953)
High School 111.385 ** (11.304) 134.169 ** (8.701)
Some College 116.602 ** (12.111) 170.217 " (9.723)
College 164.755 ** (12.298) 220.187 " (10.031)
No. of observations 71,095 86,043

R-squared 0.21 1 0.552

Data: March CPS 1992-1994 and 1996-2001 which are tax years of 1991-1993 and 1995-2000.
The sample is unmarried women aged 25-55 years old
The robust standard errors are in parentheses. Regressions are weighted with March CPS weights.

All speciﬁcations include year dummies.
” Statistically signiﬁcant at 5% level
" Statistically signiﬁcant at 10% level

Table 1.8 presents the marginal effects from a Tobit model of total hours worked.
In general, the Tobit estimates are larger (more positive) than OLS estimates in Table 1.7
(Column 2). This suggests that OLS estimates might be biased towards zero. In Table 1.8,
the coefﬁcient of interest on After] 994,*T woPlus, in Column (1) suggests that after 1994,
relative to the no children group, the two or more group increased their total hours of
work by 120 hours (compared to 61 hours from the OLS estimate). When the sample is
categorized by levels of education, the marginal effects in Column (2) through (5)
suggest that when compared with the no children group, the two or more children group

who have less than high school, high school, and some college education increased their

29

total hours of work by 102, 126, and 85 hours, respectively. These estimates are
statistically signiﬁcant. However, I do not ﬁnd any statistically signiﬁcant change in
hours worked among those with a college degree.

The estimates of demographic characteristic variables have expected signs. All
else equal, women with a preschool child worked an average of 136 hours fewer than
those without a preschool child. Non-white women worked 55 fewer hours than white
women. Women who are older or have higher education worked a greater number of
hours. The waivers have a statistically positive effect on total hours worked by high

school dropouts and high school graduates.

3O

.1. 33V.
.2. 832.
V; 5.2V.
V... $3.
I «3.8.

”VV.V

V VVVVV.
V VVVVV
I 339

VVVVVN

VVV
VuV=5

V
V V.VV.VV.
V .. VVV.VV.

V.VVVVV ..VVVVV4

V
V. VV.VVV .. VVV.VVV.
VV

VVVVV .. VVVVV.
VV.VVVV ..VVV..VV.
VVVVVV ..V.VV.VV.
VV.VV.VVV VVVVV
VVVV.V.VV V.VV.V.
VVV.V.VVV

VVVV..VVV

.. VVV.V.V
VVVV.
VV.VV..VVV VVVVV
VVVVVVV VVVVV
VVVV VVV VVV.V.V

VV.V
V35 V25

VVVVVV .. VVVVV.
VV

V.V..V.VV .. VVV.VVV.

VV VV.VVV .. VVVVVV
VVVVVV ..V.VV.VV.
VV.V.VVV .. VVVVV.
VV.VVVVV .. VVVVV
VVVVVVV VVVVV
VV VV.VVV .. VVVVVV
VVVVVVV .VVVVV.
VVVVVVV VVVV.
VVVV.V.VV VV.V.V.V4
V V VVVV
VVV

VVVVVVVVVVVV

3m. VN

3:: Es. .252 V VVVVVVVVV» VV.V—V5.3
VVVVVV VVVVVVV VV.VV VVVVVVVVV VV.VVVVVV.

VVVVVV
VVVV VVV
VV VV.VV
VVVVVV
V

VV.VVVV VVVVV VV.VV E

1 32m.
1. ~33.
V V3.2.
1 32m.
1 83.
I 23%
mm;
1. 832
SN:
VV.V. Sm?
”VVEVN.
$92
VVV

1 £5?

a 9.3:.

V... 22%.
I S: _.
1. 33V.
V... 3a.:

3.3
I 832
I ”3.3

EV.
V... 33%.

83-
VVV

VVVVVVVVVVz

VVVVVVVVVV VVVVVVVVVVV VV VVVVVVVVz
VV.V VV.V. VV.VVVVVVVV VVVVVVVV
VVVVVVV .VVVVVV as

as VVVVVVVVVVVVVVVVVV

VVVVVVVV

V25

VVVVVVVEVVVVVVVVVVVV
VVVVVUVVOVVVVVVVVE
VVVVVVVVVV

VVVVVVVE

2525

31

ogwm

VVVVVVV VVVVV
93:5

m3.—
mama

VVVVV.VV .. VVV..VV.
VV.VVVVV .. V.VV.VV
V VV.VVV .. VVVVV
VVVVVVV .. VVV.VV
VVVVVVV .. VVVVV

VVV
09:5

.22 «.2 VV V§VVV=VVV VVVaVVVVVsV .
VsaV .VV VV V§VV=VVV VVVSVVVVVVVV ..

.385 ES. V .3 838:8 085 8 25V VVEVE.» __V 8205 .30 .3 838:8 gouge =<
8%.. V8» 3565 VVVEVSEBVVV =< .3532: m5 Ea: 5? 32325 V5 VVVoVVVoVwom

2o a8» 2.3 Bwa 889$ BEVESV VV 295V 2:.

$8.32 ea 32-82 25 5a: 5:5

VVoVVVtoVVVo .8 .3252

VVVVVVoBHangVVVc<

VVVVVEVVVVV 3ch 2:3 V: .25

Ba: REV VVVV..V_ V3.3

VVVVVVV VVV.VV VVVV.V.VV .. VVV..VV VVVVVVV .. VVV..VV VVVV.V.VV .. VVVVV
VVQ=6 2% VsVVV VV.V: V8.3 .3: a... V3 .2

VVV.V VVV.V VVV.V VV.V.V.V

VVVVV VVV.VV VVV..VV VV.V.VV

VVVV..VV .. V.VV.VVV

VVVV..VV a VVV.VV.V

VVVV..VV ..VVV.VVV

VVVVVVV..VVV.VV VVVV.VVV...VVV.VV VVVVVVV ..VVV.VV VVV.V.VV :VVVVV

VV.VVVVV ..VVV.VV VVVV.V.VV :VVVVV VV.VVVVV ..V.VV.VV VVVVVV :VVVVV

VV.VVVVV..VVV.VV VVVV.VVV..VVV.VV VVVVVVV ..VVV..VV VVVVVV :VVVVV

VVV.VVVV..VVV.VV VVVV.VVV..VVV.VV VVVVVVV ..VV.V.VV VVVV..VV ..V.VV..VV

VVVVVVV..VVV.VV VVVV.VVV..VVV..VV. VVVVV.VV :VVVVV VVVVVV :VVVVV
VV.V VVV VVV VVV
VVVVVVU 2.5V .8ng .3: V8.3 as a... as .2

3:33 2:2. .255. V «Eat; VV.V—Eamon
32.5288 .252 V55. .3 38:2 3:332 3935.

80:5: VVVVVVEVVV 858823
coVVVtoVVVo Vo 2:52

03:8
03:5 08%
VaVVV VVVVV
VV-VV VaVV.
VV-VV. VVVVV
VV.-VV.V..V<
VV.VV VV.VV.
VV-VV a?

32

1.7 CONCLUSION

In this study, I use a differential increase in the maximum values of the EITC to
identify the effect of the EITC expansion in 1993 on both margins of the labor supply of
unmarried women. Using March CPS data, my ﬁndings are consistent with previous
works. Unmarried women with two or more children increased their participation rate by
3 percentage points relative to those with one child (from a base of 78 percent).
Assuming that an unmarried woman has income falling in the ﬂat range before and after
the EITC expansion in 1993, an increase in maximmn credits causes unmarried women
with two or more children to have $906 more than those with one child. Their average
income is $15,209. From a rough calculation, the upper bound of the elasticity of labor
force participation with respect to earned income for unmarried women is 0.64. The
elasticity of labor force participation is larger for those who did not have a college
degree: 1.20 for the Less than High School group and 0.92 for the High School group.23

For the EITC effect on the intensive margins, I use the estimates ﬁ'om a Tobit
model, which corrects for the inconsistency problem. Following the same method, I ﬁnd
that the upper bound of the elasticity of total annual hours of work with respect to earned
income is 1.10. Like in the participation margin, when categorized by education levels,
the elasticity of total hours of work is larger in the lower education groups: 0.79 for the
Less than High School group, 0.99 for the High School group, and 1.05 for the Some
College group. For the College group, I do not ﬁnd statistically signiﬁcant increase in

their total hours of work.

 

23 Hotz and Scholz (2003) estimate the elasticity of labor force participation from previous studies. The
range is between 0.69-1.16. Focusing on single parents on welfare program between 1991 and 2000,
Hotz et al. (2005) ﬁnd that their employment elasticity with respect to disposable income is 1.3.

33

CHAPTER 2

THE INCIDENCE OF THE EARNED INCOME TAX CREDIT

2.1 INTRODUCTION

The Earned Income Tax Credit (EITC) is a means-tested program that has
subsidized low-income families in the United States since 1975. With a reduced role and
relevance of the traditional welfare programs after welfare reform in 1996, the EITC has
become the largest cash-transfer program for low-income families in the United States. In
2003, the EITC program paid $37.5 billion in income beneﬁts to 20.8 million families
and individuals (Center on Budget and Policy Priorities, 2005). The literature suggests
that the EITC expansions substantially increased the labor force participation of
unmarried women (Dickert, Houser, and Scholz 1995; Eissa and Liebman 1996; Meyer
and Rosenbaum 2000; Darragh 2002; Hotz, Mullin, and Scholz 2006; Adireksombat
2007)‘

Interest in whether the EITC incidence is captured by the worker or the employer
follows from the EITC’s larger role in increasing low-income workers’ labor supply. In
theory, with an increase in labor supply, all else equal, the equilibrium wage will fall
unless labor demand is perfectly elastic. With a lower wage cost, theory predicts that
employers capture a share of the EITC beneﬁts.

The purpose of this study is to examine the effect of the 1993 EITC expansion on
the gross wages of unmarried women. Because the effectiveness of the EITC of

beneﬁting low-income families depends on its economic incidence, understanding the

 

1 See Hotz and Scholz (2003) for a summary of labor supply response to the EITC.

34

effect of the EITC on gross wages helps policymakers to have a more complete analysis
to evaluate the EITC program.

To investigate the incidence, I use cross-sectional and cross-time variation in the
federal and state EITCs of the 1993 EITC expansion. The 1993 EITC expansion
substantially increased the maximum credit available to a family with two or more
children as compared to a family with no children and another family with one child.
Differential increases in the credits imply that women with different numbers of children
also experience differential changes in their marginal tax rates (MTRs). Using variation
in the differential expansion by number of children, this study and other recent studies
ﬁnd that the 1993 EITC expansion increased the labor force participation of unmarried
women with two or more children as compared to those with no children and those with
one child (Hotz, et al. 2006; Adireksombat 2007). Due to the structure of the EITC, these
labor supply increases are most highly concentrated in lower education labor markets
(Dahl and Lochner 2005; Baughman and Dickert-Conlin forthcoming). Therefore, I
consider whether there are differential effects on wages by education.

I use data from the 1992 to 2001 March Current Population Survey (CPS) and the
Merged Outgoing Rotation Group Files (MORG). The sample includes unmarried
women aged 25-55 years. Using only the sample with observed wages to estimate a wage
equation, Ordinary Least Squares (OLS) estimation generally produces estimators biased
toward zero due to the sample selection problem (Hausman and Wise 1977). In my
empirical method, I ﬁrst test for selection bias in estimating wage equations. To my
knowledge, no previous studies of the incidence of the EITC address this problem. I ﬁnd

no evidence of a sample selection problem and therefore I use OLS to estimate the wage

35

equation. Results suggest that there are no statistically signiﬁcant decreases in wages
received by unmarried women who are in markets most heavily affected by an increase in
labor supply. With a positive effect on labor supply but no adverse effect on wages, this
study supports the EITC as an effective anti-poverty instrument for unmarried women.
Chapter 2 proceeds as follows: Section 2.2 discusses the structure of the EITC and
its effect on wages. Section 2.3 reviews literature on the labor supply response of the
EITC and the tax incidence. Section 2.4 describes data and the empirical method. Results

are presented in Section 2.5 and section 2.6 concludes.

2.2 THE EITC AND ECONOMIC INCIDENCE
A. The EIT C Structure and Its History

The credit equals a speciﬁed percentage of earned income up to a maximum
dollar amount over the “phase-in range.” Over a range of income termed the “ﬂat range,”
taxpayers receive the maximum credit. The credit then diminishes to zero over the
“phase-out range.” 2 The EITC is refundable. Claimants are paid regardless of whether
the credit-qualiﬁed taxpayer has any federal income tax liability. The EITC payment is
typically made once a year as an adjustment to tax liabilities or refunds.3 For those who
have children and want to claim the EITC, their children need to pass an age,
relationship, and residence tests to qualify. For example, the age test in 2000 requires the

qualifying child to be under 19 years old or under 24 years old if she/he is a full-time

 

2 Table A1 in Appendix shows the federal ElTC parameters between 1990 and 2000.

3 An “advance payment” option was added in 1978 so that workers would be able, if they so chose, to
receive the credit incrementally throughout the year. However, in 1998, only 1.1 percent of EITC
recipients with children used the advance payment option (Hotz and Scholz 2003).

36

student, or any age if she/he is completely disabled. The relationship test requires the
claimant to be the parent or the grandparent of the qualifying child.

Since it began in 1975, inﬂation eroded the EITC payments until the Tax Reform
Act of 1986 (TRA-86) increased the maximum credit in 1987 to have a real value equal
to that of the credit in 1975, and indexed the EITC value for inﬂation. The Omnibus
Budget Reconciliation Act of 1990 (OBRA-90) introduced differential credit rates and
maximum credits available to a family with one child and a family with two or more
children.

