The Impact of Mandated Maternity Benets on the Gender
Dierential in Promotions: Examining the Role of Adverse
Selection∗
Mallika Thomas†
September 6, 2016
Abstract
This paper examines how mandated maternity leave policies impact the gender gap in pro-
motions. I present a model of the gender gap in promotions where rms must choose whether
to invest in the training of their employees, but they are uncertain about their employees' future
choice of hours of work. If women are more likely than men to reduce their hours of work during
childrearing years, rms will invest less in women early in their careers, leading to a gender gap
in promotions. In the presence of asymmetric information about workers' future preferences,
mandated maternity leave policies can exacerbate this gap. Using the Multi-City Study of Urban
Inequality and the Panel Study of Income Dynamics, I test the predictions of the model in the
context of the Family and Medical Leave Act of 1993 (FMLA). Women hired after the enactment
of the FMLA are ve percent more likely to remain employed but eight percent less likely to be
promoted than those who were hired before the FMLA. Furthermore, I nd evidence suggesting
that information asymmetry, in addition to selection, is driving the increase in the gender gap in
promotions.
∗I am grateful to Gary Becker, Edward Lazear, James Heckman, and Alexander Frankel for invaluable advice and feedback
at all stages of this project. I also thank Marianne Bertrand, Alessandra Voena, Pierre-Andre Chiappori, Samuel Scholhofer-
Wohl, Bruce Meyer, Elisa Olivieri, Daryl Fairweather, James Marrone, Seth Blumberg, Stefano Mosso, and numerous seminar
participants for many useful comments. I gratefully acknowledge the nancial support of the American Economic Association's
Dissertation Fellowship, the George Stigler Dissertation Fund, University of Chicago's Center for the Study of Gender and
Sexuality, and the University of Chicago's Division of Social Sciences.
†Cornell University Department of Economics, mallika.thomas@cornell.edu
1
1 Introduction
Over the past 20 years, most economically advanced countries have enacted an array of maternity leave ben-
ets, and these countries have witnessed a large increase in female labor force participation in comparison
to the United States. However, employed women in the United States are almost three times as likely to
reach managerial level positions and much more likely to work full-time than in advanced countries with
more comprehensive maternity leave policies.1 A key question in designing policies that are intended to
improve women's labor market outcomes is whether such policies may be facilitating participation in entry
level positions but simultaneously reducing the opportunities for women to attain upper level positions.
This paper evaluates whether maternity leave policies can contribute to a widening of this managerial-entry
level gap, by changing the incentives for employers to invest in their workers. Employers are uncertain
about whether workers will reduce their productivity in the future, and in particular, whether employees will
reduce their hours of work when they have children. However, they must choose whether to invest in their
employees, by training or mentoring them early on in their careers, before learning what their labor supply
choices will be in the presence of children. These investments are more protable to the rm when workers
work longer hours. If women, on average, work fewer hours during child-rearing years than men, and rms
cannot perfectly distinguish between workers who will work fewer hours and those who will not, the rm
will be less likely to invest in and train women during the early years of their careers. Mandated maternity
benets can exacerbate the existing information problem and widen the gender dierence in promotions, by
changing the distribution of types of women who select into the labor force. Therefore, the expected return
on investment in any individual woman may be lower, even among those women who would have otherwise
worked as many hours as men.
I formalize this idea using a simple model where workers are heterogeneous in their future preferences for
leisure, or time with their family, but information on these preferences is private. Because workers cannot
internalize the cost of their training, and rms prot from their ability to sort workers, the private informa-
tion gives rise to a signaling game. Employers use observed productivity as a signal of future productivity. I
show that after a mandated maternity leave policy is enacted, however, observed productivity becomes less
predictive of future productivity. Precisely because of the increase in the expected career tenure of family-
oriented workers, signals become less informative, and the cost of extracting information about workers'
types increases. As a result, rms must set a higher standard in order to protably promote female workers.
The model predicts that while employment and labor force participation among women will increase after
the enactment of a maternity leave mandate, the likelihood of promotions for young women will decrease.
Moreover, the framework makes clear that the decrease in the likelihood of promotion is generated by both
the selection into the labor force of workers with higher labor supply costs as well as a reduced expected
1Blau and Kahn (2013) details these ndings in a cross-country analysis.
2
return on rm investment in all female workers, even those with the same ex-ante preferences as men.
I then determine the impact of mandated maternity leave policies in the context of the Family and Medical
Leave Act of 1993 (FMLA), a federal mandate in the United States. I exploit the variation in the fertility
of women by age and the large decrease in fertility for women at age 40 as well as the variation in state
legislation that preexisted the federal mandate, using two dierent datasets: the Panel Study of Income
Dynamics (PSID) and the Multi-City Study of Urban Inequality (MCSUI). I nd that women hired after
the FMLA are ve percent more likely to remain employed, conditional on job tenure, but they are eight
percent less likely to be promoted in comparison to those hired before the FMLA. Furthermore, the widening
of the existing gender dierential in promotions is only observed for women under the age of 40 and is larger
among women in age groups with the highest likelihood of conceiving a child.
Consistent with a model of asymmetric information, I nd that all women of childbearing age face a reduced
likelihood of promotion, including those women who never have children. Furthermore, I take advantage of
the detailed data available in the MCSUI, and I construct for each rm, measures of the cost of training. I
nd the that the widening of the gap in promotions is greater among rms with high training costs, a result
that is dicult to reconcile with a model of symmetric information. Finally, I examine several signals of
workers' future productivity, including early career hours, indicators of job perfomance, and the choice of
potential wage prole. For women hired after the Family and Medical Leave Act, I nd a decrease in the
return to these signals, in terms of the likelihood of promotion, and moreover, a decrease in the return to
these signals specically for fertile women alone, whereas the return to signaling is unchanged for women
over the age of 40, for a wide variety of signals. This is strong evidence that information asymmetry about
workers' private labor supply costs, in addition to selection based on these costs, is driving the increase in
the gender gap in promotions.
Distinguishing between the implications of a symmetric information model and a model in which asymmetric
information plays a role has important welfare implications. In a world of symmetric information, where
employers can fully discern which workers will work long hours in the future and which will not, training will
be allocated as eciently as before the enactment of a maternity leave mandate. However, if asymmetric
information plays a role, then the cost of increasing female labor force participation is borne, in part, by
the high-hours women, in terms of a loss of human capital and wage growth over the course of their lifecycle.
3
1.1 Related Literature
While there is an extensive literature examining the short-term impact of maternity leave policies on female
employment and labor supply,2 a much smaller but growing strand of literature investigates the dynamic
eect of maternity leave policies on the long-term human capital accumulation and wage growth of women.
Mukhopadyay (2012) , Erosa, Fuster, and Restuccia (2010), and Sanchez-Marcos (2014), for example, all
examine the erosion of dierent forms of human capital that are incentivized as a result of maternity leave
policies. 3 Adda, Dustman, and Stevens (2011) analyze the eect of policies that incentivize childrearing on
not only unearned wages and the loss of human capital, but also on the change in the path of wage growth
due to selection into more child-friendly occupations. To date, however, such work has tended to address
the direct eect of leave policy on the incentives for women of a given type, and the eect due to selection
has largely been ignored.
This paper not only accounts for selection on the basis of heterogeneity in labor supply costs, but it em-
phasizes a distinction between adverse selection, where unobserved heterogeneity preexists the policy change
and aects the optimal responses of agents, and the direct response of individuals to the incentive structure
created by the policy. Much of the previous literature only examines the latter. Furthermore, such literature
nds relatively small eects on wages and wage growth from the leave policy itself, especially policies where
the length of the leave period is short.4 Furthermore, the majority of approaches that examine the eect
of maternity leave polices through counterfactual analyses attribute, by design, the variation in wages and
2Several U.S. studies suggest that time o work is associated with inreases in employment and wages, such as Dalto
(1989), SpalterRoth and Hartmann (1990), Waldfogel (1994), (1997), Ruhm (2003), and Higuchi and Abe (1999). Klerman
and Leibowitz (1997), Waldfogel (1998), and Baker and Milligan (2008) show that these policies have a substantial eect on
female labor force participation. Waldfogel, Higuchi and Abe (1999) nd that family leave coverage increases the likelihood
that a woman will return to her employer after childbirth in the US, Britain and Japan. There is evidence both in the US
and Britain (Waldfogel (1998a)) that women who maintain employment continuity over childbirth have higher wages than
those who do not. Ruhm (1998) nds that in nine European countries parental leave legislation is associated with increases in
women's employment but wth reductions in their relative wages when leave is mandated at extended durations. Gruber (1994)
examines the eect on wages of a US mandate requiring job-protection as well as employer-based coverage for the medical costs
of pregnancy among rms providing such benets for temporariliy disabled workers. This paper nds that this policy, which
aects the cost of hiring a member of an identiable group, reduces wages by the cost to the employer, but has little eect on
employment.
3Mukhopadyay (2012) builds a dynamic model of labor force participation with the feature that a maternity policy that
increases current participation aects future wages through increased work experience as well as future labor force participation.
Erosa, Fuster, and Restuccia (2010) assess the impact of mandatory parental leave policies by developing a general equilibrium
model of fertility and labor market decisions, considering the eect on women's retention of job-specic human capital. Sanchez-
Marcos (2014) argues that mothers on leave do not accumulate human capital, so long lasting leaves may erode their wage
prospects, aecting future labor force participation choices as well. Lalive and Zweimuller (2009) nds that leave policies induce
a fertility delay in the short-term, and a delay in fertility can lead to substantial increases in career earnings.
4Waldfogel (1996) found that the FMLA had a slightly positive employment eect and no discernible wage eect. Adda,
Dustman, and Stevens (2011) found relatively small eects on wage growth due to a pro-fertility policy in Germany, examining
the channels of the direct incentive of the policy on fertility, and subsequently, occupational choice in anticipation of a change
in fertility choices. They nd the small estimated impact unsurprising, given that the long-term impact of the policy on
fertility is small, even in the context of a paid maternity leave policy. Ruhm (1998) examines mandated paid parental leave in
nine European countries and nds that parental leave is associated reductions in their relative wages when only when leave is
mandated at extended durations.
4
employment to a change in fertility or labor force participation incentives for individuals of xed character-
istics. This strand of literature does not allow for a role of asymmetric information.
Thus, this paper also builds on the literature on statistical discrimination, pioneered by Arrow (1972) and
Phelps (1972) and formalized by Coate and Loury (1993). This literature emphasizes that group dierences
can arise endogenously, even without any ex-ante dierences in across groups. More recent literature has
incorporated private information into the context of the gender-wage gap and its evolution over the course
of the lifecycle. Albanesi and Olivetti (2009) show that in the presence of private information about worker's
labor market attachment, rms oer labor contracts with lower earnings and performance pay to female
workers. The work closest to this paper, in addressing the long-term evolution of the gender gap in human
capital accumulation and wages resulting from private information is Gayle and Golan (2012), which formu-
lates a model of labor supply and human capital accumulation in which workers have private information
about their labor market participation costs, and employers use the observed labor supply decisions as a sig-
nal of the worker's private information. This paper, however, uses the implications of a model of asymmetric
information to analyze the eect of introducing a maternity leave policy on the equilibrium labor market
outcomes and human capital accumulation of women.
Finally, this papers speaks to a mechanism through which the gender wage gap itself evolves over the course
of the lifecycle. There is growing evidence that the wage gap is relatively small when workers are young,
and it increases over the course of the lifecycle (Bertrand, Goldin, and Katz (2010), Wood, Corcoran, and
Courant (1993)). Moreover, in spite of the signicant decline in the gender gap in earnings in the United
States over the past 40 years, there is large evidence of a persistent gap among the high ranking and higher
earning positions (Blau and Kahn (2006), Bertrand and Hallock (2001), Wolfers (2006)). A large body of
literature addresses the array of explanations for this gap (see Altonji and Blank, 1999, for a survey), from
human capital explanations (Becker (1985), Mincer and Polachek (1974)) to comparative advantage (Lazear
and Rosen (1990)) and occupational choice (Polachek (1981), Adda, Dustman, and Stevens (2011)). The
model presented in this paper and the evidence substantiating it adds to this literature by nding support for
one of the mechanisms through which the gender wage gap itself can arise. It species a model that suggests
that information asymmetry is one of the driving forces behind the increase in the gender earnings gap at
the upper end of the earnings distribution and addresses the implications for the impact of a mandated
maternity leave policy.
The rest of the paper is organized as follows. Section 2 describes the institutional details of the Family and
Medical Leave Act. Section 3 develops the theoretical framework. Section 4 presents the main results and
discusses alternative explanations. Section 5 concludes.
5
2 Background and Institutional Detail
Passed in August of 1993, the Family and Medical Leave Act (FMLA) is a federal mandate in the United
States that requires covered employers to allow eligible employees to take up to 12 weeks of unpaid leave per
12-month period for the birth and care of a newborn child. 5 Once their leave is over, employees are entitled
to reinstatement. The mandate was designed to impose a minimal cost to the employer6, and it does not
require that employees continue to receive any form of wages or compensation while on leave. 7 In order to
be eligible for the leave, employees must have worked for a covered employer for 12 months, have worked at
least 1,250 hours over the past year, and worked at a location where at least 50 employees were employed
within 75 miles.
Following FMLA leave, an employee has the right to be returned to the same position the employee held
before or to an equivalent position - one that is identical to the employee's former position in terms of pay,
benets, and working conditions, including privileges and status. The equivalent position must involve the
same or substantially similar duties and must require substantially equivalent skill, eort, responsibility, and
authority. 8
The national FMLA was not the rst legislation of its kind. Several states had passed similar legislation
prior to Congress' 1993 output. Some states, such as California, passed legislation that very closely mirrored
the federal FMLA policy. Many states passed legislation well before 1993, but such legislation was not as
expansive as what was eventually mandated federally. Other states supplied legislation that was in some
ways more progressive than the federal counterpart. Maine, for example, passed a family and medical leave
policy in 1987 that applies to rms of 15 or more employees. Oregon only passed a leave requirement in
1995, but it applies to employers of 25 or more persons and only requires that an employee as been employed
for 180 and worked an average of 25 hours or more per week during that 180 day period.
