1|

141
139
THS

 

m 1.30)

2008

LIBRARY
Michigan State
University

This is to certify that the
dissertation entitled

THREE EMPIRICAL STUDIES OF
HUMAN CAPITAL, LABOR SUPPLY, AND HEALTH CARE

presented by
Merve Cebi

has been accepted towards fulfillment
of the requirements for the

Doctoral degree in Economics

 

 

Jfl [VFW/[r15

 

Major Professor’s Siylature

I? MAY zccr

 

Date

MSU is an afiinnative-action, equal-opportunity employer

.. - --—.--o-.-.--a-c--

 

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.lProj/Acc&Pres/ClRC/Date0ue indd

 

THREE EMPIRICAL STUDIES OF
HUMAN CAPITAL, LABOR SUPPLY, AND HEALTH CARE

By

Merve Cebi

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

THREE EMPIRICAL STUDIES OF
HUMAN CAPITAL, LABOR SUPPLY, AND HEALTH CARE

By

Merve Cebi

Locus of Control and Human Capital Investment Revisited

Locus of control (LOC) is a psychological concept that measures the extent to
which an individual believes she has control over her life (internal control) as opposed to
believing that luck controls her life (external control). Findings from the early empirical
literature suggested that internal LOC is related to higher educational attainment and
earnings. However, a key concern in the early literature is that LOC could merely be a
proxy for unobserved ability, which could itself increase education and earnings. To
distinguish between the effects of LOC and the effects of ability, Coleman and DeLeire
(2003) present a model of human capital investment that incorporates LOC. I test the
predictions of the Coleman-DeLeire model using data from the National Longitudinal
Survey of Youth. My findings fail to support Coleman and DeLeire’s predictions and
suggest that LOC is not a significant determinant of educational outcomes once cognitive

ability is controlled for; however, LOC does lead to higher earnings later in life.

Employer-Provided Health Insurance and Labor Supply of Married Women
This work presents new evidence on the effect of husbands’ health insurance on
wives’ labor supply. Previous cross-sectional studies have estimated a significant

negative effect of spousal coverage on wives’ labor supply. However, these estimates

potentially suffer from bias because wives’ labor supply and the health insurance status
of their husbands are interdependent and chosen simultaneously. This paper attempts to
obtain consistent estimates by using several panel data methods. In particular, the likely
correlation between unobserved characteristics of husbands and wives affecting labor
supply—such as preferences for work—can be captured using panel data on intact
marriages, and potential joint job choice decisions can be controlled using fixed-effects
instrumental variables methods. The findings, using data from the Current Population
Survey and the National Longitudinal Survey of Youth, suggest that the negative effect of
spousal coverage on labor supply found in cross-sections results mainly from spousal
sorting and selection. There is only a small estimable effect of spousal coverage on

wives’ labor supply.

Health Insurance Tax Credits and Health Insurance Coverage of Low-Income
Single Mothers

The Omnibus Budget Reconciliation Act of 1990 introduced a refundable tax
credit for low-income families who purchased health insurance coverage for their
children. This health insurance tax credit (HITC) existed during tax years 1991, 1992,
and 1993, and was then rescinded. We use Current Population Survey data and a
difference-in-differences approach to estimate the HITC’s effect on private health
insurance coverage of low-income single mothers. The findings suggest that during 1991-
1993, the health insurance coverage of single mothers was about 6 percentage points

higher than it would have been in the absence of the HITC.

ACKNOWLEDGEMENTS

I would like to acknowledge several people for their help and support which made
this dissertation possible. First and foremost, I would like to thank my advisor, Stephen
A. Woodbury, for his generous time and commitment. He provided valuable insights and
instructive comments at every stage of the dissertation process. I am grateful for his
expert guidance and encouragement.

I wish to thank the members of my dissertation committee, Jeff Biddle, John
Goddeeris, and Dale Belman for their helpful suggestions and advice. I am also indebted
to Thomas DeLeire whose comments and insights substantially improved this study.

I owe a special note of gratitude to Joanne Lowery at the W.E. Upjohn Institute
for Employment Research who provided several editorial suggestions after meticulous
readings of this dissertation.

I extend many thanks to my colleagues and friends, especially Pamela Ortiz,
Neslihan Yilmaz, Laura Donnet, Lourdes Martinez, and Deborah Foster for their
continual encouragement.

Finally, I would like to thank my family. My mother instilled in me the desire and
skills to obtain the Ph.D; and my father was a constant source of support throughout. I am
especially grateful to my sister and best friend, Muge, for her love, inspiration, and

patience throughout this process.

iv

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................................ vii
LIST OF FIGURES ........................................................................................................... ix
CHAPTER 1
LOCUS OF CONTROL AND HUMAN CAPITAL INVESTMENT REVISITED .......... 1
1.1 . Introduction .................................................................................................................. 1
1.2. Data .............................................................................................................................. 4
1.3. Estimation Method and Results ................................................................................... 7
1.3.1. Locus of Control and Educational Attainment ................................................ 7
1.3.2. Locus of Control and Occupational Expectations ......................................... 11
1.3.3. Locus of Control and Wages .......................................................................... 14
1.4. Discussion and Conclusion ........................................................................................ 16
CHAPTER 2
EMPLOYER-PROVIDED HEALTH INSURANCE AND LABOR SUPPLY OF
MARRIED WOMEN ........................................................................................................ 19
2.1. Introduction ................................................................................................................ 19
2.2. Previous Research ...................................................................................................... 21
2.3. Empirical Methodology ............................................................................................. 23
2.4. Data ............................................................................................................................ 25
2.5. Empirical Findings ..................................................................................................... 32
2.5.1 Cross-Sectional LPM and Probit Estimates ....................................................... 32
2.5.2 Cross-Sectional IV Estimates ............................................................................ 33
2.5.3 Panel Estimates .................................................................................................. 42
2.6. Conclusion ................................................................................................................. 45
CHAPTER 3
HEALTH INSURANCE TAX CREDITS AND HEALTH INSURANCE COVERAGE
OF LOW-INCOME SINGLE MOTHERS ....................................................................... 48
3.1 . Introduction ................................................................................................................ 48
3.2. The Health Insurance Tax Credit, 1991-1993 ............................................................ 51
3.3. Approach to Estimation ............................................................................................. 53
3.4. Data ............................................................................................................................ 56
3.5. Empirical Findings ..................................................................................................... 63
3.5.1. Main Findings — Single Women with Less than a High School Education.... 63
3.5.2. Findings for Single Women with More Education ........................................... 66
3.5.3. Findings Disaggregated by Year ....................................................................... 67
3.6. Sensitivity Tests ......................................................................................................... 70
3.6.1. Medicaid Crowd-Out ........................................................................................ 70
3.6.2. State-Level Economic Conditions and State Fixed Effects .............................. 72
3.6.3. Welfare Reform and State EITCs ..................................................................... 72

3.7. Conclusion ................................................................................................................. 75

APPENDIX 1 .................................................................................................................... 79
APPENDIX 2 .................................................................................................................... 85
REFERENCES ................................................................................................................. 88

vi

LIST OF TABLES

Table 1.1. Summary Statistics for Key Variables by Education Level .............................. 8
Table 1.2. Marginal Effects of Locus of Control on Educational Attainment from Probit

Models ................................................................................................................................. 9
Table 1.3. Predicted Occupational Expectations at Age 35 .............................................. 13
Table 1.4. Effects of Locus of Control on Adult Wages .................................................. 15

Table 2.1. Labor Supply of Married Women by Spousal Health Insurance Coverage,
2000 CPS and NLSY ........................................................................................................ 29

Table 2.2. Summary Statistics for Married Women by Spousal Health Insurance
Coverage, 2000 CPS and NLSY ....................................................................................... 30

Table 2.3. Summary Statistics for Working Married Women by Spousal Health Insurance
Coverage, 2000 CPS and NLSY ....................................................................................... 31

Table 2.4. Linear Probability and Probit Estimates of Married Women's Labor Force
Participation, 2000 CPS and NLSY .................................................................................. 34

Table 2.5. Linear Probability and Probit Estimates of Working Married Women's Full-
Time Work, 2000 CPS and NLSY .................................................................................... 36

Table 2.6. OLS and ZSLS Estimates of the Effect of Spousal Coverage on Labor Supply:
2000 CPS .......................................................................................................................... 40

Table 2.7. OLS and ZSLS Estimates of the Effect of Spousal Coverage on Labor Supply:
2000 NLSY ....................................................................................................................... 41

Table 2.8. Panel Estimates of the Effect of Spousal Coverage on Labor Supply: 1989-
2000 NLSY ....................................................................................................................... 44

Table 3.1. Health Insurance Coverage Rates for Low-Education Working Single Mothers
and Low-Education Working Single Women without Children ....................................... 59

Table 3.2. Summary Statistics: Low-Education Working Single Mothers and Low-
Education Working Single Women without Children ...................................................... 61

Table 3.3. Summary Statistics: Low-Education Working Single Mothers and Low-
Education Working Single Women without Children, Before and During HITC ............ 62

vii

Table 3.4. Private Health Insurance Coverage Rates of Low-Education Working Single
Mothers and Low-Education Working Single Women without Children ........................ 65

Table 3.5. Difference-in-Differences Estimates of the Effect of the HITC on Private
Health Insurance Coverage of Low-Education Working Single Mothers from Probit
Estimates of Equation (3.1) ............................................................................................... 65

Table 3.6. Difference-in-Differences Estimates from Alternative Treatment and
Comparison Groups .......................................................................................................... 68

Table 3.7. Difference-in-Differences Estimates of the Change in Private Health Insurance

Coverage Relative to 1990, Low-Education Working Single Mothers ............................ 69
Table 3.8. Sensitivity Tests for HITC Effects on Low-Education Working Single Mothers
........................................................................................................................................... 74
Table A1.1. Earned Income Tax Credit Parameters ......................................................... 79

Table A1.2. Major Legislative Changes Affecting Low-Income Women, 1987-94 ........ 80
Table A1.3. Full Results for Table 3.5 Estimates: Results from Probit Models ............... 81

Table A1.4. Full Results for Table 3.7 Estimates: Results from Probit Models ............... 82

viii

LIST OF FIGURES

Figure 1.1. Relationships between Expected Outcomes and Locus of Control for High
School Graduates and Dropouts in the Coleman-DeLeire Model (top) and when Locus of
Control is a Proxy for Ability (bottom) .............................................................................. 3

Figure 3.1. Health Insurance Coverage Rates for Low-Education Working Single
Mothers and lbw-Education Working Single Women without Children ........................ 60

Figure A1.1. Earned Income Tax Credit and Health Insurance Tax Credit Schedules,
1991 ................................................................................................................................... 84

ix

CHAPTER 1

LOCUS OF CONTROL AND HUMAN CAPITAL INVESTMENT REVISITED

1.1. Introduction

The determinants of educational attainment have been the subject of intensive
research. A consensus has emerged that certain variables affect education, including
socioeconomic variables, family background measures, and personal attributes such as
cognitive and noncognitive skills. In an attempt to identify the impact of noncognitive
skills, a strand of literature has focused attention on the social-psychological concept of
“locus of control,” which measures the extent to which an individual believes she has
control over her life (internal control) as opposed to believing that luck controls her life
(external control).

The early empirical literature was limited to including locus of control in wage or
educational attainment regressions along with measures of cognitive skill. (See, for
example, Andrisani 1977, 1981). Findings from this literature suggested that internal
locus of control is related to higher educational attainment and higher earnings. However,
a key concern in the early literature is that internal locus of control could merely be a
proxy for unobserved ability, which could itself increase education and earnings.

To distinguish between the effects of locus of control and the effects of ability,
the subsequent literature has begun to explore the mechanism by which locus of control
affects educational outcomes. In particular, Coleman and DeLeire (2003) present a model
of human capital investment that explicitly incorporates locus of control. This model

distinguishes among four groups of teenagers:

0 Internal Graduates — teenagers who graduate from high school and believe that
graduating will lead to higher wages and higher-skill occupations
0 External Graduates — teenagers who graduate from high school but do not
believe that graduating will lead to higher wages or higher-skill occupations
0 Internal Dropouts — teenagers who drop out and believe that dropping out will
lead to lower wages and worse occupational outcomes
0 External Dropouts — teenagers who drop out and do not believe that dropping out
will lead to lower wages or worse occupational outcomes
Coleman and DeLeire’s model implies that, among high school graduates, internal
teenagers will say they expect higher earnings in the future than external teenagers —
that is, Internal Graduates will have higher earnings expectations than External
Graduates. However, among dropouts, the model implies the opposite —— that is, Internal
Dropouts will have lower earnings expectations than External Dropouts (Coleman and
DeLeire 2003, equation 5). The intuition behind this asymmetry, which is depicted in the
top panel of Figure 1.1, is that internal teenagers perceive a relationship between their
current actions and future outcomes, whereas external teenagers do not.

Coleman and DeLeire contrast their model with an alternative model in which
locus of control is simply a proxy for ability. This alternative model does not produce the
asymmetric effect of locus of control on expected outcomes (conditional on educational
attainment). Rather, if locus of control is simply an aspect of ability, internal teenagers
will expect better outcomes than external teenagers regardless of whether they graduate

from high school, as shown in the bottom panel of Figure 1.1. Thus, Coleman and

Expected

 

 

 

 

Outcomes
E( ' )
Graduates
Dropouts
External Internal
Locus of Control
Expected
Outcomes
E( ' )
Graduates
Dropouts

 

 

 

 

External Internal
Locus of Control

Figure 1.1. Relationships between Expected Outcomes and Locus of Control for
High School Graduates and Dropouts in the Coleman-DeLeire Model (top) and
when Locus of Control is a Proxy for Ability (bottom)

DeLeire’s model and the alternative ability-based model offer distinct and empirically
testable implications.

Using data from the National Education Longitudinal Study (N ELS), Coleman
and DeLeire find evidence that supports their model. Consistent with the predicted
pattern of expectations, Internal Dropouts expect to receive lower wages and to be in
lower-skilled occupations than do External Dropouts.

This study reexamines the effect of locus of control on educational attainment and
tests the predictions of Coleman and DeLeire’s model using data from the National
Longitudinal Survey of Youth (N LSY). First, I investigate whether locus of control is an
important predictor of educational attainment for a teenage sample of 10th and 11th
graders in 1979. Second, given information on these teenagers’ educational attainment
three years later, I examine the effect of locus of control on their occupational
expectations. Third, the NLSY provides an opportunity to study the subsequent labor
market outcomes of the teenage sample. Because the respondents are between the ages of
37 and 45 as of the 2002 survey, it is possible to examine the impact of teenagers’ locus

of control on their adult earnings.

1.2. Data

The NLSY is a sample of 12,686 young men and women between the ages of 14
and 22 at the time of the first interview in 1979. Since their first interview, they have
been reinterviewed annually until 1994, and biennially from 1996 to the present.

The NLSY consists of three subsarnples: a representative sample of the
noninstitutionalized civilian youths; an oversample of blacks, Hispanics, and

economically disadvantaged whites; and a sample of respondents who were enlisted in

the military. In this study, I use the nationally representative sample of 6,111 respondents
in order to derive estimates using a random sample. Observations are included if (1)
respondents had valid measures of education for the years 1979-1982; (2) information on
respondents’ locus of control scale was available; (3) respondents were in the 10th or
11th grade in 1979.1 Applying these restrictions resulted in a final sample of 1,737
individuals.

The Rotter Intemal-Extemal Locus of Control Scale, collected in the 1979 survey,
is a four-item questionnaire designed to measure the extent to which individuals believe
they have control over their lives (internal control) as opposed to believing that luck
controls their lives (external control). Respondents were asked to select one of each of
four paired statements,2 and then decide if the selected statement was much closer or
slightly closer to their opinion of themselves. A four-point scale was generated for each
of the paired items, and the resulting scores are individually standardized. The average of
the standardized scores is used to create the locus of control scale. Higher scores indicate
greater internal control, whereas lower scores indicate greater external control.

In 1980, the NLSY data were supplemented by a series of achievement tests
known as the Armed Forces Vocational Aptitude Battery (ASVAB). The scores for

selected parts of the ASVAB are then used to construct a composite Armed Forces

 

' Ninth graders are not included in the sample because although most students would have graduated from
high school, they would not be old enough to attend college by the time of the 1982 survey. The results for
high school graduation are robust to the inclusion of 9th graders. For sake of brevity, these results are not
reported but are available from the author upon request.
2 1. (a) What happens to me is my own doing; or (b) Sometimes I feel that I do not have enough control
over the direction my life is taking.
2. (a) When I make plans, I am almost certain that I can make them work; or (b) It is not always wise to
plan too far ahead, because many things turn out to be a matter of good or bad fortune anyhow.
3. (a) In my case, getting what I want has little or nothing to do with luck; or (b) Many times, we might
just as well decide what to do by flipping a coin.
4. (a) Many times, I feel that l have little influence over the things that happen to me; or (b) It is
impossible for me to believe that chance or luck plays an important role in my life.

Qualifications Test (AFQT) score for each respondent. The NLSY provides the raw and
standard scores for each subset of the ASVAB, as well as two percentile scores: an
AFQT80 and an AFQT89.3

The percentile scores are the most widely used measures of ability by researchers.
However, Blackburn (2004) discusses that the AFQT percentile ranking is not a correct
measure of ability since ability follows a normal distribution while a percentile follows a
uniform distribution. He advises the use of raw or standard scores as a more appropriate
measure of the AFQT performance. In this study, the AF QT measure is constructed as the
sum of standard scores for the verbal, math knowledge, and arithmetic reasoning subtests
of the ASVAB.

The implications of Coleman and DeLeire’s model concern the labor market
expectations of teenagers conditional on educational attainment. It thus is essential to
have information on expectations collected after the decision about educational
attainment has been made. In the 1979 and 1982 surveys, NLSY respondents were asked
about their “Occupational Expectations at Age 35 (Census 3-Digit).” Since the sample
used in this analysis consists of 10th and 11th graders in 1979, occupational aspiration of
teenagers is extracted from the 1982 survey along with information on their graduation

and college enrollment status.4

 

3 The two AFQT measures differ in the methods used to calculate the scores. The AF QT80 measure is
constructed as the sum of the following subtests of the ASVAB: word knowledge, paragraph
comprehension, arithmetic reasoning, and numerical operations. Beginning in 1989, a new formula has
been used to calculate a revised percentile score called the AFQT89. The three subtests used in the 1989
scoring version of the AFQT score are verbal, math knowledge, and arithmetic reasoning. Rest of the
ASVAB includes the following subtests: mechanical comprehension, general science, electronics
information, auto and shop information, and coding speed. Attachment 106 to the NLSY documentation
describes the ASVAB subtests in detail.