OBRA-93 substantially increased the maximum credit available for a family with
two or more children, relative to that with no children and that with one child. OBRA-93
also expanded the beginning and ending incomes of the “phase-out range.” Figure 2.1
presents the maximum credits available to a family with two or more children, a family
with no children, and a family with one child in 2005 dollars for the period from 1991 to
2000. The maximum credit available to a family with two or more children increased
substantially between 1994 and 1996, becoming fairly constant afterward, as the reform
was fully phased in (from $3,230 to $4,400). For a family with one child and a family
with no children, the maximum credits increased after OBRA-93 was implemented, and
became constant after 1994. The difference between maximum credits available to a
family with two or more children and a family with no children rose from $2,042 in 1993
to $3,233 in 1996. Relative to the one child group, the difference rose from $104 to
$1,404. In addition, OBRA-93 introduced the EITC to childless taxpayers aged between

25 and 65 years.

37

Figure 2.1 The EITC Maximum Credits duringthe 19905
S (in 2005 dollars)

 

4500 r
4000
3500
3000
2500
2000 ”
1500
1000 ‘
500 .

 

 

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

Year

 

 

No Children in One Child I the Children 1

 

The success of the federal EITC led a number of states to enact their own EITCs
during the 19905.4 5 In 1989, only four states had EITCs and by 2000, the last year of my
data, fourteen states and the District of Columbia had EITCs: Colorado, Illinois, Iowa,
Kansas, Maine, Maryland, Massachusetts, Minnesota, New Jersey, New York, Oregon,
Rhode Island, Vermont, and Wisconsin. The state EITCs are very similar to the federal
EITC. These states use federal eligibility rules and calculate the state credit as a speciﬁed
percentage of the federal credit.6 Therefore, the state EITCs also provide the same
incentives to work and hence the same effects on wages as the federal EITC. The main
difference between the federal and state EITCs is the reftmdable feature. In 2000, ﬁve of

these states had non-refundable state EITCs.7 The credit rates also vary across states. For

 

4 See Johnson (2000) for a summary of the state EITCs during the 19905.

5 In 2006 tax year, eighteen states and the District of Columbia had state ElTCs (Nagel and Johnson 2006).

6 Using federal eligibility rules, Minnesota has its own credit structure that has major elements parallel to
the federal structure.

7 Those ﬁve states were Illinois, Iowa, Maine, Oregon, and Rhode Island.

38

example, in 2000, Kansas’s EITC was 10 percent of the federal EITC and Wisconsin’s
EITC varied with the number of children and was 43 percent of the federal EITC for
families with three children.

B. The Effects on the Labor Supply of Unmarried Women

From the static labor-leisure model, the EITC expansion, resulting in changes in
the MTR, affects both intensive and extensive margins of the labor supply of unmarried
women. Therefore, in theory, the expansion leads to changes in the labor supply of those
who were out of and those who were already in the labor force.

For a non-worker who was out of the labor force before the expansion, the EITC
expansion results in only a positive substitution effect but no income effect and hence
increases the incentive to work.

For a worker who was already in the labor force, the effect of changes in the MTR
on her labor supply depends on the EITC range in which her income falls before and after
the EITC expansion. If her income falls in the “phase-in range,” with a negative MTR,
there will be a positive substitution effect and a negative income effect. Thus, when
leisure is assumed to be a “normal good,” the net effect on labor supply is ambiguous. If
her income falls in the “ﬂat range,” the MTR is zero. Thus, there will be only a negative
income effect. As a result, the effect of the EITC will lead to a decrease in hours of work.
If her income falls in the “phase-out range,” a positive MTR implies a diminishing
available credit and hence a lower effective wage relative to the absence of the EITC.
This negative substitution effect results in a reduction in hours of work. Moreover, there

is a negative income effect that leads to a further reduction in hours of work. Finally, if

39

her income is beyond the credit region, she may decide to reduce her hours of work to be
eligible for the credit.
C. The Eﬂects on the Wages of Unmarried Women

As discussed in Part B, the EITC expansion results in changes in unmarried
women’s incentives to work. With mixed effects on individual decisions to work, existing
research suggests that the EITC expansion increased the labor supply of unmarried
women (Dickert, et al. 1995; Eissa and Liebman 1996; Meyer and Rosenbaum 2000;
Hotz, et al. 2006; Adireksombat 2007), but decreased the labor supply of married women
(Dickert et a1. 1995; Eissa and Hoynes 2004). In addition, among unmarried women, the
EITC expansion substantially increased the labor market participation of those with two
or more children as compared to those with no children and those with one child (Hotz, et
al. 2006; Adireksombat 2007). All else equal, the change in unmarried women’s labor
supply might lead to a wage drop. However, because unmarried women workers are only
part of the labor market, the effect of the EITC expansion on wages also depends on the
overall effect on the labor supply of other workers (for example, married women and
men), their elasticity of labor supply, and their elasticity of labor demand.

Figure 2.2 demonstrates a simple labor demand and labor supply model and the
changes in the labor supply of those women after the 1993 EITC expansion. I assume that
there are three types of employees in the labor market: those with two or more children
(Twoplus), those with no children and those with one child (N oChildren and OneChild),
and the married women group.8 The ﬁrst three diagrams show changes in the labor

supply of married women, the Nochildren and OneChild group, and the Twoplus group,

 

8 I do not include male workers in this model. The literature suggests that wage changes have a very small
effect on male workers’ work decisions (See, for example, Pencavel I986; Triest 1990).

40

respectively. The last diagram shows the overall effect of the changes in labor supply on
the gross wage. The model predicts that with an increase in the aggregate labor supply
due to the EITC expansion, the gross wage is driven down ﬁom W0 to WI. With a
decreasing wage, employers enjoy paying lower wage costs. All employees receive a
lower gross wage at W1. However, with the EITC, the Nochildren and OneChild group
and the Tw0plus group earn a higher net wage at W”.9 Only the married women group
suffers from a lower gross wage, but do not receive a higher net wage.10

This simple model predicts that the equilibrium wage will be driven down aﬁer
the EITC expansion for everyone in the same labor market. This implies that employers

capture a share of the EITC beneﬁts.

 

9 Note that, with a smaller EITC available to the Nochildren and OneChild group, their net wage is lower
than that of the Twoplus group.

1° For simplicity, I assume that married women do not receive the EITC. Baughman and Dickert-Conlin
(forthcoming) ﬁnd that among married women in their sample only 15.6 percent of them are eligible for
the EITC and most of their incomes are in the “phase-out range.”

41

Figure 2.2 Labor Demand and Labor Supply

 

 

 

 

 

4’ Ls‘
\ WN
WO
WI
\ Ls0
Married NoChildren and
Women OneChild

 

 

 

 

v
V

TwoPlus Labor
Market

42

2.3 LITERATURE REVIEW
A. Labor Supply Response of the E1 T C

Due to the substantial increase in labor force participation and a desire to avoid
the complexity of analyzing the joint labor supply decisions of husband and wife, most
studies focus on the labor supply decisions of unmarried women. Using different amounts
of credits available to families with children and those with no children as the
identiﬁcation strategy, the early studies of the labor supply response of the EITC ﬁnd that
the EITC expansions increased the labor force participation of single mothers. However,
they do not ﬁnd statistically signiﬁcant effects on hours worked (Dickert, et al. 1995;
Keane 1995; Eissa and Liebman 1996; Keane and Mofﬁtt 1998; Ellwood 2000; Meyer
and Rosenbaum 2001; Darragh 2002).

With a substantial increase in the maximum credits available to a family with two
or more children under the 1993 expansion, more recent studies compare the labor force
participation of unmarried women with two or more children with those with no children
and those with one child.

For example, Hotz, et a1. (2006) use longitudinal data between 1991 and 2000
from California’s Medi-Cal Eligibility Data System (MEDS) and the California
Employment Development Department (EDD) Base Wage Files as well as data from
federal tax returns to examine the effect of the EITC on labor market participation of
single-parent families on welfare.ll Taking advantage of the longitudinal data, their
empirical approach controls for covariates and household-speciﬁc ﬁxed effects. Their

preferred result shows that the 1993 EITC expansion increased employment by as much

 

” MEDS provides information on welfare program participation and demographic characteristics. EDD
contains employer-reported taxable wage payments, labor force participation, employment, and earning.
Finally, federal tax return data provide tax ﬁling and EITC claiming information.

43

as 3.4 percentage points for families with two or more children relative to those with one
child. In addition, the EITC accounted for 11.8 percent of the average increase in
employment over this period.

Using the same identiﬁcation strategy, Adireksombat (2007) uses national survey
data, the March CPS, ﬁ'om 1991 to 2000 to examine the effect of the 1993 EITC
expansion on the labor supply of unmarried women. He ﬁnds that the expansion resulted
in an increase in the probability of the labor force participation of unmarried women with
two or more children by 3.6 percentage points relative to those with no children and 3.0
percentage points relative to those with one child. In addition, Adireksombat examines
the effect of the 1993 EITC expansion on hours of work. Like in the participation margin,
he ﬁnds that unmarried women with two or more children increased their total annual
hours worked, compared to those with no children and those with one child. However,
when only those who were already in the labor force were included in the sample, he
does not ﬁnd a statistically signiﬁcant change in their annual hours worked.

As discussed in Section 2.2, married women are another group that might be in
the same labor market as unmarried women. The existing research suggests that the EITC
expansions decreased the labor force participation of married women (Dickert et a1. 1995;
Eissa and Hoynes 2004).

Due to the substantial increase in the labor force participation of unmarried
women after the 1993 EITC expansion, it is especially important to correct for the

decision to work in a wage regression, when we examine the effect on wages.

44

B. Tax Incidence

The early works of tax incidence studies use theoretical and numerical simulation
models to analyze the economic incidence of the tax policy. 12 More recent studies employ
econometric techniques and micro-datasets, such as data ﬁ'om the CPS and the Survey of
Income and Program Participation (SIPP), to estimate tax burden and their distribution
across individuals. The advantage of micro data is that it allows researchers to calculate
tax burden across demographic characteristics and to analyze the tax incidence by taking
into account the interaction among taxes and government programs.

Recent studies, including this one, employ reduced-form models or natural
experiments (commonly referred to as the Difference-in-Diﬁ’erence (DID) approach).13
The advantage of the DID approach is simplicity and transparency in the assumptions
that allow the identiﬁcation of key parameters. Unlike a true experiment, in which
treatment and comparison groups are randomly chosen, in natural experiments the
treatment group (affected) and the comparison groups (not affected) arise from the
particular policy change. Therefore, to control systematic differences between the
treatment and comparison groups, two periods of data (before and after the policy
change) are needed.

Using variation in state legislation over time with data from 1979-1981 and 1987-
1988 CPS and the Bureau of Labor Statistics, Gruber and Krueger (1991) examine the
economic incidence of the workers’ compensation insurance. They ﬁnd a negative
relationship between workers’ compensation insurance rates and wages. For example, 86

percent of increases in workers’ compensation costs are shifted to workers in terms of

 

'2 See Fullerton and Metcalf (2002) for a summary of tax incidence literature.
‘3 See, for example, Gruber (1994) and Kubik (2004)

45

lower wages. Gruber (1994) also uses variation across states and CPS data to estimate
incidence in a DID framework. He examines the economic incidence of mandated
maternity beneﬁts and ﬁnds that there was a 4.3 percent wage decrease for childbearing
women in states that passed the law relative to women in states that did not.

Several studies use a tax reform to identify the economic incidence. Using
Chilean data, Gruber (1997) employs a dramatic reduction in the payroll tax burden on
Chilean ﬁrms between 1979 and 1986 to identify the payroll tax incidence, which he
ﬁnds is fully borne by workers. Bingley and Lanot (2002) propose a model of the
determination of equilibrium wages and labor supply in the presence of income tax.
Using Danish longitudinal data, they ﬁnd that the elasticity of gross wage with respect to
the income tax rate is 0.44. This implies that the tax incidence is not fully borne by
workers, but partially shifted to employers. Kubik (2004) uses the CPS data to study the
short-run incidence of personal income taxation in the United States by examining how
the wage structure was shifted aﬁer TRA-86. He ﬁnds that in an occupation where the
median MTR decreased by 10 percent, the median wage decreased by 1.3-2.5 percent.

With a much larger body of research on the incidence of income and payroll
taxes, the incidence of the EITC has not been widely explored yet.

One exception is Leigh (2005), who uses data from the CPS and the MORG
between 1989 and 2002 to examine the effect of the EITC on the hourly wages and the
labor supply of male and female workers (both married and unmarried), aged between 25
and 55. Using variation in state EITCs, he does not ﬁnd a statistically signiﬁcant effect of
changes in the average tax rates due to the EITC expansions on wages. Using another

measured source of variation, Leigh calculates the fraction of EITC-eligible employees in

46

each labor market and uses them as simulated instruments for the EITC parameters. He
ﬁnds that a 10 percent increase in the fraction of EITC eligibles leads to a 6 percent
decrease in wages when labor markets are categorized by three-digit occupation codes
and a 4 percent decrease when categorized by the employees’ gender, age, and education.
In his sample, Leigh uses only individuals who were already in the labor force. This
might lead to a sample selection problem, producing biased and inconsistent estimators
(Wooldridge 2002).

