n .-
7 .41. H .. a...

. ...VA.
.. If... ....\.

 

 

 

"‘«sszs

A

-l

1‘ 1') .5

This is to certify that the

dissertation entitled

Essays on the Impact of Maternity Leave Policy and
Educational Policy on the Labor Market
Outcomes of Workers

presented by

Heeseon Choi

has been accepted towards fulﬁllment
of the requirements for

PhD degree in Economics

Major professor

ﬂ7/ 66 r
DML

MS U i: an Afﬁrmative Action/Equal Opportunity Institution 0- 12771

 

_ LIBRARY
Michigan State
University

 

 

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 c:/CIRC/DateDue.p65-p.15

 

Essays on the Impact of Maternity Leave Policy and Educational Policy on the

Labor Market Outcomes of Workers

by

Heeseon Choi

A DISSERTATION

Submitted to
Michigan State University
in partial ﬁilﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Economics
2002

ABSTRACT

ESSAYS ON THE IMPACT OF MATERNITY LEAVE POLICY AND
EDUCATIONAL POLICY ON THE LABOR MARKET OUTCOMES OF
WORKERS

By

Heeseon Choi

This dissertation is composed of two chapters that explore the labor market
outcomes of the policy: Chapter I about the maternity leave policy in the United
States and Chapter II about the college education expansion policy in South Korea.

Chapter I investigates the effects of maternity leave beneﬁt on the labor
market outcomes of mothers with newborn babies. Until the FMLA (The Family and
Medical Leave Act) has passed in 1993, the general policy approach towards
maternity leave was to attempt to increase the number of employees with maternity
leave coverage by legislative mandates. Existing literature, therefore, examines
whether the implementation of a certain mandates was efﬁcient by showing that
there was no disemployment effect on the beneﬁted group. Chapter I provides an
argument in favor of this approach by showing that existing maternity leave program
seems to beneﬁt the women they covered.

I estimate the effect of maternity leave beneﬁt after controlling for substantial
intrinsic differences between maternity leave covered workers and uncovered

workers. The ﬁxed effects estimation method is used to analyze the wage effect.

Also, more explanatory variables are added to designate heterogeneity for the
analysis of the turnover and employment effects.

The results of the Chapter I support the hypothesis that maternity leave
coverage is beneﬁcial on women’s labor market outcomes. Maternity leave
signiﬁcantly lessens turnover one year after childbirth, and is closely related with
higher employment one or two year after childbirth. Also, it is found that women
who had a covered job maintain steeper wage proﬁles before and after childbirth.

Chapter II explores the effects of the college education expansion policy on
wages and employment during the 19803 and the 1990s in South Korea, including a
sudden increase in the college entrance quota in 1981.

First, changes in the returns to the college education before and after a large
increase in the quota in 1981 are explored by using the difference-in-difference
method. Second, a time series analysis provides the overall impact of the college
education expansion policy. The estimation results of the reduced form equation
indicates that the relative wages and employment of all workers including both
young and old responds to the policy with the time lag. Third, the estimation result
of the determinants of college enrollment, aﬁer controlling for trend and the ﬁrst
order autocorrelation, shows that the declining wages of college graduates have
discouraged the college enrollment. It implies that the market incentives are still
working in private decisions to enroll in college in South Korea where the

government has regulated the behavior of college enrollment.

ACKNOWLEDGEMENTS

I would like to thank my advisor, the chair of the committee, Professor Jeff
Biddle for his advises and encouragements. Without his generous and devoting
efforts, I would not have been able to write this dissertation. He is a respectable
person as well as a great teacher. I also wish to thank my committee, Professor David
Neumark, Professor Byron Brown, and Professor Karen Roberts for helpful
comments and constructive criticism. However, I am responsible for any remaining
mistakes.

I am thankful for my friends, colleagues, teachers, especially Professor Jong-
Hyun Kim, Professor Byong-Jick Ahn, Dr. Kyu Uck Lee, and former employer,
Korea Development Institute.

I am also grateful for the love and support from my family, my husband and
son, Seongjin J eong and Hogyeong Jeong, and my mother and father, Dongryul Park

and late Byunghyun Choi.

iv

TABLE OF CONTENTS

LIST OF TABLES
LIST OF FIGURES

CHAPTER I

THE EFFECTS OF MATERNITY LEAVE BENEFITS ON LABOR MARKET
OUTCOMES

I - 1. Introduction

..................................................................... 1
I - 2. Issues in the Maternity Leave Policy: A Historical Perspective ...... 6
I- 3. Critical Review of Previous Studies ....................................... 10
Redistribution toward Women ....................................... 10

Efﬁciency Costs of the Mandates ...................................... 14

I - 4. Data .............................................................................. 20
I - 5. Empirical Analysis ............................................................ 23
5-1. Effects vs. Heterogeneity .............................................. 23

5-2. Effects on Turnover and Employment ............................... 27

Model and Variables ................................................... 27

Findings for Turnover .................................................. 30

Findings for Employment .............................................. 33

5-3. Effects on Earning Proﬁle ............................................. 36
Model and Hypotheses ................................................ 36

Findings ................................................................. 40

V

I - 6. Conclusion ..................................................................... 46

REFERENCES ........................................................................ 48
TABLES AND FIGURES ............................................................ 53
CHAPTER II

EFFECTS OF THE COLLEGE EDUCATION EXPANSION POLICY ON
THE RELATIVE WAGES OF COLLEGE GRADUATES IN KOREA

11-1. Introduction ..................................................................... 70
11-2. Critical Review of Previous Studies ....................................... 73
11-3. Data ............................................................................... 78
[14. Overview of Changes in the Wage Structure ............................ 81
The Background of the College Education Expansion Policy ....... 81
Declining Relative Wages of College Graduates: Changes
in the Wage Structure during 1980-1995 ........................... 81
11-5. The Effect of the College Education Expansion Policy on Relative
Wages of College Graduates : Empirical Analysis .................... 86
DD Estimation for the Effect of the Policy in 1981 ................... 87
Time Series Regressions ................................................... 89
Determinants of the College Enrollment Rate .......................... 97
11-6. Summary and Policy Implications ........................................... 101
REFERENCES ........................................................................... 103
TABLES AND FIGURES ............................................................. 105

LIST OF TABLES

Table I-1 Summary Statistics: Women who had Birth during 1988-92 and Worked
just before Childbirth .................................................... 53
Table I-2 Turnover One Year after Childbirth (All Workers) ..................... 54
Table I-3 Turnover One Year after Childbirth (College Educated Workers) 55
Table I-4 Turnover One Year after Childbirth (High School Educated Workers) 56
Table I-5 Employment One Year after Childbirth (All Workers) ................ 57
Table I-6 Employment One Year after Childbirth (College Educated Workers) 58
Table I-7 Employment One Year after Childbirth (High S. Educated Workers) 59
Table I-8 Employment Two Years after Childbirth (All Workers) .............. 60
Table I-9 Employment Two Years after Childbirth (College Educated Workers) 61
Table I-lO Employment Two Years after Childbirth (High S. Educated Workers) 62
Table I-11 The Effects of Coverage at the Time of Childbirth on the Wage Proﬁle
(Hourly Wages) ........................................................... 63
Table I-12 The Effects of Coverage at the Time of Childbirth on the Wage Proﬁle

(Weekly Earnings) ....................................................... 64

vii

Table II-l

Table II-2

Table II-3

Table II-4

Table II-S

Table II-6

Table II-7

Table II-8

A Summary Statistics of All Workers Aged 19-59 ................ 105
A Summary Statistics of Young Workers Aged 19-59 ............ 107
Changes in the Returns to the College Education

before/after the Policy .............................................. 109
Changes in the Returns to College Education (including four-year

college and two-year college) before/after the Policy ......... 1 10
The Relationship between the Quota and the Enrollment Rate ... 1 11
The Determinants of Relative Employment (the Reduced Form

Equation) ........................................................... 112
The Determinants of Relative Wage (the Reduced Form

Equation) ........................................................... 1 13

The Sensitivity Analysis (Coefﬁcients for the Lagged Enrollment Rates)

Table II-9 The Relative Wages and the Relative Supply (the Demand Curve) 115

Table II-lO The Determinants of the Enrollment Rates .......................... 116

viii

LIST OF FIGURES

Figure I-1 Proportion of Working (Non-Zero Earnings) by Groups .......... 65
Figure I-2 Average Work Hours per Week by Groups ........................ 65
Figure I-3 Simple Wage Proﬁles of All Workers ................................ 66
Figure I-4 Simple Wage Proﬁles by Groups ...................................... 66
Figure I-5 Predicted Hourly Wage Proﬁles (All Workers) ...................... 67
Figure I-6 Predicted Weekly Earnings Wage Proﬁles (All Workers) ......... 67

Figure I-7 Predicted Hourly Wage Proﬁles (College Educated Workers) 68
Figure I-8 Predicted Weekly Earnings Proﬁles (College Educated Workers) 68
Figure I-9 Predicted Hourly Wage Proﬁles (High S. Educated Workers) .. . .. 69
Figure I-10 Predicted Weekly Earnings Proﬁles (High S. Educated Workers) 69
Figure IL] The Effect of the College Education Expansion Policy

on the Wages of College Graduates .................................. 74
Figure II-2 Actual Relative Wages, College vs. High School Graduates 117
Figure II-3 Detrended Relative Wages, College vs. High School Graduates 117
Figure II-4 Actual Relative Supply, College vs. High School Graduates 118

Figure II-S Detrended Relative Supply, College vs. High School Graduates 118

ix

CHAPTER 1

THE EFFECTS OF MATERNITY LEAVE BENEFITS ON LABOR MARKET
OUTCOMES

I. Introduction

The job-reinstatement right that maternity leave policy provides is believed to
affect women’s earnings proﬁles by extending their pre-birth job tenure. By
removing fertility-related career interruption, this right can narrow the wage gap
between working mothers and single women workers as well as the wage gap
between male and female workers (Waldfogel 1998). Getting maternity leave
covered jobs in their earlier careers is considered to be important for young women
to ﬁnd equal footing with their male counterparts in the workplace.

The job-reinstatement right has become an essential part of maternity leave
beneﬁts. Earlier versions of maternity leave policy such as the 1978 Pregnancy
Discrimination Act (the PDA) required eligible employers to cover the medical
insurance costs for pregnancy or childbirth. Although the guidelines of the PDA
prohibited dismissals due to pregnancy or childbirth, guaranteeing the job-
reinstatement right was not the main concern of PDA. Indeed, maternity leave policy
practiced by ﬁrms was largely informal and discretionary before the late 19805.
(Kamerrnan, Kingston and Kahn 1983). Most ﬁrms that permitted such a leave did
not provide a formal guarantee of job protection. Thus, guaranteeing a job-

reinstatement right became the core issue in the implementation of the state

maternity/parental/family leave statutes during the late 1980s. After the Family and
Medical Leave Act (the FMLA) was passed in 1993, job-reinstatement rights became
an essential part of the maternity leave beneﬁt on a national scale.1

Since the job-reinstatement right is at the core of maternity leave beneﬁts
after the 19805, we can ask whether it has signiﬁcantly increased employment of
mothers of new-born babies by helping them to return to pre-childbirth employers.
This paper will investigate the effect of ﬁrms’ provision of a maternity leave beneﬁt
on labor market outcomes of mothers with newborn babies. A maternity leave
program is expected to remove involuntary job-separation at the time of giving birth,
so could contribute to a steeper wage proﬁle. Also, a maternity leave program, by
guaranteeing the future job-reinstatement right, could encourage young women to
participate more in the labor force, to invest more resource in the education and
training before giving birth, and therefore to increase their earnings.

However, there is a serious barrier to attempts to estimate the size of these
effects. Jobs that offer maternity leave coverage may be correlated with higher
wages. In other words, women with more unobserved ability or greater labor force
attachment may self-select into covered jobs. Then, an unobserved heterogeneity
between maternity leave covered women and not-covered women would result in

biased estimates of the effect of coverage. In this paper, I try to estimate the effect of

 

' The FMLA covers women who are employed by ﬁrms with more than 50 employees and who have
worked more than half-time for their employer for at least one year. Around 45 percent of women
workers are classiﬁed into this category (Waldfogel 1998). Prior to the FMLA, estimated percentage
of actual coverage varies by studies. Waldfogel (1998) estimated that 40 percent of working women
had employers who offered maternity leave in 1991. According to my calculation using the NLSY,
maternity leave coverage was 65.1 percent in 1988, 66.9 percent in 1991, and 65.2 percent in 1996.
This calculation includes all workers including part-time workers.

maternity leave beneﬁts, after controlling for substantial intrinsic differences
between maternity leave covered workers and not-covered workers.

Existing studies on the effects of maternity leave were largely inspired by the
growth of women’s rights movement. Their main argument is that the maternity
leave policy has exerted signiﬁcant and positive effects on women’s wages and
employment. Representative works are Waldfogel (1993; 1997; 1998), Albrecht,
Edin, Sundstrom, and Vroman (1998), Ruhm (1998), Trzcinski (1991), and Spalter-
Roth and Hartman (1991). Among them, Waldfogel (1998) is closest in spirit to this
paper, as it attempted to evaluate the effect on women of working in a job that
provided maternity leave beneﬁts, rather than the effect on outcomes of all women
of childbearing age of legislated maternity leave mandates that leave many women
uncovered. Waldfogel found a large wage premium for those who were covered by
and who used maternity leave at the time of giving birth, a premium large enough to
offset the negative wage effect from childbearing.

However, other studies cast doubt on the large wage premium attributed to
maternity leave by Waldfogel. For example, Mincer and Polachek (1974), Mincer
and Ofek (1982), Sandel and Shapiro (1982), and Corcoran, Duncan, and Ponza
(1983) investigated the effect of an interrupted career on women’s wages and found
the wage “rebound” phenomenon that took place with a quick return to the labor
force. Malkiel and Malkiel (1973), Garvey and Reimers (1980), Filer (1993), and
Light and Ureta (1995) also argued that if wages were more closely related with

actual experience rather than with tenure, the alleged beneﬁcial effects of maternity

leave policy on wages would be depreciated. 2 In addition, Waldfogel did not
consider the fact that women with higher unobserved skills may self-select into
maternity leave covered job, a possibility I will explicitly explore in this paper

Contrary to above studies which evidenced the positive effects of maternity
leave for women, another stream of studies, like, Summers (1989), Klerman and
Leibowitz (1997), Ruhm (1997), Kane (1997), and Bond (1991), Baum (1998),
Gardecki (1997), Gruber (1994), emphasized the “efﬁciency cost” of maternity leave,
that is, the disemployment effect of the policy on the beneﬁted group. It is argued
that “efﬁciency cost” happens when women do not value the beneﬁt of maternity
leave, so wages are not lowered aﬁer the implementation of the mandates.

In this paper, I use the sample of women who gave births and worked at the
time of giving births during 1988-1992 from the National Longitudinal Survey of
Youth (NLSY) data set. This sample is appropriate for evaluating the effect of
maternity leave beneﬁt or more concretely the job-reinstatement right of that beneﬁt
on women’s labor market performance. The treatment group is maternity leave
covered women and the control group is not-covered women. Clear deﬁnition of the
treatment and control group is an advantage of my sample. The unclear control group
of some studies about the effect of enactment of maternity leave legislation oﬁen

brings them ambiguous results.

 

2 Malkiel and Malkiel (1973), Garvey and Reimers (1980), Filer (1993), and Light and Ureta (1995)
argued that actual experience is more important than tenure in the earlier career. They argued that
quick returnees to the labor force suffer little detrimental effect from their time out of labor force.
However, that does not mean that the wage level of returnees can catch that of continuous workers.
That is, since these studies is about the effect of job separation in general, their results cannot be
directly compared with studies which argues that fertility-related involuntary job separation might be
signiﬁcantly detrimental.

This paper proceeds as follows. Section II brieﬂy summarizes the issues in
the maternity leave policy. Section III reviews previous studies and Section IV
covers data. Section V performs empirical analysis for the effect on turnover,

employment, and wage proﬁles. Concluding remarks follow in Section VI.

II. Issues in the Maternity Leave Policy: A Historical Perspective

Maternity leave policy was ﬁrst conceived as a means of protecting the health
of pregnant women workers and their infants. Maternity leave policy was originally
advocated in the movement to improve working conditions of women in the early
stages of industrialization in Europe, though a major component of the protective
legislation also included the limitation of work hours for women. Maternity
protection was ﬁrst introduced in Germany in 1883 as a part of the social insurance
program by Bismarck. This measure has been regarded as the origin of such policies
(Karnerman 1988, Frank and Lipner 1988).

However, a more powerful motivating force for the policy was the pro-
natalistic movement in the 19305 in Europe, giving rise to the second
conceptualization of the policy. In that period, Europe suffered from an
unprecedented low birth rate, a trend that had begun about a century before (van de
Walle and Knodel 1980). Therefore, some authors argued that the historical and
ideological origin of the maternity leave policy began in Sweden. At that time,
Sweden aimed to foster fertility of working women (Gustafsson and Stafford 1995).
During that period, maternity leave policy was extended toward infant care leave
policy. In contrast, the birth rate in the United States in that period was relatively
high, which might explain why the introduction of the nationwide maternity leave

policy in this country was delayed until 1993 (Allen 1988).

The third and most recent purpose of maternity leave policy was the
promotion of equality between genders in the work place. As female labor force
participation increased dramatically, gender equality in the work place became a
crucial issue. The benchmark event was the parental leave law that was passed in
Sweden in 1964. Permitting fathers to take a leave of absence was expected to
promote equality between genders in the work place as well as in the function of
child rearing.3 Establishment of gender-neutral family leave would guarantee that
working women would not be penalized ﬁom child rearing.

In the United States, the Pregnancy Discrimination Act (the PDA) in 1978,
the ﬁrst nationwide maternity leave policy, was passed in the spirit of anti-
discrimination. 4 However, the most controversial aspect of the PDA was its
requirement that pregnancy and childbirth coverage be included in employer
sponsored group health insurance programs. Moreover, the PDA did not require the

job-reinstatement rights

 

3 The effect of the parental leave on the distribution of childrearing effort has been largely symbolic
in that the total amount of parental leave taken by men is low. (Allen 1988)

4 The 1978 PDA, as an amendment of Title VII of the 1964 Civil Rights Act, explicitly prohibited the
discrimination against pregnancy and childbirth in accessing the fringe beneﬁt program. The text
follows:

The terms “because of sex” or “on the basis of sex” include, but are not limited to, because of or on
the basis of pregnancy, childbirth, or related medical conditions; and women affected by pregnancy,
childbirth, or related medical conditions shall be treated the same for all employment-related
purposes, including receipt of beneﬁts under fringe beneﬁt program, as other persons not so affected
but similar in their ability and inability to work, and nothing in section 703(h) of this title shall be
interpreted to permit otherwise.

In sum, if a ﬁrm provides health insurance, disability insurance and leave, then it should cover
pregnancy and childbirth.

The guideline of the 1978 PDA prohibited dismissal because of pregnancy or childbirth in view of
antidiscrimination against pregnancy or childbirth. However, it is different from the job-reinstatement
right that was included in the state maternity/parental/farnily leave statutes in the late 19805.
According to Kamcrrnan, Kingston, and Kahn (1983), the policy is informal and probably
discretionary in most of the ﬁrms permitting such a leave but not providing a formal guarantee of job
protection. Therefore, they called the PDA an implicit national maternity policy. Frank and Lipner
(1988) also did not regard the PDA as maternity leave policy, while Trzcinski (1991) did.

While maternity leave policies with the protective or pro-natalistic rationales
contributed to improve the working conditions in the early industrial periods, they
proved to be sometimes contradictory to the advance of women’s equality in the
work place. As Ferber (1997) mentioned, “the maternity leave policy which insisted
pregnant women and women after childbirth not to work for some period of time
often intended to restore the traditional family and to ﬁx the sex roles.” Also,
evidence from Europe showed that substantially longer leaves might actually have
discouraged women from participating in work for pay (Allen 1988). In fact, in the
United States, equality in the work place was one of the most crucial issues for the
women’s rights activists.‘5 As early as the 19205, when women’s suffrage amendment
was passed, leaders of the women’s movement began to question the impact of the
protective laws on working women’s advancement (Frank and Lipner 1988). They
argued “protection, no matter how benevolent in motive, unless applied alike to both
sexes, amounts to actual penalization” (Rothman 1978). 7 Instead, the job-
reinstatement right was viewed as the most important concept in the issue of equality
in the work place. It was believed to promote an equal employment opportunity for
women along with social equality. In fact, the term “leave (of absence)” conveyed a

meaning of a guarantee of returning to a job. Job-protected leave would help women

 

6 The point of view of American Women’s Movement is different from that of European Women’s
Movement. While the latter was more willing to focus on the condition and needs of working
mothers, the leader of the former rather focused on legal equality (Frank and Lipner 1988). For
example, women’s rights activists during the implementation of the maternity leave legislation in
Sweden expected that the leave policy would help to improve the living standards of working class
and let them to have more children while in stable jobs.