" I use the revised version of the Highest Grade Completed variable to identify high school graduates and
college attendees.

Table 1.] presents descriptive statistics for teenagers by their educational level in
1982. Out of 1,737 observations, 1,370 graduated from high school but only 545 had
attended college as of 1982. Teenagers who graduated from high school come from
higher income families and have a significantly higher locus of control score than
teenagers who dropped out of high school, 0.043 versus -O.172. Similarly, the locus of
control score of teenagers who attended college (0.141) is significantly higher than of
teenagers who did not (-0.068). Both mothers and fathers of teenagers who attended
college obtained more education on average and were more likely to have worked as a

professional or manager than those of teenagers who did not attend college.

1.3. Estimation Method and Results
1.3.1. Locus of Control and Educational Attainment

In Table 1.2, the left-hand panel shows the estimated marginal effects of locus of
control on high school graduation from probit models. The right-hand panel reports the
estimated marginal effects on college attendance. The basic specification is presented in
Column 1. It includes dummy variables indicating race, ethnicity, gender, age, residence
in an SMSA, and residence in an urban area as controls. According to these estimates,
locus of control is an important predictor of educational attainment for teenagers. A one-
standard-deviation increase in locus of control is estimated to increase the probability of
high school graduation by 5.4 percent, and the probability of college attendance by 7.4
percent.

Column 2 adds indicators of parental education as controls to the basic model.
The estimated marginal effect of locus of control remains both economically and

statistically significant. In particular, a one-standard-deviation increase in locus of control

Table 1.1. Summary Statistics for Key Variables by Education Level

 

 

Entire High High Attended Did not
Sample School School College Attend
Grads Dropouts College
High School Graduate 0.789 1 0 1 ' 0.692
(0.408) (0) (0) (0) (0.462)
Attended College 0.314 0.398 0 1 0
(0.464) (0.490) (0) (0) (0)
Locus of Control -0.003 0.043 -0.172 0.141 -0.068
(0.574) (0.562) (0.587) (0.543) (0.576)
AFQT 195.857 203.610 166.009 221.361 184.018
(35.277) (32.164) (30.535) (26.551) (32.453)
Family Income 20,718 22,615 13,528 26,999 17,879
(14,028) (14,116) (11,069) (15,838) (12,110)
Father's Education
Less than High School 0.401 0.336 0.643 0.167 0.508
(0.490) (0.472) (0.480) (0.373) (0.500)
High School 0.331 0.353 0.248 0.308 0.341
(0.471) (0.478) (0.432) (0.462) (0.474)
Some College 0.104 0.115 0.063 0.156 0.081
(0.306) (0.320) (0.243) (0.363) (0.272)
College and beyond 0.164 0.196 0.046 0.369 0.070
(0.370) (0.397) (0.210) (0.483) (0.256)
Mother's Education
Less than High School 0.376 0.310 0.621 0.145 0.482
(0.485) (0.463) (0.486) (0.352) (0.500)
High School 0.445 0.481 0.311 0.495 0.422
(0.497) (0.500) (0.463) (0.500) (0.494)
Some College 0.089 0.103 0.038 0.154 0.060
(0.285) (0.304) (0.192) (0.361) (0.237)
College and beyond 0.090 0.106 0.030 0.206 0.037
(0.286) (0.308) (0.171) (0.404) (0.189)
Father's Occupationa 0.219 0.249 0.106 0.406 0.133
(0.414) (0.433) (0.309) (0.491) (0.340)
Mother's Occupationa 0.083 0.094 0.041 0.163 0.046
(0.276) (0.292) (0.198) (0.370) (0.210)
Number of Observations 1,737 1,370 367 545 1,192

 

Notes: a. A dummy variable indicating adult male (or female) in household worked
as a professional or manager when the respondent was 14 years old. Standard

deviations are reported in parentheses.

£399.28 5 88.582 Ba Bebe 23x85 .3 we own at S
89> .35 :33 Emu .089: a 22 use £3396: .mafiwwafi 8338.. basswo— conEoE 9:5 a .5 Eowmcofi @585: $38.23
bee—6 mews—us £5 «EOE .559: new 85m... Eon me :3338 use .0885 >558 mug—as 232:5 bian— docmosvo $559:
was mafia .68 £868 can AS 3802 .<m2m an 5 8.8862 can own $532: SEES basic 08205 83850on =< $802

 

 

 

43 m3 3o 43 _ 3 Bo Rd 8o «3 cud
own; 32 83 RE RE cm? 8m; 33 RE RE
8% 02 02 02 02 mo», 02 oz 02 02
BS mo> oz 02 oZ mo> 8% oz 02 oZ
mo> mo> mo> oz 02 mo> mo> mo> 02 OZ
mo> mo> mo> mo> oZ mo> mo> mo> mo.» 02
88.8 88.8 :88 88.8 28.8 88.8 88.8 €88 $88 68.8
888 Rod 3.3 Red Sod was. $3. ~88- 48.? 89°.
$8.8 $8.8 888 @888 88.8 88.8 28.8 28.8 85.8 88.8
83 N88 53 Rod Bed 33 :58 98¢ 23 88¢
:88 €88 $8.8 $8.8 33.8 88.8 98.8 88.8 €88 38.8
33 ”NS 23 83 Sad. 83 _ _ 3 53 ME; 888
88.8 28.8 $3.8 $5.8 9.88 38.8 28.8 28.8 88.8 38.8
$3 83 33 em; ”8.? God 83 Sod- 88.? Rod.
#88 98.8 38.8 88.8 88.8 35.8 85.8 85.8 63.8 85.8
Sod 83 83 Rod cm; 888 Sod E3 835 wood
5 A8 A8 6 8 5 A8 6 A8 3

02.853. 82.5 aéaeeo .255 .3:

255mm 258m
8033030

5”?

85 go:
0.5585 SEE
cases—om 3:988

:35
£958
£523:
xosm

35:00.? 903

 

€82 .32.. as... a 2:532 .8385 .8 .235... 285 E 388.... 35...: .2 oz“...

is estimated to increase teenagers’ probability of graduating from high school by 4.6
percent, and their probability of attending college by 5.7 percent. The addition of family
income and parents’ occupation status in Column 3 produces very similar results to those
obtained from previous specifications.

Column 4 adds dummy variables indicating whether teenagers received
magazines, newspapers, and held a library card at age 14. The estimated marginal effects
are essentially similar and still statistically significant. A one-standard-deviation increase
in locus of control is associated with a 3.8 percent increase in teenagers’ likelihood of
graduating from high school.

Column 5 adds teenagers’ AF QT score as a control for cognitive ability. With this
addition, the estimated marginal effect of locus of control drops and becomes much less
significant. The estimated marginal effect of locus of control on high school graduation is
0.026, with a t-statistic of 1.44 (p-value = 0.15). This implies that a one-standard-
deviation increase in locus of control increases the probability of high school graduation
by 1.5 percent. The marginal effect of locus of control on college attendance is 0.04
which is significant only at the lO-percent level. This implies that a one-standard-
deviation increase in locus of control increases the probability of college attendance by
2.3 percent. The results in Column 5 suggest that locus of control is capturing the
marginal effect of the AF QT score on educational attainment in Columns 1-4. The locus-
of-control estimates in Columns 1-4 suffer from omitted variable bias. The simple

correlation between locus of control and AFQT is 0.28.

10

1.3.2. Locus of Control and Occupational Expectations

I follow Coleman and DeLeire and estimate the following by OLS,

(1.1) occexp35 = X ,6 + 51 internal Xgrad + 82 average >< grad + 53 external xgrad

+ 54 internal Xdropout + 55 average Xdropout + 56 external ><dropout + u,

(1.2) occexp35 = X ,8 + 51 internal Xcollege + 52 average Xcollege + 53 external Xcollege

+ 54 internal chollege + 55 average chollege + 56 external chollege + u,

where the dependent variable is a dummy variable indicating that the teenager expects to
work in a high-skilled occupation at age 35. Controls include race, ethnicity, gender, age,
residence in an SMSA and in an urban area, and AF QT.

The locus-of-control score of teenagers measured in 1979 is used to construct
three dummy variables. The variable internal equals 1 if the teenager is above the 75th
percentile of the locus of control range. The 25th to 75th percentiles are defined as
average, and values below the 25th percentile are defined as external. The variables grad
and dropout are indicators of whether the teenager had graduated from high school or
dropped out of high school as of 1982. Similarly, college and ncollege are dummy
variables indicating whether the teenager did or did not attend college.

Table 1.3 reports the predicted expectations of being in a hi gh-skilled occupation
at age 35 that result from estimating Equation 1.1 in the top panel and Equation 1.2 in the
bottom panel. Each panel shows predicted occupational expectations for six groups of
teenagers: high school graduates (or college attendees) with internal, average, or external
locus of control and dropouts (or non-college attendees) with internal, average, or

external locus of control. For each group, three predictions are shown: those from (1) a

11

specification that includes no control variables; (2) a specification that controls for race,
ethnicity, gender, age, and residence in an SMSA and in an urban area; and (3) a
specification that controls for AF QT in addition to the controls in Specification 2.

The results in Table 1.3 suggest that lntemal Dropouts have basically the same
occupational expectations as External Dropouts. Similarly, internal non-college attendees
have roughly the same expectations as external non-college attendees. This implies that
the pattern predicted in Coleman and DeLeire’s model is not apparent in these data.

Another way to distinguish between the two models is to test whether the gap
between lntemal and External Graduates’ expectations differs from that between Internal
and External Dropouts. To see this, recall first that in both the Coleman-DeLeire model
and the alternative ability-based model, we should observe only a small (if any)
difference in expectations between External Graduates and External Dropouts (compare
the top and bottom panels of Figure 1.1). Recall next that in the Coleman-DeLeire model,
lntemal Graduates have higher expectations than External Graduates, and lntemal
Dropouts have lower expectations than External Dropouts (as shown in the top panel of
Figure 1.1), so there is a large gap between the expectations of lntemal Graduates and
lntemal Dropouts. 1n the alternative model, in contrast, lntemal Graduates have higher
expectations than External Graduates, and lntemal Dropouts will have higher
expectations than External Dropouts, so there is only a small expectations gap between
lntemal Graduates and lntemal Dropouts (bottom panel of Figure 1.1). Together, these
predictions suggest that if we estimate the difference-in-differences between External
Graduates and External Dropouts, and lntemal Graduates and lntemal Dropouts, an

estimate statistically different from zero would support the Coleman-DeLeire model.

12

 

 

5.3 .83 5.3
ocd No.0- V0.0- muofluhocmn— fiasco-5.55a—
333 30.3 :33 833 $33 $33 552m 25 5835
85 85 So 85 3o 85 .5223 85535
$33 5.3 5.3 $33 5.3 5.3 .83 5.3 5.3
mg m2 35 N3 85 85 $3 53 85 253 552m
:33 5.3 5.3 :53 5.3 5.3 :33 5.3 5.3
mg 2.2 $5 3.5 one $5 35 38 3o 253 52552
5.3 5.3 5.3 $33 5.3 5.3 $33 5.3 53
N3 :3 52 one 25 as one woo 3o 253 5535
8:20me owe—BU owe—BU oocouobmn owozou owe—BU 3:80me omo=o0 owe—BU
oz 52 oz
.55.. ”3555 22:50 5 .53. ”35.35 22:50 5 222.8 2 8
moot—5.37:: Z can moo—533‘ owe—BU
333 :53 :33
mod mod cod noun—onuumn— 5.023.359
5.3 :33 5.3 :33 5.3 .33 5552.2 25 5253
85 8.5 8.0 :5 :5 :.o 5233 8553a
5.3 5.3 5.3 5.3 833 833 $33 333 5.3
3.5 one :3 25 35 one a; :5 one 253 552m
:33 5.3 5.3 :33 833 833 :33 333 5.3
35 £5 33 Ed was 33 w; 53 $3 253 52252
5.3 5.3 5.3 5.3 833 5.3 $33 633 5.3
8:20me Shaman $80 8:08th 3:03on mun—O 8:0qume 85030.5 .680
mm mm m: mm mm m:

 

3.: 235.35 28.25 6 22.55 52 A:

muacaen can moan—.139 32.9w Jam:

.53. ”5.5.2: 29550 5

 

mm ow< «a muccauooaum— 1:833:30 6833.5 .mA 035,—.

13

Table 1.3 presents the findings on differences in occupational expectations for
high school graduates and dropouts. The third column in each panel shows the difference
between high school graduates and dropouts for each group (internal, average, and
external teenagers). The fourth row reports the difference between internal and external
teenagers for both high school graduates and dropouts. Finally, the difference-in-
differences estimates are presented in the last row. Parallel findings for college attendees
and non-attendees are given in the bottom panel of Table 1.3.

In all specifications, the difference-in-differences estimate is close to zero and
statistically insignificant. Once again, the predictions of Coleman and DeLeire’s model
are not borne out in these data.

1.3.3. Locus of Control and Wages

To examine the relationship between teenagers’ locus of control and their wages
later in life, I estimate a human capital earnings function similar to that estimated by
Andrisani (1977). The dependent variable is log hourly wages measured at the time of the
2000 interview. The estimation results are presented in Table 1.4. Control variables in
Column 1 include years of education, race, ethnicity, gender, marital status, residence in
an SMSA and in an urban area, a quadratic in age, and a set of occupational dummies.
Column 2 adds locus of control, Column 3 adds the AFQT score, and Column 4 includes
both.

Based on the estimates in Table 1.4, the return to a year of education without
controlling for any measures of ability is 7 percent. As shown in Column 2, adding locus
of control does not affect this estimate. However, with the addition of the AFQT score in

Column 3, the estimated return falls to 5 percent, which reflects the familiar ability bias

14

Table 1.4. Effects of Locus of Control on Adult Wages

 

 

(1) (2) (3) (4)
Locus — 0.057 — 0.036
(0.015) (0.015)
AF QT — — 0.003 0.003
(0.001) (0.001)
Education 0.074 0.072 0.050 0.049
(0.004) (0.004) (0.005) (0.005)
Black -0. 172 -0. 168 -0.057 -0.060
(0.025) (0.025) (0.028) (0.028)
Hispanic -0.041 -0.038 0.022 0.022
(0.032) (0.032) (0.032) (0.032)
Female -0.312 -0.309 -0.304 -0.302
(0.017) (0.017) (0.017) (0.017)
Married 0.081 0.080 0.065 0.065
(0.020) (0.020) (0.020) (0.020)
Urban 0.045 0.045 0.042 0.042
(0.020) (0.020) (0.020) (0.020)
Observations 4,278 4,27 8 4,137 4,137
R-squared 0.29 0.29 0.31 0.31

 

Notes: The dependent variable is the log(hourly wage) in 2000. In addition to
the variables shown, all specifications include a quadratic in age, a set of
occupational dummy variables, and a dummy for residence in an SMSA.
Heteroskedasticity-robust standard errors are in parentheses.

in the estimated returns to schooling. Adding locus of control in Column 4 leaves the
estimates unchanged relative to those in Column 3. The coefficient on locus of control is
0.036 and is statistically significant at the 5-percent level. A one-standard-deviation
increase in locus of control increases hourly wages by 2.1 percent, while a-one-standard-
deviation increase in the AFQT score leads to an 11.5 percent increase in hourly wages.
These results suggest that locus of control is in fact capturing a distinct aspect of ability
not related to cognitive ability as measured by the AFQT. Combined with the findings in
Section 1.3.2, the results suggest that, although locus of control is not a significant
determinant of educational outcomes, it is rewarded in the labor market through higher

wages.

15

1.4. Discussion and Conclusion

Using data from the NLSY, the analysis in this paper yields three main findings.
First, there is no evidence that locus of control predicts high school graduation and little
evidence that it predicts college attendance once the AF QT score is included in models of
educational attainment (Section 1.3.1). Second, lntemal Dropouts have basically the same
occupational expectations as External Dropouts, as do internal non-college attendees and
external non-college attendees (Section 1.3.2). Third, locus of control measures a distinct
skill not captured by the AF QT, and this skill brings a reward in the labor market
(Section 1.3.3).

The finding that locus of control does not predict educational attainment conflicts
with Coleman and DeLeire’s results. Using data from the NELS, they find that locus of
control strongly affects educational attainment, presumably by influencing teenagers’
assessments of the returns to education. One possible reason for the difference between
their findings and mine could be that the cognitive ability tests available in the NELS
differ from those in the NLSY. The NELS contains standardized scores in math, reading,
science, and history given when students were in the 8th grade. To make the cognitive
ability tests in the NLSY as close as possible to those in the NELS, I do two things. First,
I use the sum of standard scores in the verbal, math knowledge, and arithmetic reasoning
subtests of the ASVAB, and I omit the rather specific subtests such as “electronics
information” or “coding speed.” Second, I include dummy variables that identify one of
eight age groups. My sample of 10th and 11th graders in 1979 who took the ASVAB
tests in 1980 consists of students who were between the ages of 15 and 22. By including

age dummies, I attempt to control for effects of age at the time the test is taken. Also, I

16

reestimate all the models using different subtests of the ASVAB to allow each to
potentially reflect a different skill. These changes, however, do not affect the findings.
Hence, there is a real puzzle — the NELS and NLSY give quite different results.

The finding that locus of control is unrelated to teenagers’ occupational
expectations in the NLSY also conflicts with Coleman and DeLeire’s findings from the
NELS. A complete test of the predictions of Coleman and DeLeire’s model requires
information on teenagers’ income expectations in addition to their occupational
expectations. However, a question about income expectations is not available in the
NLSY, so I am unable to test whether locus of control affects income expectations, and
my empirical test is incomplete. It follows that my findings in section 1.3.2 hardly
provide a convincing rejection of the Coleman-DeLeire model. This is particularly true
given the fact that Coleman and DeLeire find stronger results for income expectations
than for occupational expectations.5

The finding that locus of control is associated with higher subsequent earnings is
consistent with the results from previous research by Andrisani (1977). The estimates
based on the NLSY data suggest that, although the return to locus of control is smaller
than the return to the AF QT, it is still substantial.