Using the same dataset as Leigh (2005), Rothstein (2007) examines the effect of
the 1993 EITC expansion on the labor supply and wages of women aged 16-64 between
1991 and 1996. Rothstein employs a semi-parametric estimation method with a DID
approach to obtain estimates of female labor supply and wage changes at each point in
the skill distribution before and after the 1993 EITC expansion. He ﬁnds that the
expansion leads to increases in the labor force participation of low- and middle-skill
single mothers. Results from his simulation model suggest that for every dollar spend
expanding the EITC, wages fall by enough to reduce single mothers’ earnings by $0.30
and those for women with no children by $0.43. In his analysis, Rothstein does not
include any non-tax means-tested transfer programs, which is another government policy
that has signiﬁcant effects on the labor supply of low income families.14

This chapter extends the existing literature by analyzing the EITC incidence using
variation from both federal and state EITCs, controlling for substantial non—tax policy

variation due to welfare reform in 1996, and accounting for the sample selection problem

 

‘4 See Schoeni and Blank (2000), O’Neill and Hill (2001), Grogger (2003), and Kaushal and Kaestner
(2005) for the effect of welfare reform on the labor supply of low-income families.

47

and the endogeneity of the EITC. In addition, instead of using estimated skill distribution,

this paper uses levels of education to categorize labor markets.

2.4 DATA AND EMPIRICAL METHOD
A. Data

I use pooled cross section CPS data from 1992—2001 (describing data in 1991-
2000 tax years). CPS is the United States govemment’s monthly household survey of
employment and labor markets. Wage data are from the MORG, extracted from the CPS
Annual Earnings File by the National Bureau of Economic Research (N BER).15 This
dataset provides consistent hourly gross wage data. However, the MORG does not
identify the number of children during the 1994-1998 survey years and incomes, which I
use to calculate the MTRs, for the whole sample period. To attain data of numbers of
children and incomes, I extract them from the March CPS Annual Demographic File. For
the matching method, I follow Madrian and Lefgren (1999) by using the household
identiﬁcation, household number, and person line number in the household.16

The sample in this study includes unmarried women aged 25-55 who were not
self-employed. I exclude self-employed women because their wage rate is too
complicated to calculate, which might lead to an inconsistent result of wage rates. The

resulting sample is 13,237 observations.

 

‘5 Every household that enters the CPS is interviewed each month for 4 months, then ignored for 8 months,
then interviewed again for 4 more months. If the occupants move, the new occupants are interviewed.
Usual weekly hours/earning questions are asked only at households in their 4'” and 8th interview. These
outgoing interviews are the only ones included in the extracts. New households enter each month, so one
fourth the households are in an outgoing rotation each month.

'6 Household identiﬁcation is a unique household identiﬁer. Household number notes which household is
living at this address. Person line number is useful in matching individuals across years. Madrian and
Lefgren (1999) suggest that the unique and consistent combination of these three variables is able to
match CPS respondents.

48

Table 2.1 presents descriptive statistics for my sample of unmarried women.

Average wages received by women in my sample vary signiﬁcantly across levels of

education ($9.57 for Less than High School, $12.58 for High School, $14.27 for Some

College, and $19.68 for College). Those with lower education are also less likely to

participate in the labor force and work fewer annual hours. In addition, other

demographic characteristics vary across levels of education. Those with lower education

tend to have more children, have less income from other sources, be less likely to be a

union member, and are more likely to be non-white.

Table 2.1 Summary Statistics

Less than High School Some College College
High School

Wage 9.57 12.58 14.27 19.68
(in 2005 dollars) [5.70] [6.49] [9.79] [10.12]
Labor Force 0.40 0.68 0.75 0.84
Participation [0.49] [0.47] [0.43] [0.35]
Annual hours 870.32 1447.25 1604.35 1844.56
of work [958.29] [915.34] [872.22] [787.23]
Number of Children 1.17 0.83 0.74 0.43
[1.37] [1.07] [1.03] [0.79]

Age 38.80 38.14 37.44 37.40
[8.99] [8.69] [8.60] [8.64]

Other income 2.64 2.79 3.57 4.41
(in thousands [5.33] [6.93] [8.45] [l 1.78]

of 2005 dollars)

Union Member 0.04 0.08 0.09 0.13
[0.21] [0.28] [0.29] [0.34]

Nonwhite 0.36 0.29 0.30 0.21
[0.48] [0.45] [0.45] [0.41]

Observation 1,764 4,473 2,855 4,145

Data: Merged Data form the 1 992-200! March CPS and MORG CPS

The numbers shown are the estimates of mean. The numbers in parentheses are standard deviations

The sample is unmarried women aged 25-5 5 years old.

Wages are in 2005 dollars. Labor force participation equal one if wage data is not missing. zero otherwise.

Means are weighted with CPS MORG weights.

49

B. Empirical Approach

To justify deﬁning my labor markets by education level, I ﬁrst illustrate how the
law change affected skill levels differentially. Using TAXSIM, a tax simulation model
prepared by the NBER”, with income data from the March CPS, I calculate the marginal
tax rates (MTRs) under federal and state income tax laws experienced by women in my
sample.l8 Tax rates depend not only on tax laws but also on various characteristics over
time. To isolate the effects of the tax reform from the effect of changes in the
characteristics and composition of the sample I use the same sample to calculate tax rates
for all years.19 I use the sample from the 1995 March CPS (describing data for the 1994
tax year) and adjust their incomes with the Consumer Price Index to convert them into
1991-2000 dollars. Then, I use the TAXSIM to calculate the MTRs and their changes
(before and aﬁer 1994) experienced by unmarried women by year, state, education, and
number of children.

Table 2.2 presents the median MTRs and their changes experienced by unmarried
women before and after 1994 by education level.20 Under OBRA-93, the EITC expansion
is the tax policy change that is most likely to affect low-income families.21 Column 3
shows changes in the median MTRs after 1994. Negative values mean that they
experienced a decrease in their median MTRs. After 1994, women with Less than High

School, High School, and Some college education (Treatment Groups) experienced a

 

'7 See F eenberg and Coutts (1993) for a description of the TAXSIM calculator.

‘8 Marginal tax rates are the effective marginal tax rates, which are calculated by adding a dollar to income.

'9 This approach is used in several tax studies, for example, Dickert-Conlin and Houser (2002) and Kubik
(2004).

2° To avoid the effect of outlier observations, 1 use median. Using the mean MTR also presents consistent
results.

2' Other policy changes are more likely to affect high-income taxpayers. For example, OBRA-93 imposed a
36 percent tax bracket for taxable incomes over $140,000 for joint ﬁlers and $115,000 for individuals
and a 39.6 percent tax bracket for both joint and individual ﬁlers for taxable incomes over $250,000. See
Steuerle (2002) for a summary of tax policy changes under OBRA-93.

50

signiﬁcant decrease in their MTR (approximately 11, 14, and 4 percentage points,
respectively). However, those with a College degree (Comparison group) experienced a
very small change in MTRs (one percentage point).

Column (4) presents the difference in changes of the median MTRs between the
treatment and the comparison groups. These estimates suggest that relative to the
comparison group, women in Less than High School, High School, and Some College

groups experienced a larger MTR decrease by 10, 13, and 3 percentage points.

Table 2.2 Median Marginal Tax Rates of Unmarried Women Before and After 1994

Before Alter Difference (2)-(1) Differender,,,,,,,,,,,,,,-Dit‘ferencec,,,,,p,,,,,on

(1) (2) (3) (4)

Less than high school -5.35 - 16.44 -1 1.09 -10.09
High school 15.00 0.80 -l4.20 -l3.20
Some college 12.33 7.93 -4.40 -3.40
College (Comparison) 21.00 20.00 -1.00 ----------

Sample: lmmanied women aged 25-55 excluding those who were self-employed from the 1995 March CPS.

Table 2.2 shows that women with only a high school education experienced the
largest decreases in their MTR, primarily due to expansion in the EITC, therefore I
expect larger changes in labor force participation rates and wages in the High School
group and smaller changes in Less than High School and Some College groups, relative to

the comparison group.

51

Table 2.3 Labor Force Participation Rates of Unmarried Women Before and After 1994

Before After Difference (2)-(1) Difference-in-Difference
(1) (2) (3) (4)

Less than High School 0.47 0.58 0.11 "* 0.10 "‘**
(0.02) (0.02) (0.03) (0.03)

High School 0.77 0.81 0.04 "** 0.03 **
(0.01) (0.01) (0.01) (0.01)

Some College 0.83 0.88 0.05 "* 0.04 **
(0.01) (0.01) (0.01) (0.02)

College (Comparison) 0.92 0.93 0.01
(0.01) (0.01) (0.01)

The numbers shown are the estimates of mean. The numbers in parentheses are standard errors.
‘” Statistically significant at 1% level

“ Statistically signiﬁcant at 5% level

" Statistically signiﬁcant at 10% level

Data: Merged Data form March CPS and MORG CPS 1 992-2001.

The sample is unmarried women aged 25-55.

Means are weighted with CPS MORG weights.

Table 2.3 presents the labor force participation rates of unmarried women. 22
Columns (1) and (2) show the average participation rates before and after 1994, and
Column (3) shows the difference between them. As expected, the treatment groups
signiﬁcantly increased their labor force participation compared with the comparison
group. While women with a college degree increased their labor force participation rate
by only one percentage point, those with Less than High School, High School, and Some
college education increased their participation rate by 11, 4, and 5 percentage points,
respectively. The Difference-in-Difference (DID) estimates in Column (4), which

compare changes of the labor force participation before and after 1994 between the

treatment group and the comparison group, suggest that the treatment groups increased

 

22 For the deﬁnition of labor force participation, I follow Eissa and Liebman (1996) who deﬁne labor force
participation as working a positive number of hours during the year. As a sensitivity check, I use wage
observability as another deﬁnition, and ﬁnd consistent results.

52

their labor force participation rate relative to the comparison group by 10, 3 and 4
percentage points, respectively.

Table 2.4 Wages for Unmarried Women before and After 1994

(in 2005 dollars)
Before After Difference (2)-(l) Difference-in-Diﬁ‘erence

(1) (2) (3) (4)

Less than high school 9.4] 9.64 0.23 —0.38
(0.35) (0.29) (0.46) (0.59)

High school 12.15 12.53 0.38 -0.23
(0.18) (0.17) (0.46) (0.45)

Some college 14.32 14.25 -0.06 -0.67
(0.25) (0.21) (0.33) (0.49)

College (Comparison) 19.28 19.89 0.61 ----
(0.28) (0.24) (0.47) -—--

The numbers shown are the estimates of mean. The numbers in parentheses are standard errors.
“"' Statistically significant at I % level

“ Statistically signiﬁcant at 5% level

"' Statistically significant at 10% level

Data: Merged Data form March CPS and MORG CPS [992-2001.

The sample is unmarried women aged 25-55 with observable wages.

Means are weighted with CPS MORG weights.

Table 2.4 shows wages in 2005 dollars received by unmarried women before and
after 1994 by education level. Column 3 presents changes in their wage rates. Negative
values mean that the group experienced a wage drop after 1994. Estimates in Column 3
suggest that most of the women in the sample did not experience a statistically signiﬁcant
change in their wage after 1994. Column 4 reports the change in the wage rate of the
treatment group compared with the comparison group. The DID estimates in Column 4
suggest that the treatment groups experienced a larger wage drop than the comparison
group. However, these estimates are not statistically signiﬁcant.

Although these descriptive statistics are informative, the groups are not
observationally the same and it is necessary to control for these differences to isolate the

effect of the reform. For example, the treatment groups tend to have more children and

53

have less income from other sources as shown in Table 2.1. In addition, major policy
changes, such as welfare reform, may have had differential effects on the labor force
participation and hence the wages of treatment and comparison groups.

To account for other characteristics and policy changes that could result in
differential changes in labor force participation and wages, I use a Probit model to
analyze individual decisions to work for women in my sample. In the wage analysis, I
start with testing for evidence of selection bias. If there is a sample selection problem, I
use a Heckit Two-Step method, if not, I use OLS to analyze changes in their gross wages.
Following is the empirical model:

Labor force participation equation
Pﬂfpir—“U = a0 + alAfter1994, + azEducationj + ajAﬁerI994,*Education,-

+ a 3 MT Rjks,+ aMaxBeneﬁtsk, +a5 Waivers, + a6 TANF.“

+ a 7Unemprs, + a,Year, +a,State, + ak/Yg'sk; + ail-5k, (1),
Wage equation
log(wage,-jsk,) = ,80 + BIAfter1994, + ,8 2Educationj + ,6,After1994,*Educationj

+ B3MTRJ-ks +B4Unemprs, + ﬂ5lamda+ B,Year, + BsState,

+ )Bk Xyskr + 8 ijskt (2),
where i indexes individuals, j indexes education levels, t indexes years, s indexes states, k
indexes numbers of children.

In the labor force participation equation, [fan is a dummy variable equal to 1 if a
woman reported working at least one hour that year, and equals 0 otherwise. Aﬁer1994,

is equal to 1 for all years after 1994, and equals 0 otherwise. Education]- is a set of

54

dummy variables for levels of education.23 Thus, a,- measures the differential effect of
education levels on the probability of labor force participation. MTR)“, is the median
MTR faced by unmarried women in state s with j education, k children in year t. All else
equal, (12 measures the effect of the median MTRs on the probability of labor force
participation.

Equation (1) also includes a set of demographic characteristic variables, Km, to
control for differences in population sampled over time. Demographic characteristics
include the number of never-married children in the family between ages 6 and 18, the
number of children in the family between ages 0 and 5, other income (in thousands of
2005 dollars), dummy variables for race, status of union membership, and age
categories.24

To account for area characteristics, I include Unemprm the unemployment rate of
the metropolitan areas. For unmarried women outside of metropolitan areas, I use state
level unemployment rates.25

Besides the EITC expansion in 1993, the major welfare reform in 1996 under the
Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) may have
increased the labor supply of low-income, unmarried women during the 19903.26
Ofﬁcially the Aid to Families with Dependent Children (AFDC) became Temporary
Assistance to Needy Families (TAN F) and the major provisions included devolution of

greater program authority to the state, ongoing work requirements, and a ﬁve year

 

23 I include three dummy variables for education levels; Less than High School, High School, and Some
College (College is the omitted category).
2‘ 1 include ﬁve dummy variables for ages; 31-35, 36-40, 41-45, 46-50, and 51-55 (25-30 is the omitted category).
25 Data of the unemployment rate are from the Local Area Unemployment Statistics, Bureau of Labor
Statistics.
2" Several studies examine the effect of welfare reform on labor supply. See, for example, Schoeni and
Blank (2000), O’Neill and Hill (2001), Grogger (2003), and Kaushal and Kaestner (2005).