7 The President’s Commision in 1963 also showed this viewpoint. It concluded that “wherever
practiced, legislation affecting labor standards should beneﬁt men as well as women, especially where
women employees might be placed at a disadvantage otherwise”(Kamerman, Kahn, and Kinston
1983)

to remain in the labor force. Of course, the maternity leave beneﬁt with a job-
reinstatement right would be used only for a few occasions in a woman’s work life
and not all covered women would use it. However, the employment guarantee
including the maternity period was crucial for preserving seniority as well as for
increasing the labor force attachment in women’s work lives. There can be an
incentive for women to enter the labor force before having a child in order to qualify
for parental leave (Ruhm 1998). Or, by reducing the conﬂict between maternity
leave and employment, this program could induce more young women to enter and
remain in the labor market. This would increase lifetime earnings of working
mothers and induce greater investment in education and training, resulting in ﬁrrther
increase in earnings and labor supply (Winegarden and Bracy).

Even after the passage of the PDA, there still existed a wide variation in the
percentage of ﬁrms in the private sector that provided job-protected leaves in the
early 19805. The fact that most surveys depended heavily on large and medium-sized
ﬁrms can be one reason for that. However, the job-compatibility also explains parts
of this variation. Although women were formally qualiﬁed for a job-protected leave,
they frequently found it impossible to return to their former employers, because their
newly allocated jobs were no longer comparable to their previous jobs. (Kamennan,
Kahn, and Kingston 1983) Therefore, most of the state maternity/parental/family
legislation in the late 19805 and the 1993 FMLA included an explicit and detailed

requirement of the job-reinstatement right.

111. Critical Review of Previous Studies

Redistribution Toward Women

A substantial number of studies have already been done about the effects of
maternity leave policies. Most studies of the 19805 and in the early 19905
concentrated on the question of the enactment of a nationwide leave policy. This is
because the enactment of the FMLA took a long time from its ﬁrst introduction in
1985 to its passage in 1993. The enactment of the FMLA in the United States was
late compared to other industrialized nations.

Many of these studies, inspired by the women’s rights movement, strongly
advocated the enactment of maternity leave. They argued that a job-protected leave
policy could enhance the labor market performance of women workers. According to
Trzcinski (1991), forcing women out of the labor market for reasons of pregnancy or
childbirth was one of the structural barriers to economic equality for women. The
chief goal of the state maternity leave statutes in the late 19805 was the promotion of
equal employment opportunities for women. This was well represented in the
majority opinion written by Justice Marshall in the US. Supreme Court decision in

the case of California Federal Savings and Loan Association vs. Guerra (1987).8

 

8 Justice Marshall wrote that “Title VII, as amended by the PDA, did not preempt the California
statute, since both the federal and the state provisions shared the goal of promoting ‘equal
employment opportunities’ for women”. That is, the purpose of such legislation is to eliminate
discrimination against women by making their opportunities for long-term employment equal to those
for men. (Trzcinski 1991)

10

Confronting the opponents to the enactment of mandated leave legislation,
advocates emphasized the long-term beneﬁts of a job-protected leave for the society
as well as for women. A strong dissident group against the enactment of leave
legislation consisted of small employers, who faced a relatively large increase in
costs. Small employers argued that a mandated leave policy would hurt optimality
based on the neoclassical economics. However, for larger ﬁrms, a leave policy did
not mean a substantial cost increase. Many of them had already provided the beneﬁts
and they had a large pool of employees to substitute temporarily for the work of
workers on leave.

Trzcinski (1991) argued that the economic dynamics behind this process for
promoting equality depended on supply-side effects. For her empirical study, she
used May 1979 and May 1983 CPS and chose individuals who resided in a state with
maternity leave statutes as a treatment group. She thought that neoclassical
economics, which usually concentrates on the short-run employment and wage
effects, does not take into consideration the supply-side effects. Maternity leave
legislation would eliminate structural barriers to labor market participation of women,
and by doing so it would break the occupational segregation that had conﬁned
women to lower paying jobs. Her empirical study of the impact of mandated leave
roughly supported her argument. Although wages and employment were not
adversely affected in states that passed the parental leave mandates, parental leave
legislation signiﬁcantly improved the labor market position of childbearing aged
women. This was especially the case in states with Temporary Disability Insurance

(TDI).

ll

Although her study was inﬂuential in the debate over maternity leave
legislation and was quoted by many authors, it had two weak points. Her result was
strongly dependent upon ﬁrm size. Also, although she emphasized that the job-
reinstatement right was the important legislative initiative and TDI should not be
directly linked to it, her empirical result was especially supportive in states with TDI.

Ruhm and Teague ( 1997) investigated the relationship between mandated
leave policies and national income and labor market outcomes. They created a
longitudinal data set for seventeen countries during the 19605-1989 period using the
International Labor Ofﬁce’ Legislative Series and 1984 global survey on protection
of working mothers, along with Social Security Programs Throughout the World by
the United States Social Secru'ity Administration. Their result supported the view
that parental leave hardly reduces economic efﬁciency, though the result was
sensitive to the length of parental leave. Instead, parental leave provided a modest
beneﬁcial impact. Short to moderate durations of parental leave raised per capita
incomes, employment-to—population ratios, and labor force participation rates. That
is, the result suggested that “short to moderate period of family leave may increase
economic efﬁciency, as measured by incomes and labor market status, while lengthy
duration could reduce it”.

Waldfogel (1997 and 1998) conﬁrmed the positive effect of maternity leave
policy on wages by showing that maternity leave covered women maintained higher
wages than maternity leave not-covered women after giving birth. Using the data sets
of the NLSY in the United States and the National Child Development Study of

Britain (NCDS), she examined the effect of past maternity leave coverage at the time

12

of a woman’s most recent birth on her current wages. Her result showed a large
maternity leave wage premium, which is big enough to offset the negative wage
effect ﬁ'om childrearing.

Waldfogel interpreted this result as implying that maternity leave policies
could be an effective recipe for the “family gap”, deﬁned as the wage differential
between working mothers and childless women. She viewed the lack of a job-
protected maternity leave as a “family barrier”, since it prevented women with
children from competing on an equal footing in the labor market. In this regard, she
found the similarity between women who have children without the right to a job-
protected maternity leave and displaced workers. Therefore, in her view, extending
the job-protected maternity leave rights would further reduce the “family gap” for
future working mothers by increasing their likelihood of returning to their current
employers after childbirth.

However, Blank (1997) criticized Waldfogel, arguing that the positive
association between wages and maternity leave coverage did not show the impact of
maternity leave uncontaminated by other effects. According to Blank, Waldfogel’s
result was not compatible with a standard labor economic theory, where wages and
fringe beneﬁts are interchangeable with each other within total compensation. If Job
A is almost identical to Job B except that Job A provides the fringe beneﬁt, then the
wages of Job A will be lower than those of Job B roughly by the amount of
pecuniary value of the beneﬁt. More important problem in her study is that she did
not fully handle heterogeneity between maternity leave covered workers and not-

covered workers. Interpreting the positive association as a causal relationship from

13

the former to the latter is problematic. It is true that maternity leave coverage at the
time of giving birth could raise wages by removing involuntary job separation at that
time. However, higher wages from maternity leave coverage can also be explained
by unobserved heterogeneity. Even after controlling for observed characteristics,
maternity leave coverage is positively associated with wages. It implies that
maternity leave covered workers may have greater unobserved ability. Maternity
leave tends to be more prevalent for “good” jobs and at large ﬁrms, and these attract
workers with more skills, observed and unobserved. Also, women with more labor

force attachment and higher potential productivity will self-select into covered jobs.

Efﬁciency Costs of the Mandates

After Summers(1989) mentioned the efﬁciency cost of an employer mandates
by reducing employment of the beneﬁted group, many studies examined the effect of
maternity leave policy following his suggestion. This group of studies, unlike the
above studies such as Waldfogel (1998), did not identify the people who were
actually covered by maternity leave beneﬁt. Instead, they looked for the effect of the
introduction of maternity leave mandates on all childbearing aged women in the
jurisdiction that passed the mandate, both covered and not-covered. For example,
when Kane (1998) examined the effect of the state maternity leave legislation, the
treatment group was all women in states that implemented this legislation and the
control group was all women in states that did not implement this legislation. Gruber

(1994) used all childbearing aged women as the treatment group and men and

14

remaining women as the control group when he examined the effect of the PDA.
Meanwhile, Waldfogel (1998) used women who were actually covered by maternity
leave as her treatment group and women who were not as her control group.
Ordinary people who think that the purpose of maternity leave policy is to
help women might be confused when they see that so many studies contended that
the enactment of maternity leave policy had negligible effect or negative
consequences on labor market outcomes for women. It is mainly because those
studies examined the immediate effect of introducing a maternity leave mandate.
Their argument can be interpreted like this: even though an existing maternity leave
program might be beneﬁcial, the marginal effect of further extension of this program
would have efﬁciency costs. For example, a maternity leave beneﬁt that is provided
by individual ﬁrms might be efﬁcient since it is voluntary, while state or federal
maternity leave mandates might not be efﬁcient. Following this line of argument,
Gardecki (1997) asserted that implementation of state maternity leave mandates
before 1979 was efﬁcient, while implementation of the federal PDA was inefﬁcient.
However, it is not clear to me whether they argued that a certain maternity leave
provision was inefﬁcient or whether maternity leave provision itself is inefﬁcient.
Originally, Summers (1989) pointed out the possibility of the latter. He mentioned,
“employer mandate, no matter how socially attractive it is, can bring an efﬁciency

cost by reducing employment of the beneﬁted group”.9 However, a mandated leave

 

9 Summers (1989) argued that in a perfectly ﬂexible market, mandated employer beneﬁt programs
would increase inefficiency, since they hamper free bargaining over the terms of compensation
packages. There are some rationales, however, for a mandated leave policy. First, there can be
positive extemalities associated with welfare of children (Kamerrnan 1991, Bravo 1991, Stoiber 1990,
Gilland 1989, Decker 1991). However, some economists found that this rationale unconvincing and
not sufﬁcient for government intervention (Surmners 1989 and Gruber 1994). The second rationale
for government intervention is based on adverse selection. If there exists adverse selection, it is clear

15

policy can also work efﬁciently if employees value the beneﬁt and if there is no
structural barrier to wage adjustment.10 Evidence of efﬁciency costs of a mandated
maternity leave policy was found in Europe, where the full-scale maternity leave
policies have been in effect longer than in the United States. In this context, the 1952
Convention on Maternity Protection of the International Labor Organization (ILO)
emphasized that employers who provided maternity beneﬁts should not have
individual liability. This revision reacted to a concern that individual employer
liability for maternity leave beneﬁts could result in employment discrimination
against women (Pizzo 1988).

Gruber (1994), in his seminal work, investigated whether Summers’
efﬁciency argument could be sustained for the group-speciﬁc mandate case, which
extends beneﬁts for a demographically identiﬁable group.ll He thought it necessary
to analyze whether the cost of the mandate is shifted to the wages of the beneﬁted
group in view of efﬁciency of these mandates. Using the CPS 1974-1982, Gruber
estimated the responses of the labor market to the enactment of the state maternity
mandates in the 19705 and the 1978 federal PDA. He used the difference-in-
difference-in-difference (DDD) estimation method, which controls for any

systematic shocks to the labor market outcomes of the treatment group in the

 

that “the market solution will not provide the desirable amount of maternity leave, even though all
young women are willing to pay more than it costs” (Summers 1989).

‘0 “Their wages would fall, offsetting the cost of the beneﬁt to the employer, and in the limit there
would be no efficiency cost” (Summers 1989).

” “In the case of maternity leave mandates, there would be less scope for the free adjustment of
wages to reﬂect the valuation of the beneﬁt by the targeted group. Anti-discrimination regulations or
workplace relative-pay norms which prohibit differential pay for the same job across groups may
affect not to adjust their wages” (Gruber 1994).

16

experimental states. '2 He reported an extensive shifting of the cost of the state
maternity mandates to group-speciﬁc wages. Meanwhile the implementation costs of
the federal PDA were not shifted and there was some disemployment effect.
Gardecki (1997), largely a replication of Gruber(1994), generally conﬁrmed his
results. Furthermore, she argued that states only adopted the level of state maternity
legislation considered efﬁcient, when she explained the disparity between the
signiﬁcant cost shifting in his state study and the strong disemployment effect in his
federal study. According to Gardecki (1997), workers in states that enacted the state
maternity mandates knew its beneﬁts very well, while in states that had to follow the
federal PDA, the cost of implementing the mandates outweighed the beneﬁts.

However, Klerman and Leibowitz (1997)’s estimate of the labor supply
effects of state maternity leave legislation is different from Gruber (1994). Using a
sample of the census Public Use Microdata Sample for 1980 and 1990, they found
that maternity leave policy had little effect on employment. The treatment group was
mothers of younger children who would be protected by a maternity leave legislation
and the control group was mothers of older children. To control pre-existing
differences between states with maternity leave legislation and without legislation,
they also used a DDD approach.

Kane (1998), using the NLSY data set, evaluated the impact of such
mandates on part—time vs. full-time employee substitution, on access to leave, and on

wages. She constructed a model that consisted of the wage offer equation and the

 

‘2 Experimental states are states which have maternity leave mandates, while the treatment group is
individuals of 20-40 year-old married women and/or their husbands. The DDD estimation method has
become a popular method that has been widely used in the studies that examined the effect of
implementing maternity leave legislation.

l7

hours of work equation with selection mechanism and estimated these equations by
Heckman’s two-step procedure. She found that with the enactment of the mandates
the employment decreased for some time, but after that eventually increased, while
the wage has increased for a time, but after that, eventually decreased.13 Based on
this result, she strongly opposed to the expansion of mandated beneﬁt.14 Baum
(1999) also tackled the same question as Kane (1998). He investigated the effect of
state maternity leave legislation and the FMLA on mothers’ labor supply after
childbirth. Although he used data from NLSY that identiﬁes who was actually
covered, he did not use that information. Instead, he followed the typical method of
existing studies that examined the effect of implementing maternity leave legislation
at the state level on all the women in the state. The treatment group was mothers who
gave births in a state that had maternity leave legislation or the FMLA during 1987-
1994. 15 His theoretical model predicted that the effect of maternity leave legislation
on a mother’s return to work decision was ambiguous. However, his empirical work
reported that maternity leave legislation slightly increased the hazard rate for
returning to pre-birth employers, though that result depended on the time of return.

On the other hand, he reported that maternity leave legislation had a negative effect

 

'3 Kane(1998)’s result was different from Gruber ( 1994), who found the negative wage effect for the
treatment group after mandate was enacted. Also, her evaluation that “women place positive value on
the availability of maternity leave”, which was the same proposition of Gruber who uses this for
determining the maternity leave mandates as the efﬁcient policy, was hardly compatible with her
strong opposition to the mandate.

'4 According to Kane(1998), “the negative wage impact suggests that advocates for the expansion of
mandated beneﬁts should temper their enthusiasm for such legislation.” Also, according to her, there
was no evidence of substituting full-time women workers with part-time women workers. State
maternity leave mandates hardly increase the demand for female part time employees relative to
female full time employees.

'5 “If a mother gives a birth in a state that has maternity leave legislation and the FMLA is in effect,
then her maternity leave legislation variable equals the number of weeks of leave guaranteed by the
more generous legislation...... Even if the data suggests that a mother does not meet the eligibility

l8

on the starting wages for women who found a new job after giving birth. The merit
of Baum (1999) is that it constructed an uniﬁed framework which could determine
the effects of maternity leave legislation on leave, employment, and wages at the
same time. However, it is the demerit of his model in that it was very structural, with
a heavy burden of identiﬁcation problems.

Since the existing literature has used different treatment and control groups as
well as different methods and samples in estimating the economic effects of
maternity leave mandates, it is sometimes hard to summarize those results. For
example, when Kane (1998) examined the effect of the state maternity leave
legislation, the treatment group was all women in states that implemented this
legislation and the control group was all women in states that did not implement this
legislation. Gruber (1994) chose all childbearing aged women as the treatment group,
while using remaining women and men as the control group when he examined the
effect of the PDA. Meanwhile, Waldfogel (1998) used women who were actually
covered by maternity leave as her treatment group and women who were not as her
control group. Generally speaking, most studies found that implementation of
maternity leave legislation discouraged turnover and boosted employment. However,
on wages, it is hard to say that existing studies reached a consensus, though most
studies had suggested that introduction of a certain maternity leave legislation had

negative effect on wages of any of the treatment groups that they chose.

 

requirement, I assign her the mandated amount of leave if she lives in a state with a mandate” (Baum,
1999)

19

IV. Data

This paper uses the National Longitudinal Survey of Youth 1979 (NLSY
1979) to estimate the effects of employer provision of maternity leave beneﬁts on
women’s labor market outcomes. The sample is women who had births during 1988-
1992 and worked just before giving births. If a woman had multiple births during
that period, the most recent birth during that period is counted. From this procedure,
every woman in the sample has only one birth during 1988-92. I refer to that birth as
the reference birth.

The 1988-1992 period is chosen, since the late 19805 are the periods when
the state maternity leave legislation was vigorously implemented. Since maternity
leave has been generalized since 1993 when the FMLA was passed. If I would
include the early 19805 or the late 19905, heterogeneity problem would become
serious. In the early 19805, large ﬁrms tended to provide maternity leave beneﬁts to
women workers, and those women may be intrinsically different from those who
received maternity leave beneﬁts after states or federal mandates. In the late 19905,
women who are not covered by maternity leave beneﬁt might be more likely part-
time workers or new entrants. Or they might be working in a small ﬁrm or in an
exempted industry.

Since I use the calendar year for childbirth, constructing a sample that
satisﬁes the condition that women worked just before giving birth is a little
cumbersome. For the base year wage or the zero year wage in my paper, I use the

pre-birth wage, which is the wage that is found in the childbirth year if the interview

20

date is earlier than the childbirth date. If interview date is later than the childbirth
date in that year, then the pre-birth wage is the previous year’s wage.‘6 The base year
or the zero year is the year when women worked just before the reference birth. All
variables such as tenure, industry, occupation, ﬁrm size, marital status, and so on, are
adjusted according to the above rule.

The advantage of a sample that consists of women who had birth and worked
just before giving birth is that these women are directly affected by the job-
reinstatement right. However, there is a disadvantage, too. In order to accurately
account for employment effect, transition ﬁ'om not-working to working as well as the
transition from working to not-working should be considered. Since my data consists
of women who worked at the time of giving birth, the employment effect cannot
include the transition from not-working to working. However, this sample is better
for showing the effect of maternity leave coverage on turnover, which is a major
issue in assessing the program. Also, even for testing the employment effect, women
who had birth can be an appropriate sample, since we have little reason to regard all
childbearing aged women as a group who are equally interested in and affected by
maternity leave program. In sum, since the purpose of my paper is not to investigate
the effect of extending maternity leave mandates but to investigate the effect of job-
reinstatement right on mothers’ labor market performance, this sample is an

appropriate one for the question.17

 

'6 For example, the base year wage for a woman who gave birth on July 1" 1988 and whose interview
date was May 1" 1988 is the wage that was reported on interview date in 1988. However, the base
year wage for a woman who gave birth on July 1" 1988 and whose interview date was September lSt
1988 is the wage that was reported on interview date in 1987.

‘7 Some state mandates includes men in a beneﬁted group for paternity leave and family and medical
leave. But I do not include men in the sample as a part or as a whole control group. According to
Kane (1998), men are not a good control group for women because their labor market outcomes are

21

Among women who gave birth during 1988-1992 and worked just before
giving birth, women who were full-time students at the entering year of my panel
data were deleted, because low wages of full-time students at the entering year will
make the slope of the wage proﬁles upward biased. Women whose highest hourly
wages are more than a hundred dollars or whose wages are less than a dollar are
excluded. Hourly wages are deﬂated using the CPI. I also exclude women who do
not have information about maternity leave coverage even though they worked at the
base year. After cleaning those observations, the number of women in the sample is

around 850.

 

affected to a smaller degree by this program. “Maternity leave program might increase the input costs
of male labor as well as female labor, but perhaps not to the same extent if men are less likely to take
advantage of the mandated leave. men do not value this program as much as women do and
therefore wages for men will be less affected than wages of women.” Kane (1998).

22

V. Empirical Analysis

V-I. Effects vs. Heterogeneity

Maternity leave is expected to give beneﬁts mainly through the job-
reinstatement right. A popular argument from existing literature that analyzed the
effects of the enactment of a maternity leave legislation is that the beneﬁted group
bears the costs after the enactment, since maternity leave gives beneﬁts in the form
of the job re-instatement. Although most studies assume the maternity leave beneﬁt
to be considerable, no major study except Waldfogel (1998) examined the size of the
beneﬁt provided by maternity leave coverage. Waldfogel argued that the effect of the
maternity leave beneﬁt on wages was big enough to cancel the negative effect of
childbearing. However, by only focusing on the wage proﬁle after giving birth, she
missed an important point: the heterogeneity between maternity leave covered
women and uncovered women.

Systematic differences in the labor market outcomes between maternity leave
covered and uncovered women could be attributed to unobserved heterogeneity as
well as the beneﬁcial effect of maternity leave coverage. The main theme of this
subsection is whether we can identify positive signiﬁcant effect of maternity leave
beneﬁts on labor market outcomes such as wages and employment even after

controlling for heterogeneity.