In this paper, we take into consideration the variation in the state laws prior to the passage of the federally
mandated FMLA. I exploit the variation in the state laws across states and over time, as summarized in
Appendix D Table 1.
5If an employee is unable to work because of pregnancy, she may take FMLA leave before the child is born. However,
she still receives 12 weeks of FMLA leave for the 12-month leave period. Reasons for leave also include placement of a son or
daughter for adoption or foster care, to care for an immediate family member with a serious health condition, or medical leave
due to an employee's own serious health condition.
6Lenho and Bell (2002)
7Employees are entitled to continue their health benets while on leave. The employer must continue to pay whatever
premiums it would pay if the employee were not on leave. If the employee voluntarily chooses not to return from leave, however,
the employer may require the employee to repay the cost of the health care premiums it paid while the employee was on leave.
8Equivalent pay includes any bonsues or payments that occurred while the employee was on FMLA leave, if those payments
were unconditional.
6
3 Model
This section provides a simple framework that formalizes the mechanism through which mandated mater-
nity benets can both increase employment of women with children and reduce the likelihood of promotions
among women of childbearing age, even among women who never have children. The model makes clear that
the decrease in the probability of promotions, conditional on employment, is generated by (i) the selection
into the labor force of workers with higher marginal costs of labor and (ii) a decrease in the expected return
on rm investment in low marginal cost types. The model highlights that in the presence of asymmetric
information, mandated maternity leave benets result in a change in the distribution of types of women who
select into and are retained in the labor force, and the likelihood of promotion for young women decreases
relative to even the symmetric information case. From this model, I derive a series of testable implications
that I take to the data in Section 4.
3.1 Economic Environment and Decision Structure
I specify a two-period model in order to capture the phenomenon of workers' privately-known but anticipated
increase in their future value of nonmarket time, in the presence of children. There are two types of agents,
rms and workers. All are assumed to be risk neutral, and there is no discounting between periods. In Period
1, workers are identical in their marginal value of leisure, but in Period 2, they vary in their marginal value
of leisure or nonmarket time, θ, which is unknown to the rm in the rst period. I assume that there are
three types of workers, A, B, and C, who have marginal values of leisure θA, θB , and θC , respectively, where
θA > θB > θC . The types are distributed with probabilities P (A), P (B), and P (C), respectively, and the
distribution from which these types are drawn is common knowledge. I begin by assuming that no worker
has children in Period 1, and all workers have children in Period 2. Later, I relax these assumptions.
At the end of Period 1, rms must choose whether to undertake a training investment in each worker, of
cost c to the rm, that generates an increase in the worker's rm-specic human capital. A worker who
receives this investment has a greater Period 2 productivity. The decision to invest in a worker is denoted
by τ (τ = 1 if the rm invests and τ = 0, if not). This training is complementary with the worker's labor
supply in Period 2, h2, in the rm's production function:
Y2(h2; τ) = α(τ)h2,
where α(τ) represents the rm's choice of the worker's human capital, and α(1) > α(0).
In Period 1, workers choose their labor supply, h1, and receive a shock, ε, to their Period 1 productiv-
ity, y1. Employers, however, observe only their productivity, a noisy signal of workers' choice of hours:
7
( )
y1 = h1 + ε, where ε ∼ N 0, σ2
Y1(h1, ε) = α(0)y1
Because training is rm-specic, workers cannot fully internalize the cost of their training. Workers sharing
the cost of training is also not sucient to perfectly sort workers, due to incomplete information on the
part of the rm about workers' abilities. Previous literature has established that imprecise information with
respect to worker quality can function similarly to a credit constraint. In this literature, rms maximize
prots by oering a uniform wage contract at the time of hire and paying for training, rather than oering
training to workers who accept lower wages.9 A discussion of how this model can be extended to more for-
mally to include a wage rigidity due to imperfect information with respect to ability is oered in Appendix
B. However, it is important to note that this implies that no worker can fully internalize the cost of the
training.10 Thus, rms form beliefs about workers' future productivity using their rst period signal.
The salary contract for each period is prespecied such that rms and workers share the rent to human
capital. The worker takes a share R1 in Period 1 and R2 in Period 2, and the rm takes shares 1−R1 and
1−R2 in Periods 1 and 2 respectively:
S1(h1; ε,R1) = R1α(0) (h1 + ε)
S2(h2; τ,R2) = R2α(τ)h2,
where 0 < R1 < 1 and 0 < R2 < 1. Firm-specic training creates a bilateral monopoly situation in wage
determination, and R2 is determined as a result of a bargaining arrangement. Since there is a match-specic
surplus generated in Period 2, this surplus will have to be shared by bargaining. This typically implies that
rms obtain a fraction of the productivity of the worker as prots.11
Labor force participation is determined by whether the worker's utility from working is greater than the
9Weiss (1980) demonstrates that when rms have imprecise information concerning the labor market endowment or ability
of workers, and the labor market endowment is positively correlated with a worker's outside option, a wage rigidity exists.
Reservation wages of workers are an increasing function of productivity, and workers will not be able to increase their probability
of securing an employment contract by lowering their reservation wages. Acemoglu and Pischke (1998) demonstrates that
asymmetric information with respect to worker ability can function similarly to credit constraints in that workers may not pay
for their training. In this case, rms are willing to pay for training and willing to make an additional payment in terms of
wages for ex-post monopsony power over workers who are revealed to be more able. In either case, imperfect information about
workers' abilities keeps wages articially high, and thus workers cannot compensate rms for training.
10Simply having a minimum wage constraint would impose an inecient allocation of training only for workers where the
minimum wage constraint is binding.
11See Hall and Lazear (1984) and Hashimoto and Yu (1980) for analyses of prespecied division of the rent to human capital.
Hashimoto and Yu (1980) relies on rm-specic human capital, but Hall and Lazear (1984) uses rm-specic capital only as
one example of a surplus generated by the rm-worker match. Both rely on some uncertainty with respect to some aspects of
the value of that trade as well as the value of their alternatives and that renegotiation of the contract is costly. Other literature
shows that the presence of matching and search costs in the labor market creates a bilateral monopoly, since it is both dicult
for workers to nd new employers as well as costly for rms to replace their employees. See Mortensen (1982), Diamond (1982)
and Pissarides (1990) for analyses of the standard search and matching model. Bargaining, induced by a match-specic surplus,
therefore compresses the wage structure.
8
utility of the outside option. Workers' utility from labor force participation in each period is based on their
salary and their cost of supplying labor, which is increasing and convex in hours of work. The marginal cost
of labor is dependent on their type. We assume that costs take a quadratic form, with C1(h1) =
1
2h
2
1 and
C (h ; θ) = 1θh22 2 2 2.
Notably, each worker has a xed disutility of labor force participation in each period, ofγ(M) in period 2,
whereM is an indicator function for whether the mandated maternity leave policy is enacted, and normalized
to 0 in period 1.12 Thus, the utility of working in each period takes the following form:
U1(h1; ε,R1) = R1α(0) (h1 + ε)−
1
h2
2 1
1
U2(h2; θ, τ, R2) = R2α(τ)h − θh22 2 − γ(M),2
where M ∈ {0, 1}. I characterize the requirement of the FMLA to hold a job open for a maternity-related
absence as lowering the xed utility cost of labor force participation in Period 2, when children are present,
so that 0 < γ(1) < γ(0). This characterization captures the fact that having a maternity leave policy in
place lowers the minimum requirements for a woman with a child to sustain a job with the same employer,
by oering the option to take more time o for physical recovery, more exibility within a 12-month period
surrounding the birth of a child, or the ability to retain one's job while temporarily unable to work.13
This approach has been used in previous literature, including Mukhopadhyay (2012), which analyzes the
eects of the 1978 Pregnancy Disrcrimination Act on female labor supply and characterizes the requirement
to hold a job open for a pregnancy-related absence as a reduction in the utility cost of supplying labor
around the period of childbirth. Although this model does not include an additional period for the time
directly around childbirth, a decrease in the Period 2 participation cost is equivalent to a multiperiod model
with a reduction in the utility cost of supplying labor around the period of childbirth (shown in Appendix
B). Importantly, in a worker's decision about whether or not to remain employed, the worker considers not
only the xed disutility cost, but the total surplus from remaining employed with the same rm, including
the wage increase from the rm-specic training. The total lifetime surplus of remaining employed with the
same rm depends on the worker's marginal cost of working and their anticipated optimal choice of future
labor supply. Therefore, the impact of a temporary reduction in the utility cost of working for new mothers
on both employment and hours of work can be captured in a two-period model by a reduction in the Period
2 participation cost.
12We consider the impact of a mandated maternity leave policy only on those individuals who participate in the labor force
in their early careers.
13For example, Ruhm (1998) examines mandated paid parental leave in nine European countries and nds that parental
leave is associated with increases in women's employment. Waldfogel (1998) nds that maternity leave raises women's retention
over the period of child-birth and allows female employees to retain good job matches. Several U. S. studies suggest that time
o for a newborn child is associated with increases in employment and wages (Dalto 1989; SpalterRoth and Hartmann 1990;
Waldfogel 1994, 1997).
9
All employees work in period 1 but remain in the labor force in Period 2 if and only if their second period
utility from working is suciently high to compensate for the greater disutility of working in Period 2. The
outside option in each period is normalized to 0.
To close the model, I also impose a free entry condition on rms in Period 1. Thus, no rm will earn positive
prots in equilibrium.
The timing is as follows:
1. Workers decide how many hours to work in Period 1, h1, knowing their Period 2 valuation of leisure,
θ.
2. Workers draw a normally distributed shock to their productivity, ε. Worker productivity is observed
by all participants at the end of Period 1.
3. Firms decide whether to undertake a training investment of cost c in each worker. If the rm invests,
workers receive rm-specic human capital of α(1) in Period 2. If not, workers' human capital remains
at its initial level, α(0).
4. Workers decide whether or not to participate in the labor market and how many hours to work in
Period 2, h2.
The objective is now to study how a reduction in the Period 2 participation cost of working impacts the
Period 1 signaling strategies of workers and the likelihood of a promotion for a worker of a given type. In
order to analyze this clearly, we assume the following restrictions on θ and γ(M):
R2α(1)2
• θ 1 2A > 2 γ(1)
2
1 R2α(1)
2 R21 2α(0)
2
• 2 γ(0) < θB ≤ 2 γ(1)
1 R
2
2α(0)
2
• θC ≤ 2 γ(0() )2
• 0 < γ(1) < α(0)α(1) γ(0)
3.2 Equilibrium Labor Supply and Firm Investment
Proposition 1 The Perfect Bayesian Equilibrium of this game consists of a unique threshold, y∗, and hours
of work, h∗1(θ; y
∗), such that all workers with productivity y1 ≥ y∗ will be promoted and only those workers
10
will be promoted.
Proof. The game is solved by working backwards.
Worker's Period 2 Problem:
The maximization problem of the worker in Period 2, for each type, given the rm's training decision can
be written as
1
maxR2α(τ)h2 − θh2
h2 2
2
− 1s.t. R2α(τ)h2 θh22 ≥ γ(M)2
The solution to the second (and nal) period problem is straightforward - workers optimally choose Period
2 labor supply as a function of their type and whether they received the rm-based investment.R2α(τ) R21 2α(τ)
h∗
, if θ ≤
2(θ; τ,M) =  θ 2 γ(M)0 , otherwise
Workers with higher marginal costs of labor optimally choose fewer hours of work. Note that although the
participation cost is the same for all types of workers, the second period participation cost of supplying labor
selectively retains the workers who have the lowest marginal costs of working. This is simply due to the fact
that the benet of working is not large enough to compensate for the xed cost of supplying labor when the
marginal cost of working is high. Therefore, the types that anticipate optimally working fewer hours in the
future do not participate in the labor force.
Note that Type C will always participate in Period 2, regardless of the maternity policy. Type A will never
participate in Period 2. Type B is at the extensive margin of labor supply, incentivized to work in Period 2
only when the maternity leave policy is in place.
Firm's Problem:
Firms choose to promote workers if the expected prot to the rm from promoting a worker, conditional on
their rst period productivity, is greater than the cost. Without a maternity leave mandate, the optimal
decision is to promote a worker who produces y1 if and only if
πCP (C|y1) ≥ c,
(1−R2)R2(α(1)2−α(0)2)
where πC = θ is the prot to the rm from promoting a Type C worker, and P (C|y1) isC
the probability the rm assigns to a worker of being Type C, conditional on observing productivity level y1
from the worker.
11
Worker's Period 1 Problem:
We can now characterize the worker's Period 1 decision, h∗1(θ; y), in response to a rm threshold, y. Only
workers who anticipate staying in Period 2 benet from receiving the training for a promotion. Because
the Period 1 signal is noisy, however, no promotion is guaranteed; rather, working additional hours only
increases the likelihood of meeting the rm's {standard. The Period 1 maximization problem of the workerin Period 1 can be written as ( ) }1 1 R2α(0)2
maxR1a0h
2
1 − h1 + AθG y − h∗ 21 + Iθ≤θ̄ ,
h1 2 2 θ M
R2(α(1)21 2 −α(0)
2)
where Aθ = 2 θ is the is the expected increase in surplus from a promotion for a worker of
R2α(0)2
type θ, and θ 1 2M ≡ 2 γ(M) is the cuto type that selects into employment in Period 2.