While the data used in this study do not fit the Coleman-DeLeire model that
explicitly incorporates locus of control into the human capital investment model, this
does not mean that we should abandon the Coleman-DeLeire model and return to the

simplistic view of the early empirical literature. Rather, future research should more fully

 

5 To see if actual wages differ between lntemal Dropouts and External Dropouts, l have included an
interaction term between locus of control and educational attainment in the wage model. The estimates (not
reported here but available upon request) suggest that lntemal Dropouts cam, on average, about $2 (in year
2000 dollars) more per hour than External Dropouts; however the estimated difference is insignificant (p-
value = 0.16) at conventional significance levels using robust inference.

17

examine the mechanisms by which locus of control affects educational attainment and
economic outcomes. Whether attitudes developed during childhood years — as captured
by a concept like locus of control —— have long-term impacts on economic outcomes is
both important and intrinsically interesting. Locus of control is potentially important in

analyzing the investments parents, schools, and the public sector make in children.

18

CHAPTER 2
EMPLOYER-PROVIDED HEALTH INSURANCE AND LABOR SUPPLY OF

MARRIED WOMEN

2.1. Introduction

Employment-based health insurance coverage is the most common form of health
insurance in the United States: In 2006, 62.2 percent of nonelderly individuals and 70.9
percent of nonelderly workers were covered by employer-provided health insurance
(Fronstin 2007). That health insurance is tied to employment in the US. health care
system has received much attention from both policymakers and researchers.
Accordingly, a substantial body of research has been devoted to examining the
relationship between health insurance and various labor market decisions, including labor
force participation, hours worked, job mobility, and retirement (see the reviews by Currie
and Madrian 1999, Gruber 2000, and Gruber and Madrian 2004).

One important group whose labor force outcomes are likely to be affected by the
availability of health insurance coverage is married women. Because employers who
provide health insurance often provide it to both employees and their families, many
married women receive health insurance coverage through their husbands (Madrian
2006). This availability of alternative health insurance may reduce wives’ demand for
insurance in their own name, and therefore affect their labor market decisions.

In the past decade, a number of studies have examined the relationship between
husbands’ health insurance and wives’ labor supply (Buchmueller and Valletta 1999;
Olson 1998, 2000; and Wellington and Cobb-Clark 2000). Using cross-section data and

estimating reduced form labor supply equations for wives, they estimate that husbands’

19

health insurance has significant negative effects on their wives’ labor supply. However,
these cross-sectional estimates potentially suffer from bias due to the simultaneity of
wives’ labor supply and the health insurance status of their husbands.

There are two main reasons for husbands’ health insurance to be endogenous to
the labor supply decision of married women. First, unobserved personal characteristics of
husbands and wives that affect labor supply—such as preferences for work—might be
correlated due to the marriage selection process (Lundberg 1988). For example, if women
with strong preferences for leisure, child rearing, or home production tend to marry men
who work long hours and hence provide health insurance to the family, then cross-
sectional regressions of wives’ labor supply on spousal coverage would overstate the
negative effect of spousal coverage.

A second possible source of endogeneity is that husbands and wives may make
joint job choice decisions, which depend on the health insurance options available to the
family (Black 2000, and Scott, Berger, and Black 1989). For example, because coverage
rates typically increase with firm size,6 husbands may sort into larger firms or into
industries—such as the manufacturing or the public sector—that are more likely to
provide health insurance coverage to their workers.7 Again, this sorting behavior would
lead the cross-sectional estimates to overstate the negative effect of spousal coverage.

This paper attempts to avoid some of the limitations of earlier empirical work,

which likely estimates a spurious negative relationship between spousal coverage and

 

6 According to the Employee Benefit Research Institute (EBRI) tabulations of data from the March 2007
Supplement to the Current Population Survey (CPS), 27.0 percent of workers in firms with fewer than 10
employees were covered by their own employer’s health plan in 2006, compared with 65.0 percent of
workers in firms with 1000 or more employees (Fronstin 2007).

7 According to the EBRI estimates of the CPS, in 2006, 67.1 percent of workers in the manufacturing
sector, and 74.2 percent of workers in the public sector were covered by their own employer’s health plan
(Fronstin 2007).

20

wives’ labor supply, by using several panel data methods. Using data from the National
Longitudinal Survey of Youth (N LSY) and the March Annual Demographic Supplements
to the Current Population Survey (CPS), I compare results obtained from three
econometric approaches: (1) cross-sectional estimation of linear probability and probit
models of labor supply, which (perhaps incorrectly) assumes exogeneity of spousal
coverage; (2) cross-sectional instrumental variables estimation to handle the potential
endogeneity of spousal coverage; and (3) panel data methods to account for the marriage
selection process and the joint job choice decisions. The findings suggest the negative
effect of spousal coverage on labor supply found in previous cross-sectional studies
results mainly from spousal sorting and selection. Once unobserved heterogeneity is
controlled, a relatively smaller estimated effect of spousal coverage on wives’ labor
supply remains.

The paper is organized as follows. Section 2.2 summarizes the empirical literature
on the effects of spousal coverage on labor supply. Section 2.3 describes the econometric
methodology and discusses how the possible endogeneity of spousal coverage can be
handled. Section 2.4 describes the data, Section 2.5 presents the main results, and Section

2.6 concludes.

2.2. Previous Research

Previous research on the effects of husbands’ health insurance coverage on wives’
labor supply has relied on the assumption that a husband’s coverage is exogenous to the
labor supply decisions of his wife (Buchmueller and Valletta 1999; Olson 1998, 2000;
and Wellington and Cobb-Clark 2000). Using cross-sectional data, these studies estimate

that husbands’ health insurance has significant negative effects on wives’ labor supply. In

21

particular, Buchmueller and Valletta (1999) use data from the April 1993 CPS and
estimate that a husband’s health insurance reduces his wife’s probability of working by
12 percent and her hours of work by 36 percent.

Buchmueller and Valletta’s key independent variable is whether or not the
husband has health insurance coverage from an employer (spouse’s insurance). Olson
(1998) points out that a wife’s labor supply decision is affected not by her husband
having his own health insurance (that is, whether or not it covers the wife, as used by
Buchmueller and Valletta 1999), but rather by whether she receives coverage through her
husband’s health insurance plan (spousal coverage). Still, using data from the March
1993 CPS, Olson finds effects similar to those estimated by that Buchmueller and
Valletta: Spousal coverage reduces the probability a wife will work by 11 percent and her
hours of work by 20 percent.

Like Olson (1998), Wellington and Cobb-Clark (2000) estimate the labor supply
effects of spousal coverage. They use the March 1993 CPS data and estimate a 23 percent
reduction in wives’ labor force participation due to coverage through husbands’ health
insurance.8 Conditional on working, spousal coverage is estimated to reduce hours of
work by 17 percent for white wives and 8 percent for black wives.

Buchmueller and Valletta (1999), Olson (1998, 2000); and Wellington and Cobb-
Clark (2000) all note that the assumption that spousal coverage is exogenous to the
wives’ labor supply is questionable. Buchmueller and Valletta use a multinomial logit

model in an attempt to account for the unobserved heterogeneity among married couples.

 

a The larger effects estimated by Wellington and Cobb-Clark (2000) may be explained by their inclusion of
both spouse’s insurance and the key variable, spousal coverage in labor supply models. Because estimated
coefficients of spouse’s insurance are positive and significant, the net effects of spousal coverage estimated
by Wellington and Cobb-Clark (that is, husbands have insurance that covers their wives) are about the same
magnitude as those estimated by Olson (1998).

22

Olson shows that the estimated effects are sensitive to econometric specification and the
underlying exogeneity assumption. Wellington and Cobb-Clark try to sign the bias that
would result from ignoring the potential endogeneity of spousal coverage. The present
study takes a different approach from earlier work in estimating the effect of spousal
coverage on wives’ labor supply, particularly along the methodological lines described in

the next section.

2.3. Empirical Methodology
An empirical model that relates the labor supply of married women to their health

insurance coverage by their husbands’ employer-provided plan can be specified:

LS,- = asPCOV, + WIFE,- ,6] + HUSBAND, 32 + FAMILY,- 53 + v,- (2.1)

where LS is the labor supply of a married woman in family i. The key variable, SPCO V,
equals one if the wife in family i is covered by her husband’s employer-provided health
insurance, and zero otherwise. WIFE is a vector of wives’ personal characteristics
(education, experience, experience squared, and race); HUSBAND denotes husbands’
personal characteristics (education, experience, and experience squared); and FAMILY
denotes family characteristics (presence of children under age 6, number of children
under age 18, region of residence, and family non-wage income). Finally, v represents
unobservable factors that affect the labor supply of wife in family 1'.

Estimating equation (2.1) by ordinary least squares (or probit) may produce
inconsistent estimates if unobservable characteristics among married couples that affect
wives’ labor supply (v) are systematically related to the availability of husbands’ health
insurance coverage (SPC 0 V). One possible solution to this endogeneity problem is to

find valid instrumental variables for SPCO V. To be valid, instruments must be correlated

23

with husbands’ employer-provided health insurance but unrelated to unobservables
affecting wives’ labor supply. Some previous studies have suggested husbands’ job
characteristics (having a part-time job, working in a small firm, and being self-employed)
as instruments for SPCOV(Abraham and Royalty 2005). These job characteristics
negatively affect the availability of husbands’ employer-provided health insurance, but
they are likely correlated with wives’ preferences for work or ability due to the marriage
selection process and the jointness of job choice decisions.

Panel data offers additional possibilities to avoid the potential problem of
endogeneity. As long as unobservables that affect wives’ labor supply remain constant
over time, panel data may solve the endogeneity problem in equation (2.1). For example,

an empirical model with unobserved effects can be written:

L5,, = 6 SPCO Vi, + WIFE” 31 + HUSBAND,~, 132 + FAMIL Y,,,83 + c,- + uit (2.2)

where i indexes families, and t indexes time; c,- represents time-invariant unobserved.

effects on wives’ labor supply; and “it represents time-varying unobserved effects on
wives’ labor supply.
If we can assume that time-invariant unobserved effect, c,-, are uncorrelated with

each explanatory variable in equation (2.2) across all time periods, then equation (2.2)
becomes a random effects (RE) model. Possible time-invariant influences on labor supply
include differences in ability or in preferences for work. It is likely that both factors are

correlated with the availability of spousal coverage because of the marriage selection

24

process. In this case, the RE estimator is inconsistent, and first-differencing (FD) or
fixed-effects (FE) methods are required for consistent estimation.9

FD and F E methods produce consistent estimates of the effect of spousal coverage
on wives’ labor supply by allowing for arbitrary correlation between unobserved
individual effects and the explanatory variables in equation (2.2). The consistency of FD
and FE estimators depends on three assumptions: (1) time-varying unobserved effects
that are uncorrelated with the explanatory variables across all time periods; (2) sufficient
variation in spousal coverage over time; and (3) strict exogeneity of explanatory
variables.

The strict exogeneity assumption rules out feedback effects from wives’ labor
supply to husbands’ health insurance coverage in subsequent years. For example, if
husbands switch to jobs which would provide health insurance to the family because their
wives decide to leave the labor force or work short hours (and hence are less likely to
receive health insurance coverage from their own employer), the strict exogeneity
assumption is violated, and the FD and FE estimators are inconsistent. In this case, one
possible approach to consistent estimation involves using instrumental variables methods

applied after a FD or FE transformation, provided valid instruments are available.

2.4. Data
To examine the effect of spousal coverage on labor supply decisions of wives, I
use data from the National Longitudinal Survey of Youth (NLSY). The NLSY is a

sample of 12,686 young men and women aged 14 to 22 at the time of the first interview

 

9 The choice between FD and FE depends on the assumptions about time-varying unobserved effects. The
FE estimator is more efficient if they are serially correlated, while the FD estimator is more efficient when
they follow a random walk.

25

in 1979. Since their first interview, they were reinterviewed annually until 1994, and then
biennially from 1996 to the present. Each survey collected information on demographic
characteristics, employment, income, and family structure. In this study, I use eight
interviews of the NLSY: 1989, 1990, 1992, 1993, 1994, 1996, 1998, and 2000. These
years are selected because the questionnaire included detailed questions on the
availability and sources of health insurance coverage only in these survey years.
Specifically, respondents were asked whether they were covered by a health plan. If the
respondent answered “yes,” the interviewer asked who paid for the plan. Responses
included current employer, previous employer, spouse’s employer, purchased directly,
and Medicaid or welfare source. If the respondent was married, the same set of questions
on health insurance coverage was asked about the wife or husband. These questions allow
me to identify the wives who had health insurance coverage through their husbands’
employer-provided health insurance plan.

I am able to follow a sample of 20,396 married women from eight interviews of
the NLSY over twelve years. An observation is included if (1) the respondent had an
intact marriage from 1989 through 2000; (2) both the respondent and her husband were
between the ages of 25 and 64; (3) information on the respondent’s source of health
insurance was available; and (4) the respondent was not covered by public health
insurance. The resulting sample, after pooling all 8 years, includes 12,822 married
women from the NLSY.

A second source of data I use is the March 2000 Annual Demographic
Supplement to the Current Population Survey (CPS), mainly for comparison purposes.

The March CPS provides information on demographic characteristics, employment,

26

income, family structure, and health insurance that are comparable to those collected in
the NLSY. Using information on both health insurance coverage and sources of that
coverage, I can differentiate between wives with and without coverage through their
husbands’ employer-provided health insurance plan. The criteria used to select the CPS
sample are similar to the NLSY: Married couples aged 25 to 64 who did not receive
public health insurance. The final sample consists of 19,515 married women from the
CPS.

I examine three alternative measures of labor supply: (1) working, a binary
variable indicating labor force participation defined as positive hours per week; (2) full-
time, a binary variable indicating hours worked per week is greater than or equal to 35;
and (3) hours, usual hours worked per week.

Table 2.1 displays the three measures of married women’s labor supply of by
spousal health insurance coverage for both the CPS (lefi-hand panel) and the NLSY
(right-hand panel). The top panel of Table 2.1 shows the labor supply of the entire sample
of married women (includes both workers and non-workers), and the bottom panel shows
the labor supply of working married women (a subsample of the entire sample of married
women because it includes only women whose weekly hours of work are greater than
zero).

The top lefi-hand panel in Table 2.1 suggests potential negative effects of spousal
coverage on married women’s labor supply in the CPS. Wives who were covered by their
husbands’ employer-provided health insurance (72.8 percent) were less likely to work
than wives who were not (83.8 percent). When they worked (bottom left—hand panel of

Table 2.1), they were much less likely to work full-time: 65.0 percent of wives with

27

spousal coverage worked full-time, compared with 83.4 percent of wives without spousal
coverage.

The negative correlation between married women’s labor supply and spousal
coverage is also apparent in the NLSY. The fraction working (top right-hand panel of
Table 2.1) was much lower for wives with spousal coverage (73.8 percent) than for wives
without spousal coverage (90.1 percent). Among working wives (bottom right-hand panel
of Table 2.1), those who had spousal coverage were 24.5 percentage points less likely to
work full-time (57.8 percent versus 82.7 percent).

Table 2.2 displays mean characteristics of the entire sample of married women by
spousal coverage in the CPS (left-hand panel) and the NLSY (right-hand panel). Overall,
the characteristics of married women in the two data sets are quite similar; however, there
are some important differences between wives who have spousal coverage and wives
who do not. Wives with spousal coverage are more likely to have children under age 6
and more children under age 18 than wives without spousal coverage. Also, they tend to
be married to more educated husbands. For example, in the CPS, 64 percent of wives
with spousal coverage are married to husbands with some college or more, compared
with 54 percent of wives without coverage. (The fractions are 59 percent versus 46
percent in the NLSY).

Table 2.3 displays mean characteristics of the sample of working married women
by spousal coverage in both the CPS and NLSY. The presentation is similar and findings

are analogous to those in Table 2.2.

28

Table 2.1. Labor Supply of Married Women by Spousal Health Insurance
Coverage, 2000 CPS and NLSY

 

 

Wife

Wife

Covered Not covered

Wife

Wife

Covered Not covered

 

All married women

 

Working (%) 72.8 83.8 73.8 90.]
(Weekly hours>0)

Full-time (%) 47.3 69.9 42.7 74.5
(Weekly hours>=35)

Hours (weeks) 25.0 32.5 25.7 36.1
Number of observations 10,019 9,496 1,076 1,113
Working married women

Full-time (%) 65.0 83.4 57.8 82.7

(Weekly hours>=35)

Hours (weeks) 34.4 38.8 34.9 40.1

Number of observations 7,337 7,907 807 1,010

 

Notes: The data are from the March 2000 Annual Demographic Supplement to the
Current Population Survey (CPS) and the 2000 interview of the 1979 National
Longitudinal Survey of Youth (NLSY). Means are tabulated using CPS March
supplement and NLSY weights.

29

Table 2.2. Summary Statistics for Married Women by Spousal Health Insurance
Coverage, 2000 CPS and NLSY

 

 

 

CPS NLSY
Wife Wife Not Wife Wife Not

Variable All Covered Covered All Covered Covered
Spousal coverage 0.52 1 0 0.52 1 0
Labor supply measures

Working 0.78 0.73 0.84 0.82 0.74 0.90

Full-time 0.58 0.47 0.70 0.58 0.43 0.74

Hours 28.61 25.04 32.55 30.67 25.74 36.10
Education

Less than high school 0.09 0.07 0.1 l 0.05 0.05 0.05

High school 0.33 0.34 0.32 0.40 0.38 0.42

Some college 0.28 0.30 0.27 0.24 0.23 0.25

College 0.21 0.21 0.20 0.18 0.20 0.15

More than college 0.09 0.08 0.10 0.13 0.13 0.14
Experience 22.57 22.21 22.96 19.74 19.74 19.75
Race

White 0.88 0.89 0.86 0.87 0.90 0.84

Black 0.07 0.06 0.08 0.07 0.06 0.09

Other 0.05 0.05 0.06 0.06 0.05 0.06
Presence of children

under age 6 0.24 0.27 0.20 0.27 0.30 0.23
Number of children

under age 18 1.09 1.24 0.92 1.83 2.02 1.61
Family nonwage income 4,089 4,468 3,671 5,277 6,080 4,415
Husband's education

Less than high school 0.11 0.07 0.15 0.08 0.06 0.10

High school 0.30 0.29 0.32 0.39 0.35 0.44

Some college 0.26 0.27 0.25 0.21 0.21 0.22

College 0.21 0.22 ‘ 0.19 0.17 0.21 0.12

More than college 0.12 0.14 0.10 0.15 0.17 0.12

Husband's experience 24.38 23.78 25.04 21.64 21.08 22.27
Number of observations 19,515 10,019 9,496 2,189 1,076 1,113

 

Notes: See Table 2.1.