55

maximum time limit (Blank 2002). To capture welfare reform effects, I include Waiver",
a dummy variable that equals 1 for the year when the state adopted a waiver to reform
AFDC, and otherwise equals 0. TANF), is also a dummy variable that equals 1 for all
years when the state had TAN F in place, and otherwise equals 0. Moreover, I include
MaxBenefitskb a sum of maximum beneﬁts from AF DC/T AN F and food stamp programs
(in hundreds of 2005 dollars), as a proxy for the outside option for those unmarried
women when they were not working.27

In the wage equation (equation (2)), the dependent variable is log(wage gm),
which is the log of wages (in 2005 dollars). The independent variables are the same as in
equation (1), but exclude MaxBeneﬁtsm, Waiver“, TANF”, and other income, which are
used as exclusion restrictions.28 To control for unobservable effects that are ﬁxed over
time within state, I also include States, a set of dummy variables for states. Moreover, I
include Year,, a set of year dummies, to control for changes in the wage structure relative
to 1991, which is the omitted year in the regression.29

As discussed earlier, in case there is evidence of selection bias, the wage equation
(equation (2)) is estimated by the Heckit Two-Step method. With this method, the wage
equation also includes the estimate of the inverse Mills Ratio, lambda, which can be used

to test for evidence of a sample selection problem.

 

27 I would like to thank Dan Rosenbaum who kindly provides the data of the maximum beneﬁts from 1991-

1997. Data from 1998-2000 are from 1998 and 2000 Green Books and the Welfare Rules Database, the
Urban Institute.

Other income is used by previous works (See, for example, Mroz 1987). For welfare reform, existing
studies suggest that the welfare program has a signiﬁcant effect on labor supply (See, for example,
Schoeni and Blank 2000; Grogger 2003). However, it does not have a signiﬁcant effect on wages
(Grogger 2005).

2” Wooldridge (2006) suggests that in the case of pooling independent cross sections across time, such as
the CPS data, to reﬂect the fact that the population may have different distributions in different time
periods, we should allow the intercept to differ across periods by including dummy variables for all but
one year.

56

The coefﬁcients of interest in the wage equation, ,6) , measure whether those
women who did not have a college degree experience a differential wage change after
1994 relative to the College group. With a greater decline in the MTRs faced by the
treatment groups as shown in Table 2.2, and assuming that the substitution effect
dominates the income effect, I expect a,- in the labor force participation equation to be
positive and ,6,- in the wage equation to be negative. In other words, I expect larger
increases in labor supply and, in response, larger declines in wages relative to the College

Educated control group.

2.5 RESULTS
A. Labor force participation response

Table 2.5 reports marginal effects of a Probit model on the labor force
participation of unmarried women. Consistent with the previous work (Hotz et al. 2006;
Adireksombat 2007), the coefﬁcients of interest on After1994 * Education,- in Column (1)
suggest that, relative to women with a College education, the 1993 EITC expansion
increased the labor force participation of women in Less than High School, High School,
and Some College groups by 4.7, 2.9, and 2.1 percentage points from a base of 40, 68,
and 75 percent, respectively. These estimates are statistically and economically
signiﬁcant.

The estimates of MT R suggest that, all else equal, a 10 percent drop in the MTR
results in an increase in the labor force participation rate by 3.3 percentage points.
Welfare reform in 1996 has signiﬁcant effects on the labor force participation. Unmarried

women who lived in states that were implementing a waiver and the TANF have a higher

57

probability of participating in the labor force by 2.5 and 4.3 percentage points,

respectively. On the other hand, the estimate of MaxBeneﬁt is not statistically

signiﬁcant.30
Table 2.5 Marginal Effects of Probit Model
Dependent Variable : Probability of Labor Force Participation
<1) (2)
Marginal Robust Marginal Robust
Effect Std. Err. Effect Std Err.
Aﬁerl994 -0.043 (0.026) Al’ter1994 -0.032 (0.021)
Aﬁerl994’Less than High School 0.047 ** (0.021) Afterl994*MTRchange -0.245 ** (0.071)
Aﬁer1994*High School 0.029 ** (0.013) —— — --—-
Afterl994‘Some College 0.021 * (0.014) —— -——
MTR -0.033 ** (0.013) —— -—--- —-—-
MaxBeneﬁt -0.001 (0.001) MaxBeneﬁt -0.001 (0.001)
TANF 0.043 ** (0.022) TANF 0.044 ** (0.022)
Waivers 0.025 * (0.013) Waivers 0.024 ** (0.012)
Less than High School 0040 *** (0.026) less than High School -0.042 *** (0.027)
High School 0023 ** (0.011) High School -0.022 ** (0.011)
Some College 0018 * (0.009) Some College 0020 * (0.011)
Other income 0006 *** (0.001) Other income 0006 *** (0.001)
Children aged under 5 -0.075 *** (0.016) Children aged under 5 -0.079 *** (0.016)
Children aged 6-18 -0.004 (0.014) Children aged 618 -0.008 (0.013)
Nonwhite -0.090 *** (0.011) Nonwhite -0.090 *** (0.011)
Union Member 0.269 *** (0.007) Union Member 0.269 *** (0.007)
Unemployment rate 0012 *** (0.003) Unemployment rate -0.012 *** (0.003)

‘" Statistically signiﬁcant at 1% level
“ Statistically signiﬁcant at 5% level
“ Statistically signiﬁcant at 10% level

Data' 1 992-2001 MORG and March CPS
The sample is married women aged 25-55 years old. All speciﬁcations include age, year, and state dummies.
Regressions are weighted with MORG weights

Demographic characteristic variables show the expected signs. All else equal,
women who did not have a college degree are less likely to participate in the labor force

than those who did. Women with a preschool child have a lower probability of

 

3° I also estimate this speciﬁcation separately for each of the three treatment groups. When the treatment
group includes only the Less than High School group, the estimate of MaxBeneﬁt is statistically
signiﬁcant at standard levels. This might be because the Less than High School group is more likely to
receive welfare beneﬁts than the High School, and Some College groups.

58

participating in the labor force by 7.5 percentage points, relative to those without a
preschool child. The probability of participating in the labor force is 9.0 percentage
points lower for non-white women compared with white women. A thousand dollar
increase in non-labor income decreases the probability of working by 0.6 percentage
points. A one percent decline in the unemployment rate raises the probability of
participating in the labor force by 1.2 percentage points.

To test whether differential decreases in the MTR result in differential changes in
labor supply, in Column (2), I estimate the labor force participation by using
After1994 *MTRchange as a policy variable. MT Rchange is the change in median MTR
(before and after 1994). This variation also captures the differential expansion of EITC
by number of children and state. I ﬁnd that women who faced larger declines in the MTR
have larger labor supply responses, as theory predicts. All else equal, a 10 percent drop in
the MTR after 1994 results in an increase in the labor force participation rate by 2.5
percentage points.
B. The eﬂect of the 1993 El TC expansion on wages

Table 2.6 presents the results for the test of selection bias. The estimate of the
inverse Mills Ratio, lambda, is not statistically signiﬁcant.3 1 This suggests that there is no
evidence of a sample selection problem in estimating the wage equation of women in my

sample. Therefore, I use OLS to estimate the wage equation. 32

 

3' Choices of exclusion restrictions might inﬂuence the results of the test of selection bias. As a
speciﬁcation check, I use other combinations of restrictions and ﬁnd consistent results.
32 With OLS estimation, the wage equation is the same as equation (2), but excludes the lamda term.

59

Table 2.6 Test of Selection Bias

Coef. Std. Err.
lambda -0.041 (0.046)

Data: 1 992-2001 MORG and March CPS.

Table 2.7 Results of Wage Equation
Dependent Variable : log(real wage)

Robust
Coef. Std. Err.

Aﬁer1994 0.016 (0.035)
Afterl994*Less than High School -0.010 (0.041)
Aﬁer1994*High School -0.008 (0.026)
Aﬁerl994*Some College 0.013 (0.028)
MTR 0.019 (0.041)
Less than High School -0.068 *** (0.034)
High School -0.043 *** (0.021)
Some College -0.025 *** (0.012)
Children aged under 5 -0.030 (0.024)
Children aged 6-18 -0.028 (0.018)
Nonwhite -0.077 *** (0.013)
Union Member 0.150 *** (0.014)
Unemployment rate -0.014 *** (0.004)
R-squared 0.25
Number of obs. 9,382

"* Statistically signiﬁcant at 1% level

"”" Statistically significant at 5% level

* Statistically signifith at 10% level

Data: 1992-2001 MORG and March CPS.

The sample is unmarried women aged 25 -55 years old.
All specifications include age, year, and state dummies.
Regressions are weighted with MORG weights.

Table 2.7 reports the results of the wage equation. Consistent with DID estimates
in Table 5, the coefficients of interest on Aﬁer1994 *Education in the Less than High
School and High School groups are negative. However, all coefﬁcients of interest, on
After1994 *Education, are not statistically signiﬁcant, suggesting that women with lower

education did not experience a statistically signiﬁcant wage drop due to their labor supply

60

increase, despite the signiﬁcant increase in the labor force participation among the
groups.

The estimate of MT R is also not statistically signiﬁcant}3 Demographic
characteristic variables have the expected signs. All else equal, women with lower
education receive lower wages. Having children does not have a statistically signiﬁcant
effect on wages. Non-white women receive a 7.7 percent lower wage than white women.
Consistent with the wage premium literature, such as Hirsch and Macpherson (2003), I
ﬁnd that a wage premium for those who are union members is 15.0 percent.34 Lastly, a
one percent decline in the unemployment rate results in a 1.4 percent wage increase.

With no evidence of a statistically signiﬁcant wage drop after the EITC
expansion, among the groups most affected by the EITC expansion, one might suspect
that other policies such as the minimum wage might prevent a wage drop, speciﬁcally,
among those workers with low education. In Table 2.8, I estimate the same wage
equation as in Table 2.7, but also include the log of the state minimum wage (in 2005
dollars) in the regression equation.35 The coefficients of interest are very similar to those
in Table 2.7 and still are not statistically signiﬁcant. The coefﬁcient on the minimum
wage is also not statistically signiﬁcant. I also estimate this speciﬁcation with only
women in the Less than High School and College groups and ﬁnd consistent results. This
might be because among women in my sample who earned positive wages, only 3.75

percent of them (352 out of 9,3 82) earn at or below the minimum wage.36

 

33 As a speciﬁcation check, I lag MTR one year and two years, and ﬁnd consistent results.

34 Using CPS data, Hirsch and Macpherson (2003) ﬁnd that the union membership wage premium for
female workers during the 19908 is between 12. 2 and 15.7 percent.
3sThese minimum wages are maximum values between the state and the federal minimum wage. I would
like to thank Raj Chetty, who kindly provides me the dataset of state and federal minimum wages.

3‘ Only 2. 7 percent of unmarried women aged 25- 54 in 2002 earned at or below the minimum wage
(Bureau of Labor Statistics, U. S. Department of Labor).

61

Table 2.8 Results of Wage Equation
(Including the Minimum Wage)
Dependent Variable : log(real wage)

Robust
Coef. Std. Err.

After] 994 0.014 (0.035)
Log MinWage -0.246 (0.193)
After] 994*Less than High School -0.01 1 (0.042)
Aﬁer1994*High School -0.007 (0.026)
After1994*Some College 0.013 (0.028)
MTR 0.019 (0.041)
Less than High School -0.067 *** (0.034)
High School -0.042 *** (0.021)
Some College -0.025 *** (0.012)
Children aged under 5 -0.03l ** (0.014)
Children aged 6-18 -0.027 *** (0.008)
Nonwhite -0.077 *** (0.013)
Union Member 0.150 *** (0.014)
Unemployment rate -0.015 *** (0.004)
R-squared 0.25

Number of obs. 9382

* ** Statistically significant at 1% level

** Statistically signiﬁcant at 5% level

* Statistically significant at 10% level

Data: 1 992-2001 MORG and March CPS.

The sample is unmarried women aged 25-55 years old.
All specifications include age, year, and state dummies.
Regressions are weighted with MORG weights.

2.6 CONCLUSION

Using the 1992-2001 March and MORG of CPS data, this study examines the
effect of the 1993 EITC expansion on the wages of unmarried women. Accounting for
the sample selection problem and the endogeneity of the EITC, I ﬁnd no evidence of

wage decreases among low skilled workers where there were larger increases in labor

 

force participation. With no statistically signiﬁcant effects on the gross wages, I conclude
that the incidence of the 1993 EITC expansion is captured by the worker.

One possible explanation for ﬁnding no statistically signiﬁcant wage drop is that
unmarried women participate in a labor market where there were other changes that

countered an increase in their labor supply. For example, the EITC expansion also

62

resulted in a decrease in the labor supply of the other groups of workers such as married
women.37 Second, with a positive effect on labor supply decisions, the effect of the EITC
on wages also depends on whether an increase in the aggregate labor supply is large
enough to signiﬁcantly affect the existing equilibrium wage. Using national survey data,
the effect in this study might be attenuated. In future work it will be interesting to
examine the effect using data from more local labor markets with high concentrations of

low-skilled workers. 38

 

37 See Dickert et al. (1995), Ellwood (2000), and Eissa and Hoynes (2004) for the effect of the EITC on the
labor supply of married women.