14 If no maternity
mandate has been enacted at the time of hire, h∗1(θ; y) solves the worker's Period 1 problem when
h∗1(θA; y) = R1α(0)
h∗1(θB ; y) = R1α(0) (
h∗(θ ; y) = R α(0) +A g y − h∗
)
1 C 1 C 1(θC ; y)
√
Note that h∗ ∗ ∗1(θC ; y) > h1(θA; y) = h1(θB ; y), since the second term is strictly positive. σ > σ =
AC
(2eπ)1/4
guarantees a unique solution.15
In words, workers take the rm's promotion standard as given and maximize utility by choosing their Period
1 labor supply. In Period 1, workers who anticipate leaving in Period 2 simply choose hours to maximize
their rst period utility. Workers who anticipate staying employed in Period 2, however, are willing to work
more hours in Period 1 than required by the rst-order condition for a static problem. In other words,
only the Type C workers are willing to take on an additional cost in utility terms in order to convey their
private information to the rm, since only these workers will stay in Period 2 . Although rms only observe
the productivity of workers and do not perfectly observe their choice of hours, workers who are committed
to the labor force in Period 2 can increase the likelihood of the rm recognizing their anticipated Period
2 participation decision by working more in Period 1. Thus, Type C workers, who plan on staying, are
14The Period 1 problem is s[im(plied because of t)he choice of three]types.[(The problem) for continuous ty]pes can be written
R2− 1 2 1 2(α(1)
2−α(0)2) R2
as max 1 2 2h R1α(0)h1 1 h +E I2 1 2 θ θ≤θ | y1 ≥ y +E α(0) IM,1 2 θ θ≤θ | y1 < y , but we use threeM,0
types here for ease of exposition.
15σ > σ ensures that the second order condition holds everywhere. When σ ≤ σ, a unique solution is still guaranteed for
y < R1α(0)+ √
A . As is shown in the Appendix, the equilibrium threshold must be such that y < R1α(0)+ √
A , so only a
σ 2eπ σ 2π
small range of y exists where there may be more than one solution to the rst order condition. However, one of these solutions
is always a maximum, even for low σ ≤ σ. The idea is that when the variance of the signal is suciently low, intermediate
choices of hours may actually be a local minimum, and the best choice may be to aim far past the others, since each additional
hour more clearly separates Type C's from B's to the rm. For a production function where a promotion generates an increase
in salary of $100 per week, the minimum required standard deviation in observed weekly productivity for a unique solution is
3.47. This is quite reasonable, considering even just the measurement error in observed hours.
12
partially separated from the Type B and Type A workers, who anticipate leaving in Period 2.
Equilibrium Threshold:
We now show that, given the optimal Period 1 strategy of workers, the rm's optimal decision is to choose
a threshold for productivity, above which it will promote workers and below which it will not, based on its
ability to distinguish Type C workers from Type B. Let y∗ be such that
πCP (C|y∗) = c ( )
If y∗ is the promotion standard, the probability of a worker of Type C producing at y > y∗1 is g y − h∗1 1(θC ; y∗) .
Hence,
( )
g y − h∗(θ ∗
| ( ) 1 1 ( C ; y ) P (C)P (C y1) = )
g y1 − h∗(θ ; y∗1 C ) P (C) + g y1 − h∗1(θ ∗B ; y ) (P (B) + P (A))
g(y ∗1−h (θB ;y∗))
is increasing in y1, since h
∗
1(θ ; y
∗) > h∗C 1(θB ; y
∗) implies that the ratio 1 is decreasing in y .
g(y1−h∗1(θC ;y∗))
1
Therefore, all workers who produce y1 ≥ y∗ will also be promoted. All workers who produce y ∗1 < y will
not be promoted for the same reason.
The proof of the existence and uniqueness of such the equilibrium is shown in Appendix A. There, I establish
that such a threshold y∗ exists and is unique as long as the cost of rm training, c, is neither so low nor
so high that information about a worker's type is uninformative to the rm's decision. When rm costs are
specie(d such that)the signal is informative to the rm's decision, the equilibrium value of y falls within a
range, ymin, ymax , specied in Appendix A. The quantity P (C|y) is strictly increasing in y for the feasible
range o(f equilibriu)m values of y . Moreover, within the range of equilibrium values of y guaranteed by rm
costs, ymin, ymax , πCP (C|ymin) < c and πCP (C|ymax) > c.
Figure 1 below shows the left- and right-hand sides of the worker's Period 1 rst-order condition, for a given
rm threshold, y. The intersection of the two curves indicates the Type C worker's optimal choice of hours
for a given y. As y increases, the marginal benet curve is shifted to the right. The gures show that the
worker's Period 1 signal is increasing in the threshold y, for y < ymax, with the maximum number of hours
obtained from a Type C worker at y = ymax. Figure 2 illustrates the probability distribution of observed
productivities for Type C workers shifts to the right as the rm threshold increases, from some y0 < ymax
to ymax, with the maximum dierence between the means of the Type B and Type C signals obtained at
y = ymax.
13
Figure 1: Type C Worker's Best Response to Promotion Standard y
(a) Solution to First-Order Condition for y < ymax (b) Solution to First-Order Condition for y > ymax
Figure 2: Response of Distribution of Observed Worker Productivity to Promotion Standard
Shift from y0 to ymax
14
Proposition 2 If a mandate is enacted that decreases the Period 2 participation cost from γ(0) to γ(1), the
Perfect Bayesian Equilibrium of this game consists of a threshold , yM , and hours of work, hM M1 (θ; y ), such
that all workers with productivity y ≥ yM1 will be promoted and only those workers will be promoted. For
rms with a suciently high cost of training, c > c, yM > y∗.
Proof.
The worker's Period 1 and 2 maximization problems remain the same, for a given rm threshold. However,
Type B workers now anticipate staying in Period 2, rather than leaving. Therefore, hM1 (θ; y) solves the
worker's Period 1 maximzation problem when:
hM1 (θA; y) = R1α(0) ( )
hM1 (θB ; y) = R1α(0) +ABg (y − hM1 (θB ; y))
hM1 (θC ; y) = R1α(0) +ACg y − hM1 (θC ; y)
Note that hM1 (θC ; y) > h
M
1 (θB ; y) = h
M
1 (θA; y), since AC > AB .
Because of the reduced participation cost to working, Type B workers act more similarly to the Type C
workers, rather than to the Type As. Now, not only the Type C workers but also the Type B workers are
willing to take on an additional cost in utility terms in order to convey their private information to the rm,
since both types of workers will stay in Period 2 . Type C workers are willing to take on a more costly
signal than Type Bs because of the complementarity between training and Period 2 hours in their salaries.
However, the signal is noisy, and the mean of the Type B workers' Period 1 productivities is now closer to the
mean of the Type Cs. By Bayes' Rule, for any given signal, the employer now assigns a higher probability
of being a Type B relative to a Type C.
Firm's Problem Under the Mandate:
Given the optimal Period 1 strategy of workers, the rm chooses a threshold for productivity, above which
it will promote workers and below which it will not, based now on its reduced ability to distinguish Type C
workers from Type B. With a maternity leave mandate in place, the rm promotes a worker who produces
y1 if and only if
πCP
M (C|y M1) + πBP (B|y1) ≥ c,
(1−R 2 22)R2(α(1) −α(0) )
where πB = θ is the prot to the rm from promoting a Type B worker, and P
M denotes
B
the probability of a worker being a given type if hired after the mandate.
15
Let yM be such that
π PM (C|yM ) + π PMC B (B|yM ) = c
If yM is the promotion standard, then the expected prot from promoting a worker who produces a signal
M | g(y1−h
M (θ′;yM ))
y1, πCP (C y1) + πBP
M (B|y1), is increasing in y1, since the ratio 1M is decreasing in y forg(y M1−h1 (θ;y )) 1
θ′ > θ.16 Therefore, the expected prot of awarding such a worker a promotion is greater than c for all
y1 > y
M , and all workers who produce a signal y1 > y
M are promoted. For the same reason, those who
produce y1 < y
M are not promoted.
For a worker who signals the pre-mandate threshold, y∗, after the enactment of the mandate, π PM (C|y∗C )+
πBP
M (B|y∗) < c since PM (C|y∗) < PM (C|y∗) , PM (B|y∗) < P (C|y∗) when c > c, and πB < πC .
πCP (C|y) + πBP (B|y) is increasing in y for all y in the feasible range of equilibrium thresholds (a range
guaranteed by the rm costs, as specied in Appendix A). Therefore, yM > y∗. The minimum rm cost, c,
and supporting details are found in Appendix A.
Because workers produce a noisy signal, a Type C is more dicult to distinguish from a Type B if hired after
the mandate. The Type B workers have something to gain from a promotion and are now willing to vie for
a promotion. Thus, the return to the signal, in terms of the probability of a promotion, decreases for workers
hired after the enactment of the maternity mandate. In equilibrium, Type Cs will have to work harder in
Period 1 in order to distinguish themselves from the Type Bs, but they will not be willing to increase their
labor supply suciently to maintain the same probability of promotion as they received before the mandate.
Figures 3 and 4 below graph the probability distributions of the observed productivities of each type of worker
on the same axis. Figure 3 illustrates how the incentive for Type B workers to increase their likelihood of
obtaining a promotion reduces the dierence in the means of the Type B and C workers' distributions,
contaminating the Type C signal. The supporting details can be found in Appendix A.
16To see this, note that πCP (C|y1)+πBP (B|y1) = [πCP (C|y1, Bor C) + πBP (B|y1, B or C)]P (B or C|y1). P (B or C|y1)
g(y −h∗(θ′;y))
is increasing in y1, P (C|y1, Bor 1C) is increasing in y1and P (B|y1, Bor C) is increasing in y1, since the ratio 1 isg(y −h∗1 1(θ;y))
decreasing in y1 for θ′ > θ.
16
Figure 3: Response of Equilibrium Threshold to Change in Signaling Strategy of Type B
Workers After Enactment of Mandated Leave
Figure 4: Equilibrium Strategies and Promotion Standard After Enactment of Mandate
We now consider the possibility that rms have the option to provide maternity benets to their workers, even
without a government-mandated policy.
Proposition 3 If rms can choose to take on a cost of providing a benet to reduce participation costs from
γ(0) to γ(1), the equilibrium salaries for workers who are hired prior to the mandate and work h∗1(θ; y
∗) is
s∗, and the equilibrium salaries for workers hired after the mandate and work h∗ M M ∗1(θ; y ) is s < s .
Proof. See Appendix A.
17
3.3 Testable Implications
The model makes predictions about the eect of a worker being hired under a mandated maternity leave
policy on the likelihood of promotions, employment, labor supply, wages. In addition, the model makes
predictions about the return to a wide range of signals about a worker's type, including measures of eort
and performance and the worker's choice of a wage contract.
The rst prediction characterizes the likelihood of promotions. Because the probability of being a Type C
worker is increasing in the signal within the feasible range of equilibrium signals, the employer must raise
the standard for promotions in order to invest in workers and promote protably, as shown in Proposition 2.
The likelihood of receiving a promotion, conditional on being employed in Period 2 then decreases for two
reasons. First, the selection into the labor force of Type B workers in Period 2, who are less likely to produce
a high signal than Type C workers, means the probability of having received a promotion decreases simply
by the retention of workers whose optimal strategy is to work fewer hours in Period 1. The rst cause is due
solely to the retention of workers who have higher marginal valuations of leisure than those who remained
employed in Period 2 before the mandate, and these workers are less likely than Type C workers to produce
a signal above the equilibrium threshold.
A potential concern is that this prediction alone would also be consistent with a model of symmetric informa-
tion. However, the asymmetric information model presented here predicts a lower likelihood of promotions
even for Type C workers, or the workers with the lowest marginal costs of supplying labor. The reason is that
the model predicts the contamination of the Type C workers' signals by the Type B's increased incentive to
obtain a promotion. The equilibrium strategy, when hired after the mandate, is such that the fraction of
Type C workers who meet the higher equilibrium promotion standard is smaller.17
T1. Women of childbearing age hired after the enactment of the mandated maternity leave policy
will be less likely to be promoted, conditional on job tenure, including those women with the
lowest marginal costs of supplying labor.
T2. Women of childbearing age will have higher employment rates after the mandated maternity
leave policy is in place.
T3. Women of childbearing age hired after the mandate will have lower late career labor supply.
∣ ∣
17 ∣ ∣Note that when y < ymax, ∣y − h∗1(θC ; y)∣ is decreasing in y and y < h∗1(θC ; y), so 1−G(y − h∗1(θC ; y)) is decreasing in
y. Also, h∗1(θ ; y
M ) <h∗B 1(θ
∗
B ; y ) < h
∗(θ ; y∗1 C ), so 1−G(y − h∗1(θB ; yM )) < 1−G(y − h∗1(θC ; y∗)).
18
The average number of hours in Period 2 is R2α(τ)θ for workers hired before the mandate, and for those hiredC
after the mandate, the model predicts that the retention of(higher margin)al cost types in Period 2 leads to
fewer hours of work, conditional on employment: R2α(τ) P (C) + P (B) < R2α(τ)P (B)+P (C) θ θ θ . Note that the numberC B C
of Period 2 hours is predicted to decrease both conditional on whether the worker received a promotion and
unconditionally.
T4. Women of childbearing age hired after the mandate will have a lower return to their rst period
signals, in terms of the likelihood of a promotion.
This is shown in Proposition 2. For each signal, the rm's expected prot from promoting a worker is lower
for those hired after the mandate, since the probability of being a Type C for a given signal is lower and
πB < πC . Therefore, workers are less likely to get promoted for a given rst period performance observed
by the rm.
T5. Women of childbearing age hired after the mandate will receive a reduced value of wages and
fringe benets.
T6. The eect of the mandate on those hired after its enactment will have larger eects on rms
whose cost of training is high.
4 Data and Empirical Analysis of the Reforms
4.1 The data
This paper uses data from both the Panel Study of Income Dynamics (PSID) and the Multi-City Study of
Urban Inequality (MCSUI). The PSID provides a long panel on individuals' labor market and childbirth
histories for overlapping cohorts. The data are taken from the Individual File, the Family File, and the
Childbirth and Adoption History File. The sample contains individuals who were classied as either the
Head or Wife of a household in the year of the interview. I only keep individuals between the ages of 16
and 65 in the sample. After keeping only those individuals with at least one year of labor force participation
and employment status data, I have a sample of 14,635 individuals surveyed between the years of 1988 and
2001, of which 55% are women.