30

Table 2.3. Summary Statistics for Working Married Women by Spousal Health
Insurance Coverage, 2000 CPS and NLSY

 

 

 

CPS NLSY
Wife Wife Not Wife Wife Not

Variable All Covered Covered A11 Covered Covered
Spousal coverage 0.49 1 0 0.47 1 0
Labor supply measures

Working 1 1 1 l 1 1

Full-time 0.74 0.65 0.83 0.71 0.58 0.83

Hours 36.68 34.41 38.85 37.60 34.89 40.06
Education

Less than high school 0.07 0.05 0.08 0.04 0.04 0.04

High school 0.32 0.33 0.31 0.41 0.40 0.41

Some college 0.29 0.31 0.28 0.25 0.24 0.25

College 0.22 0.22 0.21 0.17 0.18 0.15

More than college 0.10 0.09 0.11 0.14 0.14 0.14
Experience 22.03 21.72 22.32 19.71 19.73 19.68
Race

White 0.87 0.89 0.86 0.86 0.89 0.84

Black 0.08 0.07 0.08 0.08 0.06 0.10

Other 0.05 0.04 0.05 0.06 0.05 0.06
Presence of children 0.22 0.24 0.19 0.23 0.25 0.21

under age 6
Number ofchildren 1.04 1.19 0.89 1.71 1.90 1.55

under age 18
Family nonwage income 3,990 4,428 3,572 4,799 5,545 4,146
Husband's education

Less than high school 0.09 0.06 0.12 0.08 0.05 0.10

High school 0.31 0.30 0.32 0.41 0.37 0.44

Some college 0.27 0.29 0.26 0.22 0.22 0.23

College 0.21 0.22 0.20 0.16 0.21 0.12

More than college 0.12 0.14 0.10 0.13 0.15 0.11
Husband's experience 24.05 23.54 24.54 21.76 21.17 22.31
Number of observations 15,244 7,337 7,907 1,817 807 1,010

 

Notes: See Table 2.1.

31

2.5. Empirical Findings

This section presents estimates of the effect of husbands’ health insurance
coverage on wives’ labor supply from three econometric approaches. These include (1)
cross-sectional estimates from linear probability and probit models; (2) cross-sectional
instrumental variable estimates; and (3) panel estimates.

2.5.1 Cross-Sectional LPM and Probit Estimates

Table 2.4 displays the estimated marginal effects of spousal coverage on labor
force participation of married women from linear probability and probit models [equation
(2.1)]. The left-hand panel in Table 2.4 shows the estimated marginal effects for married
women in the CPS, and the right-hand panel shows the estimated marginal effects for
married women in the NLSY. Controls include wives’ personal characteristics
(education, experience, experience squared, and race); husbands’ personal characteristics
(education, experience, and experience squared); and family characteristics (presence of
children under age 6, number of children under age 18, region of residence, and family
non-wage income).

The estimate of main interest is the marginal effect of spousal coverage on labor
force participation. This is essentially similar in size and statistical significance in both
models and both data sets. The estimated marginal effect, about -0.11, suggests that
spousal health insurance coverage reduces wives’ probability of participation by 11
percentage points. This represents a reduction of 12 percent compared with wives who do
not have spousal coverage.

These participation estimates are very similar to those obtained by Buchmueller

and Valletta (1999) and by Olson (1998), who estimated that spousal coverage reduced

32

married women’s labor supply by 12 percent and 11 percent respectively. However, they
are smaller than the 23 percent reduction (19.5 percentage points) estimated by
Wellington and Cobb-Clark (2000).

Table 2.5 displays the estimated marginal effects of spousal coverage on full-time
employment of working married women from linear probability and probit models. The
estimates are conditional on positive labor force participation. The controls used in the
estimation are the same as those in Table 2.4. In both the CPS and NLSY, the estimated
marginal effect of spousal coverage is about -0.16, which suggests that the probability of
working full-time is 16 percentage points (22 percent) lower for working wives with
spousal coverage than for working wives without spousal coverage.

2.5.2 Cross-Sectional IV Estimates

Table 2.6 shows the coefficients on spousal coverage in three labor supply models
estimated by OLS and ZSLS using data from the CPS. Because coefficients of control
variables are of limited interests, they are not reported. The controls included are the
same as before (see the table notes). Again, three alternative measures of labor supply are
used as dependent variables: (1) a binary variable indicating labor force participation
(Working); (2) a binary variable indicating hours worked per week is greater than or equal
10 35 (full-time); and (3) usual hours worked per week (hours).

The instruments for spousal coverage are a dummy variable indicating whether
the husband is self-employed and a dummy variable indicating whether the husband
Works part-time. This choice of instruments is based on the assumption that a wife’s
c0Verage by her husband’s employer-provided health insurance is negatively correlated

With the husband’s having a part-time job or being self-employed, but unrelated with

33

Table 2.4. Linear Probability and Probit Estimates of Married Women's
Labor Force Participation, 2000 CPS and NLSY

 

 

 

Dependent Variable: working CPS NLSY
Variable LPM Probit LPM Probit
Spousal coverage -0.103 -0. 109 -0.106 -0. l 15
(0.006) (0.006) (0.018) (0.017)
High school 0.117 0.087 0.148 0.112
(0.013) (0.010) (0.045) (0.030)
Some college 0.168 0.132 0.160 0.112
(0.014) (0.010) (0.048) (0.027)
College 0.195 0.149 0.182 0.121
(0.015) (0.009) (0.054) (0.022)
More than college 0.249 0.172 0.262 0.138
(0.017) (0.007) (0.061) (0.017)
Experience 0.013 0.01 1 -0.009 -0.012
(0.002) (0.002) (0.025) (0.023)
Experience squared 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.001) (0.001)
Black 0.053 0.056 0.046 0.048
(0.011) (0.012) (0.022) (0.021)
Other race -0.036 -0.041 0.046 0.045
(0.014) (0.015) (0.023) (0.020)
Midwest 0.034 0.039 0.039 0.038
(0.008) (0.008) (0.028) (0.023)
South -0.016 -0.016 -0.005 -0.011
(0.008) (0.009) (0.026) (0.024)
West -0.018 -0.016 -0.008 -0.011
(0.009) (0.009) (0.030) (0.028)
Presence of children under age 6 -0.106 -0.1 19 -0.1 17 -0.120
(0.009) (0.010) (0.023) (0.024)
Number of children under age 18 -0.036 -0.035 -0.035 -0.033
(0.003) (0.003) (0.008) (0.007)
Family nonwage income 0.005 0.005 -0.002 -0.001
(0.001) (0.001) (0.000) (0.000)

34

Table 2.4. (cont'd)

 

 

 

Dependent Variable: working CPS NLSY
Variable LPM Probit LPM Probit
Husband's education
High school 0.022 0.021 0.011 0.010
(0.012) (0.011) (0.030) (0.030)
Some college 0.023 0.022 0.019 0.022
(0.012) (0.012) (0.033) (0.032)
College 0046 -0.054 0.009 0.004
(0.014) (0.015) (0.040) (0.038)
More than college -0.092 -0.117 -0.046 -0.049
(0.016) (0.019) (0.045) (0.050)
Husband's experience 0.001 0.001 0.000 0.000
(0.002) (0.002) (0.007) (0.007)
Husband's experienced squared 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)
Number of observations 19,515 19,515 1,763 1,763
R-squared 0.103 0.099 0.141 0.161

 

Notes: The data are from the March 2000 Annual Demographic Supplement to the

Current Population Survey (CPS) and the 2000 interview of the 1979 National
Longitudinal Survey of Youth (NLSY). Figures are estimated changes in the

probability of labor force participation from linear probability and probit models.
Robust standard errors are in parentheses. The R-squared for the probits is the

pseudo-R-squared.

35

Table 2.5. Linear Probability and Probit Estimates of Working Married
Women's Full-Time Work, 2000 CPS and NLSY

 

 

 

Dependent Variable: full-time CPS NLSY
Variable LPM Probit LPM Probit
Spousal coverage -0. 153 -0. 156 -0.169 -0.174
(0.007) (0.007) (0.023) (0.024)
High school -0.044 -0.043 0.083 0.085
(0.015) (0.017) (0.055) (0.055)
Some college -0.031 -0.029 0.053 0.059
(0.016) (0.018) (0.059) (0.058)
College -0.010 -0.006 0.052 0.061
(0.018) (0.020) (0.068) (0.063)
More than college 0.059 0.066 0.113 0.111
(0.020) (0.020) (0.076) (0.059)
Experience -0.001 -0.002 -0.035 -0.036
(0.002) (0.002) (0.028) (0.032)
Experience squared 0.000 0.000 0.001 0.001
(0.000) (0.000) (0.001) (0.001)
Black 0.128 0.133 0.097 0.104
(0.012) (0.012) (0.027) (0.028)
Other race 0.107 0.104 0.115 0.116
(0.015) (0.014) (0.028) (0.025)
Midwest 0.025 0.024 -0.005 -0.007
(0.010) (0.010) (0.037) (0.036)
South 0.068 0.068 0.067 0.069
(0.010) (0.010) (0.032) (0.032)
West 0.023 0.022 -0.080 -0.095
(0.011) (0.010) (0.039) (0.042)
Presence of children under age 6 -0.039 -0.047 -0.036 —0.037
(0.011) (0.011) (0.028) (0.029)
Number of children under age 18 -0.057 -0.057 -0.033 -0.035
(0.004) (0.004) (0.010) (0.010)
Family nonwage income -0.002 -0.002 -0.002 -0.002
(0.001) (0.001) (0.000) (0.000)

36

Table 2.5. (cont'd)

 

 

Dependent Variable: full-time CPS NLSY
Variable LPM Probit LPM Probit
Husband's education
High school -0.006 -0.006 -0.010 -0.003
(0.013) (0.015) (0.038) (0.044)
Some college -0.025 -0.025 0.035 0.041
(0.014) (0.016) (0.042) (0.047)
College 0068 -0.070 -0.045 -0.037
(0.016) (0.019) (0.051) (0.058)
More than college -0.107 -0.117 -0.020 -0.008
(0.018) (0.023) (0.055) (0.060)
Husband's experience 0.003 0.003 0.000 -0.002
(0.002) (0.002) (0.009) (0.01 1)
Husband's experienced squared 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)
Number of observations 15,244 15,244 1,473 1,473
R-squared 0.083 0.076 0.142 0.131

 

Notes: The data are from the March 2000 Annual Demographic Supplement to the

Current Population Survey (CPS) and the 2000 interview of the 1979 National
Longitudinal Survey of Youth (NLSY). Figures are estimated changes in the

probability of full-time work from linear probability and probit models. Robust
standard errors are in parentheses. The R-squared for the probits is the pseudo—R-

squared.

unobservable factors affecting the wife’s labor supply decisions. The latter of these
assumptions is tenuous because the husband’s having a part-time job or being self-
employed is likely correlated with the wife’s preferences for work if there is spousal
sorting and selection. However, the purpose is to see whether an IV approach that has

been used in the past yields plausible findings.

37

The top panel of Table 2.6 displays the estimated effects of spousal coverage on
the participation of all married women. The OLS estimate repeats the main finding from
Table 2.4. The ZSLS estimate is essentially similar (-0.091). As expected, the standard
error of the ZSLS estimate is larger than the OLS standard error (0.024 versus 0.006), but
the estimate is still significant at conventional levels (p-value= 0.000).

In the model of working married women’s full-time work (middle panel of Table
2.6), the OLS estimate is -0. 153, which suggests that spousal coverage reduces wives’
probability of full-time work by 15.3 percentage points (a 21.9 percent reduction). But
the estimate is quite different when we use ZSLS: The estimated spousal effect is now
0.236 (p-value= 0.000). Similarly, in the hours equation for working married women
(bottom panel of Table 2.6), the OLS estimate (-3.772) suggests that wives with spousal
coverage work almost 4 hours less per week than wives without spousal coverage (p-
value= 0.000), but ZSLS estimate is 3.413 with a p-value of 0.000.

Table 2.7 shows the estimated effect of spousal coverage on labor supply for
women in the NLSY. The presentation is similar to that of Table 2.6. Because
information on whether the husband is self-employed is not available in the NLSY,
husbands’ part-time work status is the only available instrument for spousal coverage.
This, however, does not affect the main finding: Once again, the OLS and ZSLS
estimates differ in significant ways.

That OLS and ZSLS estimations produce such different results suggests the
importance of testing for the endogeneity of spousal coverage in the labor supply
equations. The regression-based Hausman test that compares OLS and ZSLS estimates

(see Wooldridge 2002, Section 6.2.1) is well-suited for this. Heteroskedasticity-robust

38

Hausman test statistics are given in both Tables 2.6 and 2.7 (see ‘Hausman test’). The test
statistic suggests strong evidence of endogeneity of spousal coverage in the three labor
supply models for both samples in both data sets (p-values = 0.000). This suggests in turn
that ZSLS is needed for consistent estimation provided the validity of the instruments,
which is what we consider next.

Because the ZSLS estimation in the CPS uses two instruments for spousal
coverage, there is one overidentifying restriction. The overidentifying restriction can thus
be tested in the model estimated using the CPS data, but not when using the NLSY. In
order to determine the validity of the instruments, I use the Sargan statistic. This is
calculated as NXR-squared from a regression of the IV residuals on the full set of
instruments (Wooldridge 2002, Section 6.2.2; Baum, Schaffer, and Stillman 2003). The
null hypothesis is that the variables used as instruments for spousal coverage (husband’s
self-employment and part-time work) are uncorrelated with unobservables affecting
wives’ labor supply.

Table 2.6 displays the results. In the model of married women’s participation, the
Sargan test rejects the hypothesis of instrument validity (p-value = 0.03 9). For the sample
of working married women, the instruments pass the overidentification test in the full-
time work equation, which suggests that the instruments are valid (p-value = 0.728), but
in the hours equation, we reject the validity assumption of the instruments at the 10-
percent significance level (p-value = 0.089).

That Sargan test produces inconsistent results is not actually surprising because it
evaluates the entire set of overidentifying restrictions, but requires that at least some of

the instruments be valid. However, as discussed before, using variables for husband’s

39

Table 2.6. OLS and ZSLS Estimates of the Effect of Spousal Coverage on Labor

Supply: 2000 CPS

 

All married women

 

 

 

 

 

Dependent variable: working OLS ZSLS
Spousal coverage -0. 1 03 -0.091
(0.006) (0.024)
First stage F-statistic 675.38
[p-value] [0.000]
Hausman test -0.103
[p-value] [0.000]
Sargan statistic 4.269
[p-value] [0.039]
Working married women
Dependent variable: full-time OLS ZSLS
Spousal coverage -0. 153 0.236
(0.007) (0.033)
First stage F-statistic 575.42
[p-value] [0.000]
Hausman test -0.177
[p-value] [0.007]
Sargan statistic 0.121
[p-value] [0.728]
Working married women
Dependent variable: hours OLS ZSLS
Spousal coverage -3.722 3.41 3
(0.186) (0.934)
First stage F-statistic 575.42
[p-value] [0.000]
Hausman test -4. 160
[p-value] [0.000]
Sargan statistic 2.889
[p-value] [0.089]

 

Notes: The data are from the March 2000 CPS. Robust standard errors are in
parentheses. In brackets are p-values. All models include wives’ characteristics;
husbands’ characteristics; and family characteristics as controls. The instruments for
spousal coverage include a dummy variable indicating whether the husband is self-
employed and a dummy variable indicating whether the husband is working part-time.
The first stage F-statistic is a test statistic for joint significance of the instruments in
the first stage regression of spousal coverage on the exogenous variables and
instrrunents. The Hausman test is a regression-based Hausman test. The Sargan
statistic is a test of overidentifying restrictions.

40

Table 2.7. OLS and ZSLS Estimates of the Effect of Spousal Coverage on Labor

 

 

 

 

 

 

Supply: 2000 NLSY
All married women
Dependent variable: working OLS ZSLS
Spousal coverage -0. 106 0.083
(0.018) (0.288)
First stage t-statistic -2.820
[p—value] [0.005]
Hausman test -0.103
[p-value] [0.000]
Working married women
Dependent variable: full-time OLS ZSLS
Spousal coverage -0. 169 0.094
(0.023) (0.331)
First stage t-statistic -2.980
[p-value] _ [0.003]
Hausman test 0171
[p-value] [0.000]
Working married women
Dependent variable: hours OLS ZSLS
Spousal coverage -3.082 1 1.740
(0.711) (11.503)
First stage t-statistic -2.980
[p—value] [0.003]
Hausman test -2.765
[p-value] [0.000]

 

Notes: The data are from the 2000 interview of the 1979 National Longitudinal Survey
of Youth (NLSY). Robust standard errors are in parentheses. In brackets are p-values.
All models include wives’ personal characteristics (education, experience, experience
squared, and race); husbands’ personal characteristics (education, experience, and
experience squared); and family characteristics (presence of children under age 6,
number of children under age 18, region of residence, and family non-wage income) as
controls. The instrument for spousal coverage is a dummy variable indicating whether
the husband is working part-time. The first stage t-statistic is a test statistic for the
significance of the instrument in the first stage regression of spousal coverage on the
exogenous variables and the instrument. The Hausman test is a regression-based
Hausman test. Sample sizes are 1,763 (all married women) and 1,472 (working
married women).

41

work status as instruments for spousal coverage raises concerns because the husband’s
having a part-time job or being self-employed is likely correlated with the wife’s
preferences for work due to spousal sorting and selection.

Given that cross-sectional IV estimation does not provide a solution to the
endogeneity problem of spousal coverage, I now turn to panel estimates.
2.5.3 Panel Estimates

Table 2.8 shows estimates of the spousal coverage effect using the NLSY. Unlike
the estimates in sections 2.5.1 and 2.5.2, these estimates take advantage of the panel
nature of the NLSY and use the 1989, 1990, 1992, 1993, 1994, 1996, 1998, and 2000
interviews of the NLSY. Table 2.8 displays estimates of four different models: (1) pooled
ordinary least squares (POLS); (2) random effects (RE); (3) fixed effects (FE); and (4)
first differencing (FD). The POLS and RE models include a full set of year dummies,
wives’ personal characteristics (education, experience, experience squared, and race);
husbands’ personal characteristics (education, experience, and experience squared); and
family characteristics (presence of children under age 6, number of children under age
18, region of residence, and family non-wage income) as controls. The FE and FD
models include the same controls, except race, which is not time varying.