23

Maternity leave coverage may not be randomly determined.‘8 Women with
more unobserved ability or labor force attachment might prefer covered jobs. This is
the concept of individual heterogeneity. And, jobs that offer coverage may be
different or better in other ways as well that are correlated with higher wages. This is
the concept of job heterogeneity in this paper.

Higher wages of maternity leave covered women can result from their own
self-selection. Assuming that chhildbearing aged women favor the maternity leave
covered job, women who have more unobserved ability or commitment to the labor
force may self-select into the maternity leave covered job. In this case, maternity
leave coverage variable is systematically related to unobserved abilities. Also,
unobserved job characteristics that are correlated with maternity leave coverage
cause the coefﬁcient of the maternity leave coverage variable to be biased. That is,
we are worried that the average value of the error term (8) depends on the coverage
variable: E(e|COV=1) st E(s|COV=0).

This paper uses two different approaches to deal with unobserved
heterogeneity. For the wage analysis, this paper adopts the ﬁxed effects estimation
method. For the turnover and employment analysis, this paper adds more explanatory
variables that designate to be a better control for unobserved heterogeneity. Adding

more explanatory variables is one way to alleviate the heterogeneity problem, since

 

'8 Some researchers have argued that maternity leave coverage can be regarded as an exogenous
variable. Klerman and Leibowitz (1998) suggested potential endogeneity of maternity leave policies.
However, although Ruhm and Teague (1998) suggested that an issue of causation is problematic and
needs more investigation, they ﬁnd that maternity leave policy can be more likely an exogeneous
variable. Changes in family leave policy precedes corresponding movements in the macroeconomic
variables. Winegarden and Bracy (1992) also found that maternity leave program variable is
exogeneous in their three equations for infant mortality rates, labor force participation rates, and
fertility rates. Some researchers have argued that maternity leave coverage can be regarded as an
exogenous variable.

24

factors in the error term that can be correlated with maternity leave coverage variable
can be controlled to a certain degree. Of course, a potential bias problem remains, if
a signiﬁcant amount of heterogeneity remains uncontrolled for even after adding
more explanatory variables. However, ﬁnding an proper instrument variable for the
maternity leave coverage variable is not easy. Also, constructing a panel data set for
ﬁxed- or random effects estimation for turnover or employment is not feasible.
Assume that the effect of maternity leave coverage on employment a year after
giving birth is estimated by the random effects logit model. Then, a hypothetical
panel data set may consist of women who had multiple births and worked at the time
of giving those births. However, this hypothetical sample may have sample selection
bias, since career oriented women who have fewer children will be counted less.
Also, whether the childbirth is, for example, the ﬁrst or the second would affect a
woman’s decision to work just after giving birth. Therefore, this paper adopts the
method of adding more explanatory variables that control for heterogeneity for labor
supply analysis.

Heterogeneity is easily detected in my sample. Figure I-l shows that the
proportion of working women of the maternity leave covered group is always, before
as well as after giving birth, higher than that of the uncovered group. X-axis
represents the time distance from the base year. This time distance uses the
assumption of time invariance. For example, a covered women A of 1989 base year
is assumed to be in the same group in 1993 with a covered women B of 1990 base
year in 1994. Y-axis shows the proportion of the sample that is employed. The large

difference just after giving birth may represent the effect of maternity leave beneﬁt.

25

However, the signiﬁcant difference before giving birth suggests a systematic
difference in labor force attachment between the two groups. The proportion of
working women of the maternity leave covered group is higher than that of
uncovered group before as well as after giving birth. Figure I-2 also suggests the
same implications as Figure H. The large difference in weekly working hours after
the base year suggests the effect of maternity leave coverage. And, weekly working
hours of the covered group was always larger than that of the uncovered group.

Figure I-3 shows the time proﬁle of the average wages of the covered and
uncovered group. Figure I-4 also shows the same thing by education level. The gap
in the wage level between covered and uncovered group after giving birth may
reﬂect an effect of maternity leave. However, signiﬁcant difference in the slope as
well as in the level of the wage proﬁle before giving birth also suggests
heterogeneity. The maternity leave covered group always maintains signiﬁcantly
higher level of wages, and the level of wages of the covered group is higher than that
of uncovered group even four years before giving birth.

The big gap in the wage level by groups suggests that observable
characteristics such as education, actual experience, or tenure might not be able to
capture all those variations in the wage regression. As is seen in Table I-l, there is
little difference in the education level, number of children, and ages between the
covered and the uncovered group, although there is some difference in experience
years past six years between two groups. All variables refer to the base year.
Difference in tenure years is substantial and tenure of the covered group is more than

double that of the uncovered group at the time of giving birth. However, it is unlikely

26

that differences in tenure can explain the entire gap between the covered and
uncovered groups. This means that there may be substantial unobserved parts in the

wage variations that cannot be easily controlled by usual observables.

V-2. Effects on Turnover and Employment

Model and Variables

The hypothesis about the effect on turnover is that maternity leave induces
women to return to their pre-birth job. The hypothesis about the effect on
employment is that there is a positive and signiﬁcant correlation between maternity
leave coverage and employment. The focus is whether the positive and signiﬁcant
effect of maternity leave on labor supply can be recognized even after controlling for
heterogeneity. The model that I use for estimating the effect of maternity leave on

women’s turnover and employment is the probit model of (1).

Y,‘ = X (p + I31 COVi + [32 MED,‘ ‘1' B3 PREWAGEi + B4 UNION,‘ + [35 812134

+Zk=ggk=9lykCOHORTik+Zk=1Fgekaﬁe, ............... (1)

where Y is a dummy variable for turnover or employment. In the case of turnover, Y
is one if a woman returned to her pre-birth employer a year after giving birth, and
zero if a woman changed her employer at that time. I assume that she returns to her

pre-birth employer if her tenure of the next year is larger than that of the base year.

27

Only the CPS job is counted. In the case of employment, Y is one if a woman
worked a year (or two years) after giving birth, and zero if she did not work at that
time. So, the sample used to estimate the turnover model includes women who
worked after childbirth, while the sample for the employment model additionally
includes women who did not work after childbirth. To avoid ambiguity in treating
“employed but absent from work” status that is found in some studies, I assume that
a woman works when her wage is observed.19

Women in the sample have a certain degree of commitment to the labor force
four years before childbirth. In other words, women in the sample had started their
career at least four years before reference birth. That is, the sample dropped women
who were full-time students four years before childbirth.

The maternity leave coverage variable (COV), a dummy variable which is
one if a woman is covered by maternity leave beneﬁt at the base year and zero if a
woman is not covered, is the main concern of this paper. The NLSY collects
maternity leave coverage information starting in 1985. NLSY questionnaire asks
whether women have a job with maternity leave that allows them to go back to their
old job or that pays the same amount of wages as the old one. Women who answered
“yes” to this question for the base year job are classiﬁed as covered.

X is a vector of individual characteristics or, in other term, a vector of control
variables for individual heterogeneity. This includes dummy variables for college
education, SMSA, marital status, and three dummy variables for children. This also
includes actual experiences and tenure. All variables are pertaining to the base year

when women worked just before the reference birth. Since I pool ﬁve years of data, I

 

'9 Ruhm (1998).

28

control for ﬁve cohorts (COHORT). Although tenure is related with job
characteristics, I include tenure in the individual characteristics following the human
capital model. Three dummy variables for children are a dummy variable for
children under age two except the reference child, that for children aged two to ﬁve,
and that for children aged six to seventeen. Three dummy variables can better control
for the number of children than one variable can, since additional young children
under age two may deter their mothers from working.

To control for job heterogeneity, I include a variable indicating whether the
job offers medical insurance (MED), ﬁrm size (SIZE), union membership (UNION),
the log of pre-birth wages (PREWAGE), industry, and occupation. 20 Medical
insurance indicates a “good” job. The ﬁrm size dummy variable is one if the number
of employees is larger than 200 and zero otherwise. Pre-birth wages (PREWAGE)
could be regarded as a measure of individual heterogeneity. Women who earned
more money with the same characteristics might have greater unobserved ability in
market work. However, higher wages (holding constant human capital variables)
may also reﬂect job heterogeneity or “match” heterogeneity, and so PREWAGE is
included in the job heterogeneity variables.

Finally, I include eight dummy variables indicating recent employment

histories (2 k=1 k: 8 D k) of part-time working or not-working during four years before

 

2'01 make seven industry dummy variables from 13 categories of industry for CPS job. Those are the
primary industries; manufacturing; transportation and communication; whole sale and retail trade;
ﬁnance, insurance, and real estate; services; and public administration. Also, I make six occupation
dummy variables from 10 categories of occupation for the CPS job. Those are: managerial and
professional specialties; technical; sales; adminisrative support/clerical; servicies; operator and
others.

29

the base year.21 These variables may be correlated with each other. For example, a
part-time worker a year before reference birth might have worked part-time two
years before reference birth. However, this variable captures the additional effect that
actual experience cannot do. Variables for recent employment histories before giving
birth might be a good indicator whether a woman has a stronger propensity to work.
For example, a woman who worked part-time a year before reference birth may have
less incentive to return to work after childbirth, even though she has the same

characteristics with others.

Findings for Turnover

I estimate equation ( 1) using the maximum likelihood probit estimation.
Table I-2 through Table 10 reports the marginal effect of the coefﬁcients. Table [-2
reports the results of estimating the model on a sample of all women who return to
work the year after giving birth, while Table [-4 focuses on workers with a high
school education or less and Table 1-3 on workers with a college education.22 The
marginal effect indicates the impact of a small change in the independent variable on
the probability of observing the outcome. In the case of a dummy variable such as
the maternity leave coverage variable, it indicates how much that variable being true

(=one) changes the outcome probability over that of that variable being false (=zero).

 

2‘ Eight variables are: not-working a year before giving birth (D'), not-working two years before
giving birth (D2), not-working three years before giving birth (D3), not-working four years before
giving birth (D4), part-time-working a year before giving birth (D5), part-time-working two years
before giving birth (D6), part-time-working three years before giving birth (D7), part-timenot-working
four years before giving birth (D8).

22 College educated workers have more than 12 years of schooling. High school edcuated workers
have 12 years of schooling or less.

30

The estimation results support the hypothesis that the maternity leave beneﬁt
helps women to return to the pre-birth employer just after giving birth. Table I-2
reports that the marginal effect from maternity leave coverage is highly signiﬁcant
and as large as 0.216 to 0.151 depending on the models.

Without any control variables in model [1], the raw probability that maternity
leave covered women stay with the former employers a year after giving birth is
0.216 higher than the probability that not covered women do so. However, this
difference decreases to 0.167 after controlling for individual heterogeneity in the
model [2]. Women who live in the SMSA are more likely return to the former job.
However, SMSA might be closely related with job characteristics, and the marginal
effect of this variable after controlling for job characteristics in the model [3] largely
decreases. Marital status is also highly correlated with the returning behavior.
Married women more likely return to the former job. Presence of additional young
children under age two is correlated with not to return to the pre-birth job.

Another important element of individual heterogeneity that affects returning
behavior is tenure. As we expected, the effect of tenure is highly signiﬁcant through
the speciﬁcations. However, unlike among college educated women as in Table I-3,
tenure is less correlated with rettu'ning to the former job among high school educated
women in Table I-4. The effect of experience is negligible after controlling for
tenure. This makes sense, since what matter for returning to the former employers is
tenure not actual experience.

In model [3] with job heterogeneity control variables, the marginal effect of

ﬁrm size is large and highly signiﬁcant. Women who worked in the large ﬁrm more

31

likely return to the former job. Having a medical insurance or union coverage that
indicates a “good” job adds little explanation to returning behavior.23 The effect may
be small, since medical insurance is correlated with maternity leave beneﬁt.

The most interesting variable in model [3] is perhaps the pre-birth wage. The
effect of pre-birth wage is large and signiﬁcant among college educated women.
College educated women who earned large sum just before giving birth are highly
likely return to the former employer after giving birth compared to college educated
women who earned small sum. However, the effect is negligible among high school
educated. Their wages may be universally low and wage difference might be too
small to induce different returning behavior.

At this point, we evaluate the effect of maternity leave coverage to returning
behavior with the model [3]. The model [3] is the main model in this paper. The
effect of maternity leave coverage on returning to the former employers is as large as
0.156 and highly signiﬁcant even after controlling for various individual and job
heterogeneity. Additionally, the model [4] shows the effect of maternity leave
coverage after including eight recent work history variables (D1 to D8). Although
these eight variables are jointly signiﬁcant at the 90 percent signiﬁcance level, they
barely affect the marginal effect for maternity leave coverage. This implies that these
variables play little role in controlling for propensity to job attachment or career
orientation.

The coefﬁcient for the maternity leave coverage variable does not change

much with the addition of the heterogeneity controls among all workers and among

 

23 Since medical insurance is signiﬁcantly correlated with maternity leave beneﬁt, the effect of
medical insurance may be small.

32

college educated workers. The raw marginal effect of maternity leave dropped
approximately 5 percent point after including individual heterogeneity controls in the
model [2], but the further drop in the model [3] and the model [4] is small among all
workers and among college educated workers. Among high school educated workers,
the amount of the raw marginal effect almost similar through the model [2] to the
model [4].

In sum, signiﬁcant and large marginal effects of additional variables that are
designed to control for heterogeneity shows that there exists heterogeneity in
returning behavior between maternity leave covered women and not-covered women.
However, after controlling for those variables, the marginal effect of maternity leave
coverage on the dependent variable, whether the woman returns to the pre-birth
employer, is still substantial and signiﬁcant. This result suggests that maternity leave

beneﬁt contribute to helping women to return to the former job just after giving birth.

Findings for Employment

This subsection shows the relationship between maternity leave coverage and
employment one year or two years after giving birth. Unlike the above subsection,
where the association between maternity leave coverage and returning behavior can
be interpreted as a direct causal relationship, the relation between maternity leave
coverage and employment is rather loose. However, this relationship can still be

thought as a causal relationship, since women who do not have a right to return to

33

their pre-birth job may have more trouble in ﬁnding new employment after giving
birth and so will more likely simply not to return to work.

Although the employment models have larger samples, since they include
additional category of women who did not work after giving birth, the amount of the
marginal effect of maternity leave coverage on employment is similar to that on
returning to the pre-birth employer in the turnover model. The marginal effect after
one year is 0.244 in the model [1] without any control variables. That is, the raw
probability that maternity leave covered women work a year after giving birth is
0.244 higher than the raw probability that maternity leave not-covered women work
that time. Alter including individual heterogeneity control variables in the model [2],
the effect of maternity leave coverage on employment almost halved. In the model
[3] with individual and job heterogeneity control variables in Table I-5, the effect
further decreases to 0.129.

The marginal effect two years after giving birth is 0.269 in the model [1] with
no controls and 0.156 in the model [3] in personal and job heterogeneity controls in
Table L8. The strong positive association alter two years implies that maternity leave
beneﬁts bring lasting effects on employment, and thereby result in longer actual
experience after giving birth. However, a part of lasting effects of maternity leave
coverage on employment could result ﬁ'om heterogeneity, too. The positive
association after one year largely comes from the turnover effect in the above
subsection. Meanwhile, positive and still large association after two years also
suggests that maternity leave covered women might have higher propensity to work

than not-covered women.

34

The association between various variables and employment in Table 1-5 and
Table I-8 are often different from the association between those and turnover in
Table I-2. Women with college education are more likely work a year after childbirth
and the association between them is more intensiﬁed after two years. However, after
controlling for heterogeneity, the effect of college education on employment is
largely decreased. The effect of experience is large and signiﬁcant especially among
high school educated women a year after childbirth. Meanwhile, the effect of tenure
on employment is still signiﬁcant and large for all women. Unlike the turnover
model, living in the SMSA has little correlation with employment. Also, as we
expected, additional young children under age two deter their mothers from working
and this phenomenon is more conspicuous among college educated women.

In the model [3] with individual and job heterogeneity control variables, pre-
birth wage is interesting. Higher pre-birth wage affects employment among high
school educated women in the employment model, while that affects returning
behavior among college educated women in the turnover model. Including a pre-
birth wage variable in model [3] with other job heteroteneity control variables makes
the effect of maternity leave coverege far smaller than that in the model [2]. Also,
additional eight variables about work history (D1 to D8) play a role to control for
heterogeneity in the model [4]. Although eight work history variables do not pass the
joint signiﬁcance test with 90 percent signiﬁcance level, including them in the model
further decreases the marginal effect of coverage to 0.114 in the Table 1-5 and to

0.121 in the Table I-8.

35

This phenomena of changes in the coefﬁcients on maternity leave coverage
through the model [1] to the model [4] in the employment model is similar with that
of the turnover model. Adding variables to control for heterogeneity works, though
not perfect. The coefﬁcient for the maternity leave coverage decreases after
including additional heterogeneity controls. But the extent of decreases through the
model [2] to the model [4] is small among all workers and college educated workers.
In the meantime, like the turnover model, the extent of decreses by adding
heterogeneity controls is large among high scholl educated.

On the whole, we can interpret the result like this: although there is a sign of
heterogeneity between maternity leave covered women and not-covered women,
after controlling for heterogeneity the effect of maternity leave coverage on

employment is still substantial and signiﬁcant.

V-3. Effects on Wage Proﬁles

Model and Hypotheses

This section will discuss the effect of maternity leave coverage at the time of
giving birth on hourly and weekly earnings of women who worked after birth. Any
difference between the results of the hourly and weekly earnings models will reﬂect
difference between covered and non-covered groups on the intensive margin of labor
supply, that is, the hours decision rather than the participation decision. The weekly

earnings variable, which has two components, that is, hourly wages and weekly work

36

hours, can explain the combined effect that comes from different work hours as well
as different hourly wages. As we have seen in Figure I-2, weekly work hours are
systematically different for covered and uncovered groups after childbirth.
Meanwhile, the hourly wage model reﬂects changes in productivity.

The general hypotheses to be tested in this subsection are summarized as
follows. The ﬁrst hypothesis is that maternity leave has a beneﬁcial impact on
women’s earnings after giving birth, chieﬂy because it extends the pre-birth job
tenure. Empirically, this hypothesis is tested by measuring how much the post-
childbearing earnings of both covered and uncovered workers deviate from their pre-
childbearing trend. The ﬁnding that the post-childbearing earnings of the covered
women did not fall below what would be expected based on their pre-childbearing
earnings path, or at least did not fall as far as the earnings of uncovered women, is
taken as evidence of a beneﬁcial effect of maternity leave coverage. However, there
is a caveat in deciding the effect of maternity leave coverage based on such a test. It
is possible that women who select jobs with maternity leave coverage differ in tastes
from those who do not. In other words, those who have jobs with coverage may on
average be less likely to decide to voluntarily reduce their labor supply after the birth
in order to spend more time caring for the child. This difference in tastes would show
up in the form of lower weekly earnings of the uncovered group after the birth, even
if coverage itself has no causal effect on earnings or labor supply.

The second hypothesis is that the covered and uncovered women have
different earnings growth prior to the childbirth. This could be due either to

heterogeneity or to different human capital investment decisions in response to

37

coverage. That is, the beneﬁcial effect of maternity leave policy partly results from
the human capital accumulated by women who, anticipating having maternity leave
coverage at the time of birth, feel secure that they will continue at their current job.
By removing involuntary fertility-related job separation, being covered by a
maternity leave policy in their earlier career can provide women with the incentive to
increase human capital investment.

This paper estimates the following ﬁxed effects model (2) for analyzing the
effect of maternity leave coverage at the time of childbirth on hourly and weekly

earnings.

wit = ﬂ +X (p + B] EXP“ + [32 EXPSQ“ + [.33 EXPit*COVi + B4 EXPSQ" *COV]

+zk=1 k=59.1)",,+§>3,r_1"=5as,.r)",-.*C0v,+2", ....................... (2)

W is the log of hourly wages or the log of weekly earnings. f is the ﬁxed effect that
summarizes intrinsic, time invariant differences among women in observed and
unobserved characteristics. X is a vector of individual characteristics such as
education, marital status, and number of children. EXP and EXPSQ are experience
years and experience years squared, respectively. Since the slope of the simple wage
proﬁle of the covered group is signiﬁcantly different from that of the uncovered
group, as we have seen in Figures I-3 and L4, I include interactions of EXP*COV
and EXPSQ*COV that only apply to the covered group.

Dk is a set of dummy variables for time passage since childbirth. These

variables measure the year distance from the year of giving birth. I include these

38

variables, since Figures I-3 and L4 show that the slope of the wage proﬁle might be
different before and after the base year. Dkn =1 if a female worker i has the base year
k years before in period t. As is in the simple wage proﬁles’ case, the model uses the
assrunption of time invariance. For example, a covered women A whose base year is
1989 is assumed to be in the same group in 1993 with a covered women B in 1994
whose base year is 1990. A dummy variable for the fourth year after the base year
(1989) is one in 1993 for worker A (D’A,1993 = l) and also a dummy variable for the
fourth year after the base year (1990) is one in 1994 for worker B (D43, 1994 = 1).
Therefore, 0 4 in the model (2) represents the effect on wages of giving birth after
four years. Since the number of observations gets progressively smaller as the proﬁle
approaches the extreme, the proﬁles in later years are less accurate.24 Wage series
are adjusted by CPI. Since a key hypothesis of this paper is that maternity leave
coverage might alter the impact on wages of giving birth, I include interactions, D k
*COV that only applies to the covered group. Finally, a is the error term that is
assumed to have constant variance and to be uncorrelated across individuals and time.