38 One potential candidate is the SIPP data. In 1996, the SIPP was redesigned to over-sample of households
from areas with high poverty concentrations (the SIPP Users' Guide 2001).

63

CHAPTER 3
EVIDENCE OF THE EARNED INCOME TAX CREDIT MARRIAGE PENALTY
RELIEF
3.1 INTRODUCTION

With an effort to have a progressive tax rate structure and the unit of taxation that
is based on family income, the income tax system in the United States is not marriage
neutral. Tax liability may increase or decrease with marriage. The larger role of the
Earned Income Tax Credit (EITC) in aiding low-income families due to the expansions
over the 19905 increases the relevance of the income tax system for low-income families.
One potentially unintended consequence of the EITC expansions is that it might affect
marriage incentives. Over the same period, there had been a signiﬁcant increase in the
number of partners choosing cohabitation as an alternative to marriage.1 Whether the
EITC is responsible for encouraging cohabitation over marriage is uncertain, the EITC
expansion in 2001 for married couples provides exogenous variation in the marriage
incentives that can be used to investigate the effect of the EITC on the marriage decisions
of cohabitors.

Designed to make the tax system more marriage-neutral by reducing or
eliminating marriage penalties, tax cuts for married couples were enacted in 2001. The
Economic Growth and Tax Relief Reconciliation Act of 2001 (EGTRRA) consists of
three provisions of marriage-penalty-relief including the expansion of the phase-out range
of the EITC for married tax ﬁlers. Relative to the phase-out range for a single ﬁler or

head-of-household, EGTRRA expands the phase-out range for married couple in a

 

1 During the 19908, the number of unmarried partner living together, cohabitation, increased by 72 percent
(U .S. Bureau of the Census 2001).

64

nominal value by $1,000 in 2002, by $2,000 in 2005, and by $3,000 in 2008. In addition,
EGTRAA increases the standard deduction for married ﬁlers to twice the size of the
standard deduction for single ﬁlers and widens of the 15 percent tax bracket for married
ﬁlers. Initially, the tax cut bill was designed to expire at the end of 2010, but they were
made permanent by the 2003 tax cut bill.

In this study, I describe the distribution of the marriage incentives arising from the
EITC expansion in 2001 to examine whether changes in the marriage incentives are
correlated with marriage decisions. I focus on the EITC because under EGTRRA the
EITC expansion is the largest source of marriage penalties and subsidies for low-income
families (Acs and Maag 2005). I use a sample of cohabiting women with children and
low-education from the 2001 Survey of Income and Program Participation (SIPP). 2
Unlike unattached single parents, cohabitors already make a certain level of commitment
to one another, and hence are most likely to be sensitive to tax incentives (Acs and
Nelson 2004). Moreover, I could observe who their partners are, which makes calculating
marriage subsidies/penalties more straightforward. I focus on women with children
because families with children receive the largest EITCs and hence are more likely to be
affected by changes in the credit. I focus on women with no college degree because EITC
eligibility studies suggest that among women who are eligible for the EITC, those with
no college degree are more likely to be eligible for the credit (Dahl and Lochner 2005;
Baughman and Dickert-Conlin forthcoming). I use TAXSIM, a tax simulation model
prepared by the National Bureau of Economic Research (N BER) to calculate each

cohabiting woman’s combined state and federal EITC (in 2006 dollars), given her family

 

2 From Center for Economic and Policy Research. 2006. SIPP Uniform Extracts, Version 2.0. Washington,
DC.

65

income and demographic characteristics in December.3 4 I then simulate a marriage
between the woman and her partner and recalculate the family’s EITC. The difference
between the calculated EITC when cohabiting and when married gives a measure of the
EITC marriage incentive. A woman would receive an EITC marriage subsidy when her
BIT C is higher when married than when cohabiting. On the other hand, she would face an
EITC marriage penalty when her EITC is higher when cohabiting than when married.

I ﬁnd that cohabiting women with children could face a signiﬁcant change in their
EITC if they married. In 2001, 6 percent of the cohabiting women would receive an EITC
marriage subsidy if they married and the average EITC marriage subsidy was $937. On
the other hand, 12 percent would receive a penalty, on average, $978 if they manied. The
remaining 82 percent would face no change in their EITC if they married. When I.inﬂate
all taxable incomes to 2002 dollars and apply the 2002 tax law, all else equal, I ﬁnd that
more women (37 percent) would receive a subsidy. The average subsidy decreases to
$292 because more women receive a smaller subsidy. In 2002, fewer women (9 percent)
would receive a penalty. However, EGTRRA did not signiﬁcantly reduce the average
EITC marriage penalty. In fact, the average penalty increased to $1,240. I also examine
the effect of EGTRRA by education level (Less than high school, High school, and Some
College). And, I ﬁnd that EGTRRA increases the number of women facing a subsidy but
decreases the number of women facing a penalty in all education levels. By race, I ﬁnd
that on average, white and non-white women would face very similar changes in their

EITC if they married.

 

3 See Feenberg and Coutts (1993) for a description of the TAXSIM calculator.

‘ Following Dickert-Conlin and Houser (2002), I estimate annual income for those women who are not in
the sample for the entire year by inﬂating reported income in the months that I observe to reﬂect 12
months.

66

To examine whether changes in the EITC have any behavioral effects on the
marriage decisions of cohabiting women, my empirical method accounts for the
individual ﬁxed effect and the endogeneity of the EITC. Therefore, I use variation in the
changes in the EITC marriage incentives within a person to estimate the correlation
between the EITC and marriage. I ﬁnd no relationship between changes in the EITC
marriage incentives and changes in the marriage decisions of cohabiting women.

Chapter 3 proceeds as follows: Section 3.2 describes the marriage incentives in
the EITC and marriage-penalty-relief provisions under EGTRRA. In section 3.3, I
describe the existing literature on the effects of tax and transfer systems on marriage
decision. Section 3.4 discusses data and descriptive evidence. Section 3.5 discusses
whether EITC marriage incentives are correlated with marriage decisions. And, I

conclude in section 3.6.

3.2 INSTITUTIONAL DETAILS

Marriage incentives in the tax system arise because of the combination of a
progressive tax system and joint, rather than individual, ﬁling by married couples for
beneﬁts and taxes. The EITC expansions over the 19905 are an important contributing
source of changes in marriage incentives for low-income tax ﬁlers (Dickert-Conlin 1999).
EITC eligibility depends on the earning of a tax ﬁling unit. Legal marital status
determines who is included in the tax unit. Married couples must ﬁle a joint return and
unmarried individuals ﬁle either head-of-household or single tax returns, depending on
whether or not they have dependents. With a small credit available to childless tax ﬁlers,

a more generous credit is available to ﬁling units with qualifying children. The credit

67

equals a speciﬁed percentage of earned income up to a maximum dollar amount over the
“phase-in range.” Over a range of income termed the “ﬂat range,” taxpayers receive the
maximum credit. The credit then diminishes to zero over the “phase-out range.” The
EITC is refundable. If a family’s credit amount exceeds their tax liability, the Internal
Revenue Service (IRS) refunds the difference. This feature makes the EITC marriage
incentives particularly relevant to low-income ﬁlers’ marriage decision. The EITC
payment is typically made once a year as an adjustment to tax liabilities or reflmds.5

Previously, the EITC was associated with substantial marriage tax cost (Eissa and
Hoynes 2000). For example, consider a couple with two children in which each parents
earns $15,000. If unmarried, in 2001 the head of household would qualify for $3,605 in
EITC beneﬁts. If the couple married, their combined income, $30,000, would qualify for
$447 in EITC beneﬁts. By marrying, this couple would have lost about 10 percent of
their combined income.

EGTRRA aims directly at reducing marriage penalties and promoting marriage.
The 2001 tax bill contains the following marriage-penalty relief provisions to prevent

most couples from owing more income taxes when married than if single:

0 an increase in the beginning and ending of the EITC phase-out range in a nominal
value by $1,000 in 2002 (by $2,000 in 2005 and by $3,000 in 2008) for married

ﬁlers

0 an increase in the standard deduction for married ﬁlers to twice the size of the

standard deduction for single ﬁlers

o widening of the 15% tax bracket for married ﬁlers

 

5 An “advance payment” option was added in 1978 so that workers would be able, if they so chose, to
receive the credit incrementally throughout the year. However, in 1998, only 1.1 percent of EITC
recipients with children used the advance payment option (Hotz and Scholz 2003).

68

Table 3.1 Marriage Penalty Relief Provisions in EGTRRA

 

 

 

 

 

 

 

 

 

 

Provision 2000 2002-2004 2005-2007 2008-2009
Increase the EITC phase-out range for married couples 2000 levels +$1,000 +$2,000 +$3,000
Increase married standard deductionaspercent of single 167% 200%
Increase size of married 15% bracketaspcrcent of single 167% 200%

 

Source: Table 2 in Burman, Maag, and Rohaly (2002) and Friedman (2004)

Table 3.1 presents the time table of each provision under EGTRRA. The 2003 tax
cut bill made the standard deduction and 15 percent tax bracket provisions fully effective
after 2004. However, it does not provide any acceleration of the EITC expansion. The
EITC expansion will not be fully effective until 2008 (Friedman 2004).

In 2002, the credit for married couples with two or more children starts to phase
out when income reaches $14,520 instead of $13,520. In addition, after 2002 the EITC
explicitly depends on ﬁling status. The expansion of the phase-out range income for
married couples might reduce the EITC marriage penalties for cohabitors if they married
and ﬁled a joint return. As a result, an increase in the credit may increase their marriage
incentives. Using the same family as in the previous example, in 2002 the couple, who
would have qualiﬁed for $669 when married, would qualify for $823 in their EITC
beneﬁts. EGTRAA reduces the marriage penalty for this couple by 23 percent of their

pre-expansion EITC beneﬁts.

3.3 EXISTING LITERATURE
The early works of income tax and the marriage decisions of low-income families
include descriptive evidence to illustrate maniage costs and subsidies and empirical

evidence of the effect of the income tax on marriage. Groeneveld, Tuma and Harman

69

(1980) use data from the negative income tax (NIT) from Seattle and Denver to consider
the effect of participation in a NIT experiment on marital dissolution. They ﬁnd that
participants had higher marital dissolution rates than non-participants. However, Cain and
Wissoker (1990) reanalyze the same data and ﬁnd that the NIT had no effect on the rate
of marital dissolution among participants. Using data from a sample of tax returns, Rosen
(1987) and Feenberg and Rosen (1995) examine the marriage tax under the Tax Reform
Act of 1986 (TRA-86) and the Omnibus Budget Reconciliation Act of 1993 (OBRA-93),
respectively. They ﬁnd that the tax reforms subsidize marriage for low-income families.
However, some low-income families faced much higher marriage tax under OBRA-93
due to the EITC expansion.

More recent studies focus on the behavioral effect of the income tax on marriage
decisions. In general, they ﬁnd that the tax system has small but statistically signiﬁcant
effects on maniage and divorce decisions. For example, Whittington and Ahn (1997) use
longitudinal data from the Panel Study of Income Dynamics (PSID) for the period of
1968 to 1992 to examine the effect of income tax on the likelihood of divorce. They ﬁnd
that the marriage tax penalty results in a small increase in the probability of divorce for
couples. Using the same data, Alm and Whittington (1999) examine the effect of the
income tax on the time to ﬁrst marriage and ﬁnd statistically signiﬁcant but small effects.
See Alm, Dickert-Conlin, and Whittington (1999) for a summary of the income tax and
marriage studies.

Early research on the effect of the transfer system on marriage decisions, for
example, Schulz (1994), ﬁnds that welfare has a signiﬁcantly negative effect on

marriage. More recent studies control for state-ﬁxed effects to correct biased welfare

70

estimates and ﬁnd statistically and economically smaller effects (Mofﬁtt 1994; Hoynes
1997; Schoeni and Blank 2000).

More recent studies focus on the effects of the EITC on marriage. For example,
Dickert-Conlin and Houser (1998) describe the distribution of the joint change in transfer
beneﬁts and tax liability associated with a change in marital status. Using a sample of
representative households from SIPP, they ﬁnd that most poor two-parent families could
signiﬁcantly increase their transfer beneﬁts if they separated, but this gain could be
largely offset by the income tax system. Among unmarried women with children, all poor
mothers could face a loss of transfers if they married, but the tax system mitigates the
decline in transfer beneﬁts. The near-poor could face smaller declines in transfer beneﬁts
but face an increase in tax liability if they married. Holtzblatt and Revelein (2000)
estimate marriage penalties and subsidies under a variety of assumptions. Their preferred
results show that in 2000 the EITC increased total marriage penalties by 10.4 percent and
reduced total subsidies by 1.5 percent. Consistent with Dickert-Conlin and Houser
(1998), they ﬁnd that the EITC results in marriage subsidies for taxpayers with low-
income, but marriage penalties for those with middle-income.

Focusing on the decision to separate, Dickert-Conlin (1999) uses a sample of
women from the 1990 SIPP to examine the behavioral effects of the marriage penalties or
subsidies imposed by the transfer and tax systems on their separation decisions.
Controlling for individual and state ﬁxed effects, Dickert-Conlin ﬁnds that couples with
more to gain from separating in the form of lower 1990 tax liability are more likely to
separate. However, the economic signiﬁcance of the effect is small. Ellwood (2000)

examines the effects of the EITC expansion during the 19908 on marriage decisions.