The POLS estimates of the effect of spousal coverage are given in column 1 of
Table 2.8. The reported standard errors are robust to serial correlation and
heteroskedasticity. The estimated effect of spousal coverage is negative and statistically
significant in the three labor supply models for both samples of married women. The
results suggest that married women with spousal coverage are 9.4 percentage points (14

percent) less likely to be working, and those who work are 18.4 percentage points (28

42

percent) less likely to work full time. In the hours equation for working married women,
the estimated effect of spousal coverage is -3.790 (p-value = 0.000), which suggests that
married women with spousal coverage work almost 4 hours less per week (a 13 percent
reduction) than married women without spousal coverage.

Column 2 in Table 2.8 displays the random effects (RE) estimates. The estimated
spousal coverage effects are still negative and statistically significant but they are smaller
in absolute value than the POLS estimates. Thus, it seems that controlling for random
unobserved effects (assuming they are uncorrelated with explanatory variables)
diminishes the negative effect of spousal coverage on wives’ labor supply.

The fixed effects (FE) model in column 3 produces estimated coefficients for
spousal coverage that are even smaller in absolute value. The FE estimates are about half
the size of the RE estimates. For example, spousal coverage is estimated to reduce
weekly hours worked by 1.6 hours for working married women, compared with 2.5 hours
in the RE model.

Because the key assumption underlying the consistency of RE is whether
unobserved effects and the explanatory variables are correlated, I test this assumption
using a Hausman test that compares random and fixed effects estimates (see Wooldridge
2002, Section 10.7.3). The test strongly rejects (p-value = 0.000) the hypothesis that
unobserved effects are uncorrelated with spousal coverage in the three labor supply
models for both samples of married women. This suggests that the random effects
estimators are inconsistent but the fixed effect estimator is consistent conditional on strict

exogeneity of the explanatory variables.

43

Table 2.8. Panel Estimates of the Effect of Spousal Coverage on Labor

Supply: 19139-2000 NLSY

 

All married women

 

 

 

 

 

Dependent variable: working (1) POLS (2) RE (3) FE (4) FD
Spousal coverage -0.094 -0.051 -0.026 -0.005
(0.009) (0.007) (0.009) (0.009)
Hausman test 125. 16
[p-valuel [0.000]
Strict exogeneity test -0.020
[p-value] [0. 120]
Number of observations 12,888 12,888 12,888 8,543
Working married women
Dependent variable: full-time (l) POLS (2) RE (3) FE (4) FD
Spousal coverage -0.184 -0.132 -0.088 -0.052
(0.012) (0.009) (0.013) (0.012)
Hausman test 99.79
[p-value] [0.000]
Strict exogeneity test -0.036
[p-value] [0.280]
Number of observations 10,584 10,5 84 10,5 84 7,072
Working married women
Dependent variable: hours (1) POLS (2) RE (3) FE (4) FD
Spousal coverage -3 .790 -2.504 -1 .596 -0.989
(0.318) (0.238) (0.338) (0.380)
Hausman test 82.81
[p-value] [0.000]
Strict exogeneity test -0.883
[p-value] [0. 196]
Number of observations 10,584 10,584 10,584 7,072

 

Notes: The data are from the 1989-2000 interviews of the 1979 National
Longitudinal Survey of Youth (NLSY). Robust standard errors are in parentheses.
In brackets are p-values. POLS and RE models include a full set of year dummies,
wives’ personal characteristics (education, experience, experience squared, and
race); husbands’ personal characteristics (education, experience, and experience
squared); and family characteristics (presence of children under age 6, number of
children under age 18, region of residence, and family non-wage income) as
controls. FE and FD models include the same controls, except race, which is not
time varying. Hausman test is based on the comparison of estimates obtained from
the RE and FE models. Strict exogeneity test is the test for feedback effects from
dependent variable to future values of explanatory variables.

44

Finally, estimates obtained using first differencing (FD) are given in column 4 of
Table 2.8. The FD estimates are even smaller in absolute value than the FE estimates but
generally the same order of magnitude. To choose between FE and FD, I test whether the

differenced errors are serially uncorrelated. The results indicate that there is substantial

negative serial correlation in the differenced errors (,5 z -0.30 with p-value = 0.000),

suggesting fixed effects is more efficient than first differencing (Wooldridge 2002,
Section 10.7.1).

As mentioned above, consistency of FE estimates requires that spousal coverage

be strictly exogenous with respect to time-varying unobserved effects—u” in equation

(2.2), after accounting for the individual effect—c,- in equation (2.2). To test for feedback

effects from wives’ labor supply to future values of spousal coverage, I generate the lead
of the spousal coverage variable and use it as a regressor in the FE model (Wooldridge
2002, Section 10.7.1). The estimated leads of spousal coverage are insignificant in the
three labor supply models for both samples of married women. This suggests that there is
no evidence against strict exogeneity of spousal coverage after netting out the individual

effect using the FE.

2.6. Conclusion

The empirical analysis in this paper yields three main findings. First, there is
strong evidence that spousal coverage is endogenous to the labor supply of married
women (section 2.5.2). This results from the simultaneity of wives’ labor supply
decisions and the health insurance status of their husbands. Second, cross-sectional
instrumental variables estimation does not provide a viable solution to the endogeneity

problem (section 2.5.2). The close link between health insurance and labor supply makes

45

it difficult to identify suitable instruments that are correlated with wives’ coverage by
their husbands’ health insurance, but unrelated with wives’ labor supply decisions. This is
confirmed by the results of the test for overidentifying restrictions (see ‘Sargan test’
presented in Table 2.6), that reject the hypothesis of instrument validity. Third, once
unobserved heterogeneity is controlled, spousal coverage has a smaller negative effect on
wives’ labor supply (section 2.5 .3).

Specifically, it appears that controlling for sorting and selection of married
couples diminishes the negative effect of spousal coverage found in cross-sections. The
FE estimates in Table 2.8 suggest that spousal coverage reduces wives’ probability of
participation by 7.7 percent. Conditional on working, spousal coverage is estimated to
reduce the probability wives’ work full-time by 15 percent, and their hours of work by
6.5 percent. In contrast, the cross-sectional estimates, which are similar to those estimated
by Buchmueller and Valletta (1999), Olson (1998, 2000), and Wellington and Cobb-
Clark (2000), suggest that spousal coverage reduces wives’ probability of participation by
14 percent; working wives’ probability of full-time work by 28 percent; and their hours
of work by 13 percent.

The results have potentially important implications for the debate over health care
reform in the United States. A major goal of most proposed reforms is to expand access
to health care and health insurance coverage. Doing this generally entails uncoupling
health insurance from employment, as would happen under universal single-payer health
care, or at least weakening the link, as occurs under a plan like that adopted by
Massachusetts in 2007. The question is whether the divorce of health insurance from

employment would reduce labor supply of workers who currently work (or have adjusted

46

their work hours) so as to acquire health insurance. The results of the present analysis
suggest that universal coverage would reduce the labor supply of married women, but not

as significantly as estimated by previous work.

47

CHAPTER 3
HEALTH INSURANCE TAX CREDITS AND HEALTH INSURANCE

COVERAGE OF LOW-INCOME SINGLE MOTHERS

3.1. Introduction

Between 2000 and 2006, the percentage of the US. nonelderly population without
health insurance coverage gradually rose from 15.6 percent to 17.9 percent (F ronstin
2007). Dissatisfaction with this level and grth of uninsurance has spurred interest in
reforming the existing system of financing health care, which is dominated among the
nonpoor and nonelderly by employer-provided health insurance. One approach, described
by Pauly (1999) and Cogan, Hubbard, and Kessler (2005) among others, is to adopt a
refundable health insurance tax credit (HITC) under the federal personal income tax.
Such a policy would grant a tax credit up to a prespecified maximum — for example,
$1,000 for an individual or $2,000 for a family — on a tax return where the filer
purchased a private health insurance policy (either provided by an employer or purchased
in the market).

An HITC would reduce the price of employer-provided health insurance and in
addition would extend the tax-favored treatment of health insurance to individuals who
do not have access to employer-provided health insurance. Accordingly, it would be
expected to increase the percentage of individuals and families covered by private health
insurance. But the extent to which the HITC would reduce the number of uninsured
individuals has been controversial. Pauly and Herring (2001 , 2002), Pauly, Song, and
Herring (2001), and Wozniak and Emmons (2000) simulated a variety of HITC policies

and found that a “reasonably generous” credit could reduce the number of uninsured

48

individuals by roughly 50 percent. However, simulations by Gruber (2000a, b) and
Gruber and Levitt (2000) suggested that the HITC might reduce the number of uninsured
by only about 10 percent. Emmons, Madly, and Woodbury (2005) replicated Gruber’s
simulation model and found (as is often true) that relatively minor changes in
assumptions could result in substantial changes in simulated impacts of the HITC. Their
conclusions echoed those of Pauly, Song, and Herring (2001): simulations of the impact
of health insurance tax credits are highly uncertain because little empirical evidence
exists to guide modelers in choosing appropriate behavioral assumptions. '0

Two strands of empirical literature do consider the effects of tax subsidies and tax
credits for health insurance. The first is a rather large literature — reviewed by Cutler and
Zeckhauser (2000) — examining how sensitive employees are to out-of-pocket premiums
when they select employer-provided health insurance. This literature suggests employees
are highly sensitive to premiums when they choose plans; for example, Cutler and Reber
(1998) estimate an elasticity of plan take-up with respect to the employee premium of
about -0.2.

A second (much smaller) empirical literature has examined the responsiveness of
employee take-up of health insurance to changes in employee premiums. Chemew, Frick,
and McLaughlin (1997) and Blumberg, Nichols, and Banthin (2001) both examined
matched employer-employee data, and both estimated the elasticity of insurance take-up

with respect to premiums to be less than -0.1. Gruber and Washington (2005) examined a

change in the tax treatment of federal employees’ health insurance premiums in which

 

'0 Pauly, Song, and Herring (2001) also note that health insurance tax credits could lead to broader changes
in health insurance markets, including greater price competition among insurers, that are not accounted for
in simulation models.

49

the employee’s share, previously paid with after-tax earnings, became payable with pre-
tax earnings. Their main finding is that federal employees’ take-up of health insurance
was minimally responsive to the change in subsidy.

Two points about the existing empirical evidence are worth noting. First, the
existing studies offer evidence on the effect of a subsidy to health insurance (at the
marginal tax rate), or on the effect of different premiums on health plan selection, rather
than evidence on the effect of a tax credit that reimburses an individual dollar-for-dollar
for health insurance premium payments. Second, the worker populations studied are quite
heterogeneous; that is, the studies focus on workers whose earnings range widely, rather
than on low-wage workers. As a result, it is difficult to draw strong conclusions about the
effects of an HITC on health insurance coverage from existing research.

In this paper, we attempt to obtain direct evidence on the impact of tax credits on
health insurance coverage of low-earnings workers by examining the impact of a
supplemental tax credit that Congress added to the Earned Income Tax Credit (EITC)
during 1991, 1992, and 1993. This policy provided a refundable tax credit of up to $428
in 1991 ($451 in 1992, and $465 in 1993) to EITC-eligible households that bought health
insurance for a qualifying child. We treat this supplemental credit as a natural experiment
and estimate its impact on the health insurance coverage of single mothers using a
standard difference-in-differences approach applied to Current Population Survey data.
We first describe the tax credit, the approach to estimation, and the data we use. We then
present the main findings, followed by several sensitivity tests and a discussion of

possible alternative explanations of the findings.

50

3.2. The Health Insurance Tax Credit, 1991-1993

When Congress passed the Omnibus Budget Reconciliation Act (OBRA) of 1990,
it added a supplemental credit for health insurance purchases to the basic Earned Income
Tax Credit (EITC) program (U .8. Government Accountability Office 1991, 1993). This
HITC was a refundable tax credit for low-income workers with one or more children who
bought health insurance — either employer-provided or private nongroup — covering the
child or children. The credit offset only the cost of health insurance and did not cover co-
payments, deductibles, or out-of-pocket health expenses. To encourage participation, the
credit was refundable, so taxpayers with no federal income tax liability could still receive
a payment from the lntemal Revenue Service. The HITC was repealed effective
December 31, 1993, so it was available only during tax years 1991, 1992, and 1993.'1

The HITC had the same eligibility criteria as the EITC: To receive a credit, a
household needed to have earnings and a qualifying child. To qualify, a child needed to
meet three requirements: (1) be a child, stepchild, grandchild, or foster or adopted child
of the taxpayer; (2) have the same place of residence as the taxpayer for more than half
the tax year; and (3) be under age 19 (or 24 if a full-time student) or be permanently
disabled. Unlike the basic EITC, the HITC remained the same regardless of the number
of qualifying children in the family.

The HITC schedule closely followed the EITC’s. (Appendix 1, Table A1.1 gives
details of the HITC schedules and of the EITC schedules in the years before, during, and
after the HITC existed. Figure A1.1 in Appendix 1 is a graphical representation of the

HITC and EITC schedules in 1991). In 1991, a taxpayer with earnings and a qualifying

 

” See US. Government Accountability Office (1991, 1993, and 1994) for discussions of why Congress
eliminated the credit.

51

child could receive a credit up to $428 if he or she bought private health insurance that
covered the child. For households with earned incomes of $1 to $7,140, the credit was 6
percent of earned income. For households with earnings between $7,140 and $11,250, the
credit was $428 (6 percent of $7,140). For households with earnings between $11,250
and $21,250, the credit phased out at a rate of 4.28 percent per marginal dollar earned and
fell to $0 for earnings at or above $21,250 (see Appendix 1, Figure A1.1). In 1991, the
maximum HITC of $428 was 36 percent of the maximum EITC of $1,192. Like the EITC
schedule, the HITC schedule was indexed to inflation.

In 1991, the HITC’s first year, the average credit was $233, or 23 percent of the
reported average annual health insurance premium of $1,029. Also in 1991, 2.3 million
taxpayers received health insurance credits of $496 million (U .8. Government
Accountability Office 1991). A US. Government Accountability Office (GAO) study
(1994) estimated that the take-up rate for the HITC in 1991 (its first year) was in the
range of 19 to 26 percent. In contrast, the take-up rate for the regular EITC was 80 to 86
percent. The GAO attributed the relatively low HITC take-up rate to two factors. First,
interviews with taxpayers at IRS service sites in six cities suggested that fewer than 30
percent of EITC-eligible taxpayers knew the credit existed. Second, the GAO suggested
the credit was too modest to induce low-income workers to buy health insurance.

The GAO’s findings appear to have played a role in persuading Congress to
eliminate the HITC in 1993 (effective 1994). However, as the evidence below suggests,
the implicit conclusion that the HITC was ineffective may have been premature. Given
the subsequent attention that has been paid to tax credits for health insurance, it is curious

that no analysis of the HITC’s impact on health insurance coverage appears to exist.

52

3.3. Approach to Estimation

We treat the HITC as a natural experiment and adopt a difference-in-differences
approach to estimating its effects on the health insurance coverage of low-education
working single mothers. The approach follows a large literature on the labor supply
effects of the EITC including Eissa and Leibman (1996), Eissa and Hoynes (2004), Hotz,
Mullin and Scholz (2005) and Meyer and Rosenbaum (2000).

The population potentially affected by the HITC was low-income working
families with children. If the HITC had any effect on private health insurance coverage,
then the coverage of low-income working families with children would have been greater
than otherwise between 1991 and 1993. For three reasons, we focus on the HITC’s
possible effect on private health insurance of working single mothers with less than a
high school education. First, working single mothers were roughly 44 percent of all
EITC-eligible households in 1990, making them the largest group of taxpayers eligible
for the EITC and hence for the HITC (Liebman 2000). Second, for households headed by
a single woman, we can plausibly ignore decisions made jointly with other family
members (Eissa and Liebman 1996). Third, by focusing on high school dropouts, we can
estimate the effect of the HITC on a group that is likely to be eligible (because it is likely
to have low earnings) without conditioning explicitly on income or earnings.
Conditioning on income or earnings is ruled out because the EITC creates incentives for
earners to change their hours of work so as to qualify for the credit (Eissa and Hoynes
2005). Accordingly, our main treatment group is low-education working single mothers.

A convincing difference-in-differences approach requires a comparison or

“control” group that is as similar as possible to the treatment group without being eligible

53

for the HITC. In particular, the treatment and control groups should (1) face common
underlying trends in economic conditions and policies (other than the HITC) that would
be expected to affect health insurance coverage and (2) be essentially comparable in the
way they could be expected to respond to changes in economic incentives (Angrist and
Krueger 1999, Blundell and MaCurdy 1999, and Meyer 1995). Following Eissa and
Liebman’s (1996) line of reasoning, we use working single women without children and
with less than a high school education as the control group. Because they do not have
children, these women are ineligible for the HITC, but they should face essentially
similar labor markets, tax policy (apart from the HITC), and other economic conditions
as low-education working single mothers (the treatment group).

Two concerns invariably arise with the difference-in—differences approach. First,
if the treatment and control groups do differ in their characteristics, each may be affected
differently by contemporaneous shocks (other than the HITC). In this case, the
difference-in-differences approach may still be valid if we can control convincingly for
observables that capture characteristics of individuals that are likely correlated with
health insurance coverage. Second, the difference-in-differences estimator may be
contaminated if the compositions of the treatment and control groups change over time.
In the present case, substantial changes in tax and welfare programs affected single
mothers and increased their work incentives between 1984 and 1996 (Meyer and
Rosenbaum 2000, 2001). If single mothers who entered the labor force in later years were
more likely to work part-time and hence less likely to have employer-provided health
insurance, then changes in the characteristics of single mothers would have blunted any

rise in private health insurance that may have occurred as a result of the HITC. We can

54

again mitigate this problem by controlling for observable characteristics using regression,
but it will also be important to examine the samples carefully to see whether and how
much they did in fact change over time.

The model we estimate can be written:
Pr(ins,-=1|°) = FLBO + ,8] treatmenti + ,32 HITC, + ,63 treatmentl-XHITC, + Xifl] (3.1)

where i indexes individuals and t indexes years;

ins is a binary indicator of private health insurance coverage;

treatment equals one for a working single woman with a dependent child and less than a
high school education, and zero otherwise;

HIT C equals one for 1991, 1992, and 1993 (years during which the HITC was in effect),
and zero for 1988, 1989, and 1990 (years before the HITC was in effect); and

treatment xHIT C captures the change in coverage rates for working single mothers,
relative to working single women without children, after the HITC took effect.