I estimate the model (2) for all women who gave birth during 1988-1992 and
worked just before giving birth. The sample is almost the same as the one that I use
for the labor supply model in the previous subsection. But in this model I do not drop
women who were students four years before childbirth. Instead, I do not use the
wage observations for years when the woman was in full-time student. If wages of

full-time college students were included in the model, then the slope of the proﬁle of

 

2’ The third year after giving birth (D3) includes only 1988, 1989, 1990, and 1991 cohorts. The fourth
year (D4) includes only 1988, 1989, and 1990 cohorts. The ﬁlth year (D5) includes only 1988 and
1989 cohorts.

39

college educated women would be exaggerated, since there are many college
students in the ﬁrst year and the second year of the observation period.25 Since the
slope of the proﬁles are noticeably different according to education groups, I also
estimate model (2) with college educated women and with high school educated

women separately.

Findings

Table I-ll shows the estimation result of the model (2) using the log of
hourly wages as the dependent variable, while Table I-12 uses the log of weekly
earnings. The experience-earnings proﬁle of the covered group is much steeper than
that of the uncovered group during the whole observation period including before
and after childbirth period. One year of actual experience raises the hourly wages of
the covered group by 9.7 percent and that of the uncovered group by 3.3 percent.26
Covered women maintain a signiﬁcantly steeper proﬁle than uncovered women in
the sample of all women, the sample of college educated women, and the sample of
high school educated women. The coefﬁcient for EXP for covered women is 0.110
and that for uncovered women is 0.031 among college educated women. Among
high school educated women, corresponding ﬁgures are 0.069 and 0.038 for covered
and uncovered women, respectively. However, the negative coefﬁcient for the

interaction with quadratic term of experience, EXPSQ*COV, suggests that the

 

25 Ahnost 20 percent of the sample is full-time and part-time college students in 1984 and 1985. After
1987, it is far less than 10 percent.

2" The coefﬁcient for covered women, 0.097, is the sum of the coefﬁcient for EXP and that of
EXP‘COV.

4O

proﬁle of the covered group will be slightly more concave than that of the uncovered
group.

The estimation results of Table I-11 and Table I-12 seems to support the
general hypotheses. To investigate this, I construct the predicted experience-earnings
proﬁles from the coefficients in Table I-11 and in Table I-12. Figure 5 through
Figure I-10 show the predicted proﬁles. These proﬁles compare two women with the
same ﬁxed effect and the same values for the observable characteristics (X’s), both
of whom entered the labor force six years before childbirth.27 Thus, based on this
assumption, both have the same earnings six years before giving birth. After giving
birth, the coefﬁcients for the time-passage-since—birth dummy variables (0,.) and
interactions between these variables with the coverage variable (8],) capture any
discrepancy from the above predicted proﬁles. For example, two years after the base
year, the log of the wage level will be (01+ [33) * 7 + (02+ [34) * 49 + 0 2 + 8 2 for
the covered group and that for the uncovered group will be [31* 7 + B; * 49 + 0 2.

Most ﬁgures show that the uncovered group falls below its pre—birth earnings
proﬁle after the birth takes place. This is also true of the covered group as are seen
ins some ﬁgures. Not being covered by maternity leave especially hurts the wages
one year after childbirth. Meanwhile, the maternity leave covered group generally

maintains the slope of the pre-birth wage proﬁle until one or two years after

childbirth.

 

27 To concentrate the difference between the proﬁles of the covered group and that of the uncovered
group, contributions of other observable characteristics except experience to the wage proﬁles are
assumed to be zero in the proﬁles in Figure I-S - Figure I-lO.

41

Based on the estimation results of Table I-11 and Table I-12, this subsection
tests several hypotheses. The ﬁrst hypothesis is that the covered group maintains
higher wage proﬁles than uncovered counterparts during the pre-childbearing period.
The ﬁgures generally support this hypothesis. The proﬁle of the covered group is
always above the proﬁle of the uncovered group. Speciﬁcally, I test the hypothesis
that if other things are equal, the difference in the log of the wages of the covered

group and the uncovered group with 5 years of experience is zero.

Hypothesis 1: Ho: 5 [32+ 25 04:0.

The null hypothesis is rejected at 90 percent signiﬁcance level for all women,
college educated women, and high school educated women with the model with
hourly wages. The null hypothesis for the model with weekly earnings is rejected
with all women and college educated women, but not with high school educated
women.

The second hypothesis is that childbirth negatively affects wages in the
absence of maternity leave coverage. Speciﬁcally, for a year after giving birth, the

null hypothesis is

Hypothesis 2: H0: 0 1 2. 0 .

The coefﬁcients for D1 are all negative in Table I-11 and Table I-12, but only for

college educated women’s hourly earnings is this coefﬁcient signiﬁcant. The null

42

hypothesis is rejected at 90 percent signiﬁcance level in that case. The subsequent
tests for 9 k generally suggests that the effect of giving birth on both hourly and
weekly earnings is negative. Most coefﬁcients (0],) are negative, several are
signiﬁcant, and taken together all the negative signs suggest a cumulative negative
effect.

The third hypothesis is that the impact of childbirth on earnings is different
for the covered group compared to the uncovered group. Speciﬁcally, for a year alter

giving birth, the null hypothesis is

Hypothesis 3: H0: 5 I = 0.

The coefﬁcients for D1* COV (5 1) are all positive in Table I-11 and Table I-12, but
signiﬁcant only for all women and college educated women with hourly wages. The
null hypothesis can be rejected in those cases. In the sample of all workers and of
college educated workers, 5., is almost always positive and often signiﬁcant. For
high school educated workers, however, the post-childbirth earnings proﬁle of the
covered workers starts above, but then falls signiﬁcantly below, that of the
uncovered workers.

The fourth hypothesis is whether childbirth causes the maternity leave
covered group to depart signiﬁcantly from their pre—childbirth earnings proﬁle.
Maternity leave coverage at the time of giving birth may cancel some of negative
effect of childbearing, but not all of them. Speciﬁcally, for a year after giving birth,

the combined effect of childbearing and maternity leave coverage can be zero.

43

Hypothesis4: Ho: 01+81=0.

In the model discussing the hourly wages of all women, this hypothesis is
rejected only for the ﬁfth year after childbirth. The proﬁle of the covered group is
ahnost on the pre-birth trajectory even after the birth except the ﬁfth year.28 With the
sample of the college educated women, the covered group seems to maintain the pre-
birth slope.29 (Column [2] in Table 1-11 and Figure 1-7). With the sample of the high
school educated women, covered workers maintain the pre-birth trajectory until the
second year after birth, but after that their proﬁle is signiﬁcantly below the pre-birth
trajectory.3O (Column [3] in Table I-11 and Figure 1-9).

In the weekly earnings model for all women, covered women do not seem to
maintain their pre-birth wage proﬁle after giving birth. The alter-birth proﬁle is
signiﬁcantly below the pre-birth trajectory in the second, fourth, and ﬁfth year are
seen in Figure I-6 except the ﬁfth year.3 I It is clear that this result is being driven by
the results for the high school educated workers, in which the weekly earnings
proﬁle of the covered workers falls below that of the uncovered workers. In the
sample of college educated workers, there is only one signiﬁcant deviation from the

pre-birth earnings. Overall, the results suggest that maternity leave coverage

 

28 However, the number of groups and observations are much smaller and therefore less exact in the
extreme year.
29 The p-value for the ﬁrst year is 0.52, for the second year is 0.05, the third year is 0.81, the fourth
year is 0.88, and for the ﬁfth year is 0.66.

0 The p-value for the ﬁrst year is 0.42, for the second year is 0.98, the third year is 0.04, the fourth
year is 0.04, and for the ﬁfth year is 0.00.

' The p-value for the ﬁrst year is 0.71, for the second year is 0.04, the third year is 0.24, the fourth
year is 0.05, and for the ﬁlth year is 0.06.

44

eliminates ahnost all of the negative effect of childbearing for college educated

workers, but not for hi gh-school educated workers.

45

VI. Conclusion

The empirical ﬁndings of this paper can be summarized as follows. First, the
results support the assertion that maternity leave coverage at the time of childbirth
has a beneﬁcial effect on women’s labor market outcomes. By preventing
involuntary job separation at the time of childbirth, it signiﬁcantly lessens turnover a
year after childbirth. Also, employment a year or two years after childbirth is
signiﬁcantly affected by maternity leave coverage at the time of the childbirth.

Analysis of wage proﬁles is also consistent with the basic hypothesis. The
women who had a covered job at the time of childbirth maintain steeper wage
proﬁles before and after childbirth, although this could reﬂect an impact of
anticipated maternity leave coverage on pre-childbearing capital accumulation or
simply heterogeneity in wage growth between women in covered and uncovered jobs.
Among women with a high school education or less, childbirth negatively affects the
wage and weekly earnings proﬁles of both covered and uncovered women. Among
college educated women, however, there is little evidence of negative effect of
childbearing on earnings for women covered by maternity leave at the time of giving
birth, but there is evidence that over a several-year period, women not covered at the
time of birth suffer negative earnings consequences. Overall, the evidence for an
impact of coverage on earnings seems weaker than the evidence for an impact of
coverage on turnover and employment.

This paper adds to the literature by analyzing the effect of actual maternity

leave coverage (as opposed to an incomplete state mandate for maternity leave

46

coverage) on turnover and employment, using explicit and direct controls for
heterogeneity. Also, by using a ﬁxed effect model and data on wages both before and
after giving birth, this paper provides more accurate estimates of the impact on
earnings of ﬁrm-provided maternity leave coverage. Although much effort has been
made to control for heterogeneity that can bias the estimation result, it is possible
that some heterogeneity between covered and uncovered women remains
uncontrolled for.

During the 1990’s, the general policy approach towards maternity leave was
to attempt to increase the number of employees with maternity leave coverage by
legislative mandates that required increasing numbers of employers to offer such a
beneﬁt. An argument in favor of this approach to expanding maternity leave
coverage would seem to require, at a minimum, evidence that existing maternity
leave programs seemed to beneﬁt the woman they covered. This paper has provided

such evidence.

47

REFERENCES

Addison, J .T. and Portugal P.(1989). “Job Displacement, Relative Wage Changes,
and Duration of Unemployment, Journal of Labor Economics, Vol.7, No.3.

Albrecht, James W, Per-Anders E, Sundstrom M. and Vroman S. B. (2000). Career
Interruptions and Subsequent Earnings: A Reexamination Using Swedish
Data, Journal of Human Resources.

Baum II, C. (1999). The Effects of Maternity Leave Legislation 0n Mothers' Labor
Suppy After Childbirth, a dissertation, the University of North Carolina at
Chapel Hill.

Blau, F. (1992). “Rising Wage Inequality and the US. Gender Gap.” American
Economic Review 84.

Blau, F. and Kahn, L. (1992). The Gender Earnings Gap: Learning from
International Comparisons,” American Economic Review 82.

Bonds, J. T. et. a1. (1991). Beyond the Parental Leave Debate: The Impact of Laws in
Four States, Families and Work Institute.

Duncan, G. J. and Rogers, W. (1990). “Lone-parent families and their economic
problems: transitory or persistent?” in Lone-Parent Families .' the Economic
Challenge, Paris, OECD.

Finn-Stenenson, M. and Trzcinski, E. (1991) “Mandated Leave: an Analysis of

Federal and State Legislation,” American Journal of Orthopsychiatrist.

48

Frank, M. and Lipner, R. (1988). “History of Maternity Leave in Europe and the
United States,” The Parental Leave Crisis: Toward a National Policy, Edited
by ER Zigler and M. Frank, Yale University Press.

Gardecki, R. (1997). “The Labor Market Effects of Maternity Mandates”, Michigan
State University.

Goldin, C. (1997). “Career and Family: College Women Look to the Past,” Gender
and Family Issues in the Workplace, Francine D. Blau and Ronald G.
Ehrenberg eds.

Greene, W.H. (1997). Econometric Analysis.

Gruber, J. (1994, June). “The Incidence of Mandated Maternity Beneﬁts,” The
American Economic Review.

Gustafsson, S. and Stafford, F. (1995). “Links Between Early Childhood Programs
and Maternal Employment in Three Countries,” The Future of Children, Vol.
5, No. 3.

Gustafsson, S. (1981). Male-Female Lifetime Earnings Differentials and Labor Force
History, Eliasson G., Holmlund B., and Stafford, F. P., The Industrial
Institute for Economic and Social Research, IUI conference Reports,
Stockholm.

Jacobsen J. and Levin, L. (1995). “Effects of Intermittent Labor Force Attachment
on Women’s Earnings,” Monthly Labor Review.

Jacobson, S.J., LaLonde, J. and Sullivan, D. (1993) “Earning Losses of Displaced

Workers,” American Economic Review.

49

Kane, K. (1998). State Mandates for Maternity Leave: Impact on Wages,
Employment, And Access to Leave, a dissertation, Boston College.

Kamerman, S, Kahn, A. and Kingston, P. (1993). Maternity Policies and Working
Women, Columbia University Press.

Kamerman, S. (1988). “Maternity and Parenting Beneﬁts: An International
Overviews,” The Parental Leave Crisis: Toward a National Policy, Edited by
E.F. Zigler and M. Frank, Yale University Press.

Klerman, J. and Leibowitz, A. (1997). “Labor Supply Effects of State Maternity
Leave Legislation,” Gender and Family Issues in the Workplace, Francine D.
Blau and Ronald G. Ehrenberg eds.

Krueger, A. (1994). “Observations on Employment-Based Government Mandates,
With Particular Reference to health Insurance,” Working Paper #323,
Princeton University.

McGladrey and Pullen, (1993). The Family and Medical Leave Act: A Practical
Guide to Cost-Effective Compliance, The Panel Business management
Series, Aspen Publishers, Inc.

Neumark, D. and Korenman, S. (1994) “Sources of Bias in Women’s Wage
Equations: Results Using Sibling Data,” Journal of Human Resources 29.

Mincer, J. and Ofek, H. (1981). “Interrupted Work Careers : Depreciation and
Restoration of Human Capital,” The Journal of Human Resources XVII.

Mincer, J. and Polachek, S. (1974). “Earnings of Women,” Journal of Political

Economy 82.

50

Pizzo, P. (1988). “Uncertain Harvest: Maternity Leave Policies in Developing
Nations,” The Parental Leave Crisis: Toward a National Policy, Edited by
E.F. Zigler and M. Frank, Yale University Press.

Rubin, D. and Rosenbaum, P. (1983). “The Central Role of the Propensity Score in
Observational Studies for Causal Effects,” Biometrika 70.

Ruhm, C.J. (1988). “The Econonric Consequence of Parental Leave Mandates:
Lessons from Europe”, A Quarterly Journal of Economics, Vol.113.

Ruhm, OJ. and Teague, J. (1997). “Parental Leave Policies in Europe and North
America,” Gender and Family Issues in the Workplace, Francine D. Blau and
Ronald G. Ehrenberg eds.

Spalter-Roth R. and Hartman, H. (1990). Unnecessary Losses: Costs to Americans of
the Lack of Family and Medical Leave, Institute for Women’s Policy
Research.

Summers, L. (1989). Some Simple Economics of Mandated Beneﬁts, AEA Papers
and Proceedings, American Economic Review.

Trzcinski, E. and Alpert, W. (1991). “Pregnancy and Parental Leave Beneﬁts in the
United States and Canada: Judicial Decisions and Legislation,” Journal of
Human Resources 29, XXIX 2.

Waldfogel, J. (1994). “Women Working for Less: Family Status and Women’s Pay
in the US and UK.” Harvard University.

Waldfogel, J. (1998). “The Family Gap for Young Women in the United States and
Britain : Can Maternity Leave Make a Difference?”, Journal of Labor

Economics, Vol.16, No.3.

51

Waldfogel, J. (1997). “Working Mothers Then and Now: A Cross-Cohort Analysis
of the Effects on Maternity Leave on Women’s Pay,” Gender and Family
Issues in the Workplace, Francine D. Blau and Ronald G. Ehrenberg eds.

Wooldridge, J. (2000). Introductory Econometrics.

52

Table I-l: Summary Statistics: Women who had birth
during 1988—92 and Worked just before childbirth

 

COVERED UNCOVERED
(All)
N of Obs 709 166
Age 28.0 27.6
(0.09) (0.19)
Education 13.1 12.4
(years) (0.07) (0.14)
Tenure 3.8 1.7
(years) (0.11) (0.18)
Pre-birth wage 1113.2 785.8
(18.50) (32.65)
N of children 0.98 1.03
(0.03) (0.08)
Experience 5.54 4.95
(years) (0.03) (0.10)
(Coll Educated)
N of Obs 344 55
Age 28.6 28.4
(0.13) (0.32)
Education 14.8 14.5
(0.08) (0.16)
Tenure 4.0 1.9
(0.17) (0.31)
Pre—birth wage 1318.6 948.9
(29.28) (65.96)
N of children 0.79 0.78
(0.04) (0.14)
Experience 5.74 5.07
(0.03) (0.17)
(High School Educated)
N of Obs 365 111
Age 27.4 27.2
(0.13) (0.23)
Education 11.6 11.5
(0.06) (0.11)
Tenure 3.6 1.8
(0.16) (0.24)
Pre-birth wage 919.7 705.0
(17.86) (32.88)
N of children 1.16 1.16
(0.05) (0.10)
Experience 5.36 4.90
(0.05) (0.12)

 

53

Table I-2: Turnover One Year after Childbirth (All Workers)

(The Probit Model)

 

[1] [2] [3] [4]
dF/dx s.err. dF/dx s.err. dF/dx s.err. dF/dx sd.err.

 

cov 0.216(0.041) 0.167(0.040) 0.156(0.048) 0.151(o.047)
DCOLL -o.014(0.024) —0.025(0.026) -0.039(0.026)
TENBASE 0.015(0 004) 0.013(0.004) 0.012(0.005)
EXPBASE —0.001(0.006) -0.002(0.006) -0.003(0.006)
DRACE -0.057(0.025) -0.049(0.025) -o.050(o.024)
SMSA 0.043(0.032) 0.021(o.030) 0.018(0.030)
MARRIED 0.060(0.033) 0.047(0.033) 0.055(0.033)
carrnr -0.072(0.036) —0.056(0.035) -0.053(0.034)
CHILDZ 0.008(0.029) 0.009(0.029) 0.019(0.028)
CHILD3 —0.039(0.026) —0.034(0.026) —0 014(0.025)
MED -0.010(0.030) —0.006(0.030)
PREWAGE 0.031(0.031) 0.028(0.031)
SIZE 0.041<o.o25) 0.036(0.024)
UNION 0.002(0.032) —0.001(0.o32)
D_4(NOTN4) 0.057(o.o34)
D-3(NOTN3) -o.27r(0.115>
0-2(NOTN2) 0.066(0.036)
D-1(NOTN1) 0.085(0.026)
D-8(PARTN4) o.oos<o.o33)
D_7(PARTN3) -O.106(0.048)
D_6(PARTN2) 0.003(0.o38)
D‘5(PARTN1) —0.008(0.038)
COHORT DUM. No Yes Yes Yes

INDUSTRY DUM. No No Yes Yes

OCCUP. DUM. No No Yes Yes

N of Obs 808 808 793 793

Sig of eight significant

Dummies*

 

* Joint significance test for eight work history dummy variables with 90
percent significance level. This is the same from Table I—2 to Table I-lO.

. .. . -1 .
** Not-working er year before grvrng birth (D ), not—working two years

before giving birth (D_2), not—working three years before giving birth (D_3),

not-working four years before giving birth (D_4), part-time-working a year
before giving birth (D-S), part-time—working two years before giving birth

— . . . . -7
(D 6), part-time-workrng three years before g1v1ng birth (D ), part-

timenot-working four years before giving birth (D_8). This is the same from

Table I-2 to Table I-10.

*** Cohort dum. represents whether the model includes cohort dummy variables.
Indusgry dum. represents whether the model includes industry dummy variables.
Occup. dum. represents whether the model includes occupation dummy variables.
This is the same from Table I—2 to Table I—lO.