The PSID contains key variables that together identify whether a respondent has been promoted since the
time of hire. In each year from 1988 onward, respondents were asked when they had started working for their
current employer. If the respondent had begun working for his current employer in the year prior to the sur-
vey or earlier, he was also asked whether his current position had changed in the previous year. The possible
19
categories for the response to this last question include promotion with higher pay, a major change in duties
but with the same pay, other, and unknown. These questions allow one to determine whether a promotion
has taken place within a rm, rather than simply a wage increase within the same position or a lateral po-
sition change that does not result in wage or salary growth, and at what tenure level the promotion occurred.
The data is then partitioned into employment spells using the method Brown and Light (1992) suggests is
most accurate. An individual is assumed to have started a new spell with an employer when the reported
starting date is after the date of the last survey.18 Tenure for each survey year is determined by using the
time elapsed between the reported date of hire and the survey date. 19An individual is then categorized as
having been promoted since the time of hire if they have been promoted in the previous year at least once
within the employment spell. Because we can only determine from the survey questions whether respondents
have been promoted in the previous year, we lack promotion information on those individuals who were hired
before 1987. Therefore, whether the respondent has been promoted since the time of hire is categorized as
missing if the respondent was hired before 1987. Some promotions will not be identied, if the respondent is
both promoted and begins a new employment spell in less than one year. This is addressed in the empirical
analysis. 20
This cumulative incidence of a promotion is used, as opposed to the promotion hazard rate because of the con-
cern of unobserved heterogeneity. The primary predictable eect of the presence of unobserved heterogeneity
is that the results will be biased toward nding a declining hazard rate with respect to job tenure. Those
with a relatively high probability of being promoted due to characteristics unobserved to the econometrician
will be promoted out of the risk pool more quickly than those with low probabilities of being promoted.
Because the goal is to examine changes in gender dierences in promotions, unobserved heterogeneity could
potentially bias downward or even reverse the sign of the coecient of interest. For example, if unobserved
characteristics (unobservable to the econometrician, but observable to the employer) contributed more to the
18If the exact date of hire is missing, an individual is assumed to have started a new spell when the reported starting year
is greater than or equal to the year of the last survey.
19Brown and Light (1992) has noted that internal inconsistency in tenure within employment spells can be a concern in
the PSID. First, misreporting of tenure data is not as problematic for the survey years used here, 1988-2001, compared to the
earlier years used in Brown and Light (1992). From 1988 onward, the PSID asks for the date of hire with the current employer,
rather than the length of time for which the respondent has been employed with his current employer. While Brown and Light
examine the correlation between the time elapsed between surveys and the reported tenure to determine consistency, the data
here only requires that the reported start dates within an employment spell are consistent. I address reporting errors in start
dates by by using the date of hire reported in the previous survey year if the date of hire diers within an employment spell for
only one survey year.
20The data used here does not suer from the same degree of underidentication of position changes as in McCue (1996)
because we only use data from the 1988-2001 survey years. Prior to 1984, for an observation to be included in the promotion
category, the respondent must have both reported positional tenure below some bound and then have indicated that he had
changed position because he was promoted. From 1988 onward, the sequence of questions was revised. When the respondent
reports having been in the current position for less than one year, he is then asked whether the position changed at any time
in the previous year. Therefore, all changes from the calendar year prior to the survey are captured, as long as the employed
respondents remain with this employer.
20
likelihood of promotion for men than for women, the negative duration dependence of the hazard rate would
be overestimated more for men than for women. In other words, the hazard rate would be biased toward
declining more steeply for men than for women. In an extreme case, where observable characteristics were
the only factors aecting the employer's promotion decision for women, but unobservable characteristics are
taken into account for men, such as in a case of statistical discrimination against women on the basis of
observable characteristics, it might, in fact, appear that women have a higher probability of being promoted
than men within a given period, especially at high levels of tenure. Because the dierential use of observable
characteristics in a promotion decision is central to the context addressed in the model, the cumulative
incidence of promotion is better able to capture the eect of interest. There may still be a bias that the
cumulative incidence of promotion is increasing less steeply for men than for women, but the measure that
is relevant is the relative likelihood of promotion for a given job tenure is aected by the policy change. An
eect of the policy change on discrimination and the use of information may not be captured by the hazard
rate, while it may be captured by the cumulative incidence, since the sign of that eect will not be aected
by the bias.
The second dataset I use, the Multi-City Study of Urban Inequality, is a cross-sectional survey designed to
broaden understanding of labor market dynamics and comprises data for two surveys: a survey of households
and a survey of employers. The employer survey records detailed information from 3510 rms in four major
metropolitan areas: Boston, Chicago, Atlanta, and Los Angeles. In particular, the survey data contains a
substantial number of questions about each rm's most recently hired worker, an these questions form the
basis for the empirical analysis. Several unique features of the data are important to the analysis. First,
the dataset contains a wide variety of rm characteristics in addition to standard information on occupation
and industry, enabling us to control for these variables while still relying on a broad, representative sample,
as opposed to a single-rm or a narrowly-dened population. For example, the survey includes an indicator
for nonprot status. The data shows that women are more heavily represented in the nonprot than in
the for-prot sector; the fraction working in the for-prot sector is 74 percent for women versus 87 percent
for men. Recent empirical work by DeVaro and Samuelson (2005) documents a pronounced dierence in
promotion rates between for-prot and nonprot organizations, with promotions less likely in nonprots.
These considerations suggest that nonprot status should be controlled in analyses of gender dierences in
promotion rates. In addition to nonprot status, I control for industry, establishment size, number of sites of
operation, whether or not the rm is a franchise, and the percentage of workers covered by collective bargain-
ing agreements. While the PSID is a nationally-representative panel with detailed worker characteristics, it is
thin on rm characteristics. Second, because the main dataset used is information asked of the employer, all
worker characteristics are based on employer knowledge or perceived knowledge, which is more relevant for an
examination of employer expectations than the actual worker characteristics, such as age. Furthermore, the
data includes include job-specic worker performance ratings, allowing us to control for performance more
precisely than commonly-used skill indicators such as educational attainment or tenure would allow. Impor-
21
tantly, it also contains employer-based performance evaluations of all workers, so the performance measure
used in the analysis is normalized based on the distribution of employer ratings. Therefore, we can also view
performance as a costly signal received by the employer. The dataset also contains detailed information not
only on wages, but on a variety of fringe benets provided to employees. Lastly, we have detailed information
on rm practices, and in particular, a measure of the cost of rm training, both formal and informal training.
As an additional resource, the MCSUI data also contain a number of variables measuring wages and the
wage growth attached to promotions. Four variables pertain to the wages of the most recently hired worker:
starting wage, current wage at the time of the survey, wage the employee is expected to receive if promoted,
and highest wage an employee could attain without a promotion in the same position as the most recently
hired employee. These variables are useful for examining whether the denition of a promotion has simply
changed over time and what eect a promotion has on wage growth.
After keeping in the dataset only rms whose last hire was aged 16 to 65, I am left with a sample of 3,510
rms, of which 1,774 rms' most recent employee hired was female. The majority of rms indicated that
their most recent employee hired was hired in 1993 (48.12% of rms hired their most recently hired employee
in 1993) , so this dataset is particularly useful because most of the variation in whether the employee was
hired before or after the enactment of the FMLA has little variation in the year of hire, but rather, the
information set of the employer.
4.2 Empirical Analysis
The Family and Medical Leave Act provides the distinct opportunity to study the impact of increasing access
to job-protected leave on the employment, human capital accumulation, labor supply, and wages of workers
whose labor force participation is most likely aected by the job-protected leave. Using the PSID, I exploit
variation in state legislation on family and medical leave prior to and since the 1993 federal legislation on
family leave. The variation in state legislation is summarized in Appendix Table 1. The U.S. Department of
Labor (DOL) provides a comparison of the FMLA and the family and medical leave legislation of 11 states
and the District of Columbia.21 In addition to the 11 states and the District of Columbia, I consider the
state legislation passed by Massachusetts, Montana, North Carolina, and Tennessee, as considered in the leg-
islative dataset constructed by Stutts and Heiland (2006), since these states also passed legislation on family
and medical leave prior to and since the federal mandate. There is some amount of variation in the exact
terms of the legislation for each state.22 I focus on the impact of increasing access to job-protected leave for
21See http://www.dol.gov/esa/programs/whd/state/fmla/index.htm. States included (as of 25 February 2006) are California,
Connecticut, Hawaii, Maine, Minnesota, New Jersey, Oregon, Rhode Island, Vermont, Washington, Wisconsin, and the District
of Columbia.
22The DOL breaks down the FMLA into 14 elements, including the types of employers covered by the legislation, employee
eligibility, leave amount, type of leave allowed, whether there is a key employee exception, whether health benets are maintained
during leave, and whether medical certication is required.
22
the birth and care of a newborn child on workers hired under this regime. I dene states as nonexperimental
or control states if state legislation before August of 1993 required job-protected leave for private sector
employees for the birth and care of a newborn child, for at least eight weeks.23 I consider those that did
not to be treatment states. These states vary in their geographic region and are broadly representative
of the country as a whole, as shown in Table 2. Observations in a few states updated their laws after the
enactment of the FMLA to become signicantly more expansive than the federal law. Because these states
varied greatly in what aspects of their legislation was designed to surpass the requirements of the federal law,
observations in the years in which state legislation expanded beyond the federal requirements are insucient
to provide accurate data for such a quasi-experiment; thus, they will not be used for causal interpretation.24
Testable Implication T1: Likelihood of Promotions
To examine the impact of the being hired after the enactment of the Family and Medical Leave Act on the
likelihood of promotion, I estimate the following equation for an individual i employed with rm j, in state
s, and in year t :
promotedi,j,s,t = β1(HiredAfteri,j · FemaleUnder40i,t · TreatmentStates) (1)
+ β2(FemaleUnder40i,t · TreatmentStates) + β3(HiredAfteri,j · TreatmentStates)
+ β4(HiredAfteri,j · FemaleUnder40i,t) + β5HiredAfteri,j + β6FemaleUnder40i,t
+ β7TreatmentStates + β8Xi,t + β9Zj + λt + cs + fi + δi,t,s + εi,j,s,t
The vector X contains a set of controls for sex, age, age squared, tenure, tenure squared, level of education
attained, marital status, marital status interacted with sex, marital status interacted with FemaleUnder40,
and a dummy for nonwhite. The vector Z contains a set of controls for the characteristics of the rm and the
23Hawaii is considered a treatment state because its state law would not apply to private sector employees until January 1,
1994, after the FMLA was passed. Montana is considered a treatment state because its law, passed in 1975, much more closely
mirrored the Pregnancy Discrimination Act of 1979, by making it illegal to terminate a woman's employment because she was
pregnant and required that employers allow the women on maternity leave any compensation or other benets for which they
would have been eligible under any other form of disability leave. After the Pregnancy Discrimination Act of 1979 amended
Title VII of the Civil Rights Act of 1964, pregnancy became a protected class under discrimination law, and discrimination
in employment on the basis of pregnancy was prohibited by law in every state. This amendment included the requirement
to provide disability benets to women "disabled by pregnancy" the same amount of compensation and for the same length
of time as they provide for any other type of disability.While Montana's 1975 legislation also mandated that employers allow
women a reasonable leave of absence for pregnancy, the terms and length of this leave were not specied and the exact terms
were left to the discretion of the employer and employee. Therefore, I do not include it in the group of states that mandated
job-protected leave for the birth and care of a newborn child.
24The North Carolina General Assembly, for example, passed two acts in relation to family leave, one, in 1993 and, the other,
in 1997. Unlike most states being examined in this paper, North Carolina's original 1993 legislation did not mirror or even
resemble the federal FMLA. North Carolina's Chapter 509, Section 1 (sections 2 and 3 of Chapter 509 addressed issues outside
of family leave) allowed parents in the State to take protected leave from work to be involved with their child's school activities.
Other states whose legislation eventually became signicantly more progressive than the federal legislation were Maine, in 1997,
New Jersey, in 1995, and Oregon, in 1995.
23
job. λt denotes year xed eects, cs, state xed eects, and fi, individual xed eects. δi,t,s denotes a set
of state-specic year xed eects and year xed eects specic to the demographic group of interest, women
under the age of 40. The dependent variable promoted is constructed to represent whether the respondent
has ever been promoted in the duration of the employment spell. I consider the likelihood of promotion for
respondents hired under two dierent maternity leave regimes. HiredAfter is a dummy variable for whether
the respondent was hired after August of 1993, when the FMLA was enacted. TreatmentState is a dummy
variable equal to 1 for the states that did not already have state legislated job-protected parental leave.
The coecient β1 captures the eect of interest. It captures the eect of being hired after the enactment of
the FMLA on the likelihood of promotion for women under the age of 40, relative to men and women over
the age of 40, in the treatment states, relative to the control states, the states that had already mandated
some form of job-protected leave. β7 captures the time-invariant characteristics of the experimental states,
β6 captures the existing dierential in the likelihood of promotion for a woman under the age of 40,
25 β5 con-
trols for the changes over time in the likelihood of promotions for contracts and employment spells initiated
at later dates, β4 captures the changes over time for women under the age of 40, in particular, β3, changes
over time in likelihood of promotions specic to the experimental states, and β2 captures the time-invariant
likelihood of promotion for women under the age of 40 specic to the experimental states.
The rst row of Table 1 presents the estimates of the third-level interaction from Equation 1, β1, using xed-
eect OLS regressions for dierent specications. Column 1 is the baseline specication, which includes the
set of controls contained in vector X as well as year xed-eects, state xed-eect and individual xed-eects.
Column 2 includes characteristics of the employment position: whether the respondent works for a private,
non-goverment company and whether the job is covered by a union contract. Columns 3 through 5 include
detailed occupation and industry controls. Column 3 controls for 12 occupational categories using the 12
standard occupational divisions used in the 1970 census occupational classication system. The occupations
in this system are organized into several large groupings of roughly descending socioeconomic status. Column
4 uses a dierent division of occupations, with 12 occupational categories, as described earlier, and divisions
made in order to capture potential selection into more child-friendly occupations, based on a literature that
measures the rates of human capital depreciation and wage growth across occupations and how these choices
are aected by planned fertility.26 Column 5 uses an even ner division of occupational categories, control-
ling for 24 standard divisions, as used in the census occupational classication system, as well as 8 standard
industry level divisions. The coecient of interest, β1, responds little to the changes in division and the
increasing neness of the division of occupational categories.