71

Using data from the 1975-1999 Current Population Survey (CPS) to predict wages and
calculate marriage penalties and bonuses, he ﬁnds that the EITC expansions increased
marriage incentives for low-wage women, but decreased the incentives for higher wage
women. To test whether these marriage penalties and bonuses have had any behavioral
effects, Ellwood compares the marriage rates of low-wage women to those of higher-
wage women and ﬁnds no evidence of differential increases in marriage rates of those
women. Creating a consistently deﬁned series on cohabitation from the CPS, Ellwood
also examines whether the EITC marriage incentives inﬂuence cohabiting decisions. He
ﬁnds an increase in the marriage rates of cohabiting couples who faced marriage
subsidies. However, the timing of the increase in the marriages rates is not closely linked
to the EITC expansions. Therefore, Ellwood suggests that these results should be treated
cautiously.

Another empirical study, Dickert-Conlin and Houser (2002) use data from the
1990-1993 panels of the SIPP to examine whether the EITC is correlated with changes in
marital status for families with children. Controlling for the individual ﬁxed effect and
the endogeneity of the EITC, they ﬁnd that married women with children who face larger
increases in their EITC are less likely to remain married. However, the economic effect is
small. They also ﬁnd that the EITC has no statistically signiﬁcant effects on the marriage
decisions of unmarried women. Using data from the PSID for the period of 1983 to 1997,
Alm and Whittington (2003) examine the effect of the federal income tax on decisions to
’ marry instead of cohabit. They ﬁnd that among cohabiting couples, the income tax has a
statistically signiﬁcant effect on their decision to make a transition from a cohabiting to a

married couple. However, the effect is small. Moreover, they ﬁnd that among partners

72

who are not living together, the effect on their decision to form either a cohabiting or a
married union is not statistically signiﬁcant. Eissa and Hoynes (2003) use data from the
1985-1998 March CPS to examine the effect of the tax and transfer systems on marriage
decisions. They ﬁnd that both systems do affect marriage behavior, but the overall effect
is relatively modest. Raising the marriage tax cost, including the EITC, by $1,000 would
lower the probability of marriage by 1.3 percentage points. When incorporating transfer
beneﬁts, the probability declines by 2.4 to 3.3 percentage points for every $1,000
increase in marriage costs.

More recent studies consider the effects of the EITC expansion for married
couples in 2001. All of them focus on the describing of the size of the relief. For
example, using a sample of low-income cohabiting couples with children from the 2002
National Survey of America’s Families (N SAP ), Acs and Maag (2005) calculate tax or
tax and transfer? marriage subsidies and penalties faced by those couples in 2003 and
2008. Results from their simulation model show that the EITC is the largest source of
marriage penalties and subsidies. In 2008, when the EGTRAA will be fully phased-in,
those couples who face marriage penalties in 2003, on average, will face a 13 percent
decrease in their penalties, while those who face marriage subsidies, on average, will face
a 58 percent increase in subsidies.

This paper extends the existing literature by using a sample of cohabiting women
from SIPP to describe the distribution of the changes in the EITC marriage incentives
associated with the EITC expansion in 2001. Moreover, I examine the behavioral effects
of the EITC marriage incentives on their marriage decisions. To my knowledge, no

previous studies examine the behavioral effects of this EITC expansion.

73

3.4 DATA AND DESCRIPTIVE EVIDENCE

I use a sample of women with children and no college degree who were
cohabiting in 2001 from the 2001 panel of SIPP. The SIPP is a multi—panel, nationally
representative dataset. The SIPP divides respondents into four staggered rotation groups.
Respondents are asked questions every fourth month about their experiences over the
prior four months. SIPP respondents are asked questions about their participation in
income maintenance programs, their household and family composition, employment and
earnings. The 2001 SIPP tracks individuals for three years and contains nine waves
(October 2000-December 2003). I include observations on women in December of each
year so that I can identify their tax ﬁling status. Therefore, I have a maximum of three
observations per person.

Cohabitors are explicitly identiﬁed because respondents have to answer: “What is
the EXACT relationship of (household member) to (household member)?” and
“(household member) is (household member)?” This variable is constant over the entire
panel. The cohabitor question was asked in the Topical Module under the second wave of
the 2001 panel.6 I identify whether women have children by using the number of children
under 18 in the family and assume that children are the women’s.7 Consistent with the
literature, I limit my sample to women aged 18-50 to focus on women who are most
likely to make marriage decisions. Because I account for state variation in EITCs and
welfare policies, I drop observations from those states that are not uniquely identiﬁed in

the SIPP.8 The resulting sample is 299 cohabiting unmarried women, representing

 

6 There are thirty-one possible responses, including Spouse, Unmarried Partner, and Room/housemate.
7 This is reasonable assumption for my sample. 96 percent of children are women’s children.
8 The 2001 SIPP aggregates Maine, Vermont, Wyoming, North Dakota, and South Dakota into two groups.

74

906,980 persons using the SIPP sampling weights, with 777 person year observations
(2.6 years per person on average).

To demonstrate how marriage affects the EITC received by cohabiting women, I
compare the EITC received by the woman when she ﬁles a tax return as head-of-
household with what would be received if she and her partner married and ﬁled a joint
return. The woman receives an EITC marriage subsidy if she receives a higher EITC
when she married. On the other hand, she receives an EITC marriage penalty if she
receives a lower EITC when she married. To analyze how EGTRAA affects the EITC
marriage subsidy/penalty, I use the sample of cohabiting women in 2001, inﬂate their
income to 2002 and 2005 dollars, and calculate their EITC marriage subsidies/penalties
in 2002 and 2005. Because I use the same sample and hold all other behaviors constant to
calculate those women’s EITC, changes in their EITC marriage subsidies/penalties are

due to changes in the tax law.

75

Figure 3.1 Weighted Number of Cohabiting Women Facing an EITC Marriage
subsidy/penalty in 2001, 2002, and 2005 (N=299; weighted N=906,980)

 

 

   

 

 

 

 

 

 

   

 

l f" "—— '
l 2001 20.2
portion) 1
107,743(12%) ISM-[Mam l
,. DSullsidyifMa‘ried ‘
amazon.) (N: 112) ,
DPellaltylfMll'ied ‘
.. l
silicon.) , (N=26) ‘l
l DNocllalgeiiMancd .
[(11:10) ,7 l
19mm.) ‘
749,169(82°/o) 1
l
2005 .
DTSubsilyifManied 7
j; 354,440, 39% (N=117) H
BPenaltyifMarricd
490,352, 54% (N‘Zl)
MDNocllllngifMarrledH
L(N=lsl)
62,188, 7%
_ __, J

 

 

Figure 3.1 shows the percentages and weighted numbers and numbers of
cohabiting women facing an EITC marriage subsidy and penalty in 2001, 2002, and
2005. Among 906,980 women, in 2001, 6 percent would receive an EITC marriage

subsidy if they married, while 12 percent would receive an EITC marriage penalty. The

76

remaining 82 percent would face no change in their EITC.9 In 2002, when changes to the
EITC for married couples went into effect, the number of women, who would receive a
subsidy if married, increases to 37 percent, while the number of women, who would face
a penalty, decreases to 9 percent. In 2005, the distribution of women facing an EITC
marriage subsidy/penalty does not change signiﬁcantly from that in 2002. The number of
women, who would receive an EITC marriage subsidy slightly increases to 39 percent,
while the number of those who would receive an EITC marriage penalty decrease to 7
percent. One possible explanation why the distribution changes signiﬁcantly from 2001
to 2002 but not from 2002 to 2005 is that the inﬂation rate that I apply to those women’s
income is larger than the inﬂation-adjusted rate for the EITC thresholds.10 As a result, in
2005 some women are not eligible for the EITC because they have too high income.
With a small sample size of women facing an EITC maniage penalty as shown in
Figure 3.1, the estimate of a marriage penalty might not be reliable. In further discussion,

I focus on an EITC marriage subsidy.

 

9 Among women who would face no change in their EITC if they married, 25 percent have zero family-
income. 58 percent have too high income to be eligible for the EITC. And 17 percent are eligible for the
same amount of credit.

'0 I inﬂate those incomes by using the all-urban consumer price index for that calendar year from
Department of Labor, Bureau of Labor Statistics. The IRS also uses the all-urban consumer price index
but they deﬁne the consumer price index for any calendar year as the average of the consumer price
index as of the close of the twelve-month period ending on September 30 of that calendar year (lntemal
Revenue Bulletin 2007-4).

77

Figure 3.2

Weighted Means of the El'l‘ C Manhge Subsidy
Sample: Conditional on Women with an EITC Subsidy

 

1200 346,692 *“ “““““““““ 354,440 “ 400,000 .
1000 350.000
300.000
300 250,000 g
a 600 200,000 2
400 150.000 =-
100,000
200 , 50,000

 

2001 2002 2005

 

[h Weighted EITC Subsidy + Weighted Number ofwow l

The EITC subsidy confronting cohabiting women is substantial. Using SIPP
weights, Figure 3.2 shows that the expansions of the phase-out income increase the
number of women facing an EITC marriage subsidy from 50,069 women in 2001 to
346,692 in 2002 and 354,440 in 2005. However, on average, the subsidy decreases from
$968 in 2001 to $298 in 2002 and $460 in 2005, respectively. These decreases occur

because more women would receive a subsidy but in smaller amounts.

78

Figure 3.3

 

Weighted Means of the EITC Marriage Subsidy by Education
Sample: Conditional on Women with an EITC Subsidy

 

 

 

2001 2002 2005

Year

Figure 3.4

 

 

Weighted Number of Women Facing an EITC Marriage Subsidy

 

160,000
131,177 136,544
140,000 ,, g 1 135,187

. _ 5 4
120,000 .. 2
100,000
80,000 ier
60,000 »# ~
40,000 -
20,000 -
0-

 

   
   
 
  

 

 

  
    

  

20,103 18,565 “399

f::;lj:_'::$:ig3§:

 

 

2001 2002 2005

 

Riggs plan ﬁgment in; School Somem

79

 

For ﬁrrther investigation, I calculate an EITC marriage subsidy by the education
levels of those women.11 Figure 3.3 suggests that in 2001 on average, women with a high
school education would receive the largest EITC subsidy at $1,479. After the EITC
expansion in 2001, the average subsidies decrease for all education levels. However, in
2002 and 2005, more women in all education levels would receive a marriage subsidy
relative to in 2001, as shown in Figure 3.4.

Figure 3.5 presents the EITC marriage subsidy and penalty by race. On average,
white and non-white women would face very similar penalties in 2001 and subsidies in
2002 and 2005. Conditional on the same education level and number of children, Figure
3.6 shows that white cohabiting women with a less than high school education and two or
more children would face signiﬁcant larger EITC subsidies if they married. This is likely
to be driven by the fact that white and non-white women’s partners have different

distribution of earnings.

 

” As discussed earlier, I focus on women without a college degree because previous studies suggest that
they are more likely to be eligible for the EITC.

80

Figure 3.5

,-_____7

Weighted Means of the EITC Marriage Incentive by Race

 

 

 

 

 

 

 

 

 

100
50 7 7 20 19—: _
a 0* 7 ﬂ
-50 “
-100
2001 2002 2005
Year

 

Figure 3.6

 

Weighted Means of the EITC Marriage Incentive by Race I
Sample: Cohabiting women with Less than High School Education .
and Two or more Children l

 

250 f
200

 

 

 

69100"

 

 

 

 

 

 

 

 

 

 

2001 2002 2005

 

lgill/bite : Two Plus Cl Non-White: Two Plus 7

81

3.5 ARE EITC MARRIAGE INCENTIVES CORRELATED WITH MARRIAGE
DECISIONS?

The descriptive evidence in the previous section suggests that there is signiﬁcant
cross-time variation in the EITC marriage incentives due to the EITC expansion. This
variation might be useful to estimate the correlation between changes in the EITC
marriage incentives and changes in marital status for cohabiting women with children. In
this section, I use variation the EITC maniage incentives over time to examine whether
the EITC marriage incentives are correlated with the marriage decision of cohabiting
women.

Because marriage is a choice, the marriage decision can be modeled using
standard preference theory. To investigate whether changes in the EITC marriage
incentives are correlated with changes in the marital status of cohabiting women, I test
the hypothesis based on Becker’s (1973, 1974) model of marriage. The model suggests
that individuals marry if the expected utility after getting married exceeds that of
remaining single. To consider the determinants of the choice of marriage, Becker deﬁnes
the utility function as
U=U(M, 2“, T”, X);

Where M is marital status, Z is a measure of household output that depends on marital
status, T is the tax-transfer consequence of marriage, and X are individual characteristics.
Assuming that the indirect utility function is linear, I can deﬁne M,(*, the difference in the
maximal utility between married and not married as:

Mn" = (I + YIEITCit + Xit'Yz + Yt'l’s + Dii'

Mn” is unobservable, however, we can observe a woman’s marital status, Mn:

82

 

M,, = I (married) if Mit*>0
= 0 (unmarried) if Mit*50.

EITC“ is the EITC marriage incentive, which is the change in combined federal and state
EITC for individual i in year t if she would marry. A Negative number means that a
woman would face an EITC marriage penalty if she married. y, is a set of three year
dummies to account for any changes, affecting all women in that particular year, for
example, federal government policies. on is the error term. To control for differences in
population sampled over time, the regression equation also includes, X“, a set of
demographic characteristic variables, which includes age, two dummies for education
levels (high school and some college), and a dummy for whether the woman lives in an
urban area. To account for state-year characteristics that might be correlated with
marriage decisions, Xit also includes the sum of maximum beneﬁts from AF DC/T AN F
and food stamp programs (in hundreds of 2006 dollars) to capture the generosity of the
welfare system, the real per capita income and the unemployment rate to capture the
business cycles.