In the basic specification we estimate, the vector of controls X includes age,
indicators of race (white, black, and other), number of children under age 6, number of
children aged 6—18, and number of children aged 19-24 and a full-time student, indicators
of work status (full-time/full-year, full-time/part-year, part-time/full-year, and part-

time/part-year),12 earned income,13 and unearned income.'4 For some of the specification

 

'2 Full-year work implies at least 50 weeks of work in the previous year. F ull-time work implies usual
weekly hours of 35 or more in the previous year.

'3 Earned income includes income from wages, salaries, and self-employment.

1’ Uneamed income includes income from unemployment compensation, worker's compensation, social
security or railroad retirement, supplemental security, public assistance or welfare, veteran payments,
survivors benefits, disability, retirement funds, interest, dividends, rent, educational assistance, child
support, alimony, contributions, financial assistance from friends, and other noneamings.

55

tests reported in Table 3.8, we include additional controls. We let F denote the standard

normal cumulative density and estimate equation (3.1) as a probit.

3.4. Data

We estimate equation (3.1) using data from the March 1989-1994 Annual
Demographic Supplements to the Current Population Survey (CPS), which provide
information for tax years 1988 through 1993. Respondents to the March 1989, 1990, and
1991 CPS constitute a before-HITC sample (tax years 1988, 1989, and 1990).

Respondents to the March 1992, 1993, and 1994 CPS constitute a during-HITC
sample (tax years 1991, 1992, and 1993). The relevant unit of observation is the tax-filing
unit, which in the CPS implies allocating primary families and subfamilies to separate
tax-filing units.

The sample includes women aged 19 to 44 who worked (had annual hours greater
than zero), were single (widowed, divorced, or never married), and had less than a high
school education. We exclude women who reported negative earnings, those in school
full-time, those who were separated from their spouse, and those who reported being ill
or disabled. The resulting sample, afier pooling all six years, includes 3,661 observations.

We allocate working single women with at least one dependent child to the
treatment group, and working single women without a child to the control group. We
consider any child in the tax-filing unit who was under age 19 (or under age 24 if a full-
time student) to be a dependent child for tax purposes. Consistent with the literature on

the EITC, we do not try to impose the support or residency test for the HITC eligibility.ls

 

'5 This is mainly due to limitations of the CPS. However, using data from the SIPP and IRS, Scholz (1994)
shows that the support test does not greatly change estimates of EITC eligibility.

56

We examine the six-year period — 1988 through 1993 — for two reasons. First,
when Congress passed the OBRA of 1993, which repealed the HITC, it enacted the
largest expansion of the EITC in the credit’s history (see Appendix 1, Table A1.1;
Baughman and Dickert-Conlin 2003). In 1993, a mother of one child with earnings up to
$7,750 could receive a credit of 18.5 percent of earned income, resulting in a maximum
credit of $1,434. In 1994, the credit rate rose to 26.3 percent of earned income, resulting
in a maximum credit of $2,038. Also beginning in 1994, eligibility for the credit was
expanded to include families with no children. For these families, the credit was 7.65
percent of earnings up to $4,000, resulting in a maximum 1994 credit of $306. As a
result, the EITC expansion of 1994 through 1996 makes it impossible to separate the
effect of eliminating the HITC from that of expanding the EITC.

The second reason for choosing the period between 1988 and 1993 is that the CPS
remained unchanged throughout this period. In March 1988, the Bureau of Labor
Statistics modified the CPS health insurance questions to capture more accurately the
insurance coverage of dependents.‘6 The next important revisions to the CPS health
insurance questions occurred in March 1995, when BLS introduced a more detailed set of
health insurance questions. In particular, previous surveys asked about employer
coverage as a subset of private coverage, but beginning in March 1995, the survey asked
separate questions about employer-provided and other types of private health insurance.
This change led to an increase in the number of persons reporting employer-provided

coverage.

 

'6 For more detailed information, see Appendix T of Unicon’s CPS Utilities (2005), which describes the
changes in health insurance questions on March file.

57

During the years we examine, the CPS health insurance questions read as follows:
75A. Other than government sponsored policies, health insurance can be obtained
privately or through a current or former employer or union. Was anyone in this
household covered by health insurance of this type at any time during l9xx [last
year]?
75B. Who was that?
75C. Was ...’s health insurance coverage from a plan in ...’s own name?
75F. What other persons were covered by this health insurance policy? Possible
answers are Spouse, Children in household, Children not in the household, Other,
and No one.

These questions allow us to define three alternative measures of health insurance:

1. private insurance coverage, defined broadly to include coverage by a privately
purchased or employer-provided health insurance plan, whether or not in the
respondent’s own name [that is, positive responses to questions 75A and 75B]

2. private insurance in the respondent’s own name [a subset of the first definition
because it implies a positive response to question 75C]

3. private insurance in the respondent’s own name that covers children in
household [a subset of the second definition because it implies a “children in
household” response to question 75F]

Table 3.1 and Figure 3.1 show average private health insurance coverage rates for
both working single mothers and working single women without children from 1988
through 1993. We report the three measures of private health insurance coverage defined

above. Both Table 3.1 and Figure 3.1 show decreases in the private health insurance

58

coverage rates of single mothers between 1988 and 1993. For example, the rate of
insurance coverage in own name fell by 5.4 percentage points (from 32.3 to 27 percent);
however, only about one-fifth of this decrease occurred after 1990. The coverage rate for
single women without children also fell between 1988 and 1990, but fell sharply even
after 1990 (from 37.8 to 20.9 percent). A likely explanation for the drop after 1990 is the
recession of 1991, which would have reduced both employment and access to employer-

provided health insurance of single women.

Table 3.1. Health Insurance Coverage Rates for Low-Education Working Single
Mothers and Low-Education Working Single Women without Children

1988 1989 1990 1991 1992 1993

 

 

Single mothers
Private insurance 0.350 0.353 0.297 0.297 0.279 0.308
Private insurance in own name 0.323 0.325 0.276 0.278 0.255 0.270

Private insurance inown name 0.221 0.249 0.218 0.209 0.197 0.202
that covers children

Single women without children
Private insurance 0.466 0.409 0.414 0.367 0.320 0.272

Private insurance in own name 0.398 0.385 0.378 0.341 0.283 0.209

 

Notes: The data are from the March 1989-1994 Annual Demographic Supplements to
the Current Population Survey (CPS). The sample contains working single women
with less than a high school education. We define "working" as having positive hours
and positive earnings during the year. We exclude women who are in school full-time,
those who are separated from their spouse, and those who report being ill or disabled.
Means are tabulated using CPS March supplement weights. Sample sizes are 2,228
(single mothers) and 1,433 (single women without children).

59

Figure 3.1. Health insurance coverage rates for low-education working single
mothers and low-education working single women without children

 

Private insurance

0-50 1 —— Single women without children
— — — - Single mothers

    

 

 

1988 1989 1990 1991 1992 1993

Private insurance in own name

0'50 _ — Single women without chikiren
0-45 ’ —---Singre mothers
0.40 5

0.35 5
0.30 5
0.25 5
0.20 5
0.15 i T 1 fl 5
1988 1989 1990 1991 1992 1993

 

 

 

 

Private insurance in own name covers children

0'50 I — Single women without children
0-45 ‘ --—-Single mothers

0.40 5x
0.35 5
0.30 5
0.25 5
0.20 5
0.15 v r u a a
1988 1989 1990 1991 1992 I993

 

 

 

 

Notes: See Table 3.1.

60

Table 3.2 displays mean characteristics of working single mothers and working
single women without children pooled for 1988 through 1993. The two groups differ in
some important ways. Relative to single women without children, single mothers are
more likely to be black (33.6 versus 14.7 percent) and less likely to work full-time, full-
year (37.8 versus 49.1 percent). Also, single mothers have average earnings that are
lower than single women without children ($7,257 versus $8,686).

Table 3.2. Summary Statistics: Low-Education Working Single Mothers and
Low-Education Working Single Women without Children

 

 

Single Single women
Variable mothers without children
Age (years) 30.5 29.7
White (%) 63.8 81.0
Black (%) 33.6 14.7
Other race (%) 2.6 4.4
Has children under age 6 (%) 48.6 0
Has children aged 6-18 (%) 69.9 0
Has children aged 19-24 2.0 0
and a full-time student (%)
Full-time, full-year (%) 37.8 49.1
Part-time, full-year (%) 8.9 9.2
Full-time, part-year (%) 31.7 29.0
Part-time, part-year (%) 21.6 12.8
Earned income ($) 7,257 8,686
Uneamed income ($) 1,833 464
Number of observations 2,228 1,433

 

Notes: See Table 3.1. Dollar amounts are converted to 1993 dollars using the
Consumer Price Index, All Urban Consumers (CPI-U).

Table 3.3 displays summary statistics for both single mothers and single women
without children in the before-HITC and during-HITC years. Overall, the characteristics

of both single mothers and single women without children appear quite stable over the

61

years in the sample. The only important change is in labor force attachment, reflecting
the recession in 1991 and 1992. For single mothers, the fraction working full-time, full-
year declined from 39.0 percent to 36.5 percent between the before-HITC and during-
HITC periods. For single women without children, the trend in the fraction working full-
time, full-year is similar to that for single mothers, falling from 50.4 in the before-HITC
period to 47.6 percent in the during-HITC period.

Table 3.3. Summary Statistics: Low-Education Working Single Mothers and

Low-Education Working Single Women without Children, Before and During
HITC

 

 

 

Single women
Single mothers without children
Before During Before During
HITC HITC HITC HITC

Variable (1988-90) (1991-93) (1988-90) (1991-93)
Age (years) 30.3 30.7 29.6 29.8
White (%) 64.6 63.0 81.8 80.1
Black (%) 32.9 34.4 15.5 13.7
Other race (%) 2.5 2.7 2.6 6.2
Has children under age 6 (%) 48.6 48.7 0 0
Has children aged 6-18 (%) 70.5 69.3 0 0
Has children aged l9-24 1.5 2.4 0 0
and a full-time student (%)
Full-time, full-year (%) 39.0 36.5 50.4 47.6
Part-time, full-year (%) 7.2 10.7 6.8 11.6
Full-time, part-year (%) 31.6 31.7 31.7 26.0
Part—time, part-year (%) 22.2 21.1 11.1 14.7
Earned income ($) 6,770 7,764 8,003 9,419
Uneamed income ($) 1,599 2,077 472 455
Number of observations 1,153 1,075 741 692

 

Notes: See Table 3.1. Dollar amounts are converted to 1993 dollars using the
Consumer Price Index, All Urban Consruners (CPI-U).

62

3.5. Empirical Findings

In the following analysis, the outcome of interest is coverage by private health
insurance defined as whether a working single woman has private insurance in her own
name that covers her child or children. We focus on this outcome because the HITC
could be used only to purchase a health insurance policy — either in the market or
through an employer or union — covering a qualifying child.

3.5.1. Main Findings — Single Women with Less than a High School Education

Table 3.4 displays the average private health insurance coverage rates for single
mothers and single women without children in the years before and during the HITC. The
first row shows that health insurance coverage for single mothers fell by 2.4 percentage
points between 1988-90 and 1991-93. The second row shows that, over the same time
period, coverage fell for single women without children by 9 percentage points. The
implication is that, after netting out the declining trend in insurance coverage, the private
health insurance coverage of single mothers was higher by 6.5 percentage points than it
would have been without the HITC. The robust standard error of this point estimate is
0.031.

Table 3.5 reports estimates of the key coefficients in equation (3.1).17 The
estimates in column 1 come from a specification that includes no control variables.
Column 2’s estimates control for age, race (white, black, and other), number of children
under age 6, ntunber of children aged 6-18, and number of children aged 19-24 and a fill]-

time student. Column 3’s estimates control in addition for work status (full-time/full-

 

'7 The procedure we use to compute the probit difference-in-differences is presented in Appendix 2.

63

year, full-time/part-year, part-time/full-year, and part-time/part-year), earned income, and
unearned income.

In column 1 of Table 3.5, the coefficient on treatment (working single mothers) is
-0.145 and statistically significant (p-value = 0.000). With the addition of demographic
characteristics in column 2, it falls to -0.080 (p-value = 0.006). In column 3, when we
control for work status and income along with demographic characteristics, it changes
slightly from -0.080 to -0.084 (p-value = 0.000). That the coefficient on treatment falls as
controls are added to the model suggests that observable characteristics other than the
presence of children are important in explaining the difference between single mothers
and single women without children in private health insurance coverage.

In column 1, the coefficient on HITC is negative (-0.090) and statistically
significant (p-value = 0.000). This is consistent with the declining trend in average health
insurance coverage for both single mothers and single women without children. Including
additional controls (columns 2 and 3) leaves this estimate essentially unchanged.

The estimate of main interest is the coefficient on the interaction term. This is
essentially similar in size and statistical significance across the three specifications. The
estimate in column 3 is 0.063 (p-value = 0.019), which suggests that private health
insurance coverage of working single mothers with less than a high school education was
higher by 6.3 percentage points than it would have been without the HITC. The finding is
consistent with the law of demand —— a drop in the price of health insurance should

induce consumers to demand more of it.

64

Table 3.4. Private Health Insurance Coverage Rates of Low-Education
Working Single Mothers and Low-Education Working Single Women without

 

 

Children
Before HITC During HITC Difference
(1988-1990) (1991-1993)
Single mothers 0.244 0.220 -0.024
(0.013) (0.013) (0.018)
[1,153] [1,075]
Single women 0.389 0.299 -0.090
without children (0.018) (0.017) (0.025)
[741] [692]
Difference -0.145 -0.080 —
(0.022) (0.022)
Difference-in-differences — — 0.065
(0.031)

 

Notes: See Table 3.1. Figures are average private health insurance coverage rates.
Robust standard errors are in parentheses. Sample sizes are in brackets.

Table 3.5. Difference-in-Differences Estimates of the Effect of the HITC on
Private Health Insurance Coverage of Low-Education Working Single Mothers

from Probit Estimates of Equation (3.1)

 

 

Dependent variable:
Covered by private health insurance (1) (2) (3)
Treatment (working single mothers) -0, 14 5 -0.080 -0084
(0.020) (0.029) (0.026)
HITC -0.090 -0.087 -0.104
(0.023) (0.022) (0.021)
Treatment ><HI TC 0.065 0.060 0.063
(0.030) (0.028) (0.028)
Ntunber of observations 3,661 3,661 3,661
Pseudo R-squared 0.020 0.061 0.256

 

Notes: See Table 3.1. Figures are estimated changes in the probability of private
health insurance coverage from probit models. Standard errors are in parentheses.
The specification in column 1 includes no control variables. The specification in
column 2 includes age, indicators of race (white, black, and other), number of
children under age 6, number of children aged 6-18, and number of children aged

19-24 and a full-time student. The specification in column 3 adds indicators of work
status (full-time/full-year, full-time/part-year, part-time/full-year, and part-time/part-
year), earned income, and unearned income to the controls in specification 2. The
marginal effects of the complete set of regressors are reported in Appendix 1, Table
A13.

65

3.5.2. Findings for Single Women with More Education

We would expect the estimated effect of the HITC on low-education single
mothers to be greater than its effect on single mothers with more education for a simple
reason: Low-education single mothers are more likely to be in low-wage jobs and hence
eligible for the EITC and HITC. For single mothers with more education, we would
expect a smaller estimated HITC effect because fewer are eligible.

We view this as an interesting hypothesis to test because it may suggest whether
the HITC estimates in Table 3.5 are convincing. No reason exists to suppose that single
mothers with more than high school (and hence higher earnings on average) should have
experienced improved health insurance coverage between 1988-90 and 1991-93.
Accordingly, finding that single mothers with high school, more than high school, and
college experienced health insurance coverage increases similar to those of single
mothers with less than high school would cast considerable doubt on the findings in
Table 3.5.

Table 3.6 displays tests of the hypothesis that single mothers with higher levels of
education — hence higher earnings — were less likely to experience an HITC-induced
improvement in health insurance coverage than single mothers with low education by
estimating equation (1) using samples of women with successively more educational
attainment. The specifications are the same as those in column 3 of Table 3.5; that is,
they include demographic characteristics, indicators of work status, earned income, and
unearned income as controls. We refer to this specification as the “basic specification.”

The first row of Table 3.6 repeats the main finding from comparing single

mothers and single women with less than a high school education (column 3 of Table

66

3.5). The second row of Table 3.6 shows the results of comparing single mothers and
single women with a high school education. The estimate on the interaction term is 0.043
with a p—value of 0.002, which suggests that private health insurance coverage rate of
single mothers with a high school education was 4.3 percentage points higher than it
would have been without the HITC. This accords with the prior expectation that the
HITC effect on single mothers should be less as educational attainment increases.

The third and fourth rows of Table 3.6 show findings from comparisons for single
mothers with some college education and with college or more. As expected, these
comparisons produce still smaller estimated effects of the HITC on health insurance
coverage. For single mothers with some college education, the estimated HITC effect on
health insurance coverage is 2.5 percentage points, but the estimate is not statistically
significant (row 3 of Table 3.6). Similarly, the estimated effect of the HITC on health
insurance coverage of single women with college or more is —0.6 percentage points
which is essentially zero. That we observe negligible responses to the HITC among single
mothers more than high school — women who are unlikely to be eligible for the HITC —
tends to increase our confidence that we are isolating an HITC effect for low-education
single women in Table 3.5.