25Note that with individual xed eects, when β6 is negative, this captures the increase in the likelihood of promotions for
the same female individual when she reaches the age of 40.
26See, in particular Adda, Dustmann, and Stevens (2011)
24
Appendix C presents a set of robustness checks on these results. The results shown in Figure 13 demonstrate
that while women under the age of 40 face a lower likelihood of promotion when hired after the enactment
of the FMLA, women over the age of 40 are unaected in comparison to men. I also show that women under
the age of 40 are unaected by the policy change in control states. Furthermore, while table 1 compared
women under the age of 40 to a control group comprised of men and women over the age of 40, I show in
Tables 11 and 12 that using either of two control additional groups yields similar results: (i) men under the
age of 40 only or (ii) women over the age of 40 only. Appendix Tables 15-17 include linear time trends to
conrm that dierential trends in the year of hire do not drive the results. The potential concern is that
if treatment states already had more steeply declining promotion rates for young women relative to men,
with respect to the date hired, than did control states, then the estimates from equation (1) would pick up a
pre-existing trend. I show that the results are robust to the inclusions of state-specic, demographic-specic,
and state-demographic-specic linear time trends in the year of hire. While women under the age of 40
do witness a slight increase in the likelihood of being promoted over time relative to men, this trend exists
for all women and is not specic to women under the age of 40. Thus, the time trends reveal that the
enactment of the FMLA retards the progress in the growth over time of women's likelihood of promotion
relative to men, for women of childbearing age. The inclusion of time trends has little eect on the estimates
when the control group is made up exclusively of women over the age of 40. Therefore, there is no evidence
of a systematic bias from omitted factors that aect fertile women dierentially over time in treatment states.
However, omitting state- or demographic-specic factors that inuence the likelihood of promotions over time
only bias the estimated eect of being hired after the FMLA if there is a systematic relationship between
the trend in promotion rates and the adoption of state legislation on family and medical leave. Inclusion
of state-demographic-specic trends may be too restrictive a specication, if there is a dynamic component
to the eects of the policy, since the year trend may control for exactly the eect we want to observe27.
Therefore, I also examine individual year-treatment eects, as opposed to year trends, in order to observe
dynamic component to the treatment eect. Figure 5 shows the results of year-treatment coecients and the
95 percent condence intervals of the estimated coecients. The estimated eect is relative to the those hired
in the pre-FMLA 1993 period (January - July 1993). The gure also shows the estimates of year-treatment
coecients when the control group consists of men under the age of 40 only or older women only, and the
results are robust to the choice of control group.
Next, I exploit the variation in fecundity for women by age and the large decrease in fecundity for women
at age 40 in examining whether the gender gap in promotions was widened for women with the highest
likelihood of having children. Figure 6 shows the estimates of age-treatment coecients and the 95 percent
condence intervals of the estimated coecients for the change in the likelihood of promotions for women
27This argument was made by Wolfers (2006) in examining the dynamic eect of the introduction of unilateral divorce laws
on divorce rates
25
relative to men. For each of the 5-year age bins, being hired after the FMLA widens the gender gap in
promotions, but only for those women of childbearing age. Among each of the older age groups, women
hired after the FMLA face no signicant eect in their likelihood of promotion. Panels (a) and (b) of Figure
6 demonstrate that the eect on each group of younger women is signicant and negative in treatment states,
but there is no signicant eect on any of the groups of younger women in control states.
Finally, the federal FMLA requires employees to have worked more than 1,250 hours over the past 12 months
before becoming eligible. An additional estimation exploits the variation in eligibility for the FMLA across
jobs, and, therefore, the variation in likely future eligibility for the FMLA. 28
promotedi,j,s,t = β1(HiredAfteri,j · EligFemaleUnder40i,t · TreatmentStates) (2)
+ β2(EligFemaleUnder40i,t · TreatmentStates) + β3(HiredAfteri,j · TreatmentStates)
+ β4(HiredAfteri,j · EligFemaleUnder40i,t) + β5EligFemaleUnder40i,t
+ β6HiredAfteri,j + β7TreatmentStates + β8Xi,t + β9Zj + λt + cs + fi + δi,t,s + εi,j,s,t
The estimates of the Equation (2), using xed-eect OLS regressions with the specications described earlier,
are presented in Table 2.
One potential concern is that the passage of the FMLA not only had an eect on those hired after the mandate
through adverse selection (unobserved heterogeneity preexists the contractual relationship and constrains its
form), but also on the behavior of a given type of worker (whereby behavior directly responds to the incentive
structure created by the contract). The latter is the eect conventionally studied in analyses of mandated
maternity and family leave policies, where individuals in a given employment relationship potentially adjust
future labor supply and fertility decisions. These individuals experience a change in the likelihood of promo-
tions in the survey years after the FMLA was passed, due to a direct eect of the mandate on the incentives
for a given type of worker. However, the information that the employer has about the worker's type is xed,
implicitly, and only the mapping from the type to the behavior changes due to a change in incentives. In
this paper, we are primarily concerned with the existence of an eect due to adverse selection. However,
ensuring that the eect that we observe in the data is not driven by the incentive eect of the policy, for a
28The federal FMLA also requires employees to have been employed for a total of 12 months and to have worked more than
1,250 hours over the past 12 months before becoming eligible. I did not use the tenure requirement in determining eligibility
because in this paper, I test the hypothesis that costly employer investment is reduced in anticipation of women's selection
into the labor force (or the selective retention of female employees) in the future, when the investment will pay o. The model
relies on a change in the average group behavior among those who are mostly likely to have reduced participation costs in
the future. Whether an employee currently has at least one year of tenure is not important for determining the risk of future
FMLA eligibility because an employee with less than one year of tenure may be eligible in the future, by the time children are
conceived. However, I assume that jobs that do not require more than 1,250 hours of work over the course of the year are likely
to continue to require fewer than 1,250 hours of work per year in the future. Although the stylized model gives workers a choice
of hours in the rst period, workers who choose job oers with hours below the minimum eligibility requirements for coverage
by FMLA are selecting into employment positions that are not aected by the FMLA.
26
given type, is a crucial task. I show that the change in the likelihood of promotion is not driven by the eect
of the mandate on a given type of worker. In the results shown in Table 1, we include year xed eects,
state-specic year xed eects, and year xed eects specic to the demographic group of interest, women
under the age of 40, but also state-demographic-specic year xed eects. This specication allows for the
possibility that the FMLA aected the likelihood of promotion over the time period in which it was enacted,
through a direct eect on the behavior of those already in employment relationships. Appendix Table 13
presents the estimates from Equation (1), but the panel data is restricted to observations from the period
after the Family Medical Leave Act was passed, still exploiting the variation in whether the respondent was
hired before or after its enactment. The restricted sample isolates the selection eect from the incentive
eect, and the estimates are consistent with those from the unrestricted sample. Appendix Tables 18 and 19
and Figure 15 address the concern about the potential impact of the FMLA on a given type and demonstrate
that there appears to be little eect of the introduction of the FMLA on the population hired before its
enactment who had already entered into their employment relationships.
In Appendix Table 14, I examine the eect of being hired after the enactment of the FMLA on the likelihood
of promotion for women under the age of 40, excluding the state-specic, demographic-specic, and state-
demographic-specic year xed-eects. This allows us to investigate the eect of the FMLA on individuals
hired before the FMLA, including a potential direct incentive eect on individuals of a given type once the
mandate is in place. We see from the rst row of Table 14 that the total change in the likelihood of promotion
is still signicantly negative for women under the age of 40 hired after the FMLA in treatment states, in
spite of a possible improvement in promotion probabilities for a given type in the years following the mandate.
While the results in Table 1 are striking, they are also compatible with an alternative model of symmetric
information, in which women who are more likely to benet from the mandate select into the labor force, but
full information about their types is known to the employer. Therefore, I examine the eect of being hired
after the enactment of the FMLA on women under 40 who do not benet from the mandate and are not
on the extensive margin of participation due to its enactment: women who never have children. I estimate
equation (1), but I include two additional demographic groups of interest: women under 40 who never have
children and women under 40 who eventually have children. By also including their respective interaction
terms29, I examine whether the mandate aects the likelihood of promotions for either of these two groups
dierently from other women under the age of 40. Columns (1) and (2) of Table 3 show that the the eect on
the likelihood of promotion is not signicantly dierent for women who never have children. In other words,
29Columns (1) and (2) of Table 3 estimates the following equation: promotedi,j,s,t =
(β1,nFemaleUnder40NeverHasKidsi,t + β1,eFemaleUnder40EverHasKidsi,t + β1FemaleUnder40) ·
HiredAfteri,j · TreatmentStates + (β2,nFemaleUnder40NeverHasKidsi,t + β2,eFemaleUnder40EverHasKidsi,t +
β2FemaleUnder40) · TreatmentStates + (β3,nFemaleUnder40NeverHasKidsi,t + β3,eFemaleUnder40EverHasKidsi,t +
β3FemaleUnder40) · HiredAfteri,j + β4,nFemaleUnder40NeverHasKidsi,t + β4,eFemaleUnder40EverHasKidsi,t +
β4FemaleUnder40EverHasKidsi,t + β5HiredAfteri,j + β6(HiredAfteri,j · TreatmentStates) + β7TreatmentStates +
β8Xi,t + β9Zj + λt + cs + fi + εi,j,s,t
27
the likelihood of promotions is decreased for all women under the age of 40, including the women who never
have children. Column (2) is restricted to a sample of women only, so the control group is comprised only of
women over the age of 40. The results here demonstrate that the likelihood of promotions decreases for all
women who have the potential to have children in the future, regardless of their realized fertility. Column
(3) is also restricted to a sample of women only, but here the treatment group is restricted to only women
under the age of 40 who never have children. The results show that there is, indeed, a reduced likelihood of
promotions for women under the age of 40 who never have children.
Testable Implication T2: Employment
To examine the impact of the being hired after the enactment of the Family and Medical Leave Act on the
likelihood of promotion, I estimate the following equation for an individual i employed with rm j, in state
s, and in year t :
employedi,s,t = β1(Aftert · FemaleUnder40i,t · TreatmentStates) (3)
+ β2(FemaleUnder40i,t · TreatmentStates) + β3(Aftert · TreatmentStates)
+ β4(Aftert · FemaleUnder40i,t) + β5Aftert + β6FemaleUnder40i,t
+ β7TreatmentStates + β8Xi,t + λt + cs + fi + εi,j,s,t
Here, we examine the likelihood of employment in the periods before and after the enactment of the FMLA.
The model implies that it is only the participation constraint that aects the employment, so here, the
information that the employer has about the worker does not aect the worker's decision to continue par-
ticipating in the employment relationship. Table 4 shows that employment among women of childbearing
age increases by 3 percent in treatment states relative to control states, after the enactment of the FMLA.
Column (1) of Table 4 controls for state xed eects, year xed eects, and individual xed eects, while
Column (2) includes state-specic and demographic-specic year xed eects. Figure 8 shows the estimates
of the year-treatment coecients from 1988 to 2001 and conrms that the rise in employment after the
FMLA was not due to pre-existing trends.
Testable Implication T3: Hours
To examine the impact of the being hired after the enactment of the Family and Medical Leave Act on the
realized labor supply of women with children, I estimate the following equation for an individual i employed
with rm j, in state s, and in year t :
28
hoursi,j,s,t = β1(HiredAfteri,j · FemaleWithKidsi,t · TreatmentStates) (4)
+ β2(FemaleWithKidsi,t · TreatmentStates) + β3(HiredAfteri,j · TreatmentStates)
+ β4(HiredAfteri,j · FemaleWithKidsi,t) + β5HiredAfteri,j + β6FemaleWithKidsi,t
+ β7TreatmentStates + β8Xi,t + λt + cs + fi + δi,t,s + εi,j,s,t
This equation is analagous to Equation (1), except that the demographic group of interest is no longer those
who are at risk of having children in the future. Here, we examine the labor supply eects on those for whom
the risk to the employer was realized. Table 5 shows the estimates of equation (4). As in the literature,
women with children do work fewer hours per week. However, the results indicate that there is a decrease in
late-career weekly hours of work among the women hired after the mandate. In other words, fertility is not
the only determining factor of labor supply, and the types of women with children who are still employed at
high levels of tenure work fewer hours than the types who were hired before the mandate. Employer beliefs
based on the specied model are rational.
Testable Implication T4: Return to Signals
In order to test whether information asymmetry and the informativity of a signal is indeed aected by the
introduction of the FMLA, I now turn to the second dataset, the Multi-City Study of Urban Inequality to
examine the impact of the Family and Medical Leave Act on the return to costly signals. This dataset is
unique in that it collects detailed information from each rm on performance evaluations of its employees.
The prediction of the model is that the signal of those women who anticipate working a high number of
hours would be contaminated by the signaling of those women who now anticipate working in the future but
anticipate working a low number of hours. Table demonstrates that these performance evaluations are, in
fact, predictive of the likelihood that a worker receives a promotion. In fact, there is a nonlinear increase
in the likelihood of promotion with performance. For women, a performance evaluation above the 75th
percentile leads to a 7 percent increase in the likelihood of promotion. The results in Column (1) of Table
7 demonstrate that at each level of performance, the marginal return to a higher performance level is lower
for fertile women hired after the mandate, while the return to signaling is unchanged for women over the
age of 40. This is a direct implication of the main mechanism of the model: signal contamination from the
anticipated future selection of women whose optimal choice of hours is lower. As a result of the increased
participation of low hours types, the information conveyed from a given signal is lower. Here, we also nd
that the return to the performance level that yielded the greatest increase in the likelihood of promotions,
the 75th percentile, is also lower for fertile women hired after the mandate. This is shown in Columns (2)-(5)
of Table 7. Lastly, we see in Column (6) that the marginal return of performing above 75th percentile is
29
reduced for women of childbearing age.