54

Table I-3:

(The Probit Model)

Turnover One Year after Childbirth

(College Educated Workers)

 

 

[l] [2] [3] [4]

dF/dx s.err. dF/dx s.err. dF/dx s.err. dF/dx s.err.
COV O.l90(0.062) O.l35(0.058) 0.121(0.066) 0.105(0.065)
TENBASE 0.022(0.006) 0.017(0.006) 0.019(0.006)
EXPBASE -O 008(0.008) 0.008(0.007) -0.006(0.007)
DRACE -0.040(0.033) 0.031(0.033) -0.040(0.032)
SMSA 0.070(0.050) 0.030(0.043) 0.023(0.042)
MARRIED 0 065(0.055) 0.035(0.050) 0.021(0.046)
CHILDl -O 012(0.043) 0.003(0.038) 0.000(0.039)
CHILDZ 0 014(0.038) 0.019(0.036) 0.026(0.036)
CHILD3 -O 011(0.037) 0.006(0.034) -0.0l7(0.036)
MED 0.013(0.038) 0.000(0.040)
PREWAGE 0.062(0.039) 0.054(0.038)
SIZE 0.034(0.032) 0.032(0.03l)
UNION 0.009(0.042) 0.020(0.039)
D-4(NOTN4) dropped
D-3(NOTN3) —0.707(0.266)
D‘ (NOTN2) 0.084(0.018)
D-1(NOTN1) 0.026(0.072)
D-8(PARTN4) 0.006(0.047)
D-7(PARTN3) —0.039(0.058)
D—6(PARTN2) 0.026(0.044)
D_5(PARTN1) 0.035(0.040)
COHORT DUM. No Yes Yes Yes
INDUSTRY DUM. No No Yes Yes
OCCUP. DUM. No No Yes Yes
N of Obs 396 396 390 378
Sig of seven not sig

dummies

 

55

Table I-4:

Workers)

(The Probit Model)

Turnover One Year after Childbirth

(High School Educated

 

 

[l] [2] [3] [4]
dF/dx s.err. dF/dx s.err. dF/dx s.err. dF/dx s.err.
COV 0.234(o.055) 0.213(0.057) O.206(0.071) 0.188(0.072)
TENBASE 0.007(0.006) 0.005(0.006) 0.002(0.006)
EXPBASE 0.006(0.008) 0.007(0.008) 0.007(0.010)
DRACE 0.085(0.036) 0.080(0.035) 0.079(0.o35)
SMSA 0.034(0.042) o.030<o.o42) 0.024(0.041)
MARRIED o.o71<0.043) 0.058(0.04l) 0.07o<o.044)
CHILDl —O.140(0.058) -o.115<0.054) 0.103(0.054)
CHILDZ —0.011(o.045) -0.010(0.o42) 0.009(0.041>
CHILD3 -0.058(0.034) -0.042(o.o32) —o.010(0.034)
MED -0.008(0.040) -o.010(0.042)
PREWAGE —o.012(0.047) —0.009(0.049)
SIZE 0.037(o.033) 0.043(0.034)
UNION -o.003(0.043) 0.009(o.043>
'4 (NOTN4) 0.015(0.058)
D’3 (NOTN3) —0.169(0.124)
D‘2 (NOTNZ) 0.030(0.068)
D"1 (NOTNl) dropped
D'8 (PARTN4) 0.004(0.043)
D'7 (PARTN3) -o.125(o.07o)
9‘6 (PARTNZ) -o.009(o.054)
D-S (PARTNl) -o.044(o.059)
COHORT DUM. No Yes Yes Yes
INDUSTRY DUM. No No Yes Yes
OCCUP. DUM. No No Yes Yes
N of Obs 412 412 392 378
Sig of seven not sig
dummies

 

56

Table I—5:

Employment One Year after

(The Probit Model)

Childbirth

(All Workers)

 

 

[l] [2} [3] [4]
dF/dx s.err. dE/dx s.err. dF/dx s.err. dF/dx s.err.
cov 0.244(0.o35) 0.158(0.035) 0.129(0.o40) 0.114(0.o39)
DCOLL o.049<o.023) -0.001(o.025) —0.003(0.026)
TENBASE 0.018(0.005) 0.012(0.005) 0.008(0.005)
EXPBASE 0.015(0.005) 0.013(0.005) 0.011(0.006)
DRACE -o.014(o.024) -0.019(0.024) —0.010(0.024)
SMSA —o.030(0.025) -0.055(0.022) —0.056(0.022)
MARRIED 0.051(0.030) 0.045(0.029) 0.040(o.o30)
CHILDl -o.057(o.033) -o.051(o.o31) —0.038(0.031)
CHILDZ —o.017(0.029) -0.009(0.028) 0.014(0.027)
CHILDB 0.012(0.024) 0.007(0.023) 0.020(0.024)
MED —0.004(0.028) -0.010(0.028)
PREWAGE 0.058(o.029) 0.046(0.029)
SIZE 0.003(0.025) 0.004(0.025>
UNION 0.064(0.024) 0.062(0.024)
'4 (NOTN4) 0.002(0.045)
D'3 (NOTN3) 0.0l4(0.046)
D'2 (NOTN2) —0.069(0.07l)
D"1 (NOTNl) -0.028(0.061)
D'8 (PARTN4) —o.oo4(o.o32)
D"7 (pARTN3) 0.018(0.030)
D”6 (PARTNZ) —o.110(0.049)
D‘5 (PARTNl) —o.o91<o.o42>
COHORT DUM. No Yes Yes Yes
INDUSTRY DUM. No No Yes Yes
OCCUP. DUM. No No Yes Yes
N of Obs 928 921 898 898
Sig of eight significant
Dummies

 

57

Table I-6: Employment One Year after Childbirth (College Educated Workers)

(The Probit Model)

 

[1] [2] [3] [4]
dF/dx s.err. dE/dx s.err. dF/dx s.err. dF/dx s.err.

 

cov 0.183(0.054) 0.121(0.051) 0.106(0.059) 0.113(0.060)
TENBASE 0.013(0.006) 0.010(0.005) 0.006(0.00S)
EXPBASE 0.008(0.007) 0.008(0.006) 0.012(0.006)
DRACE -0.006(0.030) -0.017(0.029) —0.020(0.028)
SMSA -o.059(0.027) -0.061(0.023) —0.062(0.020>
MARRIED 0.039(0.046) 0.039(0.047) 0.030(0.044)
CHILDl -0.083(0.046) -0.074(0.044) -o.067(o.043)
CHILDZ —0.025(0.037) -0.024(0.o36) 0.004(0.031)
CHILD3 0.034(0.029) 0.023(0.028) 0.023(0.027)
MED -0.005(0.035) —o.013<0.031)
PREWAGE 0.007(0.034) 0.012(0.033)
SIZE 0.005(0.029) 0.005(0.027)
UNION 0.044(0.027) 0.039(0.026)
D-4 (NOTN4) 0.054(o.022)
D‘3 (NOTN3) —o.205(0.202)
D’2 (NOTN2) 0.032(0.046)
D'1 (NOTNl) —o.o37(0.091>
D’8 (PARTN4) 0.043(0.026>
D'7 (PARTN3) -o.oo1(0.044)
D’6 (PARTNZ) -o.178(0.095)
D's (PARTNl) —0.021(0.047)
COHORT DUM. No Yes Yes Yes
INDUSTRY DUM. No No Yes Yes

OCCUP. DUM. No No Yes Yes

N of Obs 432 428 419 419

Sig of eight not sig

Dummies

 

58

Table I-7:

Employment One Year after Childbirth
(High School Educated Workers)

(The Probit Model)

 

 

[1] [2] [3] [4]
dF/dx s.err. dF/dx s.err. dF/dx s.err. dF/dx s.err.
cov O.269(0.046) 0.181(0.047) 0.156(0.056) 0.118(0.056)
TENBASE 0.024(o.oo7) 0.014(o.oo7) 0.007(0.008)
EXPBASE 0.020(0.008) 0.014(0.008) 0.011(o.o10)
DRACE -0.023(0.038) -0.023(0.038) —o.010(o.o39)
SMSA ~0.018(0.039) -o.055(o.033) -o.os4(o.o34>
MARRIED 0.068(0.o4o) 0.070(o.o41) 0.068(0.042)
CHILDl -0.029(0.046) —o.024(o.o44) —o.o13<o.o45)
CHILDZ -0.006(0.043) —o.oo4(o.042) 0.022(0.041)
CHILD3 -o.002(0.o35) -o.o1o(o.o34) o.oos<o.o37)
MED —o.018<o.041) -0.016(0.044)
PREWAGE 0.141(o.o49) 0.124(0.050)
SIZE 0.009(0.038) 0.023(0.038)
UNION 0.080(0.037) 0.085(0.037)
’4 (NOTN4) -0.076(0.080)
D‘3 (NOTN3) 0.071(o.o49)
D‘2 (NOTN2) —o.111(o.104)
D_1 (NOTNl) -0.011(0.082)
0‘8 (PARTN4) —o.045(o.051)
D‘7 (PARTN3) 0.044(o.o41)
D"6 (PARTNZ) -0.088(0.064)
D’5 (PARTNl) —O.154(0.067)
COHORT DUM. No Yes Yes Yes
INDUSTRY DUM. No No Yes Yes
OCCUP. DUM. No No Yes Yes
N of Obs 496 493 472 472

Sig of eight

Dummies

significant

 

59

Table I-8:

(The Probit Model)

Employment Two Years after Childbirth

(All Workers)

 

 

[l] [2] [3] [4]

dF/dx s.err. dF/dx s.err. dF/dx s.err. dE/dx s.err.
cov 0.215(0.037) 0.138(0.037) 0.113(0.044) 0.121(0.045)
DCOLL 0.123(0.028) 0.070(0.031> 0.050(0.o32)
TENBASE 0.022(0.005) 0.016(0.005) 0.013(0.005)
EXPBASE 0.021(0.006) 0.019(0.006) 0.008(0.007)
DRACE -o.043(0.029) -0.059(0.030) -0.058(o.030)
SMSA —0.002(0.033) —0.034(0.o31) -0.031(0.032)
MARRIED 0.000(0.032) 0.004(0.033) 0.007(0.034)
CHILDl -0.041(0.036) -o.041<0.036) -0.038(0.036)
CHILDZ -0.001(o.o33> —0.003(0.034) 0.013(0.033)
CHILD3 0.057(0.028) 0.048(0.029) 0.066(0.029)
MED 0.015(0.038) -0.011<0.037)
PREWAGE 0.100(0.036) 0.090(0.036)
SIZE 0.024(0.030) 0.031(0.029)
UNION —0.011(0.039) —0.009(0.039)
D'4(NOTN4) -0.112(0.073)
D‘3(NOTN3) -0.065(0.075)
D'2(NOTN2) o.oo4<o.oe7)
D'1(NOTN1) -O.105(0.088)
D'8<PARTN4) -o.1o4(o.047)
D'7(PARTN3) 0.030(0.038)
D'6(PARTN2) —0.o32(0.047)
D'5(pARTN1) -0.051(0.046)
COHORT DUM. No Yes Yes Yes
INDUSTRY DUM. No No Yes Yes
OCCUP. DUM. No No Yes Yes
N of Obs 928 921 898 898
Sig of eight significant

Dummies

 

60

Table I—9: Employment Two Years after Childbirth
(College Educated Workers)

(The Probit Model)

 

 

[l] [2] [3] [4]

dF/dx s.err. dF/dx s.err. dF/dx s.err. dF/dx s.err.
COV 0.153(0.056) 0.081(0.049) 0.137(o.070) 0.145(0.072)
TENEASE 0.021(0.006) 0.021(0.006) 0.020(0.006)
EXPBASE 0.013(0.007) 0.009(0.007) 0.002(0.007)
DRACE 0.013(0.034) —0.021(o.033) -0.023(0.032)
SMSA 0.012(0.043) 0.007(0.041) 0.010(0.041)
MARRIED —o.010(0.044) 0.011(0.046) 0.012(0.046)
CHILDl -0.135(0.052) ~0.145(0.054) ~0.136(0.054)
CHILD2 0.010(0.038) 0.016(0.033) 0.027(0.032)
CHILD3 0.098(0.029) 0.078(0.027) 0.090(0.027)
MED —0.018(0.038) -0.019(0.038)
PREWAGE 0.043(0 038) 0.028(0.038)
SIZE -o.043(0.035) -o.033(0.034)
UNION —o 096(0.053) —0 096(0.054)
D'4(NOTN4) -o.o91(o.131)
D'3(NOTN3) —O.l30(0.165)
D'2(NOTN2) —o.1o9<o.142)
D‘1(NOTN1) -0.026(0.086)
D'8(PARTN4) —0.018(0.056)
D-7(PARTN3) 0.021(0.041)
D‘6(pARTN2) 0.011<o.o47)
D'5(pARTN1) o.o14<o.o42)
COHORT DUM. No Yes Yes Yes
INDUSTRY DUM. No No Yes Yes
OCCUP. DUM. No No Yes Yes
N of Obs 432 428 419 419
Sig of eight not sig

Dummies

 

61

Table I—lO: Employment Two Years after Childbirth
(High School Educated Workers)

(The Probit Model)

 

[l] [2] [3] [4]
dF/dx s.err. dF/dx s.err. dF/dx s.err. dF/dx s.err.

 

COV O.232(0.048) 0.178(0.051) 0.098(0.060) 0.094(0.062)
TENBASE 0.023(0.008) 0.012(0.008) 0.005(0.008)
EXPBASE 0.027(0.010) 0.021(0.010) 0.006(0.012)
DRACE —0.088(0.047) -0.086(0.048) -0.083(0.048)
SMSA -0.025(0.048) -0.081(0.045) -0.071(0.046)
MARRIED 0.008(0.046) 0.014(0.047) 0.019(0.049)
CHILDl 0.056(0.049) 0.057(0.049) 0.059(0.049)
CHILD2 -o.021(0.052) -0.009(0.052) 0.000(0.053)
CHILD3 0.012(0.043) 0.000(0.044) 0.025(0.046)
MED 0.033(0.057) -0.004(0.057)
PREWAGE 0.194(0.061) O.200(0.063)
SIZE 0.079(0.045) 0.094(0.045)
UNION 0.067(0.057) 0.078(0.056)
D"4(NOTN4) —o.202(0.100)
D'3(NOTN3) —o.043(0.095)
D‘2(NOTN2) o.o33<o.o90)
D-1(NOTN1) —o.046(0.114)
D'8(PARTN4) —o.173(0.067)
D-7(PARTN3) 0.020(0.060)
D” (PARTNZ) —0.038(0.068)
D'5(PARTN1) —0.095(0.068)
COHORT DUM. No Yes Yes Yes

INDUSTRY DUM. NO No Yes Yes

OCCUP. DUM. No No Yes Yes

N of Obs 496 493 479 479

Sig of eight significant

Dummies

 

62

Table I-ll: The Effects of Coverage at the Time of Childbirth
on the Wage Profiles (Hourly Wages)

(The Fixed Effects Estimation)

 

 

Dep. Var. = Hourly Wages

[l] [2] [3]

All Workers Coll Wkrs High S. Wkrs

Coeff. std.err. Coeff. std.err. Coeff. std.err.
EDU 0.070 (0.007) 0.073 (0.008) 0.007 (0.019)
NUMCD -0.033 (0.008) —0.069 (0.011) 0.006 (0.011)
MARR_D -0 002 (0.014) 0.006 (0.022) 0.000 (0.019)
MARR_N -0.016 (0.013) -0.012 (0.016) —0.002 (0 020)
EXP 0.033 (0.009) 0.031 (0.013) 0.038 (0.014)
EXP*COV 0.064 (0.010) 0.079 (0.013) 0.031 (0.016)
EXPSQ -0.001 (0.001) 0 000 (0 001) —0.002 (0 001)
EXPSQ*COV —0.004 (0 001) —0.005 (0 001) -0.001 (0.001)
D1 —0.051 (0.031) -0.089 (0.048) -0.023 (0.041)
D1*COV 0.066 (0 035) 0.102 (0.052) 0.043 (0.048)
02 —0.020 (0.036) —0 028 (0.052) —0 020 (0.049)
02*cov 0.042 (0.039) 0.071 (0.056) 0.020 (0.055)
03 —0.079 (0.041) —0.063 (0.060) —0.095 (0.058)
D3*COV 0.054 (0.045) 0.069 (0.065) 0.035 (0.064)
D4 -0.093 (0.044) —0 088 (0.065) —0 100 (0.059)
04*cov 0.058 (0.049) 0.084 (0 071) 0 033 (0.068)
D5 —0.035 (0.048) -0.081 (0.072) —0.004 (0.064)
D5*cov —0.026 (0 054) 0.066 (0.079) —0.103 (0.074)
CONSTANT 5.555 (0.095) 5.519 (0.115) 6.280 (0 216)
N of Groups 1025 559 466
N of Obs 9708 5437 4271
Obs per Grp (avg) 9.5 9.7 9.2

F(18,8665)=75.43 F(18,4860)=68.80 F(18,3787)=ll.90

 

* The number of groups and observations are progressively smaller after the
third year. The third year after giving birth (D3) includes only 1988, 1989,
1990, and 1991 cohorts. The fourth year (04) includes only 1988, 1989, and

1990 cohorts. The fifth year (D5) includes only 1988 and 1989 cohorts. This
is the same in Table I—12.

63

Table I-12:

(Weekly Earnings)

The Effects of Coverage at the Time of Childbirth
on the Wage Profiles

 

 

(Fixed Effects Estimation)

Dep. Var. = Weekly Earnings

[1] [2] [3]

All Workers College Wkrs High 3. Wkrs

Coeff. std.err. Coeff. std.err. Coeff. std.err.
EDU 0.180 (0.014) 0.215 (0.016) —0.010 (0.032)
NUMCD -0 099 (0.014) -0.138 (0.020) -0.042 (0.020)
MARR_D 0.078 (0.025) 0.097 (0.041) 0.075 (0.031)
MARR_N 0.043 (0.022) 0.018 (0.030) 0.106 (0.033)
EXP 0.083 (0.018) 0 066 (0.025) 0.133 (0.026)
EXP*COV 0.088 (0.019) 0.118 (0.026) 0.001 (0.029)
EXPSQ -0.004 (0.001) —0.003 (0.002) -0 006 (0.002)
EXPSQ*COV -0.005 (0 001) —0.006 (0.002) 0.001 (0.002)
D1 -0.100 (0.065) -0.098 (0.094) -0.108 (0.089)
D1*COV 0.110 (0.071) 0.091 (0.101) 0.143 (0.097)
02 -0.129 (0.079) -0 129 (0.114) —0 149 (0.106)
D2*COV 0.060 (0.085) 0.050 (0.122) 0.101 (0.117)
D3 -0 104 (0.076) -0.131 (0.116) -0.105 (0.096)
D3*COV 0.065 (0.082) 0.139 (0.124) 0.008 (0.106)
D4 -0.128 (0.079) -0 330 (0.118) 0.034 (0.102)
04*cov 0.053 (0.087) 0.342 (0.129) —0 204 (0.115)
05 0.081 (0.096) 0.000 (0.141) 0.144 (0.126)
D5*cov —0.164 (0.106) -0.010 (0.155) —0.299 (0.140)
CONSTANT 7.469 (0.180) 6.870 (0.227) 9.822 (0.380)
N of Groups 1020 557 463
N of Obs 7597 4331 3266
Obs per Grp (avg) 7.4 7.8 7.1

F(18,6559)=56.34

F(18,3756)=48.91

F(18,2785)=10.67

 

64

Figure I-l: Proportion of Working (Non-Zero earnings) by Groups

 

    

1.2

1
D!
.E
x
g 0.8
‘5 0.6 l+ 00" ” i
S l+' uncov
E 0.4.
O.
2
D- 0.2

n4yr n3yr n2yr n1yr Oyr 1yr 2yr 3yr
years

 

 

 

cov: covered group

uncov: uncovered group

Data for 3yr only includes women with reference birth year of 1988, 1989, 1990, 1991, since
the 1994 data is the latest one. Data for 4yr only includes women with reference birth year of
1988, 1989, 1990.

Figure I-2: Average Work Hours per Week by Groups

 

41
40
39
38
37
36
35
34
33
32
31

COV

+
.... uncov

avg wk hours per week

 

n4yr n3yr n2yr niyr Oyr 1yr 2yr 3yr 4yr

yea rs

 

 

 

cov: covered group

uncov: uncovered group

Data for 3yr only includes women with reference birth year of 1988, 1989, 1990, 1991, since
the 1994 data is the latest one. Data for 4yr only includes women with reference birth year of
1988, 1989, 1990.