Table 3.2 presents descriptive statistics for my sample. Initially being unmarried,
8 percent got married by the third year in the sample. In 2001, the average EITC for these
women is $1,412. And the average EITC marriage incentive is negative, suggesting that
on average cohabiting women in my sample would face a marriage penalty at $42.18 if

they married.

83

Table 3.2

DESCRIPTIVE STATISTICS
MEANS AND (STANDARD DEVIATIONS)

 

Married 0.08 (0.27)
EITC 1412.30 (1456.40)
EITC Marriage Incentive -42.18 (632.42)
Age (years) 31.93 (7.55)
Number of Children 1.96 (1.11)
Non-White 0.43 (0.49)
Less than High School 0.27 (0.44)
High School 0.34 (0.47)
Some College 0.38 (0.48)
Urban 0.53 (0.49)
State Unemployment Rate 5.04 (1.08)

 

Sample: Women with children and no college degree who were cohabiting in 2001
from the 2001 panel of SIPP. Means are weighted with SIPP weights.

Two concerns arise in my empirical method. First, because the EITC beneﬁts
depend on marital status and family earnings, but family earnings depend on marital
status, these correlations potentially lead to the endogeneity problem, resulting in biased
and inconsistent estimators (Wooldridge 2006). To mitigate this potential endogeneity
problem, I isolate the effects of the EITC expansion from the effect of changes in the
characteristics and earnings. I use each woman’s family demographic and income

characteristics from the ﬁrst year she is in the sample, adjust her income with the

 

consumer price index, and then calculate her EITC in following years, holding other i i
behaviors constant. '2 This method provides a proxy of policy variable that is correlated

with actual changes in the EITC but exogenous to the marriage decision.

 

'2 This approach is used in several tax studies, for example, Dickert-Conlin and Houser (2002) and Kubik
(2004).

84

Second, although I observe a number of variables that are likely to affect the
marriage decision, unobserved individual characteristics, such as attitudes toward
cohabitation and marriage, are also likely to inﬂuence the marriage decision. Using
Ordinary Least Squares (OLS) to estimate longitudinal data without accounting for
unobserved effects results in biased and inconsistent estimators.

To account for the endogeneity of the EITC and individual ﬁxed effects, I
estimate the following empirical model:

Mn" = 0i + YIEITCsl 'l’ Xit'Yz ‘l' yt'73 + ”it,

where at is the individual ﬁxed effect, which captures all unobserved, time-constant
factors that affect marriage, such as attitudes toward marriage. EITCS, is a proxy variable
that is correlated with actual changes in the EITC but exogenous to the marriage decision.
The coefﬁcient of interest, 71, measures the effect of changes in the EITC marriage

incentive on the probability of marriage within individuals.

85

Table 3.3

LINEAR PROBABILITY MODEL FOR ALL WOMEN WITH CHILDREN WHO ARE COHABIIING IN 2001

 

 

DEPENDENT VARIABLE: MARRIED
OLS-Actual OLS-Proxy Fixed Effect-Actual Fixed Effect-Proxy
(1) (2) (3) (4)
Robust Robust Robust Robust

Coef. Std Ell. Coef. Std Err. Coef. Std Err. Coef. Std FJT.
EllC(100) 0.01711“ (0.00401) 0.00202 (0.00242) 0.01383 *** (0.00398) 0.01151 (0.01792)
Age 0.00243 ** (0.00111) 0.00251 (0.00163) 0.03485 ** (0.01616) 0.03088 *** (0.00924)
High School (l=yes) 0.02853 (0.02043) 0.03608 (0.02264) 0.01437 (0.09994) 0.03285 (0.10177)
Some College (l=ycs) 0.02142 (0.02032) 0.02712 (0.02228) 0.01147 (0.14479) 0.00086 (0.17479)
Urban(1=yes) 0.00550 (0.01809) 0.00687 (0.02002) 0.01999 (0.12325) 0.05934 (0.14309)
StateMaximumAFDC+Food
Stalnp(100) 0.00443 0 (0.00113) 0.00839*** (0.00231) 0.04012*** (0.01092) 0.05000“l (001109)
State Real Per Capita Income 0.00001 (0.00002) 0.00001 (0.00001) 0.00001 (0.00002) 0.00001 (0.00002)
Stale Unemployment Rate 0.01781 (0.01158) 0.03435 ** (0.01412) 0.01104 (0.03457) 0.01709 (0.03629)
2001(l=yet) 0.08426 *** (0.02032) 0.09981*** (0.01998) 0.03230 (0.01982) 0.03276 (0.01936)
2002(l=yes) 0.16725 *** (0.03294) 0.21213 *** (0.03586) 0.02520 (0.02815) 0.01666 (0.03287)
2003(1=yes) 0.13591*** (0.04309) 0.19831*** (0.05954) (dropped) (dropped)

 

Sample: Women with children and no college degree who were cohabiting 102001 hon 11102001 panel 01 SIPP
Number of observation : 299 women with 777 person your observations

“i Statistically signiﬁcant at 7% level

“ Statistically signiﬁcant at 5% level

‘ Slalisticnllysigmﬁcanlal 10% level

Table 3.3 presents results for the linear probability model for a sample of women
with children and no college degree who were cohabiting in 2001. Column 1 shows the
results when I use OLS estimation and the estimate of the actual changes in the EITC as a
policy variable. This speciﬁcation does not account for individual effects or the
endogeneity of the EITC. The coefficient of interest on the EITC variable has an
unexpected sign, it is negative and statistically signiﬁcant at the 1% level. However, the

estimate is very small (-0.01711), suggesting that all else equal, a $100 increase in the

86

EITC marriage incentive lowers the probability of being married by 1.7 percentage point.
At the mean of the probability of being married (8/ 100) and the average EITC marriage
incentive (-$42.18), this coefﬁcient translates into an elasticity of -0.09.

In Column 2, when I use a proxy variable for the actual EITC, the coefﬁcient of
interest on the EITC variable is also negative, but becomes less economically and
statistically signiﬁcant. At the mean the estimated elasticity is -0.0l. Some of the
coefﬁcients on covariates also have unexpected signs. All else equal, age is negatively
correlated with marriage at statistically signiﬁcant levels. On the other hand, education
and welfare beneﬁts are positively correlated with marriage. The unemployment rate is
negatively correlated with marriage at statistically signiﬁcant levels.

Accounting for individual ﬁxed effects and using the estimate of the actual EITC
as the policy variable, the coefﬁcient of interest in Column 3 is slightly greater than those
in Column 1. At the mean the estimated elasticity is -0.07. The coefﬁcients on covariates
have the expected signs. Age is positively correlated with marriage at statistically
signiﬁcant levels. The coefficients on education variables are negative, but are not
statistically different from zero.13 Consistent with Dickert-Conlin and Houser (2002), the
coefﬁcient of welfare beneﬁts is negative and statistically signiﬁcant, suggesting that an
increase welfare generosity reduces cohabiting women’s incentive to marry.

In Column 4, when I control for individual ﬁxed effects and use a proxy variable
for the actual EITC, the coefﬁcient of interest on the EITC is not statistically signiﬁcant,
consistent with Ellwood (2000) and Dickert-Conlin and Houser (2002). This suggests

that, controlling for individual ﬁxed effects and the endogeneity of the EITC, all else

 

'3 This might be because there is not much variation in education variables. Among 299 women in my
sample, only 25 women attained more education.

87

equal, the EITC marriage subsidy does not affect the marriage decisions of cohabiting

women.

For further investigation, I also estimate similar regression equations on three

subsarnples of women with children who were cohabiting in 2001: those with less than

high school, high school and some college education.14 Results are shown in Table 3.4. In

this discussion, I focus on results in Column 4. Consistent with results in Table 3.3, the

coefficients of interest in all panels are not statistically signiﬁcant at standard levels,

suggesting that there is no statistically signiﬁcant relationship between the EITC and

 

 

marriage.
Table 3.4
LINEAR PROBABILHV MODEL FOR COHABITING WOMEN WITH CHILDREN BY EDUCATION LEVEL
DEPENDENT VARIABLE: MARRIED
OLS-Actual OLS-Proxy Fixed Effect-Actual Fixed Effect-Proxy
(I) (2) (31 (4)
Robust Robust Robust Robust
Coef. Std Err. Coef. Std Err. Coef. Std. Err. Coef. Std Err.
Panel A: Less than High School
ETTC(100) 0.00737 (0.00882) 0.00363 (0.00245) 0.00956 (0.00810) 0.11110 (0.14730)
Panel B: Ihgh School
EITC(100) 0.02000 *** (0.00497) 0.00244 (0.00280) 0.01566 *** (0.00522) 0.01662 (0.02786)
Panel C: Some College
EITC(100) 0.02051 *** (0.00859) 0.00114 01.00672) 0.00990 (0.00824) 0.00958 (0.02369)

 

”’ Statistically signiﬁcant at 1% (ml
” Statistically signiﬁcant at 5% (ml
‘ Statistically signrﬁcantat 10% Incl

 

'4 In these regression equations because I estimate them by education level, the independent variables are
the same as the equation in Table 3, but exclude the education variables.

88

3.6 CONCLUSION

Using a sample of women with children who were cohabiting in 2001 from the
2001 SIPP, this study describes the distribution of the marriage incentives arising from
the EITC expansion in 2001 for married couples. Descriptive results suggest that the
EITC expansion substantially increases the number of women facing an EITC marriage
subsidy, but marginally reduces the number of women facing a penalty. Among those
facing a subsidy, the average subsidy decreases signiﬁcantly after the EITC expansion,
because more women receive a smaller subsidy.

Using cross-time variation in the EITC marriage incentives within a person, I
examine whether changes in the marriage incentives are correlated with maniage
decisions. In my empirical method, I account for the endogeneity of the EITC and
individual ﬁxed effects. I ﬁnd that the EITC expansion in 2001 has no statistically
signiﬁcant effect on the marriage decision.

One possible explanation for my ﬁnding is that an increase in the tax incentive
due to the EITC expansion might not be signiﬁcant enough to encourage those cohabiting
couples to get married. In an effort to relieve marriage penalty in the income-tax system,
EGTRRA contains three marriage penalty relief provisions for low-, middle-, and high-
income families. While provisions for middle- and high-income families were made to
become fully effective after 2004, the EITC expansion in 2001 for low-income couples

was left behind to be gradually phasing in and will not be ﬁally effective until 2008.

89

APPENDIX

Table A.1: Federal Earned Income Tax Credit Parameters, 1990-2000

Year Kids Phase-In Phase-In Maximum Phase-Out Phase-Out Range
Rate (%) Range (3) Credit (3) Rate (%) (S)
1990 1+ 14 0-6.810 953 10 10,730-20,264
1991 1 16.7 0-7,140 1,192 11.93 11,250-21,250
2+ 17.3 0-7, 140 1.235 12.36 11250-21250
1992 1 17.6 0-7.520 1,324 12.57 ll,840-22,370
2+ 18.4 0-7,520 1,384 13.14 11,840-22,370
1993 1 18.5 0-7,750 1,434 13.21 12,200-23,050
2+ 19.5 0-7.750 1,511 13.93 12,200-23,050
1994 0 7.65 04000 306 7 .65 5,000-9,000
I 26.3 0-7,750 2,038 15.98 1 l,000-23,755
2+ 30 0-8.425 2,528 17.68 11,000-25,299
1995 0 7.65 0-4,100 314 7.65 5,130-9,23O
l 34 0-6,160 2,094 15,98 1 1,290-24,396
2+ 36 0-8.640 3,1 10 20.22 1 1,290-26,673
1996 0 7.65 0-4,220 323 7.65 5,280-9,500
1 34 0-6,330 2,152 15,98 11,610-25,078
2+ 40 0-8,890 3,556 21.06 11,610-28,495
1997 0 7.65 0-4,340 332 7.65 5,430-9,770
1 34 O-6.500 2,210 15.98 ll,930-25,76O
2+ 40 0-9,14O 3,656 21.06 1 l,930-29,29O
1998 0 7.65 0-4,460 341 7.65 5,570-10,03O
1 34 0-6,680 2,271 15.98 12,260-26,473
2+ 40 0-9,390 3,756 2 1.06 12,260—30,095
1999 O 7.65 0-4,530 347 7.65 5,670-10,200
1 34 0-6,800 2,312 15.98 12,460-26,928
2+ 40 0-9,54O 3.816 21.06 12,460-30,580
2000 O 7.65 0-4.610 353 7.65 5,770-10,380
1 34 0-6,920 2,353 15.98 12,690-27,415
2+ 40 0-9.720 3,888 21.06 12,690-31,152

Sources: Baughman and Dickert-Conlin (forthcoming) and The House Ways and Means
Committee Green Book (2004), P13-3 7-13-38

90

 

 

 

 

 

Figure A.l: Labor force participation of unmarried women between 1991
and 2000

 

1.00
0.90

 

0.80 —

 

 

0.70

 

 

 

0.60
0.50

 

0.40

 

 

0.30 -__.

 

0.20
0.10

 

 

 

0.00 f 1 1 l T T r r r
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

Year

 

 

P - — No Children -I- One Child -I-—Two or more children I

 

 

 

 

Figure A.2: Labor force participation of Unmarried women with two
or more children

1.00

 

 

 

0.90

 

0.80 —

0.70 . /
0.60 /
0.50

0.40 W

 

 

 

 

 

 

 

 

 

0.30 p , p e . a , . ,
1991 1992 1993 1994 1995 1996 1997 199a 1999 2000

Year

 

 

 

[+— All + Less than High School— — High School ->"- Some College+ College

 

91

 

 

 

 

 

 

 

 

Figure A.3: Labor force participation of Unmarried women with one

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

, child
1.00
0.90
0.80 v
0.70
0.60 W
0.50
0.40 7 7 ' T Y r r 7
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Year .
(+ All —-— Less than High School — — High School «1+»— Some College +College __ I ((
Figure A.4: Labor force participation of Unmarried women with no
chidldren
1.00
0.90
0.70
0.60 W
0.50
0.40 , . e . 1 a . . 1
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Year
1—9— All -I— Less than High School A High School 2g... Some College -)K-— College J

 

 

 

 

92

BIBLIOGRAPHY

Acs, Gregory and Elaine Maag. “Irreconcilable Differences? The Conﬂict between
Marriage Promotion Initiatives for Cohabiting Couples with Children and
Marriage Penalties in Tax and Transfer Systems.” Assessing the New Federalism
Policy Brief No. B-66. Washington, DC: The Urban Institute 2005.