3.5.3. Findings Disaggregated by Year

The estimates in Tables 3.5 and 3.6 are restrictive because they lump together the
three before-HITC years and the three during-HITC years. We would like to know
whether it is reasonable to restrict the estimated effect of the HITC to be the same in the
three during-HITC years (1991, 1992, and 1993). Accordingly, we estimate a model that

includes a set of year dummies (for 1988, 1989, 1991, 1992, and 1993), a treatment group

67

Table 3.6. Difi'erence—in-Differences Estimates from Alternative Treatment and
Comparison Groups

 

 

Dependent variable: Treatment
Covered by private health insurance Treatment HIT C XHIT C
1. Treatment group: -0.084 -O.104 0.063
Single mothers with less than high school (0.026) (0.021) (0.028)
[N=2,228]
Control group:
Single women with less than high school
[N=l ,433]

2. Treatment group: -O.l66 -0.107 0.043
Single mothers with high school (0.016) (0.011) (0.016)
[N = 6,794]

Control group:
Single women with high school
[N=6,608]

3. Treatment group: —O. 173 -0.098 0.025
Single mothers with some college (0.020) (0.012) (0.021)
[N=3,630]

Control group:
Single women with some college
[N=5,060]

4. Treatment group: -0. l 84 -0.048 -0.006

Single mothers with college (0.034) (0.009) (0.024)

[N=1 ,179]

Control group:

Single women with college
[N=5,736]

 

Notes: Estimates in row 1 are identical to those in column 3 of Table 3.5. Estimates

in rows 2, 3, and 4 come from applying the same model to different samples of

women, as indicated.
dummy, and interactions of the treatment and year dummies. We estimate this model for
our main group of interest —- working single women with less than a high school
education, and report estimates from a specification that is analogous to that in the basic

specification. The estimates are year-specific difference-in-differences estimates of the

HITC effect on private health insurance coverage rates.

68

Table 3.7. Difference-in-Differences Estimates of the Change in Private Health
Insurance Coverage Relative to 1990, Low-Education Working Single Mothers

 

Dependent variable: Covered by private health insurance

 

Treatment (working single mothers) -0. 104
(0.03 2)
1 988 Xtreatment 0.025
(0.042)
1 989 Xtreatment 0.036
(0.044)
1 990 Xtreatment —
1991 ><treatment 0.064
(0.046)
1992 Xtreatment 0.064
(0.046)
1993 xtreatment 0. 1 10
(0.046)
Number of observations 3,661
Pseudo R-squared 0.264

 

Notes: See Table 3.]. Figures are estimated changes in the probability of private
health insurance coverage from a probit model. Standard errors are in parentheses.
The specification includes age, race, number of children, work status, earned
income, unearned income, and a set of year indicators. The marginal effects of the
complete set of regressors are reported in Appendix 1, Table A1.4.

The estimates, reported in Table 3.7, suggest that the 1991, 1992, and 1993
coverage rates of single mothers with less than high school were higher by 6.4, 7.2, and
11 percentage points than they would have been in the absence of the HITC (although
only the estimate for 1993 has a p-value less than 0.05). The pattern of these estimates ——
increasing over time — has at least two possible interpretations. First, it could be that the
impact of the HITC increased as the existence of the program became known and its

implications better understood. This would be consistent with the pattern of participation

in several social programs (Madrian and Shea 2001; Remler and Glied 2003; Currie

69

2006). Alternatively, it could be that we have omitted one or more variables that affected
the health insurance coverage of low-education mothers (but not other low-education

single women). We investigate the latter possibility next.

3.6. Sensitivity Tests

This section describes findings from several sensitivity tests that attempt to
control for changes in the early 1990s other than the HITC and other possible influences
on private health insurance coverage. These include (1) expansion of the Medicaid
program, (2) state-level economic conditions and state fixed effects, and (3) welfare
reform and the introduction of state-level EITCs.
3.6.1. Medicaid Crowd-Out

During the years we are examining, eligibility for Medicaid expanded
substantially to include low-income pregnant women and children with no ties to the
AFDC Program (see Appendix 1, Table A1.2). Because Medicaid and private health
insurance are potential substitutes — they offer similar health coverage, and Medicaid is
much less costly —— Medicaid expansion may have drawn some low-income single
mothers out of private health insurance and into Medicaid. Cutler and Gruber (1996) first
referred to such substitution as “crowding out,” and to the extent it exists, the estimates in
Tables 3.5, 3.6, and 3.7 may give downward-biased estimates of the HITC’s impact on
private health insurance coverage. Alternatively, the availability of the HITC could
conceivably have drawn some low-income single mothers out of Medicaid and into
private health insurance, in which case the estimates in Table 3.5 would overstate the net

impact of the HITC on health insurance coverage.

70

To address these concerns, we estimate variants of equation (3.1) that use two
alternative dependent variables: medicaid, a dummy indicator of whether a woman had
Medicaid coverage, and insured, a dummy indicator of whether a woman had coverage
from either Medicaid or private health insurance. In the first case, the question addressed
is whether Medicaid coverage of low-education single mothers changed (relative to low-
education single women without children) during the HITC years. In the second case, the
question addressed is whether overall health insurance coverage of low—education single
mothers changed (relative to low-education single women without children) during the
HITC years.

Row 2 of Table 3.8 displays findings from the model in which medicaid is the
dependent variable. The estimates offer only marginal evidence that relative Medicaid
coverage of low-education single mothers was higher during the HITC period than in the
prior years. (The point estimate is 1.4 percentage points, but the standard error is large.)
We interpret this finding as evidence that, if the Medicaid expansions did crowd out
private health insurance during the HITC years, crowd-out was slight.

Row 3 of Table 3.8 displays findings from the model in which insured (by either
private health insurance or Medicaid) is the dependent variable. The estimates suggest
that during the HITC period, relative net health insurance coverage of low-education
single mothers was higher by 8.4 percentage points (p-value = 0.007) than in the
preceding years. Estimates from the basic specification in row 1 (which repeats the main
findings from Table 3.5, column 3) suggest that the HITC was the main factor in this net
increase in health insurance coverage, and that Medicaid expansion played a relatively

minor role. Specifically, rows 1, 2, and 3 suggest that 6.3 percentage points of the 8.4

71

percentage points were due to an increase in private coverage, and 1.4 percentage points
were due to an increase in Medicaid coverage (although the latter is not statistically
significant). A relatively small residual (0.7 percentage points) cannot be explained by
either the HITC or changes in Medicaid.

3.6.2. State-Level Economic Conditions and State Fixed Effects

It is also possible that changes in private health insurance coverage of low-
education single mothers were related to state-level economic conditions or to state fixed
effects. To account for these possibilities, rows 4, 5, and 6 of Table 3.8 report estimates
from models that successively add to the basic specification the year-specific
contemporaneous state unemployment rate (to control for cyclical labor market
influences on health insurance coverage), a set of state dummy variables (to control for
time-invariant state-level effects on health coverage), and both the year-specific state
unemployment rate and state fixed effects.

The findings reported in rows 4,5, and 6 of Table 3.8 suggest that including these
state-level controls in the model produces estimates of the HITC effect that are
essentially similar to the basic specification — an increase in single mothers’ private
health insurance coverage of roughly 6.5 percentage points (with p-values less than 0.02).
3.6.3. Welfare Reform and State EITCs

Although the findings in Table 3.5 are consistent with the HITC increasing
private health insurance coverage of low-education single mothers, an alternative
explanation for this increase is the welfare reforms adopted by some states after 1990.
California, Michigan, New Jersey, Oregon, and Utah all implemented a welfare waiver in

1993 (DeLeire, Levine, and Levy 2006). These waivers changed the nature of the AFDC

72

by imposing time limits on AFDC participation and introducing work requirements for
AFDC participants (Meyer and Rosenbaum 2000). If these changes moved single
mothers into the labor market and onto private (employer-provided) health insurance, the
estimates in Tables 3.5, 3.6, and 3.7 could overstate the HITC’s impact on private health
insurance coverage.

To address the above possibility, we restrict the sample to low-education single
women in states that did not have welfare waivers. Row 7 of Table 3.8 shows that
restricting the sample in this way leaves the estimated HITC impact on single mothers
essentially unchanged — an increase in the private health insurance coverage of single
mothers of 6.3 percentage points (p-value = 0.03 5).

In addition to implementing welfare reforms, several states made changes in their
EITCs during the period we examine. By 1993, six states had their own EITCs:
Minnesota, Vermont, and Wisconsin had refundable tax EITCs, while Iowa, Maryland,
and Rhode Island had non-refundable EITCs (Baughman 2005). When we restrict our
sample to single women in states that did not have state-level EITCs, we again see no
major change in the estimated effect of the HITC. The estimates reported in row 8 of
Table 3.8 show a relative increase in the coverage rate of single mothers of 6.5
percentage points (p-value = 0.017), suggesting that changes in state EITCs were not
responsible for the HITC effect we observed earlier.

Row 9 of Table 3.8 shows the findings when we restrict the sample to women in
states with neither welfare reforms nor state-level EITCs. Again, the main findings are

essentially unchanged.

73

8805553 E 08 £88
ESE-8m 28:88 v.8 0:885 353:: 98 .8885 358 8388 #83 62220 80 .5985: .38.. awn 88205 2 Bob
5:358on 883 23. .m 5:200 .m.m 033- mm 888 05 28 €282.38on 283v _ 38 3 82883-8 .m.m 033. com 8802

 

 

74

6:8 9%
c 8.8 $8.8 98.8 5983 99 88 9 98%; 89:83 9283

888 88.8 888 888- 88- 889... 9 5983 988 9 89999 989mm 8
28.8 88.8 :88 6:8 9% 989?

888 88.8 .888 88 8- 888- ,8988 9 599» 989m 9 89959 9958 8
$88 38.8 88.8 .82 9 92883 89:83 8983

888 98.8 888 888- 888- 388 9 5983 289.... 9 898989 98988 8
$8.8 28.8 :88 8988888... 998 98 9 99

888 88.8 888 828- 888- 85938985 9% 99 99988 988 83. 8
88.8 28.8 :88 8999.88.89w

888 88.8 888 888- 888- 998 98 9 99 95938985 99a 83.. .m

38.8 88.8 :88

888 68.8 888 9: 8- 888- 89888888 998 98 9 889938 9% 83. 8
88.8 98.8 €88 @8882 8 8559 99E 598 Eta-5

888 88.8 888 £88- 888- 9 9989; 988888 :5 docsccam 99m .m
98.8 88.8 88.8 9898: .3 89968 999.3

88 88.8 :88 888 888 9 9985, 9888898 98 aozsaoam 99m 8

38.8 28.8 68.8

888 88.8 888 9:8- .888. 89888898 9am ._

twang-m 3039.830 DRE-fix 3.85338 DRE ~595ch €80: Ecoxov 8:5wa 833$ 83.2% 3 83:8er

038m mo 838: Z ”038:? Eowaomoa

 

E2332 0.35 uni-83 gangsta-Beu— :e 38...:— Uh—fl ..8 88,—. 5352.5 .w.m via-H

3.7. Conclusion

The ongoing debate over how best to increase health insurance coverage in the
United States has led to a range of proposals for reform. It seems unlikely that an HITC
would receive serious consideration unless it could be expected to increase the health
insurance coverage of low-wage workers.

The Health Insurance Tax Credit of 1991 through 1993 offers a natural
experiment that we have used to examine the effectiveness of tax credits for health
insurance. The main findings reported in Table 3.5 suggest that the HITC increased the
private health insurance coverage of low-education single mothers by 6.3 percentage
points on average during 1991, 1992, and 1993. This estimate implies an elasticity of
health insurance take-up with respect to tax credits of roughly 1.1. The calculation is
straightforward: The HITC we are examining covered 23 percent of the reported average
annual health insurance premium in 1991 (US. Government Accountability Office
1991). The HITC-induced 6.3 percentage points increase in health insurance coverage of
single mothers during 1991, 1992, and 1993 implies a 25.8 percent increase in private
coverage (because 24.4 percent of single mothers had private health insurance coverage
during the before-HITC years). Accordingly, a 23 percent reduction in the price of health
insurance led to a 25.8 percent increase in health insurance coverage, which implies an
elasticity of 1.1.

It is not entirely surprising that the findings offered here differ from previous
empirical findings. Gruber and Washington’s (2005) study found convincing evidence
that extending the tax subsidy on the employee’s share of health insurance premiums to

federal employees had little or no effect. The policy change Gruber and Washington

75

(2005) examine differs in two ways from the HITC of 1991-1993. First, it was a tax
subsidy, as opposed to a refundable tax credit, and hence was less valuable in dollar
terms. Second, it applied to a group of relatively high-wage workers, most of whom were
already covered by health insurance, as opposed to a low-income group of workers most
of whom were (presumably) not previously covered. Accordingly, Gruber and
Washington’s findings do not necessarily conflict with those presented here, although
they do appear to at first blush.

The findings presented here appear favorable to the idea of a health insurance tax
credit, in that they suggest enough eligible low-wage workers with children would take
up a credit to significantly increase the percentage of such workers who are covered by
health insurance. Still, questions about the feasibility of a health insurance tax credit
remain. In particular, during the existence of the HITC, reports emerged that some
insurers offered policies of little real value that happened to cost the same amount as the
credit. Indeed, a Congressional investigation found that some insurers sold policies for
children covering only “cancer, heart attacks, strokes, and other diseases that few
children have” (Solomon 2007). Senator Lloyd Bentsen, the HITC’s original sponsor,
considered these abuses serious enough that he led efforts to repeal the HITC. Such
abuses raise serious concerns about tax credits for health insurance. It is an open question
whether such concerns could be mitigated by the existence of a well-developed private
health insurance market existed, or failing that, could be addressed with appropriate
regulation of health insurance.

Although the results presented here are consistent with the HITC increasing

health insurance coverage substantially, the analysis is hampered by small sample sizes.

76

It seems possible that further research using different data sources — such as the Survey
of Income and Program Participation and the National Longitudinal Survey, both of
which cover the years in question — could give more convincing evidence because both
would allow us to observe the same individuals before and after the adoption of the

HITC.

77

APPENDICES

78

APPENDIX 1

Table Al.l. Earned Income Tax Credit Parameters

 

Basic Tax Credit parameters, 1987-1996, nominal dollars

 

 

Tax year Phase-in Phase-in Maximum Phase-out Phase-out
rate (%) range ($) credit ($) rate (%) range ($)

1987 14.00 0-6,080 851 10.00 6,920-15,432
1988 14.00 0-6,24O 874 10.00 9,840-18,576
1989 14.00 0-6,500 910 10.00 10,240-19,340
1990 14.00 0-6,810 953 10.00 10,730-20,264
1991

One child 16.70 0-7,l40 1,192 11.93 11,250-21,250

Two children 17.30 0-7,140 1,235 12.36 11,250-21,250
1992

One child 17.60 0-7,520 1,324 12.57 11,840-22,370

Two children 18.40 0-7,520 1,384 13.14 11,840—22,370
1993

One child 18.50 0-7,750 1,434 13.21 12,200-23,050

Two children 19.50 0-7,750 1,511 13.93 12,200-23,050
1994

No child 7.65 0-4,000 306 7.65 5,000-9,000

One child 26.30 0-7,750 2,038 15.98 11,000-23,755

Two children 30.00 0-8,245 2,528 17.98 11,000-25,296
1995

No child 7.65 0-4,100 314 7.65 5,130-9,500

One child 34.00 0-6,160 2,094 15.98 11,290-24,396

Two children 36.00 0-8,640 3,110 20.22 11,290-26,673
1996

No child 7.65 0-4,220 323 7.65 5,280-9,500

One child 34.00 0-6,330 2,152 15.98 11,610-25,078

Two children 40.00 0-8,890 3,556 21.06 11,610-28,495

 

Health Insurance Tax Credit parameters, 1991-1993, nominal dollars

 

 

Tax year Phase-in Phase-in Maximum Phase-out Phase-out
rate (%) range ($) credit ($) rate (%) range ($)
1991 6.00 0-7,140 428 4.28 11,250-21,250
1992 6.00 0-7,520 451 4.28 11,840-22,37O
1993 6.00 0-7,750 465 4.28 12,200-23,050

 

Source: US. House of Representatives (1998); Government Accountability Office
(1991,1994).

79

Table Al.2. Major Legislative Changes Affecting Low-Income Women, 1987-94

 

Tax year

Earned Income Tax Credit

 

1988

1991

1992

1993

l 994

The beginning and end of the phase-out range increased by about
$3,000.

The credit rates rose by 2 percentage points.

Additional credits established for families with two or more children.
The new increment to the maximum credit for a second child was
$43.

Supplemental credits added for child health insurance premiums and
children under age 1.

The credit rates rose by 1 percentage point.
The increment to the maximum credit for a second child rose to $60.

The credit rates rose by 1 percentage point.
The increment to the maximum credit for a second child rose to $77.

The credit rates rose by about 10 percentage points.

The increment to the maximum credit for a second child rose to $490.
Supplemental credits for health insurance and children under age 1
repealed.

Small credits established for taxpayers without children and between
the ages of 25 and 65.

The IRS began to notify eligible taxpayers of the advance payment
option, which had been available since 1979.

 

Tax year

Medicaid

 

April 1937
July 1988

October 1988

July 1989
April 1990

July 1991

States permitted to extend Medicaid coverage to children under age 2
in families below 100 percent of the poverty line

States permitted to extend Medicaid coverage to children under age 5
in families below 100 percent of the poverty line.

States permitted to extend Medicaid coverage to children under age 8
in families below 100 percent of the poverty line, and to children
under age 1 in families below 185 percent of the poverty line.

States required to extend Medicaid coverage to children under age 1
in families below 75 percent of the poverty line.

States required to extend Medicaid coverage to children under age 6
in families below 133 percent of the poverty line.

States required to extend Medicaid coverage to children under age 19
in families below 100 percent of the poverty line.

 

Source: US House of Representatives (1998); Meyer and Rosenbaum (2000);
General Accountability Office (1993).

80

Table A1.3. Full Results for Table 3.5 Estimates: Results from Probit Models

 

Dependent variable:

 

Covered by private health insurance (1) (2) (3)

Treatment -0.145 -0.080 -0.084
(0.020) (0.029) (0.026)

HIT C -0.090 -0.087 -0. 104
(0.023) (0.022) (0.021)

Treatment XHI TC 0.065 0.060 0.063
(0.030) (0.028) (0.028)

Age — 0.011 0.005
(0.001) (0.001)

White — -0.012 -0.009
(0.035) (0.031)

Black — -0.057 -0.033
(0.036) (0.033)

Number of children under age 6 — -0.046 -0.004
(0.015) (0.014)

Number of children aged 6-18 — -0.030 -0.019
(0.010) (0.009)

Number of children aged 19-24 — -0.034 -0.029
and a full-time student (0.053) (0.047)
F ull-time, full-year —— — 0.259
(0.034)

Part-time, full-year — — 0.077
(0.035)

F all-time, part-year —- — 0.138
(0.027)

Part-time, part-year — — —

Earned income — — 0.01 8
(0.002)

Uneamed income — — 0.005
(0.003)

Number of observations 3,661 3,661 3,661
Pseudo R-squared 0.020 0.061 0.256

 

Notes: Figures are estimated changes in the probability of private health insurance
coverage from probit models. Standard errors are in parentheses.