Testable Implication T5: Eect on Wages and Fringe Benets
The MCSUI is unique in that it contains detailed information on each rm's last hire and the wages and
fringe benets oered to an employee of this position. I examine the eect of being hired after the enact-
ment of the FMLA on these various forms of compensation. Among starting wage, starting salary, health
insurance provision, family health insurance provision, dental coverage, pension plan contributions, bonuses,
and other benets given to employees of the position in question, there is no eect on women of childbearing
age relative to older women. However, among two particular benets, exible hours and day care provision,
we see a large reduction in the likelihood of those benets being provided when the last hire was a female
under the age of 40, relative to older women. This provides evidence that the positions for which women
under the age of 40 are hired, after the enactment of the FMLA, provide lower compensation in terms of
fringe benets, conditional on starting salary and starting wage.
Testable Implication T6: Eect on Firms with High Costs of Training
An important implication of the model is that not all rms raise the standard of promotions when the
retention of women with high marginal costs of labor is increased. In particular, the cost to the rm of
training a worker in the early career must be high enough that the rm suers a loss when investing training
in a worker who will optimally choose to work part-time in the future. If this were not the case, the prot
to the rm of training a worker who will work part-time in the future would be lower, but as long as the
prot were positive, the expected prot to the rm increases under the mandate because the rm retains the
worker and reaps some prot in the future. However, it is only when the cost of training is suciently high
that the rm optimally chooses a threshold to separate the full-time workers from the part-time, rather than
the stayers from the leavers. When the information on hours, conditional on future labor force participation
is more informative to the rm's promotion decision than the participation decision, then the rm raises its
promotion standard. In other words, only when rm costs are suciently high that retention of part-time
workers does not benet the rm does the introduction of a maternity leave mandate have a negative eect
on the likelihood of promotion for women of childbearing age, conditional on observing their rst-period
behavior. Table 9 shows the estimated coecients from the following regression:
promotedi,j,t = β1(HiredAfteri,j · FemaleUnder40i ·HighTrainingj) (5)
+ β2(FemaleUnder40i ·HighTrainingj) + β3(HiredAfteri,j ·HighTrainingj)
+ β4(HiredAfteri,j · FemaleUnder40i) + β5HiredAfteri,j + β6FemaleUnder40i
+ β7HighTrainingj + β8Xi + λt + cc + εi,j,t
30
The rst row of Table 9 demonstrates that the reduction in the likelihood of promotions among women under
the age of 40 is signicantly larger among rms with high costs of training, whereas there is no signicant
eect among rms with low costs of rm-sponsored training. This is a particularly important empirical
nding because it refutes the alternative explanation that the decrease in likelihood of promotion is entirely
due to selection. If this were the case, we should not expect to see a larger eect among rms with a high
cost of training or mentorship.
5 Concluding Remarks
This paper sheds light on the hetereogeneous welfare consequences of mandated maternity leave policies. I
show that maternity leave mandates can both increase employment and, yet, decrease promotion rates for
women. I then show, in two dierent datasets, that women who were hired after the enactment of a minimal
maternity leave mandate were ve percent more likely to be employed but eight percent less likely to be
promoted than those who were hired before its enactment. Moreover, I nd a series of empirical results that
would be dicult to reconcile with alternative explanations: the likelihood of promotions decreases even
among women who do not benet from the mandate, women who never have children, the labor supply
among women with children is reduced by more than ten percent, even among those who were promoted,
and that the widening of the gender gap in promotions is largest among rms where the cost of training is
high. Finally, using a series of measures of signals of future productivity, including job performance, early
career hours of work, and importantly, the selection of wage contracts, I nd that the return to this signal,
in terms of the likelihood of promotion, decreased for women hired after the enactment of a maternity leave
mandate. Moreover, it decreases for women under the age of 40 alone. This set of evidence is dicult to
reconcile with an explanation based on purely on selection and strongly suggests that information asymmetry
between rms and workers is at the heart of the problem.
An innovative feature of the model is that it is precisely because the policy lengthens the expected ca-
reer tenures of some women that the information problem is exacerbated. Women who expect to completely
separate from their employers upon having children have little incentive to invest or take on costly signals
in order to advance within the rm. Because the maternity policy reduces participation costs for women
who would have otherwise left the labor force, such workers are retained. These workers have an additional
incentive, under a maternity policy, to take on costs in order to increase the likelihood of obtaining a pro-
motion. However, these workers are also more likely to work part-time than those who were willing to vie
for a promotion before a maternity policy. The increased incentive for the less committed types to obtain
a promotion as well muddies the signals from the most career-oriented workers and increases the cost of
sorting workers.
The empirical results are consistent with these predictions and have powerful welfare implications for such
31
policies. Because of the limited ability of rms to sort workers and allocate training eciently, mandated
maternity leave policies can result in a lower rate of human capital accumulation and wage growth for all
women over the course of their lifecycle, and fewer advancement opportunities within the rm. While such
policies may enhance welfare for women who would not have otherwise participated in the labor market,
the cost of increasing the labor force participation of women is borne, in part, by the most career-oriented
women, through a loss of their human capital accumulation and lower wage growth over the course of their
lifecycles. Studies examining only the short-term impact of maternity leave policies may not fully capture
the welfare consequences of maternity leave policies, and empirical work focused only on the eect due to
changes in the direct incentives for workers' choice of human capital and occupation miss a crucial welfare
consequence of such policies.
References
[1] Abbring, Jaap H., Pierre-Andre Chiappori, James Heckman, and Jean Pinquet (2003), Adverse Selec-
tion and Moral Hazard in Health Insurance: Can Dynamic Data Help to Distinguish? Journal of the
European Economic Association, 1, 2-3, 512-521.
[2] Adda, Jerome, Christian Dustman and Katrien Stevens (2011), The Career Costs of Children, IZA
Discussion Paper No. 6201.
[3] Acemoglu, Daron and Jorn-Steen Pischke (1998), Why Do Firms Train? Theory and Evidence, The
Quarterly Journal of Economics, 113, 1, 79-119.
[4] Acemoglu, Daron and Jorn-Steen Piscke (1999), Beyond Becker: Training in Imperfect Labour Mar-
kets, The Economic Journal, 109, 453, F112-F142.
[5] Albanesi, Stefania and Claudia Olivetti (2008), Home Production, Market Production and the Gen-
derWage Gap: Incentives and Expectations, Review of Economic Dynamics, 12, 1, 80107.
[6] Altonji, Joseph and Rebecca Blank (1999), Race and Gender in the Labor Market, in Handbook of
Labor Economics edited by Orley Ashenfelter and Albert Rees, 3, 31433259.
[7] Arrow, Kenneth J. (1973), The Theory of Discrimination in Discrimination in Labor Markets, edited
by Orley Ashenfelter and Albert Rees. Princeton: Princeton University Press, 3-33.
[8] Autor, David (2001), Why Do Temporary Help Firms Provide Free General Skills Training? Quarterly
Journal of Economics, 116, 4, 1409-1448.
[9] Attanasio, Orazio, Hamish Low and Virginia Sanchez-Marcos (2008), Explaining Changes in Female
Labor Supply in a Life-Cycle Model, American Economic Review, 98, 4, 15171552.
32
[10] Barron, John, Dan Black and Mark Lowenstein (1993), Gender Dierences in Training, Capital and
Wages, Journal of Human Resources, 28, 2, 343364.
[11] Becker, Gary (1962), Investment in Human Capital: A Theoretical Analysis, Journal of Political
Economy, 70, 5, 9-49.
[12] Becker, Gary (1971), The Economics of Discrimination, 2nd ed, Chicago: The University of Chicago
Press.
[13] Becker, Gary (1985), Human Capital, Eort, and the Sexual Division of Labor, Journal of Labor
Economics, 3, 1, S33-S58.
[14] Blau, Francine D. and Jed DeVaro (2006), New Evidence on Gender Dierences in Promotion Rates:
An Empirical Analysis of a Sample of New Hires, NBER Working Paper 12321, National Bureau of
Economic Research.
[15] Blau, Francine D. and Lawrence M. Kahn (1997), Swimming Upstream: Trends in the Gender Wage
Dierential in the 1980's, Journal of Labor Economics, 15, 1, 1-42.
[16] Blau, Francine D. and Lawrence M. Kahn (2006), The U.S. Gender Pay Gap in the 1990's; Slowing
Convergence, Industrial and Labor Relations Review, 60, 1, 45-66.
[17] Blau, Francine D. and Lawrence M. Kahn (2013), "Female Labor Supply: Why is the US Falling
Behind?" NBER Working Paper 18702, National Bureau of Economic Research
[18] Blau, Francine D. and Lawrence M. Kahn (2013), "Female Labor Supply: Why is the US Falling
Behind?" American Economic Review, 103, 3, 251-256.
[19] Brown, James and Audrey Light (1992), Interpreting Panel Data on Job Tenure, Journal of Labor
Economics, 10, 3, 219-257.
[20] Chiappori, Pierre-Andre and Bernard Salanie (2002), Testing Contract Theory: Some Recent Work,
CESIFO Working Paper No. 738.
[21] Chiappori, Pierre-Andre and Bernard Salanie (2000), Testing for Asymmetric Information in Insurance
Markets, Journal of Political Economy, 108, 1, 56-78.
[22] Coate, Stephen and Glenn C. Loury (1993), Will Armative Action Policies Eliminate Negative Stereo-
types? The American Economic Review, 83, 5, 1220-1240.
[23] Dalto, Guy (1989), A Structural Approach to Women's Hometime and Experience-Earnings Proles:
Maternity Leave and Public Policy, Population Research and Policy Review, 8, 3, 247-266.
33
[24] Eckstein, Zvi and Kenneth Wolpin (1989), Dynamic Labor Force Participation of Married Women and
Endogenous Work Experience, The Review of Economic Studies, 56, 3, 375-390.
[25] Erosa, Andres, Luisa Fuster and Diego Restuccia (2002), Fertility Decisions and Gender Dierences in
Labor Turnover, Employment, and Wages, Review of Economic Dynamics, 5, 4, 856-891.
[26] Erosa, Andres, Luisa Fuster and Diego Restuccia (2010), A General Equilibrium Analysis of Parental
Leave Policies, Review of Economic Dynamics, 13, 4, 742-758.
[27] Francesconi, Marco (2002), A Joint Dynamic Model of Fertility and Work of Married Women, Journal
of Labor Economics, 20, 2, 336-380.
[28] Gayle, George-Levi and Limor Golan (2012), Estimating a Dynamic Adverse-Selection Model: Labour-
Force Experience and the Changing Gender Earnings Gap 1968-1997, Review of Economic Studies, 79,
1, 227-267.
[29] Gayle, George-Levi and Robert Miller (2006), Life-Cycle Fertility Behavior and Human Capital Accu-
mulation, 2006 Meeting Papers 784, Society for Economic Dynamics.
[30] Gibbons, Robert and Lawrence Katz (1991), Layos and Lemons, Journal of Labor Economics, 9, 4,
351-380.
[31] Gruber, Jonathan (1994), The Incidence of Mandated Maternity Benets, The American Economic
Review, 84, 3, 622-641.
[32] Kahn, Lisa (2013), Asymmetric Information between Employers, American Economic Journal: Ap-
plied Economics, 5, 4, 165-205.
[33] Kahn, Lisa and Fabian Lange (2014), Employer Learning, Productivity and the Earnings Distribution:
Evidence from Performance Measures, Review of Economic Studies, 81, 4, 1575-1613.
[34] Hart, Oliver and John Moore (1988), Incomplete Contracts and Renegotiation, Econometrica, 56, 4,
755-785.
[35] Hall, Robert and Edward Lazear (1984), The Excess Sensitivity of Layos and Quits to Demand,
Journal of Labor Economics, 2, 2, 233-257.
[36] Hashimoto, Masanori and Ben Yu (1980), Specic Capital, Employment Contracts, and Wage Rigidity,
The Bell Journal of Economics, 11, 2, 536-549.
[37] Heckman, James and Thomas Macurdy (1980), A Life-Cycle Model of Female Labour Supply, Review
of Economic Studies, 47, 1, 47-74.
34
[38] Laont, Jean-Jacques and David Martimort (2002), The Theory of Incentives: The Principle-Agent
Model, Princeton: Princeton University Press.
[39] Laont, Jean-Jacque and Jean Tirole (1988), The Dynamics of Incentive Contracts, Econometrica,
56, 5, 1153-1175.
[40] Lazear, Edward and Sherwin Rosen (1990), Male-Female Dierentials in Job Ladders, Journal of Labor
Economics, 8, 1, S106-S123.
[41] Lalive, Rafael and Josef Zweimuller (2009), How Does Parental Leave Aect Fertility and Return to
Work? Evidence From Two Natural Experiments, Quarterly Journal of Economics, 124, 3, 1363-1402.
[42] Low, Hamish and Sanchez-Marcos, Virginia (2014), Female Labor Market Outcomes and Maternal
Leave Policies, Working Paper.
[43] McCue, Kristin (1996), Promotions and Wage Growth, Journal of Labor Economics, 14, 2, 175-209.
[44] Mukhopadhyay, Sankar (2012), The Eects of the 1978 Pregnancy Discrimination Act on Female Labor
Supply, International Economic Review, 53, 4, 1133-1153.
[45] Pallais, Amanda (2014), Inecient Hiring in Entry-Level Labor Markets, The American Economic
Review, 104, 11, 3565-3599.
[46] Phelps, Edmund S. (1972), The Statistical Theory of Racism and Sexism, The American Economic
Review, 62, 4, 659-661.
[47] Ruhm, Christopher (1998), The Economic Consequences of Parental Leave Mandates: Lessons From
Europe, The Quarterly Journal of Economics, 113, 1, 285-317.
[48] Spence, Michael (1973), Job Market Signaling, The Quarterly Journal of Economics, 87, 3, 355-374.
[49] Thomas, Jonathan and Tim Worrall (2007), Limited Commitment Models of the Labor Market,
CESIFO Working Paper No. 2109.
[50] Waldfogel, Jane (1998), Understanding the 'Family Gap' in Pay for Women with Children, The Journal
of Economic Perspectives, 12, 1, 137-156.
[51] Weiss, Andrew (1980), Job Queues and Layos in Labor Markets with Flexible Wages, Journal of
Political Economy, 88, 3, 526-538.