65

Figure I-3: Simple Wage Proﬁles of All Workers (CPI adjusted)

 

1400
1200
1000

800

cents

600
400

200

 

n4 n3 n2 n1 zero p1 p2 p3 p4

years

 

 

 

cov: covered group
uncov: uncovered group

Figure I-4: Simple Wage Proﬁles by Groups (CPI adjusted)

 

1400
1 200
1 000

800

cents

600
400

200

 

n4 n3 n2 n1 zero p1 p2 p3 p4

years

 

 

 

cov cl: college educated covered group

uncov cl: college educated uncovered group

cov hs: high school educated covered group
uncov hs: high school educated uncovered group

66

Figure I—5: Predicted Hourly Wage Proﬁles (All Workers)

 

 

 

 

 

’ + c6v
+ uncov
yr-6 yr-5 yr-4 yr-3 yr-2 yr-1 er yr1 yﬂ YI3 yr4 yr5

X axis: years, Y axis: log of hourly wages

cov: covered group, uncov: uncovered group

Figure I-6: Predicted Weekly Earnings Proﬁles (All Workers)
:co 1
l... uncov!

yr-6 yr—5 yr4 yr-3 yr-2 yr-1 er yr1 yr2 yrs yr4 yr5

 

 

 

X axis: years, Y axis: log of weekly earnings
cov: covered group, uncov: uncovered group

 

 

Figure I-7: Predicted Hourly Wage Proﬁles (College Educated Workers)

 

7?

j + COV
+ UHCOV

 

yr-6 yr-5 yr-4 yr-3 yr-2 yr-1 er yr1 yI’Z yr3 yr4 yr5

 

 

 

X axis: years, Y axis: log of hourly wages
cov: covered group, uncov: uncovered group

Figure I-8: Predicted Weekly Earnings Proﬁles
(College Educated Workers)

 

1

0.9

0.8

0.7

0.6 i.-. .1

0 5 + covTi
' + uncov

0.4
0.3
0.2
0.1 .
0

 

yr-8 yr-5 yr-4 yr-3 yr-2 yr-1 yIO yr1 yr2 yIG yr4 yr5

 

 

 

X axis: years, Y axis: log of weekly earnings
cov: covered group, uncov: uncovered group

Figure I-9: Predicted Hourly Wage Proﬁles
(High School Educated Workers)

 

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

0

igﬁ i
... uncov

 

yr-6 yr-5 yr-4 yr-3 yr-2 yr-1 er yr1 yr2 yrs yr4 yr5

 

 

X axis: years, Y axis: log of hourly wages
cov: covered group, uncov: uncovered group

Figure I-lO: Predicted Weekly Earnings Proﬁles
(High School Educated Workers)

 

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

0

+ GOV —l
+ uncovl

 

yr-6 yr-5 yr-4 yr-3 yr-2 yr-1 er yr1 yr2 yrs yr4 yr5

 

 

 

X axis: years, Y axis: log of weekly earnings
cov: covered group, uncov: uncovered group

69

 

CHAPTER 2

EFFECTS OF THE COLLEGE EDUCATION EXPANSION POLICY ON
THE RELATIVE WAGES OF COLLEGE GRADUATES IN KOREA

I. Introduction

Mass college education beginning in the 1980s in Korea has contributed to
sustaining economic growth, comparable to the effect of universal primary education
that had provided the basis to initiate industrialization in the 19603.32 The college
education expansion policy in 1981 opened the epoch of mass college education by
abruptly increasing the college education entrance quota. This policy was thought to
cause an eruption of the supply of young college graduates in the labor market after
the mid-19805. Therefore, a major development in the wage structure in that period,
that is, a sharp decline of relative wages of college graduates, was thought to result
from the college education expansion policy. Although the Korean government
already changed the rule that restricted the quota in the late 1970s, increase in the
quota in 1981 was sudden and as large as 50 percent.33

The Korean experience seems to provide fertile ground to study how a

college education expansion policy of the government affects the labor market. But

 

32 World Bank (1998/99) recognized Korea as one of the countries that successfully linked education
to consistent economic growth. It is thought to help to upgrade the average skill level of Korean
workers in the international division of labor, thereby helping to raise their average wages in the
world economy in the long run.

33 58.6 percent increase in the quota of four-year college. 48.7 percent increase in the quota including
two-year college. The Korean government has regulated the number of college entrants by setting the

70

only a few studies have been devoted to that topic. Previous studies documented a
severe decline in the relative wages of college graduates and an increase in within-
group inequality aﬁer the sudden increase in college entrance quota in 1981 (Ihm
1990, Kim and Topel 1995, Choi 1996). These studies attempted regression analysis
of the relationship between the relative employment of college graduates and the
relative wages of them. However, the data on employed college graduates would be a
poor measure of the supply of college graduates if the unemployment rate for them
is unusually high as in Korea after the mid-19803. Moreover, the data on the supply
of college graduates also would not serve for a good independent variable in
estimating the effect of government educational policy on the relative wages Of
college graduates, because it reﬂects the demographic and overall socioeconomic
changes as well as the effect of the government educational policy.

This paper explores the college education expansion policy on wages and
employment during the 19803 and the 19903, including a sudden change in the
college entrance quota in 1981. First, this paper explores the changes in the returns to
the college education before and after a sudden increase in the quota in 1981 by
using the difference-in—difference (DD) method. A DD estimate shows that the
relative wages for all workers as well as young workers quickly dropped
immediately after large inﬂux of college graduates, that is, between 1982-83 and
1987-88.

Second, considering that the DD strategy does not allow one to analyze the

effect of the policy throughout the 19803 and 19903, this paper uses time series

 

college entrance quota every year, based on population increase and the forecast for manpower
demand during the 19703.

71

analysis to examine the policy of regulating the college entrance quota. The reduced
form equations for the relative supply and the relative wages as well as the demand
curve are estimated. The reduced form equations include the policy variable such as
the quota or the resulting target variable such as the enrollment rate as an
explanatory variable. The estimation results indicate that the relative wages and
employment of all workers including both young and old responds to the policy with
a time lag.

Third, this paper estimates the enrollment equation that includes the current
relative wages as an explanatory variable. The current relative wages is positively
and signiﬁcantly associated with the enrollment rate, aﬁer controlling for trend and
ﬁrst order autocorrelation. This means that the market incentives were not totally
paralyzed in the private decisions to enroll in college during that period.

This paper is organized as follows. Section 11 reviews previous studies.
Section 1H describes the data used in this paper. Section IV overviews the
background of the college education expansion policy and the changes in wage
structure during the 19803 and 19903. Section V explains the empirical strategy for
estimating the impact of the college education policy on the relative wages of college
graduates and reports the estimation results. Conclusion and policy implications

follow in Section VI.

72

11. Critical Review of Previous Studies

About the effect of the college education expansion policy on the relative
wages, a handful of related studies exist. Among them, Ihm(1990), Kim and
Topel(1995), Choi(1996), Kim and Yoo(1994) are worthy of discussion.

Figure II-l depicts the conceptual framework that helps us to understand the
transmitting process of the college education expansion policy on relative wages.
The college entrance quota is the policy variable set by the government and the
college enrolhnent rate is the target variable. Conceptually, abrupt increase in the
quota can be viewed as a supply shock. Under the chronic excess demand for higher
education, increase in the quota is expected to be directly reﬂected in the enrollment.
Therefore, the process (A) might be automatic and this paper conﬁrms that. And the
increase in the quota results in an increase in college enrollment, leading to increase
in the supply of college graduates and their relative employment, exerting downward
pressure on their relative wages. In this process, there exists a time lag between the
increase in college enrollment and the increase in the supply of college graduate (B).
And the demand works in the process (C). Only some of the college graduates who

enter in the labor market will be hired.

73

Figure II-l: The Effect of the College Education Expansion Policy on the Wages of

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

College Graduates
E
V
Entrance College Graduates Graduates Relative
Quota f A—D Enroll fB—f Supply f C—'> Employed T D—> Wage j
E 1

If the technical identity can be obtained in the relation (A), that is, between
the quota and the number of enrolled college students, as was evidenced by Kim and
Yoo(1992), we can use the college enrollment rate as well as the college entrance
quota for a policy variable that reﬂects the college education expansion policy.34

Most of the previous studies on the effect Of the policy concentrated only on
the relation (D) or the relation (C) and (D), that is, relations between relative supply
or relative employment and their relative wages (Ihm 1990 and Choi 1996).
Ihm(1990) found that supply of college graduates exceeded the industrial demand,
even within the context Of a rapidly developing economy, using the 1986
Occupational Wage Survey. He also found a strong sign of occupational ﬁltering and
educational upgrading in the 19803. While the earnings premium associated with
college education has continued to grow in absolute amounts, the rate of growth

declined during the period.

 

3‘ Indeed, the college entrance quota set by the government has been almost exactly translated into
the college enrollment in Korea during 19803 and 19903 when the excess demand for college
education existed. In Figure II-l, relation (B) between the number of enrolled college students and the
college graduates that constitute the labor supply in the college labor market can also be assumed

74

Kim and Topel(1995) also identiﬁed the changes in relative supply as a main
determinant for explaining the changes in the wage structure after the mid-19803.
They analyzed the overall labor market performance during the 19703-19903, the
period of rapid growth and structural changes. The upshot of their argument is that
there was little evidence of the demand shifts in affecting relative wages.35

Choi(1996) also suggested the supply shock. The rapid increase in the
number of college graduates among the total economically active population since
1985 has resulted from the college education expansion policy in 1981. He referred
this supply shift of college graduates as exogeneous. (Choi 1996). From investigating
the experience-eamings proﬁle, he found that the large increase in the supply of
young college graduates during the 19803 reduced their relative wages. However, he
did not refer the sudden increase in relative supply of young college graduates as a
prime cause of declining relative wages of college graduates. Instead, he emphasized
that changes in the relative wages have been small, despite the substantial increase in
the number of college graduates. And he thought that factor biased technological

changes favored more educated workers can be behind this.

 

technically identical, especially in the situation where the college drop-out ratio were negligible, like
Korea in the 19803-19903.

35 They argued that wages converge over observable characteristics such as schooling and age, but
relative demand shifts played a trivial role. They were surprised at one of their ﬁndings that industrial
policies, with which the government purposely pushed toward manufacturing and exports, did not
raise the relative demand for less skilled workers. However, the apparent small changes in the relative
demand for less skilled workers in this period could be attributed to the defectiveness of the demand
shift measure. According to Katz and Murphy (1992), the demand shift index could understate the
"true" between-sector growth for college graduates relative to high school graduates. Indeed, it is
widely acknowledged in the studies on the wage structure in the 19803 in the United States that the
shift in the industrial composition of labor demand has little effect on the relative wage (Katz and
Murphy 1992, Juhn, Murphy, and Pierce 1993, Johnson and Bound 1992). Choi (1996) also indicated
that effect of labor demand changes brought by product demand changes across industries on the
relative wages was negligible.

75

However, these studies overlook one important point. The unemployment
rate of college graduate labor market entrants was extremely high in Korea in the
19803 and 19903, and the supply of college graduates is the sum Of employed and
unemployed college graduates. As their analysis is limited to the relation (D), they
cannot be viewed as studies on the effect of the college education expansion policy
on the relative wages of college graduates. Also, there is a problem of the studies
such as Kim and Topel (1995). Those studies depend on their methods of singling
out the supply effects on the relative wages, since only changes in employment
which combine the supply effects and the demand effects are observed in the labor
market.

Unlike previous studies, the time series analysis of this paper uses the
reduced form equation for relative wages. In that equation, the supply shock is
incorporated into the quota or the enrollment rate as an explanatory variable. This
reduced form equation estimation also informs how long it took to transmit the
college education expansion policy to changes in relative wages.

Moreover, this paper attempts to discuss one of the hot issues in that period,
that is, whether the economic incentives worked in private decisions to enroll in
college. There may be a feedback effect of resulting changes in the relative wages of
college graduates on the demand for college education revealed in the college
enrollment rate.

Kim and Yoo(1994) studied a similar issue. They tried to ﬁnd determinants
of demand for college education which was reﬂected in the number of college

applicants. They examined whether abrupt increase in college entrance quota brought

76

about the rapid increase in demand for college education. They used the
Occupational Wage Survey and the Annual Report of Educational Statistics for
1962-92 period. The results indicated that the college entrance quota as well as
relative wages of college graduates and several other determinants had signiﬁcantly
affected the demand for college education.36

However, from the analysis of the enrollment equation, this paper ﬁnds that
the enrollment rate for college education did respond to change of relative wages,
after controlling for time trend and the ﬁrst order serial correlation. This result is
contrary to previous studies which attributed the puzzling phenomenon in the 19803,
that is, ever-increasing enrollment rate despite continuously decreasing relative
wages of college graduates, to the so-called “excessive fever for higher education”,

originating ﬁ'om Korean culture (Kim and Kim 1992),

 

36 It is controversial whether a decrease in college-high school wage differentials would lead to a
decrease in the demand for college education. According to Kim and Yoo(1992), the abrupt increase
in the college entrance quota during the early 19803 has led to enormous increase in the number of
college applicants far beyond the increase in the quota. On the other hand, the number of college
applicants has decreased signiﬁcantly as entrance quota was ﬁxed during 1982-86. When the
government increased the college entrance quota once more after the late 19803, the number of
college applicants also increased more rapidly than the college entrance quota.

77

III. Data

The main data of this paper mainly is the Occupational Wage Survey (OWS)
that is reported armually by the Ministry of Labor of Korea. Another important data
source is the Annual Report of Education Statistics published by the Ministry of
Education of Korea. The original OWS data set includes various kinds of wages,
employment, and demographic information for about 500,000 workers per year.
Although the number of observations is large, the number of variables is quite small
compared to the Current Population Survey (CPS) in the United States. The OWS
data also differs from CPS data, in that the sampling units are ﬁrms rather than
households. The survey includes only ﬁrms with at least ten employees, therefore the
manufacturing sector is over-represented and retail trade industry is under-
represented in the sample. As a result, it omits roughly one-third of the non-
agricultural labor force in a typical year (Kim and Topel 1995). Another major
drawback is that the public sector is excluded from the survey. Therefore, OWS data
does not provide accurate information about actual wage structure by industries.

This paper uses a random ten percent sample from the original OWS data.
The number of observations is roughly 25,000 to 50,000 for each year. This paper
also uses yearly aggregated data that is calculated from the original OWS data and
published in the printed version of OWS. Wages include regular payments and
overtime payments. I summed up all types of monthly payments and also included

one-twelfth of the special payment that is usually paid once a year at the end of the

78

year. I calculated the hourly wages for full-time workers who work more than 180
hours and less than 300 hours per month and then took a natural logarithm of the
hourly wage rates.

The education level is classiﬁed into ﬁve groups. 1 = elementary school
graduates, 2 = middle school graduates, 3 = high school graduates, 4 = junior college
graduates, and 5 = college graduates. I give actual years for each category.
Elementary school graduates including elementary and middle school dropouts are
given six years. Middle school graduates including high school dropouts are given
nine years. High school graduates including junior college dropouts are given twelve
years. Junior college graduates including vocational school graduates and college
dropouts are given fourteen years. College graduates including more than sixteen
years of schooling are given sixteen years.

Table II-l is a summary of statistics for workers aged 19-59 from the ten
percent sample of the OWS data in each period. Table II-2 is an equivalent of Table
II-l for young workers aged 19-29. Monthly wages are adjusted by the consumer
price index (CPI). Although the policy was implemented in 1980, the effects began
to be observed after the mid-19803 when the ﬁrst cohort following the policy
graduated and entered into the labor market. I chose 1982-83 for the before-the-
effect period, and 1987-88 for the aﬁer-the-effect period. Many male college
graduates who entered college in 1981 may have entered the labor market around
1987, because most of them had to complete about three years of military service. A
difference in average age between young male college graduate workers and their

female counterparts in Table II-l mainly reﬂects the gap due to military service. A

79

sudden large inﬂux of female college graduates began to enter the labor market after
1985. However, their number was still far smaller than their male counterparts as we
have seen in Table II-2. A short time interval of around ﬁve years allows us to avoid
considering the signiﬁcant structural changes of the economy. Workers in the 1987-
88 samples can be regarded as part Of the ﬁrst generation that received the mass

college education. Data for 1995—1996 are added for updated information.

80

IV. Overview of Changes in the Wage Structure

The Background of the College Education Expansion Policy

The epoch of mass college education began with the unprecedented
expansion of the college entrance quota by the Korean government in 1981. In Korea,
the government has regulated the college entrance quota, deciding the entrance quota
level each year. Especially before 1981, the increase in the entrance quota was
restricted approximately according to the increase in population (Choi 1996). With
the college education expansion policy announced in 1980, the Korean government
dramatically enlarged the college entrance quota of 1981 by 48.7 percent.37 The
enrollment rates for four-year-college including teachers’ colleges increased about 5
percent points annually in the 1980-1984 period. 38

The Korean government proclaimed that the Objective of the expansion of
college entrance quota was to fulﬁll the increasing demand for high-skilled
manpower. The previous mode of economic growth based on low-wage and low-
skilled labor had faced its limits during the late 19703, and skill-based industries
were needed for ﬁirther growth. Also, the “Baby Boom Generation” who arrived at
the gates of higher education in the late 19703 suffered from limited admission to

39

colleges. However, political consideration cannot be ignored in this sudden

 

37 48.7 percent increase in the quota including two-year college. 58.6 percent increase in the quota of
four-year college.

38 The enrollment rate of all higher education including colleges, universities, teachers colleges,
junior colleges and miscellaneous schools, also increased about 5 percentage points in this period.

39 The “Baby Boom Generation” usually has a strong demand for higher education, as they received
the universal elementary school education unlike the previous generation.

81

expansion of college entrance quota in 1980. Indeed, the Korean government initially
announced that they would increase only the college entrance quota, while keeping
the number of graduates at the same level as before the increase of quota. But the
government soon abandoned the initial policy of restricting the number of graduates,
as it became clear that government could not technically control college graduation.
Thereafter, the increase of college entrance quota by the government ahnost directly

led to the increase of college graduates whatever the initial intent of policy-makers.

Declining Relative Wages of College Graduates: Changes in the Wage Structure

during 1980-1995

The economic effect of the abrupt expansion of the college entrance quota
began to be felt after the mid-19803 when the unprecedented large ntunber of college
graduates ﬂooded the labor market. Since the Korean economy was in the middle of
the recession in the early 19803 after the fast growth of two decades, the economy
could not absorb a sudden inﬂow of many college graduates. The unemployment rate
of college graduates soared. The sharp increase in the unemployment rate of college
graduates in the 19803 coined the word “the educated unemployed” which referred to
the unemployed college graduates. For the ﬁrst time since the rapid industrialization
from the 19603, the unemployment rates of college graduates in Korea started to

exceed that of high school graduates from 1985.40 During the mid-19803, some Of the

 

4° The employment rates of all college graduates who had just entered the labor market dropped from
73.0 percent in 1980 to 52.1 percent in 1985. The employment rate of female college graduates
dropped more severely: it was 55.2 percent in 1980 and 31.7 percent in 1985, while that of male
college graduates was 83.4 percent in 1980 and 67.6 percent in 1985. However, the low employment

82

college graduates who could not ﬁnd managerial or professional jobs rushed into
technical or clerical jobs that had been previously regarded as jobs for high school
graduates.41 This phenomenon was referred to as “overeducation” or “downward
employment”. However, from a dynamic point of view, the growth of the education
level of the labor force might be seen as “occupational upgrading”.

The inﬂux of unprecedented large numbers of college graduates also resulted
in a substantial decline in their relative wages, as is seen in Table H-3 and Figure 11-
2. The labor market conditions for college graduates were directly reﬂected in the
monetary returns of college graduates. As coefﬁcients for a dummy variable for
college graduation vs. non-college graduation (column (1), (2) and (3) in Table II-3)
show, the relative returns to college education experienced a severe downward trend
for young workers aged 19-29 and all workers aged 19-59, respectively. Young
college graduates earned 61.3 percent more compared to young non-college graduate
workers in 1982-83, 49.1 percent more in 1987-88, and 30.5 percent more in 1995-

1996. 42 Young male college graduates earned 67.8 percent more than young male

 

rate of female college graduates did not persist for a decade. Aﬁer the 19903 the employment rate of
female college graduates restored ahnost to the level of the 19703. In 1995, the employment rate of
female new entrants was 50.0% and little lower than 55.2 % in 1980. The employment rate of male
new entrants was 69.2 % in 1995, much lower than 83.4 % in 1980. The remarkable recovery of the
employment of female workers was largely caused by the cyclical boom in the late 19803 which was
triggered by “three low prices”, low oil price, low international interest rate, and low Won (especially
compared to high Yen). In addition, the peculiar recruitment practice that hampers the accumulation
of unemployed female college graduates made the recovery of the employment rate of female college
graduates relatively easy, compared to their male counterparts.

' As Ihm (1990) pointed out, increasing proportions of college graduates have taken occupations of
lower status and lesser prestige. He argued that same level of schooling is not rewarded equally in all
occupations. For example, higher level education is best rewarded in higher level education-speciﬁc
occupations, particularly those in the professional/technical category. This implies that downward
employment has been one of the causes of the decrease of relative wage of college graduates.

‘2 Since the schooling data is right censored, the estimate of the returns to college education could be
an overestimate. The number of people who earned M.A. or PhD. degree has increased enormously
recently. The number of people who newly earned M.A. degrees was 1,978 in 1970, 5,028 in 1980,
and 25,787 in 1994. The number of people with newly earned Ph.D. degrees was 172 in 1970, 524 in
1980, and 3,818 in 1994. (Ministry of Education, Annual Report of Education.) However, there is no

83

non-college graduate workers in 1982-83 and 54.4 percent more in 1987-88. Young
female college graduate workers earned 63.5 percent more in 1982-83 and 48.6
percent more in 1987-88.

Figure II-2 also shows the decline in the relative wages of college vs. high
school graduates in the actual and detrended relative wage series. In Figure II-2,
relative wage of college vs. high school graduates among all workers stayed at a
stable level until 1987, dropped abruptly in 1988, and then declined consistently until
1995. In the case of younger workers, relative wages were consistently declining
throughout the 19803 and the early 19903. This was particularly severe during the
late 19803. In both cases of younger and all workers, there exists a strong association
between the abrupt increase in the college entrance quota in 1981 and declining
relative wages after the mid-19803.