Acs, Gregory and Sandi Nelson. “What “1 do” 3 do? Potential Beneﬁts of Marriage for
Cohabiting Couples with Children.” Assessing the New Federalism Policy Brief
No. B-59. Washington, DC: The Urban Institute 2004.

Adireksombat, Kampon. 2007. “The Effect of the 1993 Earned Income Tax Credit
Expansion on the Labor Supply of Unmarried Women.” unpublished Ph.D.
Dissertation, Michigan State University.

Alm, James, and Leslie Whittington. “For Love or Money? The Impact of Income Taxes
on Marriage.” Economica 66 (1999): 297—3 16.

Alm, James, and Leslie Whittington. “Shacking Up or Shelling Out: Income Taxes,
Marriage, and Cohabitation.” Review of Economics of the Household 1 (2003):
169—1 86.

Alm, James, Stacy Dickert—Conlin, and Leslie Whittington. “Policy Watch: The Marriage
Penalty.” Journal of Economic Perspectives 13 (1999): 193—204.

Baughman, Reagan and Stacy Dickert-Conlin. Forthcoming. “The Earned Income Tax
Credit and Fertility. Journal of Population Economics.

Becker, Gary. “A Theory of Marriage: Part1.” Journal of Political Economy 81 (1973):
813—46.

Becker, Gary. “A Theory of Marriage: Part 11.” Journal of Political Economy 82 (1974):
511—26.

Bingley, Paul and Gauthier Lanot. 2002. “The Incidence of Income Tax on Wages and
Labor Supply.” Journal of Public Economics 83: 173-194.

Blank, Rebecca M. 2002. “Evaluating Welfare Reform in the US.” Journal of Economic
Literature 40: 1105—1166.

Cain, Glen G., and Douglas A. Wissoker. “A Reanalysis of Marital Stability in the

Seattle—Denver Income—Maintenance Experiment.” The American Journal of
Sociology 95 (1990): 1235—69.

93

Cogan, John, 1980, “Married Women’s Labor Supply: A Comparison of Alternative
Estimation Procedures, in Smith, editor, Female Labor Supply: Theory and
Estimation, 113.

Dahl, Gordon and Lance Lochner. 2005. “The Impact of Family Income on Child
Achievement: Evidence from the Earned Income Tax Credit” Working paper,
University of Rochester.

Darragh, Scott B., 2002, “The Impact of the Earned Income Tax Credit on Poverty, Labor
Supply, and Human-Capital Accumulation”, Dissertation for the Degree of Ph.D.,
Michigan State University.

Dickert, Stacy, Scott Houser, and John Karl Scholz. 1995. “The Earned Income Tax
Credit and Transfer Programs: A Study of Labor Market and Program
Participation.” In Tax Policy and the Economy, vol. 9, edited by James M.
Poterba, 1-50. Cambridge, MA: MIT Press.

Dickert-Conlin, Stacey and Scott Houser. 2002. “The EITC and Marriage.” National Tax
Journal 55:25—40.

Dickert-Conlin, Stacy and Scott Houser. “Taxes and Transfers: A New Look at the
Marriage Penalty.” National Tax Journal 51 (1998): 175-218.

Dickert-Conlin, Stacy. “Taxes and Transfers: Their Effects on the Decision to end a
Marriage.” Journal of Public Economics 73 (1999): 217-240.

Eissa, Nada and Hilary Hoynes. “Explaining the Fall and Rise in the Tax Cost of
Marriage: the Effect of Tax Laws and Demographic Trends, 1894-97.” National
Tax Journal 53 (2000): 683-712.

Eissa, Nada and Hilary Hoynes. “Good News for Low-Income Families” Tax-Transfer
Schemes, and Marriage.” Unpublished Manuscript. Berkeley, CA: University of
California, Berkeley, 2003.

Eissa, Nada and Hilary Williamson Hoynes, 2006, “Behavioral Responses to Taxes”, Tax
Policy and the Economy, James M. Poterba (ed.), National Bureau of Economic
Research and the MIT Press, 20, 73-110.

Eissa, Nada and Hilary Williamson Hoynes. 2004. “Taxes and the Labor Market
Participation of Married Couples: the Earned Income Tax Credit.” Journal of
Public Economics 88: 1931-1958.

Eissa, Nada and Jeffrey B. Liebman. 1996. “Labor Supply Response to the Earned
Income Tax Credit.” Quarterly Journal of Economics 111: 605-637.

94

Ellwood, David T. 2000. “The Impact of the Earned Income Tax Credit and Social Policy
Reforms on Work, Marriage, and Living Arrangements.” National Tax Journal
53(4): 1063-1105.

F eenberg, D. R., and E. Coutts. 1993. “An Introduction to the TAXSIM Model.” Journal
of Policy Analysis and Management 12: 189-194.

Feenberg, Daniel and Harvey Rosen. “Recent Developments in the Marriage Tax.”
National Tax Journal 48 (1995): 91-101.

Freidman, Joel. “Revisited House Tax-Cut Bill Swells Deﬁcit and Continues Less

Favorable Treatment of Lower-Income Couples.” Center on Budget and Policy
Priorities (2004).

Fullerton, D. and G. Metcalf. 2002. “Tax Incidence”. In Handbook of Public Economics,
edited by AJ. Auerbach and M. F eldstein, Oxford, Elsevier.

Groeneveld, Lyle R, Nancy Brandon Tuma, and Michael T. Harman. “The Effects of
Negative Income Tax Programs on Marital Dissolution.” Journal of Human
Resources 15 ( 1980): 654—74.

Grogger, Jeffrey. 2003. “The Effects of Time Limits, The EITC, and Other Policy
Changes on Welfare Use, Work, and Income Among Female-Headed Families.”
Review of Economics and Statistics 85(2): 394-408.

Grogger, Jeffrey. 2005. “Welfare Reform, Return to Experience, and Wages: Using
Reservation Wages to Account for Sample Selection Bias.” National Bureau of
Economic Research Working Paper 11621.

Gruber, J. 1994. “The Incidence of Mandated Maternity Beneﬁts.” American Economic
Review 84 (3): 622-641.

Gruber, J. 1997. "The Incidence of Payroll Taxation: Evidence from Chile." Journal of
Labor Economics 15: 72-101.

Gruber, J., and Alan B. Krueger. 1991. “The Incidence of Mandated Employer-Provided
Insurance: Lessons from Workers Compensation Insurance.” In Tax Policy and
the Economy, vol. 5, edited by Bradford, Cambridge, MA: MIT Press.

Hausman, J. A., and DA. Wise. 1977. “Social Experimentation, Truncated Distributions
and Efﬁcient Estimation.” Econometrica 45: 319-339.

Heim, Bradley T., 2005, “The Impact of The Earned Income Tax Credit on The Labor

Supply of Married couples: Structural Estimation and Business Cycle
Interactions”,Unpublished manuscript, Duke University.

95

Hirsch, T. Barry, and David A. Macpherson. 2003. “Union Membership and Earnings
Data Book: Compilations from the Current Population Survey (2003 Edition)”
Washington, DC, Bureau of National Affairs.

Holtzblatt, Janet and Robert Rebelein. “Measuring the Effect of the Earned Income Tax
Credit on Marriage Penalties and Bonuses.” National Tax Journal 53 (2000):
1107-1134.

Hotz, Joseph V. and John Karl Scholz. 2003. “The Earned Income Tax Credit.” in
Means- Tested Transfer Programs in the United States, edited by Robert Mofﬁtt,
Chicago: University of Chicago Press and National Bureau of Economic
Research.

Hotz, Joseph V., Mullin, Charles H., and John Karl Scholz. 2006. “Examining The Effect
of The Earned Income Tax Credit on The Labor Market Participation of Families
on Welfare.” National Bureau of Economic Research, Working Paper 11968.

House Ways and Means Committee Green Book., 2004, P13-37 -13-38.
http://frwebgate.accessgpo.gov/cgi-
bin/getdoc.cgi?dbname=108 greenibook&docid=f:wm006 13.ndf.

Hoynes, Hilary Williamson. “Does Welfare Play Any Role in Female Headship
Decisions?” Journal of Public Economics65 (1997): 89—117.

Internal Revenue Service, 2005b. EITC Threshold and Tax Law Updates.
(http://www.irs.gov/individuals/articIe/0,,id=1 505 13,00.html)

Johnson, Nicholas. 2000. “A Hand Up: How State Earned Income Tax Credits Help
Working Families Escape Poverty in 2000.” Center on Budget and Poverty
Priorities, Washington, DC.

Kaestner, Robert and Kaushal, Neeraj, 2005, “Immigrant and Native Responses to
Welfare Reform”, Journal of Population Economics, 18, 69-92.

Keane, Michael and Robert Mofﬁtt, 1998, “A Structural Model of Multiple Welfare

Program Participation and Labor Supply”, International Economic Review, 39(3),
553-589.

Keane, Michael, 1995, “A New Idea for Welfare Reform”, Federal Reserve Bank of
Minneapolis Quarterly Review, 19(2), 2-28.

Kubik, Jeffrey D. “The Incidence of Personal Income Taxation: Evidence from the Tax
Reform Act of 1986.” Journal of Public Economics 88 (2004): 1567-1588.

96

Leigh, Andrew. 2005. “Who Beneﬁts from the Earned Income Tax Credit? Incidence
Among Recipients, Coworkers and Firms.” Center for Economic Policy Research,
Australian National University, Discussion Paper 494.

Madrian, Brigitte C., and Lars John Lefgren. 1999. “A Note on Longitudinally Matching
Current Population Survey (CPS) Respondents.” National Bureau of Economic
Research, Technical Working Paper 247.

Meyer, Bruce D. 2002. “Labor supply at the extensive and intensive margins: the EITC,
welfare and hours worked.” American Economic Review, Papers and
Proceedings, May, 373-379.

Meyer, Bruce D. and Dan T. Rosenbaum, 2000, “Making Single Mothers Work: Recent
Tax and Welfare Policy and Its Effect”, National Bureau of Economic Research
Working Paper #7491.

Meyer, Bruce D. and Dan T. Rosenbaum. 2001. “Welfare, the Earned Income Tax Credit,

and The Labor Supply of Single Mothers.” Quarterly Journal of Economics
116(3): 1063-1114.

Mofﬁtt, Robert. “Welfare Effects on Female Headship with Area Effects.” Journal of
Human Resources 29 (1994): 621—636.

Mroz, TA. 1987. “The Sensitivity of an Empirical Model of Married Women’s Hours of
Work to Economic and Statistical Assumptions.” Econometrica 55: 765-799.

Nagel, Ami and Nicholas Johnson. 2006. “A Hand Up: How State Earned Income Tax
Credits Help Working Families Escape Poverty in 2006.” Center on Budget and
Poverty Priorities, Washington, DC.

O’Neill, Jane E. and Hill, Anne M., 2001, “Gaining Ground? Measuring the Impact of
Welfare Reform on Welfare and Work”, Civic Report No.17, Center for Civic
Innovation, New York, Manhattan Institute.

Pencavel, J. 1998. “Assortative mating by schooling and the work behavior of wives and
husbands.” American Economic Review Papers and Proceedings 88: 326—329.

Rosen, Harvey S. “The Marriage Tax Is Down but Not Out.” National Tax Journal 40
(1987): 567—76.

Rothstein, Jesse. 2007. “The Unintended Consequences of Encouraging Work: Is the
EITC As Good As an NIT?” Unpublished manuscript, Princeton University.

97

Schoeni, Robert F. and Rebecca M. Blank. 2000. “What Has Welfare Reform
Accomplished? Impacts on Welfare Participation, Employment, Income, Poverty,

and Family Structure.” National Bureau of Economic Research Working Paper
7627.

Scholz, John Karl, 1994, “The Earned Income Tax Credit: Participation, Compliance, and
Anti-Poverty Effectiveness”, National Tax Journal Vol 47, No.1 (March 1994),
63-87.

Schultz, T. Paul. “Marital Status and Fertility in the United States: Welfare and Labor
Market Effects.” Journal of Human Resources 29 (1994): 637-669.

Steuerle, E. 2002. “Tax Policy from 1990 to 2001,” in American Economic Policy in the
1990s, edited by J. Frankel, and P. Orszag, 139-169. Cambridge: MIT Press.

Triest, R.K. 1990. “The effect of income taxation on labor supply in the United States.”
Journal of Human Resources 25: 491— 516.

Whittington, Leslie and Alm, James. “Til Death or Taxes Do Us Part: The Effect of
Income Taxation on Divorce.” Journal of Human Resources 32 (1997):388-412.

Wooldridge, Jeffry M., 2002. “Econometric Analysis of Cross Section and Panel Data”,
The MIT Press.

Wooldridge, Jeffrey M. “Introductory Econometrics A Modern Approach” (3rd ed.),
Thomson South-Westem, 2006.

98

 

I‘leﬁt-um,isrant-nifqtﬂnilﬁi' _ u .-'2 um. it n . V l

   

11000000000