[52] Wolfers, Justin (2006), Did Unilateral Divorce Laws Raise Divorce Rates? A Reconciliation and New
Results, The American Economic Review, 96, 5, 1802-1820.
35
6 Tables and Figures
Table 1: Probability of Promotion Since Hire (T1)
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Hired After*Female Under 40* -0.06** -0.08*** -0.08*** -0.08*** -0.08***
Treatment State [0.024] [0.026] [0.025] [0.025] [0.026]
Female Under 40* 0.02 0.02 0.02 0.02 0.03
Treatment State [0.023] [0.025] [0.025] [0.026] [0.026]
Hired After*Treatment State 0.04*** 0.05*** 0.05*** 0.05*** 0.05***
[0.011] [0.013] [0.014] [0.014] [0.014]
Hired After*Female Under 40 0.02 0.05** 0.06*** 0.06*** 0.05**
[0.019] [0.021] [0.019] [0.020] [0.020]
Female Under 40 -0.08*** -0.07*** -0.07*** -0.07*** -0.07***
[0.019] [0.020] [0.021] [0.021] [0.020]
Hired After -0.01 -0.03** -0.03** -0.03** -0.03**
[0.011] [0.014] [0.014] [0.014] [0.014]
Treatment State 0.02 0.09 0.03 0.03 0.03
[0.035] [0.059] [0.037] [0.038] [0.039]
Tenure 0.05*** 0.05*** 0.05*** 0.05*** 0.05***
[0.002] [0.002] [0.002] [0.002] [0.002]
Tenure Squared -0.00*** -0.00*** -0.00*** -0.00*** -0.00***
[0.000] [0.000] [0.000] [0.000] [0.000]
Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 57,269 49,108 48,987 48,987 48,637
Individual Fixed Eects 13,913 12,904 12,889 12,889 12,857
Mean 0.07 0.07 0.07 0.07 0.07
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: The coecients reported are the estimated coecients from Equation (1). This sample is restricted to respondents hired
in 1987 or later. All regressions control for age, age squared, tenure, tenure squared, education level, marital status, marital
status interacted with sex, marital status interaction with a dummy for female under 40, and a dummy for nonwhite. The
excluded education level is high school graduate. All regressions include year, state, and individual-xed eects. All regressions
also include state-specic, demographic group-specic, and state-demographic specic year xed eects, where the demographic
group of interest is women under the age of 40. The dependent variable is whether the respondent has been promoted since the
time of hire. Column (2) includes controls for whether the respondent was a member of a union and whether the employer was
private. Occupations were divided into a set of 12 standard categories in Column (3) and 12 categories reecting variation in
human capital depreciation in Column (4). Column (5) include3s6both 24 standard occupation divisions and 8 standard industry
divisions.
Figure 5: Eect of Introduction of FMLA On Promotion Rate of Women Under 40
98
7
98
8 89 0 1 2 3 49 99 99 99 99 99 99
5 96 97 989 9 9 99
9 0
00 00
1
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2
Year of Hire
Parameter estimate 95% Confident Limit
Notes: The coec∑ients βτ,1 and 95% condence intervals reported are obtained from estimating
promoted 2001i,j,s,t = τ=1987 βτ,1(HiredY earτi,j∑· FemaleUnder40i,t · TreatmentStates)
+β∑2(FemaleUnder40i,t · TreatmentState ) + 2001s τ=1987∑βτ,3(HiredY earτi,j · TreatmentStates)
+ 2001τ=1987 βτ,4(HiredY earτ · FemaleUnder40
2001
i,j i,t) + τ=1987 βτ,5HiredY earτi,j
+β6FemaleUnder40i,t + β7TreatmentStates + β8Xi,t + β9Zj + λt + cs + fi + δi,t,s + εi,j,s,t.
Standard errors are clustered at the state level. The sample is restricted to respondents hired in 1987 or later. The regression
controls for sex, age, age squared, tenure, tenure squared, level of education attained, marital status, marital status interacted
with sex, marital status interacted with FemaleUnder40, and a dummy for nonwhite. The regression also controls for job
characteristics (private rm, union), 24 standard occupational categories, and 12 standard industry categories from the census
occupational classication system.
87 88 89 90 1 29 9 9 9 99 99 99
3 94 95 96 97 89 9 9 9 99 99
9 00 01
87
0 0
9 98
8 9 0 1 2 3 4 5 6 7 8 9 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2
1 1 19
8 99 9 91 19 19 19
9 99 99 99 99 991 1 1 1 1 19
9 00 00 372 2 Year of Hire
Year of Hire
Parameter estimate 95% Confident Limit
Parameter estimate 95% Confident Limit
(b) Control Group = Women Over 40 Only
(a) Control Group = Men Under 40 Only
Regression Coefficients
 
−.4 −.3 −.2 −.1 0 .1
Regression Coefficients
 
−.4 −.2 0 .2
Regression Coefficients
 
−.6 −.4 −.2 0 .2
Figure 6: Eect of Mandate on Gender Dierences in Promotions Relative to Older Women
(I) PSID
16−20 21−25 26−30 31−35 36−40 41−45 45−50 51−65
Age
Parameter estimate 95% Confident Limit
Notes: The coec∑ients βa,1 and 95% condence intervals reported are obtained from estimating
pr∑omoted 45i,j,s,t = a=15(5) βa,1(FemaleAged a to (a+ 4)i,t ·HiredAfteri,j · TreatmentStates)
+∑45a=15(5) βa,2(FemaleAged a to (a+ 4)i,t · TreatmentStates) + β3(HiredAfteri,j · TreatmentStates)
+∑45a=15(5) βa,4(FemaleAged a to (a+ 4)i,t ·HiredAfteri,j) + β5HiredAfteri,j
+ 45a=15(5) βa,6(FemaleAged a to (a+ 4)i,t + β7TreatmentStates + β8Xi,t + β9Zj + λt + cs + fi + δi,t,s + εi,j,s,t.
Standard errors are clustered at the state level. The sample is restricted to respondents hired in 1987 or later. The regression
controls for sex, age, age squared, tenure, tenure squared, level of education attained, marital status, marital status interacted
with sex, marital status interacted with FemaleUnder40, and a dummy for nonwhite. The regression also controls for job
characteristics (private rm, union), 24 standard occupational categories, and 12 standard industry categories from the census
occupational classication system.
16−20 21−25 26−30 31−35 36−40 41−45 45−50 51−65 16−20 21−25 26−30 31−35 36−40 41−45 45−50 51−65
Age Age
Parameter estimate 95% Confident Limit Parameter estimate 95% Confident Limit
(a) Treatment States (b) Control States
Notes: The coec∑ients βa,4 and 95% condence intervals reported are obtained from estimating
pr∑omoted 45i,j,s,t = a=15(5) βa,4(FemaleAged a to (a+ 4)i,t ·HiredAfteri,j) + β5HiredAfteri,j
+ 45a=15(5) βa,6(FemaleAged a to (a+ 4)i,t ++β8Xi,t + β9Z
3j8+ λt + cs + fi + δi,s + εi,j,s,t.
Standard errors are clustered at the state level. The sample is restricted to respondents hired in 1987 or later. The regression
controls for sex, age, age squared, tenure, tenure squared, level of education attained, marital status, marital status interacted
with sex, marital status interacted with FemaleUnder40, and a dummy for nonwhite. The regression also controls for job
characteristics (private rm, union), 24 standard occupational categories, and 12 standard industry categories from the census
occupational classication system.
Regression Coefficients
−.3 −.2 −.1 0 .1
Regression Coefficients
−.4 −.3 −.2 −.1 0 .1
Regression Coefficients
−.2 −.1 0 .1 .2
Table 2: Probability of Promotion Since Hire In Jobs Eligible for FMLA Coverage
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Eligible Female Under 40* -0.07** -0.11*** -0.10*** -0.10*** -0.10***
Hired After*Treatment State [0.033] [0.034] [0.033] [0.033] [0.034]
Eligible Female Under 40* 0.02 0.03 0.03 0.03 0.04
Treatment State [0.028] [0.031] [0.030] [0.030] [0.031]
Eligible Female Under 40* 0.03 0.08*** 0.07*** 0.07*** 0.07**
Hired After [0.027] [0.028] [0.027] [0.027] [0.028]
Eligible Female Under 40 -0.10*** -0.10*** -0.10*** -0.10*** -0.11***
[0.025] [0.029] [0.027] [0.027] [0.027]
Tenure 0.05*** 0.05*** 0.05*** 0.05*** 0.05***
[0.003] [0.003] [0.003] [0.003] [0.003]
Tenure Squared -0.00*** -0.00*** -0.00*** -0.00*** -0.00***
[0.000] [0.000] [0.000] [0.000] [0.000]
Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 38,431 32,952 32,878 32,878 32,638
Individual Fixed Eects 11,436 10,484 10,475 10,475 10,439
Mean 0.10 0.10 0.10 0.10 0.10
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: The coecients reported are the estimated coecients from Equation (2). "Eligible Female Under 40" is dened as
women under the age of 40 who have worked at least 1250 hours in their current employment position in the previous year.
The control group consists of men and women over the age of 40. The sample is restricted to workers hired in 1987 or later
and to those respondents who had a non-missing response to the number of hours worked over the previous 12 months. All
regressions control for age, age squared, tenure, tenure squared, education level, marital status, marital status interacted with
sex, marital status interaction with a dummy for female under 40, and a dummy for nonwhite. The excluded education level is
high school graduate. All regressions include year, state, and individual-xed eects. All regressions also include state-specic,
demographic group-specic, and state-demographic specic year xed eects, where the demographic group of interest is women
under the age of 40. The dependent variable is whether the respondent has been promoted since the time of hire. Column (2)
includes controls for whether the respondent was a member of a union and whether the employer was private. Occupations were
divided into a set of 12 standard categories in Column (3) and 12 categories reecting variation in human capital depreciation
in Column (4). Column (5) includes both 24 standard occupation divisions and 8 standard industry divisions.
39
Figure 7: Eect of Mandate on Gender Dierences in Promotions Relative to Older Women
16−20 21−25 26−30 31−35 36−40 41−45 45−50 51−60
Age
Parameter estimate 95% Confident Limit
Notes: The coe∑cients βa,1 and 95% condence intervals reported are obtained from estimating
pr∑omoted 45i,j,t = a=15(5) βa,1(FemaleAged a to (a+ 4)i,t ·HiredAfteri,j) + β2HiredAfteri,j
+∑45a=15(5) βa,3(FemaleAged a to (a+ 4)i,t
+ 45a=15(5) βa,3(Aged a to (a+ 4)i,t + β8Xi,t + β9Zj + λt + cc + εi,j,t.
All regressions control for years of age, age squared, education level, a dummy for nonwhite, a standardized performance level,
24 standard occupational categories, and 8 standard industry divisions. The excluded education level is high school graduate.
Effect of Mandate on Gender Difference in Promotions
Relative to Effect on Gender Difference for Older Women
20 23 26 29 32 35 38 1 4 7 0 0
8− 1− 4− 7− 0− 3− 6− 9−
4 −4 4 5 6
1 2 2 2 3 3 3 3 42 45
−
48
− −
51
Age
Parameter estimate 95% Confident Limit
Data: MCSUI
Notes: The coe∑cients βa,1 and 95% condence intervals reported are obtained from estimating
pr∑omoted 45i,j,t = a=18(3) βa,1(FemaleAged a t∑o (a+ 2)i,t ·HiredAfteri,j) + β2HiredAfteri,j
+ 45a=18(3) βa,3(FemaleAged a to (a+ 2) +
45
i,t a=18(3) βa,3(Aged a to (a+ 2)i,t + β8Xi,t + β9Zj + λt + cc + εi,j,t.
All regressions control for years of age, age squared, education level, a dummy for nonwhite, a standardized performance level,
24 standard occupational categories, and 8 standard industry divisions. The excluded education level is high school graduate.
40
Regression Coefficients Regression Coefficients
 −.6 −.4 −.2 0 .2
−.6 −.4 −.2 0 .2
Table 3: Eect of Mandate on Probability of Promotion
Among Women Under 40 Who Never Have Children
(1) (2) (3)
VARIABLES Promoted Promoted Promoted
Female Under 40 Never Has Kids* 0.01 0.01 -0.08*
Treatment State*Hired After [0.053] [0.054] [0.048]
Female Under 40 Ev. Has Kids* -0.03 -0.04
Treatment State*Hired After [0.040] [0.042]
Female Under 40* -0.09*** -0.09**
Treatment State*Hired After [0.024] [0.033]
Tenure 0.05*** 0.05*** 0.04***
[0.003] [0.003] [0.010]
Tenure Squared -0.00*** -0.00*** -0.00***
[0.000] [0.000] [0.001]
Year Fixed Eects Yes Yes Yes
State Fixed Eects Yes Yes Yes
Firm Characteristics Yes Yes Yes
Occupation Controls 24 24 24
Industry Controls 8 8 8
Observations 28,496 14,475 4,364
Individual Fixed Eects 10,054 5,295 1,828
Mean 0.09 0.08 0.07
Robust standard errors in brackets
Standard errors clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: This sample is restricted to respondents hired in 1987 or later and year observations from August 1993 or later. All
regressions control for age, age squared, tenure, tenure squared, education level, marital status, marital status interacted with
sex, marital status interaction with a dummy for female under 40, and a dummy for nonwhite. The excluded education level is
high school graduate. All regressions include year, state, and individual-xed eects. All regressions also control for 24 standard
occupation divisions and 8 standard industry divisions. The dependent variable is whether the respondent has been promoted
since the time of hire. Columns (2) and (3) restricts the sample to women only.