Table Al.4. Full Results for Table 3.7 Estimates: Results From Probit Models

 

Dependent variable: Covered by private health insurance

 

Treatment (working single mothers) -0. 104
(0.03 2)
1 988 0.036
(0.03 6)
I 989 0.012
(0.03 9)
1 990 —
1 991 —0.037
(0.034)
1 992 -0.094
(0.034)
1 993 -0. 144
(0.034)
1988 ><treatment 0.025
(0.042)
1 989 Xtreatment 0.036
(0.044)
1 990 Xtreatment —
1991 Xtreatment 0.064
(0.046)
1992 Xtreatment 0.072
(0.049)
1993 Xtreatment 0.1 10
(0.046)
Age 0.005
(0.001)
White -0.008
(0.03 0)
Black -0.03 1
(0.034)
Number of children under age 6 -0.004
(0.013)
Number of children aged 6-18 -0.019
(0.009)
Number of children aged 19-24 -0.027
and a full-time student (0.055)

82

Table Al.4. (cont’d)

 

Dependent variable: Covered by private health insurance

F ull-time, full-year 0.257
(0.034)
Part-time, full-year 0.077
(0.039)
F ull-time, part-year 0.138
(0.028)
Part-time, part-year —
Earned income 0.01 8
(0.002)
Uneamed income 0.006
(0.003)
Number of observations 3,661
Pseudo R-squared 0.264

 

Notes: Figures are estimated changes in the probability of private health insurance
coverage from a probit model. Standard errors are in parentheses.

83

 

 

 

 

 

 

Health
Insurance Earned
Tax Credit Income .
(3) Tax Credit
Phase-in range Plateau Phase~out range ($)
428 1,192
$10 = r l =
400 p; ' S 0pc r2 1.000
3 00 800
600
200
400
100 200
0 0
7.140 11.250 21.250
Household Earnings ($)

Figure A1.1. Earned Income Tax Credit and Health Insurance Tax Credit
Schedules, 1991

84

APPENDIX 2

Calculating Difference-in-Differences Estimates in Linear Probability and Probit
Models

In a linear probability model, calculating the difference—in-differences (DD)
estimate is trivial. Let y be the outcome variable of interest, which takes on two values:
zero and one. In the simplest case, we have two time periods (pre and post) and two

groups (control and treatment). The linear probability model can be written as

Pr( y =1 l 0) = '60 + filtreatment + ,szost + fl3treatment x post + X ,6 (A21)

where X denotes a vector of explanatory variables. The linear probability DD estimate is
simply the OLS estimator of ,6}, the coefficient on the interaction between treatment and

post.
In a probit model, the coefficient on the interaction term no longer has the simple

interpretation in equation (A2.1). To illustrate, consider the following probit model
Pr( y = 1 | 0) = <D(,80 + ,Bltreatment + ,62 post + fl3treatment x post + Xfl) (A22)

where (l) is the standard normal cumulative distribution function. The marginal effect of

the interaction term is

APr(y =1 l') _ @(flo + filtreatment + ,szost + ,63 + Xfl)-

_ A2.3
Atreatment x post <I>(,B0 + filtreatment + '82 post + X ,8) ( )

 

85

As discussed by Ai and Norton (2003), Norton, Wang, and Ai (2004), and DeLeire
(2004), many authors incorrectly interpret equation (A2.3) as the DD estimate. However,

the DD estimate from the probit is

My =1|,) _ A[(D(,60 + ,6] + flzpost + ,B3post + Xfl) — owe + ,82post + Xfl)J
AtreatmentApost - Apost

 

 

= {id’iflo “561 +52 +133 + Xfl)‘ (”(150 + fl] + Xflli‘} ((42.4)

[M + ta + x13)- W. + Xflll

Equations (A2.3) and (A24) clearly show that the marginal effect of a change in the
interaction term is not equal to the marginal effect of a change in both interacted
variables.

Following DeLeire (2004), we calculate the DD estimates from probits by taking
the discrete double difference of the standard normal cumulative distribution function. In

particular, we first estimate the probit model (equation 3.1)
Pr(insi =1|0)= <D(,B0 + ,Bltreatmenti + ,BZHITC, + fl3treatmenti x HITC, + Xifl)

Then, we predict four counter factual probabilities for each observation in the sample,
plugging in the observed values for each explanatory variable. The predicted probabilities

are:

1. Predicted probability of private health insurance coverage for the treatment group

in the post-HITC period: (IDLE), + ,3] + ,32 + .33 + Xpé)

2. Predicted probability of private health insurance coverage for the treatment group

in the pre-HITC period: (MA) + 3] + X ,- fi )

86

 

3. Predicted probability of private health insurance coverage for the control group in

the post-HITC period: (Difio + ,6} + X {,3 )

4. Predicted probability of private health insurance coverage for the control group in

the pre-HITC period: (Di/30 + X13)

Using the predicted probabilities, we calculate the DD for each observation as

= {[(Di'éo 433! +32 “533 +Xné)‘ (13(50 +fi1+Xifi)]_}
W. + I}. + m)— an, . ml

The probit DD estimate is the average of DD across all observations:

[We +31 +32 +83 ”(1'13“)“ (”(30 +131 +Xifli‘}

_ —1 '7
._ n Zz=l{[¢(fl“o +fi2 +Xiié)— (D060 + Xifi»

87

 

REFERENCES

88

REFERENCES

Abraham, Jean, and Anne Royalty. 2005. “Does Having Two Eamers in the Household
Matter for Understanding How Well Employer-Based Health Insurance Works?” Medical
Care Research and Review 62(2): 167-186.

Ai, Chunrong, and Edward Norton. 2003. “Interaction Terms in Logit and Probit
Models.” Economics Letters 80(1): 123-129.

Andrisani, Paul. 1977. “Internal-Extemal Attitudes, Personal Initiative, and the Labor
Market Experience of White and Black Men.” Journal of Human Resources 12(3): 308-
328.

. 1981. “Internal-Extemal Attitudes, Sense of Efficacy, and Labor Market
Experience: A Reply to Duncan and Morgan.” Journal of Human Resources 16(4): 658-
666.

 

Angrist, Joshua D., and Alan B. Krueger. 1999. “Empirical Strategies in Labor
Economics.” In Handbook of Labor Economics, ed. Orley Ashenfelter and David Card.
Amsterdam: Elsevier Science, Vol.3A: 1277-1366.

Baughman, Reagan A. 2005. “Evaluating the Impact of the Earned Income Tax Credit on
Health Insurance Coverage.” National Tax Journal 58(4): 665-684.

Baughman, Reagan, and Stacy Dickert-Conlin. 2003. “Did Expanding the EITC Promote
Motherhood?” American Economic Review 93(2): 247-251.

Baum, Christopher R, Mark E. Schaffer, and Steven Stillman. 2003. “Instrumental.
variables and GMM: Estimation and Testing.” Stata Journal 3(1): 1-31.

Black, Dan. 2000. “Family Health Benefits and Worker Turnover.” In Employee Benefits
and Labor Markets in Canada and the United States, ed. William T. Alpert and Stephen
A. Woodbury. Kalamazoo, Michigan: W.E. Upjohn Institute for Employment Research.

Blackburn, McKinley. 2004. “The Role of Test Scores in Explaining Race and Gender
Differences in Wages.” Economics of Education Review 23(6): 555-576.

Blumberg, Linda J ., Len M. Nichols, and Jessica S. Banthin. 2001. “Worker Decisions to
Purchase Health Insurance.” International Journal of Health Care Finance and
Economics 1(3-4): 305-325.

Blundell, Richard, and Thomas MaCurdy. 1999. “Labor Supply: A Review of Alternative

Approaches,” in Handbook of Labor Economics, ed. Orley Ashenfelter and David Card.
Amsterdam: Elsevier Science, Vol.3A: 1560-1695.

89

Buchmueller, Thomas, and Robert Valletta. 1999. “The Effect of Health Insurance on
Married Female Labor Supply.” Journal of Human Resources 34(1): 42-71.

Chemew, Michael, K. Frick, and Catherine McLaughlin. 1997. “The Demand for Health
Insurance Coverage by Low-Income Workers: Can Reduced Premiums Achieve Full
Coverage?” Health Services Research 32(4): 463-470.

Cogan, John F., Glenn R. Hubbard, and Daniel P. Kessler. 2005. Healthy, Wealthy, and
Wise: Five Steps to a Better Health Care System. Washington, DC: AEI Press.

Coleman, Margo, and Thomas DeLeire. 2003. “An Economic Model of Locus of Control
and the Human Capital Investment Decision.” Journal of Human Resources 38(3): 701-
721.

Currie, Janet, and Brigitte C. Madrian. 1999. “Health, Health Insurance, and the Labor
Market.” In Handbook of Labor Economics, ed. Orley Ashenfelter and David Card.
Amsterdam: Elsevier Science, Vol.3C: 3309-3416.

Currie, Janet. 2006. “The Take Up of Social Benefits.” In Public Policy and the Income
Distribution, ed. Alan Auerbach, David Card, and John Quigley. New York: Russell Sage
Foundation, 80—148.

Cutler, David, and Jonathan Gruber. 1996. “Does Public Insurance Crowd Out Private
Insurance?” Quarterly Journal of Economics 1 1 1(2): 391-430

Cutler, David, and Richard Zeckhauser. 2000. “The Anatomy of Health Insurance.” In
Handbook of Health Economics, ed. Anthony Culyer and Joseph Newhouse. Amsterdam:
Elsevier Science, Vol.1A: 563-643.

Cutler, David, and Sarah Reber. 1998. “Paying for Health Insurance: The Trade-off
Between Competition and Adverse Selection.” Quarterly Journal of Economics 113(2):
433-466.

DeLeire, Thomas, Judith Levine, and Helen Levy. 2006. “Is Welfare Reform Responsible
for Low-Skilled Women’s Declining Health Insurance Coverage in the 19905?” Journal
of Human Resources 41(3): 495-528.

DeLeire, Thomas. 2004. “A Note on Calculating Difference-in-Differences Using Probit
Models versus Linear Probability Models.” Unpublished manuscript. Michigan State
University.

Eissa, Nada, and Hilary Hoynes. 2004. “Taxes and the Labor Market Participation of

Married Couples: The Earned Income Tax Credit.” Journal of Public Economics 88(9-
10):1931-1958.

90

Eissa, Nada, and Hilary Hoynes. 2005. "Behavioral Responses to Taxes: Lessons from
the EITC and Labor Supply." NBER Working Paper 11729. Cambridge, MA: National
Bureau of Economic Research.

Eissa, Nada, and Jeffrey B. Liebman. 1996. “Labor Supply Response to the Earned
Income Tax Credit.” Quarterly Journal of Economics 111(2): 605-637.

Emmons, David W., Eva Madly, and Stephen A. Woodbury. 2005. “Refundable Tax
Credits for Health Insurance: The Sensitivity of Simulated Impacts to Assumed
Behavior.” Working Paper 05-119. Kalamazoo, MI: W.E. Upjohn Institute for
Employment Research.

Fronstin, Paul. 2007. “Sources of Health Insurance and Characteristics of the Uninsured:
Analysis of the March 2007 Current Population Survey.” Employee Benefit Research
Institute Issue Brief No. 310, October.

Gruber, Jonathan. 2000. “Health Insurance and the Labor Market.” In Handbook of
Health Economics, ed. Anthony Culyer and Joseph Newhouse. Amsterdam: Elsevier
Science, Vol. 1 A: 645-706.

Gruber, Jonathan. 2000a. “Tax Subsidies for Health Insurance: Evaluating the Costs and
Benefits.” NBER Working Paper 7553. Cambridge, MA: National Bureau of Economic
Research.

Gruber, Jonathan. 2000b. “Microsimulation Estimates of the Effects of Tax Subsidies for
Health Insurance.” National Tax Journal 53(3, Part 1): 329-342.

Gruber, Jonathan, and Brigitte C. Madrian. 2004. “Health Insurance, Labor Supply and
Job Mobility: A Critical Review of the Literature.” In Health Policy and the Uninsured,
ed. Catherine G. McLaughlin. Washington, DC: Urban Institute Press: 97-178.

Gruber, Jonathan, and Ebonya Washington. 2005. “Subsidies to Employee Health
Insurance Premiums and the Health Insurance Market.” Journal of Health Economics
24(2): 253-276.

Gruber, Jonathan, and James Poterba. 1994. “Tax Incentives and the Decision to
Purchase Health Insurance: Evidence from the Self-Employed.” Quarterly Journal of
Economics 109(3): 701-733.

Gruber, Jonathan, and Larry Levitt. 2000. “Tax Subsidies for Health Insurance: Costs and
Benefits.” Health Aflairs 19(1): 72-85.

Hotz, V. Joseph, and John Karl Scholz. 2003. “The Earned Income Tax Credit.” In

Means-Tested Transfer Programs in the United States, ed. Robert Moffitt. Chicago:
University of Chicago Press and NBER.

91

Hotz, V. Joseph, Charles Mullin, and John Karl Scholz. 2005. “Examining the Effect of
the Earned Income Tax Credit on the Labor Market Participation of Families on
Welfare.” Mimeo. UCLA and Wisconsin.

Liebman, Jeffrey B. “Who Are the Ineligible EITC Recipients?” National Tax Journal
53(4, Part 2): 1165-1186.

Lundberg, Shelly. 1988. “Labor Supply of Husbands and Wives: A Simultaneous
Equations Approach.” The Review of Economics and Statistics 70(2): 224-235.

Madrian, Brigitte C. 2006. “The US. Health Care System and Labor Markets.” NBER
Working Paper 11980. Cambridge, MA: National Bureau of Economic Research.

Madrian, Brigitte C., and Dennis F. Shea. 2001. “The Power of Suggestion: Inertia in
401(k) Participation and Savings Behavior.” Quarterly Journal of Economics 116(4):
1149-1187.

Meyer, Bruce D. 1995. “Natural and Quasi-Experiments in Economics.” Journal of
Business and Economic Statistics 13(2): 151-161.

Meyer, Bruce D., and Dan T. Rosenbaum. 2000. “Making Single Mothers Work: Recent
Tax and Welfare Policy and its Effects.” National Tax Journal 53(4, Part 2): 1027-1062.

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.

Norton, Edward.C. Hua Wang, and Chunrong Ai. 2004. “Computing Interaction Effects
and Standard Errors in Logit and Probit Models.” Stata Journal 4(2): 103-116.

Olson, Craig. 1998. “A Comparison of Parametric and Semiparametric Estimates of the
Effect of Spousal Health Insurance Coverage on Weekly Hours Worked by Wives.”
Journal of Applied Econometrics 13(5): 543-565.

Olson, Craig. 2000. “Part-Time Work, Health Insurance Coverage, and the Wages of
Married Women.” In Employee Benefits and Labor Markets in Canada and the United
States, ed. William T. Alpert and Stephen A. Woodbury. Kalamazoo, Michigan: W.E.
Upjohn Institute for Employment Research.

Pauly, Mark V. 1999. “Extending Health Insurance through Tax Credits.” Kaiser Project
on Incremental Health Reform. Menlo Park, CA: Henry J. Kaiser Family Foundation.

Pauly, Mark V., and Bradley Herring. 2001. “Expanding Coverage Via Tax Credits:
Trade-Offs and Outcomes.” Health Affairs 20(1): 8-26.

92

Pauly, Mark V., and Bradley Herring. 2002. Cutting T axes for Insuring: Options and
Effects of Tax Credits for Health Insurance. Washington, DC: American Enterprise
Institute for Public Policy Research.

Pauly, Mark V., David Song, and Bradley Herring. 2001. “Tax Credits, the Distribution
of Subsidized Health Insurance, and the Uninsured.” NBER Working Paper 8457.
Cambridge, MA: National Bureau of Economic Research.

Remler, Dahlia K., and Sherry A. Glied. 2003. “What Other Programs Can Teach Us:
Increasing Participation in Health Insurance Programs.” American Journal of Public
Health 93(1): 67-74.

Scholz, John Karl. 1994. “The Earned Income Tax Credit: Participation, Compliance, and
Antipoverty Effectiveness.” National Tax Journal 47(1): 63-87.

Scott, Frank A., Mark C. Berger, and Dan A. Black. 1989. “Effects of the Tax Treatment
of Fringe Benefits on Labor Market Segmentation.” Industrial and Labor Relations
Review 42 (2): 216-229.

Solomon, Judith. 2007. "The False ‘Public versus Private’ Choice for Children’s Health
Coverage.” Washington, DC: Center on Budget and Policy Priorities, July 21.

US. Government Accountability Office. 1991. “Tax Administration: Administrative
Aspects of the Health Insurance Tax Credit.” Fact Sheet for the Subcommittee on Health,
Committee on House Ways and Means, US House of Representatives. Washington, DC:
United States Government Accounting Office, GAO/GGD-91-1 10F S, September.

US. Government Accountability Office. 1993. “Earned Income Tax Credit:
Effectiveness of Design and Administration.” Testimony before the Subcommittee on
Select Revenue Measures and Human Resources, Committee on House Ways and Means,
US. House of Representatives. Washington, DC: United States Government Accounting
Office, GAO/T-GGD-93-20, March.

US. Government Accountability Office. 1994. “Tax Administration: Health Insurance
Tax Credit Participation Rate Was Low.” Report to the Subcommittee on Health,
Committee on House Ways and Means, US. House of Representatives. Washington, DC:
United States Government Accounting Office, GAO/GGD-94-99, April.

US. House of Representatives, Committee on Ways and Means. Green Book,
Background Material and Data on Programs within the Jurisdiction of the Committee on

Ways and Means. Washington DC: US. Government Printing Office, various years.

Unicon Research Corporation. 2005. CPS Utilities. College Station, Texas: Unicon
Research Corporation.

93

Wellington, Allison, and Deborah Cobb-Clark. 2000. “The Labor Supply Effects of
Universal Coverage: What Can We Learn from Individuals with Spousal Coverage?”
Research in Labor Economics 19: 315-344.

Wooldridge, Jeffrey. 2002. Econometric Analysis of Cross Section and Panel Data.
Cambridge, MA: MIT Press.

Wozniak, Gregory D., and David W. Emmons. 2000. “Tax Credit Simulation Project

Technical Report.” Discussion Paper 00-1. Chicago: American Medical Association,
Center for Health Policy Research.

94

llllllllllllllll.lllllllllllllllHtlllllllllllllllllil
3 1293 02956 7777