These ﬁgures show two things. First, the outstanding feature in the changes
of the wage structure during the 19803 was the decline in the relative wages of
college graduates. Indeed, the relative wages of college vs. less than high school
graduates consistently declined, though the relative wages of high school vs. less
than high school graduates was fairly constant during that period. Second, the
detrended relative wages moved inversely with the detrended relative supply
throughout the 19703 and 19903. Figure II-4 and Figure II-5 show the actual and
detrended relative supply of college vs. high school graudates, respectively. This
implies that change in relative supply of college graduates was an important factor in

explaining the change in relative wages in that period.

 

comprehensive data that classiﬁes more than 16 years of education and their wages. In the OWS data,
workers whose schooling years are equal to or over 16 years are put together into the category of

84

The coexistence Of falling relative earnings and declining employment rates
shows that focusing only on changes in earnings would understate the declining
position of college graduates (Eckhaus 1973, Chiswick and Mincer 1972, and Blaug,
Layard and Woodhall 1969). 43 In fact, substantial decline of relative wages of
college graduates and sharp increase in unemployment rates lead to concern about
“overeducation”.

However, the detrended relative wage movement in Figure II-3 shows that
relative wage of college graduates had consistently increased for more than a decade
before the policy effect revealed in the mid-19803. In this perspective the decrease of
relative wages during the late 19803 and the early 19903 should be understood in the
context of the long-term adjustment of relative wages in the labor market. Despite
the above qualiﬁcations, it is beyond doubt that the decrease in relative wages of
college graduates since the mid-19803 was the result of sudden increase of college

entrance quota by the government in 1981.

 

college education.

‘3 Especially in developing countries a crude estimate of returns to schooling can overestimate the
proﬁtability of investment in education. For example, Blaug, Layard and Woodhall (1969) showed
that chronic unemployment resulted in lower returns to higher education in India. About the
possibility of underestimation of the returns to schooling owing to the existence of hidden
unemployed, see Filer, Hamerrnesh and Rees (1996). They also indicated that the ofﬁcial
unemployment rate does not include discouraged workers who would have searched jobs if they had
not thought of the searching efforts as hopeless.

85

V. The Effect of the College Education Expansion Policy on Relative Wages of

College Graduates : Empirical Analysis

This subsection is about the empirical analysis of the effect Of the college
education expansion policy on relative wages of college graduates. First, by using a
DD model, this subsection estimates the effect of the college education expansion
policy in 1981 which abruptly and enormously raised the college entrance quota by
approximately 60 percent. The DD estimation method is widely used for assessing
the policy effects by comparing them before and after the introduction of the policy.
To avoid the fundamental structural and demographic changes during the
comparison period, a relatively short time span (ﬁve years) is considered. The
before-the—policy period is 1982-83 and the aﬁer—the-policy period is 1987-88 in this
subsection.

Second, by using a time-series regression and directly introducing a policy
variable or a policy target variable, this subsection estimates the effect of the college
education expansion policy through the 19803 and 19903.

Third, by examining the determinants of college enrollment, this subsection
explores a much-discussed issue about the college education expansion policy:
whether the market mechanism worked in affecting the private decision to enroll in

college during that period.

86

DD Estimation for the Effect of the Policy in 1981

This subsection estimates the impact of the college education expansion
policy, more speciﬁcally, the abrupt increase in the college entrance quota in 1981,

on the returns to the college education by using the following DD estimation method.

1n‘Vit= Bo + Bl COLLi + 132 AFTER: + 53 C0L14*AFTER: 1“ X115 + 8n (1)

COLL) is a dummy variable for four-year college education vs. non-college
education. AFTERt is a dummy variable for the after-the-policy period (1987-88) vs.
the before-the policy period (1982-83). Changes in economic conditions such as
changes in aggregate demand or relative demand shift between college graduates and
high school graduates are incorporated into the variable AF TERt. COLL1-*AFTER( is
an interaction between COLLi and AFTERt. It represents the policy effect that
reﬂects changes in the relative returns to college education after the policy. A vector
of all other control variables X" include tenure years, industries, marital status, and
gender.

The estimation results are shown in column (3) in Table II-3. The DD
estimate, that is, the coefﬁcient on the interaction term COLL(*AFTER. was -0.150.
This indicates that the returns to four-year college education declined approximately
15 percentage points aﬁer the effects of the policy Operated. The returns to the

college education among young workers age of 19-29 declined 19.8 percent point.

87

This implies that the impact of the educational expansion policy on the returns to
college education was larger among young workers than among all workers.

Another conspicuous result is that the returns to college education of female
college graduates dropped more than that of their male counterparts. The DD
estimate for male workers was -0.093, while it was -0.253 for female workers in
Table II-3. Among young workers, it was -0.138 and -O.266, respectively. DD
estimates were little different between young female college graduates and total
female college graduates, since most female workers in the work force were young.

Table II-4 reports that returns to the four-year college education dropped
more than that of two-year college education in that period. 45 The returns to the
four-year college education dropped 15.1 percentage points, while those to the two-
year college education dropped 12.6 percentage points. Table II—4 shows that this
phenomenon is more conspicuous among young workers and among female workers.
Especially among young female workers, the returns to the four-year college
education dropped 29.0 percent point, while those to the two-year college education
dropped 23.7 percent point.

Figure II-2 also shows a relatively large policy effect on female workers.
During 1987-88, actual relative wages of college graduate female workers dropped

more severely than those of their male counterparts.

 

45

111w“: Bo + B) COLLi + 52 1C1 + Br AFTER + B4 COLLi’AFTERr + 135 JC,*AFTER.+ X163 + Bit (1’)
Where all variables except JC or JC‘AFTER are the same as the model (1). JC, is a dummy variable
for two-year college graduation. JCR'AFTER. is an interaction between JC, and AFT ER.. It represents

the policy effect that reﬂects changes in the relative returns to two-year college education after the
policy along with COLLi‘AFTEK.

88

In sum, the impact of the abrupt expansion of the college entrance quota in
1981 was larger among young workers than all workers, and larger among female

workers than male workers.46 And, it was larger for four-year college graduates than

two-year college graduates.

Time Series Regressions

The above DD estimation examined the effect of the college education policy
on wages by comparing wages before and after the abrubt increase in the quota in
1981. This subsection analyzes the effect of the college education expansion policy
through 1976-1998 by directly introducing policy variables.

As we have seen in Figure II-l, the college education policy affects wages
through a process composed of a couple of links. This subsection analyzes the whole
process in a demand-supply framework, focusing on whether the policy variable
directly affects the supply of college graduates and whether the change in relative
wages can be understood as a consequence of the supply shock owing to the policy
change.

The ﬁrst hypothesis is that the policy directly and explicitly affects the

relative supply of college graduates. The second hypothesis is that the effect of the

 

4" However, aﬁer one and a half decade of an exogenous increases in college entrance quota, the DD
estimates for the 1982-83 vs. 1995-96 period indicates that the returns to college education dropped
40.5 percent points. Among male workers in age of 19-59, it dropped 31.1 percent point and, among
female workers in those ages, it dropped 53.5 percent point. However, among young workers, a
gender difference was not conspicuous. Irrespective of gender, returns to college education for young
workers dropped about 40 percent during that period. This means that, although other factors should
also be considered, generally speaking, the college education expansion policy affects female workers
more in the short period, but those effects is gender neutral after one and half decades.

89

policy variable can be identiﬁed directly and explicitly in the changes in the relative
employment and the relative wages. The college entrance quota and the college
enrollment rate are referred as the policy variable and the policy target variable,
respectively.

In order to test the ﬁrst hypothesis, I examine whether the college entrance
quota directly affects the college enrollment rate and whether increase in the college
enrollment rate affects the relative supply of college graduates. In this case, the
relative employment is examined instead of the relative supply.47 Analysis of the
relationship between the entrance quota and the enrollment would be the ﬁrst step for
studying the effect of the college education expansion policy, since the immediate
target of the entrance quota policy is the control of college enrollment. 48 The
following model captures the effect of the college entrance quota on the enrollment

rate.

10g (ENROLL). = 8. + (3. log QUOTA. + 82 TREND. + 3., (2)

In the model (2), QUOTA is the college entrance quota divided by the

number of high school graduates for every year49 ; ENROLL is the enrollment rate,

the ratio of the number of college entrants to the number of high school graduates for

 

‘7 Although changes in relative employment include changes in relative demand as well as changes in
relative supply, it is difﬁcult to identify the changes in relative demand (Kim and Topel 1994 and
Katz and Murphy 1992).

‘8 The entrance quota for each college is still the upper limit over which the college is not allowed to
admit new entrants. However, some local private colleges have had difﬁculties in ﬁnding enough new
entrants to ﬁll their enhance quota in the 1990s.

‘9 College includes two-year junior college.

90

every year.49 The number of enrolled college students cannot exceed the number of
college entrance quota, and the college enrollment has always ﬁlled up the college
entrance quota under the chronic excess demand for college education. Choosing the
college enrollment rate as a policy target variable has a merit for my study, as it is
reported by gender. Meanwhile, the college entrance quota has been allotted to each
college in lump suum without gender distiction. The model includes a time trend
(TREND) in order to eliminate the problem of spurious regression. This can happen
when we do not detrend the time-series processes, or do not include a time trend
variable in the regression. In many cases, two trending variables appear to be
correlated simply because each variable is growing over time.

Table II-5 shows that, with the trend variable, the coefﬁcient for the log of
QUOTA is 0.812 and highly signiﬁcant. However, there remains another problem.
Durbin-Watson (1 statistic is 0.483 and this strongly suggests the problem of positive
autocorrelation. To correct the serial correlation, Corchrane-Orcutt AR(1) regression
method is used. The Corchrane-Orcutt iterated estimate is 0.736 and also highly
signiﬁcant. After Corchrane-Orcutt transformation, Durbin-Watson d statistic is
1.372.

As the relationship between the college entrance quota and the enrollment
rate is established in the above discussion, we can proceed to examine the
relationship between the enrollment rate and the relative employment in the labor
market. Since the object of this subsection is to study the effect of the enrollment rate
as a supply shock on the relative employment and the relative wages, it estimates the

reduced form equations. From the structural equations which consist of the relative

 

‘9 College includes junior college.

91

demand for college graduates and relative supply of them, two reduced form
equations are derived: the relative employment equation and the relative wage
equation.“ In this speciﬁcation, the college education expansion policy affects the
dependent variables Of the reduced form equations via the lagged enrollment rate.
That is, the lagged enrollment rate which is included in the relative supply equation
as an exogenous supply shock affects the relative employment and the relative wages.
The following model is the reduced form relative employment equation. This model
is estimated by the Cochran-Orcutt procedure to correct the ﬁrst order

autocorrelation of the error term.

10g (ch/ Sns)( = Bo + B] 10g ENROLL (-7 + [32 TREND: +Zﬂ+ 8t , (3)

The dependent variable is the log of relative supply of college graduates (log
(SCG/ Sus)). Although college includes two-year junior college, the time lag to
transmit the exogenous supply shock is based on the time it takes for four-year
college graduates to enter the labor market. Approximately two thirds of college
students graduated four-year college. A seven-year lag to ENROLL is chosen, for it
takes about seven years for changes in the college enrollment rate to affect the
relative supply of college graduates (four years of college attendance plus about
three years of military service for male college graduates). For female workers’ case,

a four-year lag for ENROLL is chosen. Z. includes the log of DINDEX and the log

 

5 ' The structural equations are:
108 (xCG/XHS)d r = a0 + a1103(‘NCGANHS)t + aZTRENDt + 33108 GDP.+ a4 DINDEX: 7“ U:
108 (Xco/Xusyt = b0 4” b1108(wcc/WH3):+ szREND: + b3108 ENROLLr—7 + 31

92

of Real GDP. DINDEX is Freeman’s ﬁxed-coefﬁcient “manpower requirements”
index of the relative demand for college graduates.52 GDP controls for changes in
aggregate demand.

Table II-6 reports the results for the relative employment. The coefﬁcient for
the seven-year lag (L7.ENROLL, column (1)) among all workers is 0.249 and highly
signiﬁcant. The coefﬁcient for the seven-year lag among male workers is 0.216 and
highly signiﬁcant, while that for four-year lag among female workers are 0.200 but
not signiﬁcant. For sensitivity analysis, Table II-8 reports the coefﬁcient for the
seven-year lag, six-year lag, ﬁve-year lag, and four-year lag enrollment rates. In all
workers’ case and male workers’ case, the coefﬁcients for the six-year lagged
enrollment rate are the largest and signiﬁcant. This indicates that the effect of the
college education expansion policy on the relative employment in the labor market
reveals itself after six years. In women workers’ case, all the coefﬁcients for the
lagged enrollment variables are positive. The effect of the policy can be detected
after four years. However, the effect is largest after six years, like male workers, and
it is puzzling why transmitting the effects does not occur more quickly for female

workers, who do not have a military service requirement.

 

52 The index is calculated as 23,- (1le j./ 23,- or sz 1., according to Freeman (1975), where 01,,- is the
proportion of workers with education level i in industry j in the base year (1978). i = 1 deﬁnes the
college group and i = 2 deﬁnes the high school group. E j. is actual employment in the industry j in
year t. t is from 1978 to 1996. I used the l—digit industry code for calculation. Since the industry code
is revised in 1993, I converted the old code to the new code from 1978 to 1992. There are 17
industries in one-digit industry code and OWS only includes 14 industries. This measure is widely
used for estimating the effect of between-sector demand shifts on relative labor demands. However, as
Katz and Murphy (1992) criticized, it may understate the relative demand shift favoring groups with
increases in relative prices.

93

The effect of changes in college enrollment rate on the relative wages of

college graduates is also identiﬁed in the following reduced form equation for the

relative wages.

108 (woo/Wm): = Bo + Bl log ENROLL t-7 + 52 TREND: +ZtY+ 8: , (4)

The dependent variable is the log of relative wages of college graduates (log
(WCG/WHS) .. A seven-year lag to ENROLL is chosen, for the same reason of the
above relative employment model. Table II-7 reports a strongly negative association
between the lagged college enrollment rate and the relative wages among all workers
and among male workers. The elasticity of relative wages with respect the the
enrolhnent rate is 0.648 for all workers and 0.416 for male workers.53 Among female
workers, the coefﬁcient for the four year-lagged college enrollment rate is positive
and insigniﬁcant, while the coefﬁcient for the lagged enrollment rate is negative after
ﬁve years as is seen Table II-8. In both male and female workers case, a seven-year
lag yields the largest effect, and this is one year longer than what was found for
relative employment.

Therefore, the estimation results of both equations, (3) and (4), suggests that
the second hypothesis that the effect of the policy variable can be identiﬁed in the
changes in the relative employment and the relative wages is plausible. Also, it

suggests that transmitting the effect of the increase in the enrollment to the relative

 

53 Since the dependent variable and independent variable are in the log, the coefﬁcient can be
interpreted as the elasticity of dependent variable with respect to the independent variable.

94

supply takes at least six years and that to the relative wages takes one more year,
seven years.

Although this paper focuses on the reduced form that contains the policy
target variable, I also estimate the usual demand schedule in order to compare my
results with previous studies. 54 I do this by substituting the current relative labor
supply variable (log (SCG/ SHS) .) for the lagged enrollment rate (log ENROLL .-7 ).
Table II-9 reports the results. It shows that the coefﬁcient for the relative labor
supply variable is negative but not signiﬁcant. It indicates that the relative wages are
not highly associated with relative supply in this period.

However, in the labor demand schedule, the relative quantity variable that is
used as a proxy variable for the relative supply variable also incorporates the
resulting demand changes. Therefore, previous studies have tried hard to isolate the
supply effect from the demand effect. Kim and Topel(1995) argued that changes in
relative wages of college graduates in the 19803 in Korea were mainly supply-driven.
They got that result by examining the relationship between the grth in relative

supply and changes in relative wages of Korea in 1971-86.55 In their study, an

 

5’ I also conﬁrm that change in relative supply was a dominant factor in explaining the change in
relative wages by using the following time-series regression that is similar with Blackburn, Bloom
and Freeman (1990).

log (Woo /WH5 ). = b0 + b. log (XCG/XHS). + szREND. + b; log GDP.+ b. DINDEX. + e.

In the above model, (Wm/Was). is the log of the relative wages of college vs. high school
graduates, (XCG/XHS). is the log of the relative supply of college vs. high school graduates. TREND. is
a measure of shifts in college vs. high school graduate relative labor demand. GDP. is the log of real
GDP. Like other studies, the relative employment in the labor market is used for relative supply.

55 Katz and Murphy(l992) also estimated the model that relates relative wages with relative labor
supply during 1963-87 in the United States. They got much larger coefﬁcient for relative supply of
college graduates than Blackburn, Bloom and Freeman(1990) got for the same period. The demand
shift index may understate the “true” growth of demand for college graduates relative to high school
graduates in the 19803 and overstate the shifts in the 19703. The overall demand shiﬁ measure is the

95

increase in relative supply of both high school graduates vs. less than high school
graduates and college graduates vs. less than high school graduates reduced their
relative wages respectively.56 It implies that the rapid upgrading of the education
level of Korean workers might be a crucial factor for changes in the wage structure.
In the United States, there are several studies that have attempted to account for
changes in relative wages of college graduates mainly by changes in relative supply
of college graduates in the 19803 (Blackburn, Bloom and Freeman 1990, Topel
1994).57 Katz and Murphy(1992) also found that changes in the returns to college
education have been driven more by ﬂuctuations in the growth of supply of college
graduates, while admitting the importance of relative factor demand growth within-
industry for explaining the changes in wage structures8

In sum, this paper conﬁrms the ﬁnding of previous studies by using the

reduced form equation that contains a policy target variable such as the enrolhnent

 

combined measure of deceleration in between-industry demand shift and acceleration in within-
industry demand shift from the 19703 to the 19803.

56 In addition, Kim and Topel(1994) argued that ahnost all the variation in the relative net supply of
workers could be explained by changing relative employment shares. According to them, differences
in industry growth rates had no conspicuous effect on relative demands for college graduates in
Korea.

57 In a study on declining position of less skilled men, Blackburn, Bloom and Freeman(1990)
concluded that the increase of wage differential between college graduates and high school dropouts
among men aged 25-34 was largely supply-driven between the 19703 (1973-79) and the 19803 (1979-
87). Topel(1994) also analyzed the impact of changing input quantities (supply) in comparison with
that of an employment-share weighted average of industry scales (demand). They concluded that
virtually all of the variation in net supply is due to variation in input quantities. Virtually none is due
to changes in the industry composition of demand.

58 In fact, it is not easy to estimate the effect of the demand shifts. Katz and Murphy(1992), using
their “demand shift index”, argued that rapid secular growth in the relative demand for “more-skilled”
workers is a key component of any consistent explanation for rising inequality and changes in the
wage structure over the last 25 years (1963-1987). However, they also found that the measured
demand shifts towards more-educated workers and towards women have been signiﬁcantly smaller
than the observed relative supply changes. Also, in a time-series regression of the log of relative
earnings on the log of relative labor supply and a time trend (the relative demand shift), the coefﬁcient
on the relative labor supply is bigger than that of the relative demand shifts.

96

rate. The education policy via the enrollment rate directly and explicitly inﬂuences

the relative supply or the relative employment, thereby, the relative wages.

Determinants of the College Enrollment Rate

Until now, we have discussed the effect of college education policy on labor
market outcomes. This subsection explores the opposite direction, that is, the
determinants Of college enrollment, especially focusing on the feedback effect of
changes of relative wages of college graduates on college enrolhnent. Since the
enrollment rate is a policy target variable and the relationship between the quota and
the enrollment rate has already been established, this section excludes the discussion
about the quota.

The following model (5) is an enrollment equation in that it explains the

determinants of college enrollment.59

log ENROLL. = Bo + B] 10g (WCG/WHS)t + [32 log GDP. + B4TREND. + 8. (5)

The dependent variable is the log of the enrollment rate (ENROLL .). Real
GDP. controls for changes in aggregate demand. Since the dependent variable and
most of independent variables have a trend, TREND. is included to control for a

time trend.60 This equation is identiﬁed because the current enrollment will not affect

 

59 Freeman(l975) constructed the similar enrollment equation with explanatory variables such as
relative earnings of college graduates and the relative demand index.

6° A coefﬁcient from the regression of ENROLL on t is 0.07 (standard error = 0.008). That of log
(Woo /WHs ). is —0.03 (standard error = 0.002). That of GDP is 0.15 (standard error = 0.002). This

97

the current wages. However, there may be correlation between current enrollment
and the past enrollment. This section estimates equation (5) using the Corchran-
Orcutt method.

Although, tuition costs would be a good candidate for an explanatory variable
in the enrollment equation, this model does not include that variable, because the
number of observations is very small. However, the omission of tuition costs
variable does not seem to cause the severely biased estimates. Indeed, some evidence
shows that growth rate of the enrollment rate was relatively high in 1982 when
tuition grew rapidly, while it was low during 1983-88 when tuition increases literally
stabilized. 6‘ Also, Kim and Yoo(1994) found that tuition cost did not affect
enrollment for 4-year college.

Table II—lO reports that all coefﬁcients for (Wco fWHs). are positive. Although
the coefﬁcients for all workers in column (1) and for male workers in column (2) are
not signiﬁcant, the coefﬁcient for female workers in column (3) are signiﬁcant. The
negative sign of the coefﬁcient on GDP is mysterious. While GDP is included to
control for the shift in relative demand for college graduates, this variable is
problematic. First of all, the causal relationship between GDP and ENROLL can run

in both direction. While the rapid growth of GDP could cause high enrollment, high

 

result indicates that real GDP has the strongest upward trend and most of variables also have a
signiﬁcant time trend.

6' Tuition cost is a very important determinant of enrolling college in the United States. Trusteeship
(1994) reports that about 30% of new college entrants chose their colleges because tuition costs of
their colleges were lower than other colleges in the United States. Since the consistent time-series data
about the tuition costs in Korea is not available, it is hard to ﬁnd out how much the enrollment
decisions are affected by the tuition costs. However, limited evidence shows that the government
ofﬁcially set the annual growth rate of tuition as follows; 1982: 15-30% (public and private university
and college); 1983: 5-7% (public), 5% (private); 1984: basically not increased; 1985: 3% (public and
private); 1986: 3% (public and private); 1987: 3% (public) 4.5% (private); 1988: 6% (public) 7%
(private). Before 1989, government set the basic level of tuition costs. After that, although

98

enrollment also could bring about the rapid growth of GDP in the long run through
the accumulation of human capital. Although the latter causal relationship seems to
be weak compared to the former one, it could not be ignored, when we remember
that Korea is a model case of the rapid economic growth based on investment in
human capital. In fact, until the 19703, Korean parents have supported the cost of
college education for their children, not because their income was high enough, but
because they expected that their children will earn more money with the college
diploma. However, in the 19803, as the demand for high-skilled manpower increased
with the industrial upgrading towards high-wage, high-technology industries, high
college enrollment rate began to reﬂect the economic growth.

In sum, this result about the relationship between current wages and the
enrollment rate suggests that the market mechanism was weak but anyway worked
during the 19803 and the 19903. Although the ever-increasing enrollment rate with
the continuously decreasing relative wages appears to be contradictory, market
incentives may still affect private decisions to enroll in college. After controlling for
trend and the ﬁrst order serial correlation, the estimation results show the weak
positive association between the college enrollment rate and the current relative
wages. Therefore, to seek the cause of changes in demand for college education only
among non—economic factors might not be appropriate. It is also not always correct

to regard excessive fever for higher education, one of the famous cultural traits of

 

government still basically set the level of tuition, the Association of College Education Institutions
could inﬂuence that process. Kim, Song, Park, and Weidman( 1996)

99

Koreans, as a main cause of the persistence of ever increasing demand for college

education as Choi(1989) argued.62

 

62 Also, although it was difﬁcult to calculate the expected returns to college education especially in
the 19803, ‘overinvestment’ in college education in the 19803 did not seem to be the result of this
difﬁculty. The expected returns to college education are calculated by the current returns to college
education multiplied by the probability of being employed. Since investors will harvest beneﬁts from
the educational investment over their lifetimes, it is not easy to calculate the expected lifetime beneﬁts
from the investment in college education.

100

V. Summary and Policy Implications

This paper found that the impact of the abrupt increase in the college entrance
quota in 1981 on the returns to the college education was enormous. The difference-
in-difference (DD) estimation shows that the returns to the four—year college
education declined 15.1 percent point between 1982-83 and 1987-88. This policy
effect was greater among young workers and female workers. The returns to the
college education of young workers declined 19.8 percent point and those of all
female workers declined 25.3 percent point. Also, in this period, the returns to four-
year college education dropped more than that of two-year college education.

This paper also conﬁrms that the college education expansion policy has
increased the relative supply and reduced the relative wages of college graduates
throughout the 1980s and 19903 by using a time series regression. Also, the reduced
form equations for the relative employment and the relative wages which include the
polciy variable of the college enrollment rate as a supply shock indicates that
transmitting the effect of the increase in the enrollment to the relative supply takes at
least six years. It takes one more year, seven years, to transmit the effect on the
relative wages.

This paper also found that the increase in college entrance quota directly led
to increase in college enrollment under the enormous excess demand for college
education. However, analysis of the determinants of college enrollment shows that
the declining relative wage of college graduates may have exerted a negative effect

on the college enrollment, though not strong enough to reverse its initial increase

101

induced by the expansion of college entrance quota. Although the inﬂuence of the
government policy has been enormous in the behavior of college enrollment, it does
not imply that the market mechanism has been absent in the college graduates labor
market in Korea.

There are some limitations in this paper. First, it does not explicitly deal with
the effect of the college graduate unemployment on the relative wages of college
graduates in relation with the college education expansion policy. As the increase of
the unemployment of college graduates contributed to the decrease of their relative
wages, any estimate that does not consider unemployment may be biased. Second,
while stressing the short-run negative effect of the college education expansion
policy on the relative wages of college graduates, this paper does not examine the
widely assumed long-run positive effect of the policy on wages through the

accumulation of human capital.

102

REFERENCES

Blackburn, M.L., D.E. Bloom and RB. Freeman (1990), “The Declining Economic
Position of Less Skilled American Men,” in A Future of Lousy Jobs? Gary
Burtlees, ed., The Brookings Institutution.

Choi, Kang-shik (1996), ”The Impact of Shifts in Supply of College Graduates:
Repercussion of Educational Reform in Korea”, Economics of Education
Review, Vol. 15, N0. 1.

Filer, R.K., D.S. Hamermesh, and A.E. Rees (1996), The Economics of Works and
Pay.

Freeman, Richard B (1975), “Overinvestrnent in College Training?” Journal of
Human Resources, Vol.10, No.3.

Ihm, Chon Sun (1990), “Cohort shifts, Occupation and Economic Beneﬁts of
Education in Korea: The Impact of Educational Expansion in a dynamic
Economy,” Doctoral thesis of Education, Harvard University.

Katz, L.F. and KM. Murphy (1992), “Changes in Relative Wages, 1963-1987:
Supply and Demand Factors,” Quarterly Journal of Economics.

Kim, DI, and R. Topel (1995), “Labor Markets and Economic Growth: Lessons
from Korea’s Industrialization, 1970-1990,” in R. Freeman and L. Katz, eds,
Differences and Changes in Wage Structure, University Chicago Press, 1995.

Kim, Mun-jo and Hye-young Kim (1992), “An Overeducation Phenomenon in the
19803 in Korea”, National Culture Study, Korea University, Vol. 25.

(Korean)

103

Kim, Song, Park, and Weidman (1996), “Costs Of College Education and Tuition”,
Education and Science Print, (Korean)

Kim, Young-Hwa and Han-KOO YOO (1994), “Time Series Analysis of the
Determinants of the Students’ 4-year College Entrance Demand in Korea,”
The Journal of Educational Research, Vol. 32, No.1. (Korean)

World Bank (1998/99), World Development Report: Knowledge for Development.

104

Table II—l: A Summary of Statistics of

All Workers Aged 19-59

 

After
Effects
(1987-88)

Recent

(1995-96)

Before
Effects
(1982-83)
(All Workers)
Total
Monthly Wages* 561.5
(1.89)
# of Obs (persons) 49572
Age (years) 29.8
Tenure (years) 3.51
Coll Grads
Monthly Wages* 1207.0
(7.96)
# of Obs (persons) 6017
Age (years) 34.7
Tenure (years) 4.96

Junior Coll Grads and Less

Monthly Wages* 472.4
(1.38)
# of Obs (persons) 43555
Age (years) 29.1
Tenure (years) 3.31

675.4
(1.94)
54381
31.0
4.30

1269.0
(7.93)
7476
34.4
5.66

580.8
(1.44)
46905
30.5
4.08

1410.7
(2.69)
76533
34.2
6.15

1893.2
(6.25)
21132
34.5
6.15

1226.7
(2.43)
55401
34.1
6.15

(continued to the next page)

105

Table II-l (cont’d)

(Male Workers)

Total
Monthly Wages* 703.9
(2.51)
# of Obs (persons) 31878
Age (years) 32.8
Tenure (years) 4.16
College Grads
Monthly Wages* 1226.0
(8.21)
# of Obs (persons) 5669
Age (years) 35.0
Tenure (years) 5.07
Junior Coll Grads and Less
Monthly Wages* 591.0
(1.86)
# of Obs (persons) 26209
Age (years) 32.3
Tenure (years) 3.96
(Female Workers)
Total
Monthly Wages* 305.0
(1.30)
# of Obs (persons) 17694
Age (years) 24.4
Tenure (years) 2.33
College Grads
Monthly Wages* 898.0
(27.63)
# of Obs (persons) 348
Age (years) 29.7
Tenure (years) 3.17
Junior Coll Grads and Less
Monthly Wages* 293.1
(1.02)
# of Obs (persons) 17346
Age (years) 24.3
Tenure (years) 2.31

819.4
(2.55)
35552
33.7
5.17

1295.0
(8.24)
6903
34.8
5.79

704.8
(1.92)
28649
33.5
5.02

403.5
(1.51)
18829
26.0
2.65

955.6
(25.54)
573
29.9
4.07

386.1
(1.11)
18256
25.8
2.60

1596.2
(3.18)
55499
35.8
6.98

1941.1
(6.59)
19019
35.0
6.35

1416.4
(3.01)
36480
36.1
7.31

921.3
(3.08)
21034
30.2
3.96

1462.2
(16.88)
2113
30.1
4.41

860.9
(2.50)
18921
30.3
3.91

 

Standard errors are in parenthesis.

* in thousand won (real price)

106

Table II-2: A Summary of Statistics of Young Workers Aged 19-29

 

Before After Recent
Effects Effects
(1982—83) (1987—88) (1995-96)
(All Workers)
Total
Monthly Wages* 390.1 481.6 993.3
(1.24) (1.28) (2.20)
# of Obs (persons) 28362 27603 28353
Age (years) 23.4 23.8 24.7
Tenure (years) 2.07 2.21 2.31
College Grads
Monthly Wages* 770.5 784.2 1192.5
(5.51) (5.37) (5.36)
# of Obs (persons) 1815 2361 6259
Age (years) 27.0 27.0 26.9
Tenure (years) 1.49 1.70 1.59
Junior Coll Grads and Less
Monthly Wages* 364.1 453.3 936.9
(1.10) (1.16) (2.24)
# of Obs (persons) 26547 25242 22094
Age (years) 23.1 23.5 24.1
Tenure (years) 2.11 2.26 2.51

(continued to the next page)

107

Table II—2 (cont’d)

(Male Workers)

Total
Monthly Wages* 495.3
(2.20)
# of Obs (persons) 13162
Age (years) 25.2
Tenure (years) 1.97
College Grads
Monthly Wages* 782.0
(5.94)
# of Obs (persons) 1575
Age (years) 27.2
Tenure (years) 1.46
Junior Coll Grads and Less
Monthly Wages* 456.3
(1.85)
# of Obs (persons) 11587
Age (years) 25.0
Tenure (years) 2.03
(Female Workers)
Total
Monthly Wages* 299.0
(1.08)
# of Obs (persons) 15200
Age (years) 21.7
Tenure (years) 2.16
College Grads
Monthly Wages* 695.5
(13.81)
# of Obs (persons) 240
Age (years) 25.4
Tenure (years) 1.66
Junior Coll Grads and Less
Monthly Wages* 292.6
(1.00)
# of Obs (persons) 14960
Age (years) 21.7
Tenure (years) 2.17

586.2
(2.06)
12874
25.8
2.13

800.2
(5.59)
2003
27.2
1.65

546.7
(1.99)
10871
25.5
2.22

390.2
(1.16)
14729
22.1
2.28

694.4
(15.82)
358
25.6
2.00

382.6
(1.04)
14371
22.0
2.29

1101.3
(3.12)
15091
26.1

1208.1
(5.95)
4946
27.3
1.56

1049.3
(3.59)
10145
25.5
2.35

870.4
(2.71)
13262
23.2
2.55

1134.0
(13.46)
1313
25.6
1.73

841.4
(2.49)
11949
23.3
2.64

 

Standard errors are in parenthesis.

* in thousand won (real price)

108

Table II-3: Changes in the Returns to College
Education before/after the Policy

The dependent variable is the log of the hourly wages.

 

 

(l) (4)
OLS DD
1982—83 1987—88 1995—96 82-83
vs 87-88
All Wbrkers
Total
COLL 0.663 0.594 0.393 —0.150
(0.005) (0.004) (0.003) (0.006)
(# Obs) 49572 54381 76533 103953
(Adj R2) 0.6693 0.6634 0.6129 0.6722
Male
COLL 0.704 0.586 0.421 -0.093
(0.006) (0.004) (0.003) (0.007)
(# obs) 31878 35552 55499 67430
(Adj R2) 0.5545 0.5686 0.5534 0.5698
Female
COLL 0.752 0.619 0.432 -0.253
(0.018) (0.013) (0.007) (0.021)
(# Obs) 17694 18829 21034 36523
(Adj R2) 0.4961 0.4898 0.5393 0.5370
Young workers
Total
COLL 0.613 0.491 0.305 —0.198
(0.009) (0.007) (0.004) (0.010)
(# obs) 28362 27603 28353 55965
(Adj R2) 0.5636 0.5353 0.4522 0.5725
Male
COLL 0.678 0.544 0.302 —0.138
(0.010) (0.008) (0.005) (0.012)
(# obs) 13162 12874 15091 26036
(Adj R2) 0.4663 0.4760 0.4536 0.4864
Female
COLL 0.635 0.486 0.361 -0.266
(0.021) (0.015) (0.008) (0.024)
(# Obs) 15200 14729 13262 29929
(Adj R2) 0.4634 0.4083 0.4064 0.5014
* Standard errors are in parenthesis.

** (l), (2) and (3): The coefficient for a dummy variable of

four-year college graduation vs. non-graduation (COLL).

(4): The coefficient for an interaction between the college
graduation dummy variable and the after-policy years (1987-88) dummy
variable(COLL*AFTER).

*** All regression models include control variables such as potential
experience years and its square term, tenure, marital status, gender,
and industry.

109

 

Table II-4: Changes in the Returns to College

Education (including four-year college and two-year college)
before/after the Policy

The dependent variable is the log of the hourly wages.

 

(l) (2) (3)
DD DD DD
Total Male Female
All Wbrkors
COLL*AFTER -0.151 —0.093 —0.270
(0.006) (0.007) (0.020)
JC*AFTER —0.126 -0.087 -0.240
(0.005) (0.013) (0.020)
(# obs) 103953 67430 36523
(Adj R2) 0.6813 0.5823 0.5554
Young WOrkers
COLL*AFTER —0.203 -0.l44 -0.290
(0.010) (0.012) (0.024)
JC*AFTER -0.l66 -0.130 -0.237
(0.013) (0.017) (0.021)
(# obs) 55965 26036 29929
(Adj R2) 0.5879 0.5043 0.5520

 

110

Table II—5: The Relationship between Quota and the Enrollment Rate

 

 

(1) (2)**
dep. var.= log of the enrollment rate
log of 0.812 0.736
QUOTA (0.100) (0.122)
TREND 0.014 0.020
(0.002) (0.008)
N of Obs 31 30 i
F(2,28) F(2,27)
=150.3 =36.32
D.W. statistic 0.483 1.372
* Standard errors are in parenthesis.

** Cochran—Orcutt AR(1) regression

 

lll

Table II-6: The Determinants of Relative Employment
(the Reduced Form Equation)

 

 

(l) (2) (3)
All Male Female
dep. var.= log of the enrollment rate
Log of lag 0.249 0.216 0.200
of ENROLL** (0.079) (0.082) (0.278) '
Log of —1.007 -0.918 -0.790
Real GDP (0.175) (0.196) (0.518)
Log of 0.039 0.051 -0.048
DINDEX (0.022) (0.024) (0.059)
TREND 0.105 0.095 0.119
(0.014) (0.016) (0.041)
N of Obs 16 16 16
F(4,11) F(4,11) F(4.11)
=727.23 =529.98 =167.75
Estimated Rho -0.103 -0.279 0.300
(0.137) (0.191) (0.130)

 

* Standard errors are in parenthesis.
** In all workers’ case and male workers’ case: L7.ENROLL
In women workers' case: L4.ENROLL

112

Table II-7: The Determinants of the Relative Wage
(the Reduced Form Equation)

 

(1) (2) (3)
All Male Female

dep. var.= log of the enrollment rate

Log of lag -0.648 -0.416 0.405
of ENROLL** (0.147) (0.122) (0.455)
Log of 0.291 0.161 -0.258
Real GDP (0.172) (0.192) (0.485)
Log of -0.036 -0.030 0.080
DINDEX (0.033) (0.036) (0.084)
TREND —0.047 -0.034 —0.025
(0.014) (0.016) (0.039)

N of Obs 18 18 18
F(4,l3) F(4.13) F(4,l3)
=38.44 =21.10 =6.87
Estimated rho 0.817 0.727 0.744
(0.043) (0.137) (0.062)

 

* Standard errors are in parenthesis.
** In all workers’ case and male workers’ case: L7.ENROLL
In women workers’ case: L4.ENROLL

113

Table II-8: The Sensitivity Analysis
(Coefficient for the lagged enrollment rates)

 

Relative Employment Equation

Coefficients for

L7.ENROLL L6.ENROLL L5.ENROLL L4.ENROLL

All .249 .275
(.079) (.109)
Male .216 .400
(.082) (.081)
Female .063 .343

(.234) (.256)

Relative Wages Equation

Coefficients for

L7.ENROLL L6.ENROLL L5

All -.648 —.001
(.147) (.219)

Male -.416 .002
(.122) (.162)

Female -1.435 -.927
(.243) (.406)

-.009
(.138)
.034
(.122)
.319
(.330)

.ENROLL L4

-.014
(.231)
.039
(.173)
-.655
(.441)

.060
.100)
.033
.090)
.200
(.278)

.ENROLL

.229
(.195)
.143
(.150)
.405
(.455)

 

* Standard errors are in parenthesis.

114

Table II-9: The Relative Wages and the Relative Supply
(The Demand Curve)

 

(1) (2) (3)
All Male Female

dep. var.= log of the relative wages

Log of rel. -0.416 -0.510 -0.383
Supply (0.419) (0.357) (0.377)
Log of -0.173 -O.413 -0.671
Real GDP (0.525) (0.482) (0.727)
Log of 0.017 0.007 0.072
DINDEX (0.051) (0.049) (0.087)
TREND -0.007 0.019 0.029
(0.052) (0.045) (0.070)
N of Obs 16 16 16

F(4,11) F(4,11) F(4,11)

=13.53 =9.29 =7.06

D.W. stat (orig.) 1.030 0.871 0.803

(trans.) 1.420 1.343 1.511

 

* Standard errors are in parenthesis.

115

Table II-10: The Determinants of the Enrollment Rates

 

dep. var.= log of the relative employment

Log of Current
Relative Wages

Log of
Real GDP

TREND

N of Obs

Estimated rho

(1)
All

0.774
(1.114)

-O.382
(0.673)
0.084
(0.049)
21
F(3,17)
=2.86

0.696
(0.159)

1.416
(1.256)

-0.477
(0.811)
0.099
(0.053)
21
F(3,17)
=4.20

0.535
(0.188)

(3)
Female

0.760
(0.411)

-0.398
(0.482)
0.091
(0.034)
21
F(3,17)
=6.73

0.731
(0.141)

 

* Standard errors are in parenthesis.

116

Figure II—2: Actual Relative Wages of College vs. High School Graduates
(All Workers)

 

 

 

__1_ EH _,_
+male (

 

+ female )

1 v 7 . , L___ 1

0.5 e ,1 . __._ _

 

 

 

767880828486889092949698

 

 

 

X axis: year, Y axis: ratio

Figure II-3: Detrended Relative Wages of College vs. High School
Graduates (All Workers)

 

0.5
0.4 -
0.3 -
0.2 - 1
0.1

 

 

t; I all
+ male
(ﬂ. female

-0.1 -
-0.2 .
-0.3 --
-0.4

 

 

 

 

 

767880828486889092949698

 

 

 

X axis: year, Y axis: ratio

117

Figure II-4: Actual Relative Supply of College vs. High School Graduates
(All Workers)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0.7

0.6

0.5

0,4 "...totaT
+male

0'3 *‘ +female’

0.2 ..

0.1 --- 9 ﬂ _

O I I f I f I T I T T T 1' " I T T T I I
80828486889092949698100
X axis: year, Y axis: ratio
Figure II-5: Detrended Relative Supply of College vs. High School
Graduates (All Workers)

0.08

0.06 -1

0.04 . ‘

0.02- Click” 1
l+male

0 * +female)

-0.02

-0.04

-0.06

 

 

 

80 82 84 86

889092949698

100

 

 

X axis:

year,

Y axis:

ratio

118