41
Table 4: Eect of Mandate on Employment and Labor Force Participation
of Women Under 40 (T2)
(1) (2) (3) (4)
VARIABLES Employed Employed LFP LFP
Female Under 40*Treatment State* 0.03** 0.03* 0.05*** 0.05***
Post-Period [0.015] [0.017] [0.012] [0.013]
Female Under 40*Treatment State -0.04*** -0.04*** -0.05*** -0.05***
[0.015] [0.015] [0.012] [0.012]
Treatment State*Post-Period -0.01 -0.00 -0.02** -0.01
[0.012] [0.011] [0.009] [0.010]
Female Under 40*Post-Period 0.00 -0.02 -0.00 -0.02
[0.014] [0.022] [0.011] [0.017]
Treatment State 0.01 0.01 0.03 0.04
[0.032] [0.034] [0.025] [0.027]
Post-Period -0.01 -0.00 -0.00 -0.01
[0.009] [0.010] [0.008] [0.008]
Female Under 40 0.01 -0.01 0.03 0.00
[0.020] [0.020] [0.018] [0.018]
Year Fixed Eects Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes
State-Specic Year F.E. No Yes No Yes
Demographic-Specic Year F.E. No Yes No Yes
Observations 129,064 129,064 129,064 129,064
Individual Fixed Eects 22,105 22,105 22,105 22,105
Mean 0.75 0.75 0.79 0.79
Robust standard errors in brackets, Standard Errors Clustered at the State Level
*** p<0.01, ** p<0.05, * p<0.1
Notes: The coecients reported are the estimated coecients from Equation (3). All regressions control for age, age squared,
education level, marital status, marital status interacted with sex, marital status interacted with a dummy for female under
40, and a dummy for nonwhite. The excluded education level is high school graduate. All regressions include year, state, and
individual-xed eects. Columns (2) and (4) also include state-specic and demographic-specic year xed eects.
42
Figure 8: Eect of Mandate on Employment Rates of Women Under 40
98
8
98
9 90 91 92 93 94 95 96 97 98 99 00 01
1 1 19 19 19 19 19 19 19 19 19 19 20 20
Survey Year
Parameter estimate 95% Confident Limit
43
Regression Coefficients
 
−.05 0 .05 .1 .15
Table 5: Eect of Mandate on Late-Career Labor Supply of Women Under 40 (T3)
(1) (2) (3) (4)
VARIABLES Weekly Hours Total Weekly Hours Part-Time Overtime
Female With Children*Hired After* -4.73** -4.09** 0.09 -0.11*
Treatment State [2.127] [1.956] [0.060] [0.060]
Female With Children* 0.43 0.04 -0.01* 0.00
Treatment State [0.650] [0.246] [0.006] [0.008]
Hired After*Treatment State 2.21* 1.39 0.01 0.01
[1.199] [1.284] [0.019] [0.042]
Female With Children*Hired After 4.25** 4.26** -0.10* 0.11**
[1.979] [1.788] [0.053] [0.053]
Female With Children -4.40*** -3.49*** 0.13*** -0.10***
[0.487] [0.441] [0.014] [0.018]
Hired After -1.61 -1.16 -0.01 -0.02
[1.032] [1.104] [0.013] [0.032]
Treatment State 0.78 2.27 0.10*** 0.15**
[2.081] [1.816] [0.019] [0.058]
Tenure -0.00 -0.01 -0.00* -0.00*
[0.034] [0.036] [0.001] [0.001]
Tenure Squared 0.00 0.00 0.00 0.00***
[0.001] [0.001] [0.000] [0.000]
Year Fixed Eects Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes
State-Specic Year F.E. Yes Yes Yes Yes
Demographic-Specic Year F.E. Yes Yes Yes Yes
Occupation Controls 24 24 24 24
Industry Controls 8 8 8 8
Observations 50,481 50,481 50,481 50,481
Individual Fixed Eects 11,747 11,747 11,747 11,747
Mean 42.33 42.85 0.10 0.21
Robust standard errors in brackets, Standard Errors Clustered at the State Level
*** p<0.01, ** p<0.05, * p<0.1
Notes: This sample is restricted to respondents with at least four years of tenure with their current employer. All regressions
control for age, age squared, tenure, tenure squared, education level, marital status, marital status interacted with sex, marital
status interaction with a dummy for female under 40, and a dummy for nonwhite. The excluded education level is high school
graduate. All regressions include year, state, and individual-xed eects, as well as state-specic and demographic-specic
year-xed eects. All regressions also control for 24 standard occupation divisions and 8 standard industry divisions. "Weekly
Hours" is dened as the average number of hours a week the respondent worked on in their main job in the previous year.
"Total Weekly Hours" is dened as the average number of hours a week the respondent worked plus the average weekly hours
of overtime (determined from the annual hours of overtime reported). "Part-Time" is dened as average weekly hours of work
is below 35 hours per week, and "Overtime" is dened as average hours of work is greater than or equal to 50 hours of work
per week.
44
Table 6: Non-Linear Return to Performance Among Female Employees
(1)
VARIABLES Promoted
50th % 0.03
[0.030]
75th % 0.07**
[0.029]
80th % -0.06
[0.040]
90th % 0.04
[0.042]
Under 40 0.08*
[0.045]
Standardized Performance -0.08
[0.116]
Observations 933
R-squared 0.11
Mean 0.07
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1
Notes: This sample is restricted to rms whose most recent hire was female. All regressions control for years of age, age squared,
education level, and a dummy for nonwhite. Regressions also all include controls for standardized performance evaluations, rm
characteristics, city dummies, 24 standard occupational categories, and 8 standard industry categories. The excluded education
level is high school graduate. The standardized performance level is equal to the employer's evaluation of worker performance
minus the average performance rating given to workers in the rm, on a scale of 0 to 1. "Xth
45
Table 7: Marginal Return to Signals Among Female Employees (T4)
Table 8 - Eect of Mandate on Return to Performance
VARIABLES Promoted
50th %*Female Under 40*After 0.06
[0.136]
75th %*Female Under 40*After -0.27**
[0.114]
80th %*Female Under 40*After 0.17
[0.333]
90th %*Female Under 40*After -0.17
[0.324]
Performance -0.06
[0.116]
Sample Females
Observations 933
R-squared 0.11
Mean 0.07
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1
Notes: This sample is restricted to rms whose most recent hire was female. All regressions control for years of age, age squared,
education level, and a dummy for nonwhite. Regressions also all include controls for standardized performance evaluations, rm
characteristics, city dummies, 24 standard occupational categories, and 8 standard industry categories. The excluded education
level is high school graduate. The standardized performance level is equal to the employer's evaluation of worker performance
minus the average performance rating given to workers in the rm, on a scale of 0 to 1. "Xth
46
47
Table 8: Eect of Mandate on Wages and Fringe Benets (T5)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
VARIABLES Starting Health Dental Pension Paid Other Additional Bonuses Flexible Day Care
Wage Insurance Vacation Benets Pay Hours
Female Under 40*Hired After 0.06 -0.08 -0.10 -0.03 -0.07 -0.05 0.05 -0.02 -0.17* -0.19***
[0.914] [0.078] [0.098] [0.097] [0.066] [0.091] [0.092] [0.097] [0.097] [0.046]
Female*Hired After -0.13 0.14* 0.08 0.06 0.05 0.06 -0.05 -0.01 0.24** 0.15***
[0.910] [0.078] [0.098] [0.097] [0.066] [0.091] [0.092] [0.097] [0.097] [0.046]
Hired After 0.25 -0.04 0.05 0.02 -0.02 -0.01 -0.05 -0.03 -0.04 0.02
[0.366] [0.031] [0.039] [0.039] [0.026] [0.036] [0.037] [0.039] [0.039] [0.019]
Female Under 40 1.17* 0.22*** 0.10 0.06 0.02 0.04 -0.07 -0.01 0.05 0.10***
[0.695] [0.059] [0.074] [0.073] [0.050] [0.069] [0.070] [0.074] [0.074] [0.035]
Female -1.90*** -0.20*** -0.02 -0.06 -0.01 0.01 0.08 0.03 -0.04 -0.05
[0.666] [0.056] [0.071] [0.070] [0.048] [0.066] [0.067] [0.070] [0.071] [0.034]
Under 40 -0.54 -0.08 0.04 0.02 -0.01 -0.07 -0.04 -0.09 -0.01 -0.03
[0.653] [0.056] [0.070] [0.069] [0.047] [0.065] [0.066] [0.070] [0.070] [0.033]
Tenure 0.49** -0.01 -0.06** -0.03 0.02 -0.03 0.06** 0.08*** 0.02 0.03**
[0.247] [0.021] [0.027] [0.026] [0.018] [0.025] [0.025] [0.026] [0.026] [0.013]
Tenure Squared -0.01 -0.00 0.01 0.00 -0.01** 0.00 -0.00 -0.00 0.00 -0.00*
[0.035] [0.003] [0.004] [0.004] [0.003] [0.003] [0.004] [0.004] [0.004] [0.002]
Observations 1,576 1,690 1,688 1,685 1,690 1,685 1,686 1,690 1,685 1,691
R-squared 0.30 0.13 0.11 0.13 0.10 0.08 0.19 0.11 0.12 0.16
Mean 8.50 0.80 0.57 0.57 0.86 0.71 0.56 0.41 0.56 0.07
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1
Notes: All regressions control for years of age, age squared, education level, and a dummy for nonwhite. Regressions also all include controls for
standardized performance evaluations, rm characteristics, city dummies, 24 standard occupational categories, and 8 standard industry categories.
The excluded education level is high school graduate. Wages and salaries are in terms of 1990 dollars.
Table 9: Eect of Mandate on Likelihood of Promotion Among Firms with High Costs of Training (T6)
(1) (2) (3) (4) (5) (6)
VARIABLES Promoted Promoted Promoted Promoted Promoted Promoted
High Training*Female Under 40* -0.14 -0.22* -0.22* -0.23* -0.26** -0.26**
Hired After [0.111] [0.124] [0.126] [0.128] [0.128] [0.128]
Female Under 40*Hired After -0.07 -0.07 -0.10 -0.10 -0.11 -0.11
[0.068] [0.072] [0.073] [0.075] [0.075] [0.077]
Female Under 40*High Training 0.02 0.03 0.02 0.03 0.05 0.05
[0.059] [0.062] [0.068] [0.069] [0.070] [0.070]
Hired After*High Training 0.13 0.19 0.20* 0.20* 0.22* 0.20
[0.104] [0.116] [0.118] [0.120] [0.120] [0.121]
Hired After 0.06 0.06 0.07 0.07 0.07 0.08
[0.065] [0.068] [0.069] [0.071] [0.071] [0.073]
Female Under 40 0.10* 0.08 0.09 0.10 0.09 0.09
[0.056] [0.059] [0.062] [0.063] [0.063] [0.064]
High Training 0.00 0.00 0.00 0.01 -0.02 -0.01
[0.055] [0.057] [0.063] [0.064] [0.065] [0.065]
Tenure 0.12*** 0.13*** 0.10*** 0.10*** 0.10*** 0.09***
[0.018] [0.020] [0.028] [0.028] [0.029] [0.029]
Tenure Squared -0.01*** -0.01*** -0.00 -0.00 -0.00 -0.00
[0.003] [0.003] [0.006] [0.006] [0.006] [0.006]
Performance Control Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
City Controls Yes Yes Yes Yes
Occupation Controls 12 24 24
Industry Controls 8
Observations 1,031 927 757 752 750 745
R-squared 0.08 0.09 0.10 0.11 0.12 0.13
Mean 0.07 0.07 0.07 0.07 0.07 0.07
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1
Notes: The sample is restricted to rms whose last hire was female. All regressions control for years of age, age squared,
education level, and a dummy for nonwhite. The excluded education level is high school graduate. The dependent variable is
whether the employer's last hire has been promoted. "High Training" is dened as the number of hours of training the last hire
received from coworkers and supervisors since the time of hire was greater than or equal to 50 hours. "Hired After" is dened
as whether the worker was hired after the enactment of FMLA. "Under 40" is dened as whether the worker was aged 40 or
less at the time of hire. Column (2) uses a standardized performance level that is equal to the employer's evaluation of worker
performance minus the average performance rating given to workers in the rm, on a scale of 0 to 1. Occupations were divided
into a set of 12 standard categories in Column (4) and 24 categories in Column (5), although no observations were in the 24th
standard category, military occupations. Column (6) includes both 24 standard occupation divisions and 8 standard industry
divisions.
48
Table 10: Eect of Mandate on Fertility of Employed Women (T7)
(1) (2) (3) (4)
VARIABLES Has Children Ever Has Children Has Child LT 5 Has Child LT 1
Hired After* Female* 0.04** 0.04** 0.04** 0.03*
Treatment State [0.018] [0.016] [0.018] [0.016]
Female* 0.01 0.01 0.01 0.00
Treatment State [0.008] [0.007] [0.008] [0.007]
Hired After*Treatment State -0.01 -0.01 -0.03** -0.02**
[0.012] [0.011] [0.013] [0.011]
Hired After*Female -0.01 -0.03* -0.05*** -0.05***
[0.016] [0.015] [0.016] [0.014]
Female 0.13*** 0.13*** 0.11*** 0.09***
[0.010] [0.009] [0.010] [0.009]
Hired After -0.00 0.02 0.03*** 0.04***
[0.012] [0.010] [0.012] [0.010]
Treatment State 0.03*** 0.02*** -0.02*** -0.01**
[0.005] [0.005] [0.005] [0.005]
Tenure 0.00 -0.00 -0.00 -0.00**
[0.000] [0.000] [0.001] [0.000]
Tenure Squared -0.00*** -0.00 -0.00** 0.00
[0.000] [0.000] [0.000] [0.000]
Observations 98,020 98,747 98,747 98,747
R-squared 0.20 0.08 0.35 0.30
Mean 0.78 0.86 0.32 0.20
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1
Notes: This sample is restricted to respondents whose employment spell began in 1987 or later. All regressions control for age,
age squared, tenure, tenure squared, education level, marital status, marital status interacted with sex, marital status interaction
with a dummy for female under 40, and a dummy for nonwhite. The excluded education level is high school graduate. All
regressions also include controls for whether the respondent was a member of a union and whether the employer was private
Occupations are divided into a set of 24 standard occupation divisions and 8 standard industry divisions.
7 Appendix A
Proof of Proposition 1
Such a threshold y∗ exists as long as the cost of rm training is neither so low nor so high that information
about a worker's type is uninformative to the rm's decision. Such a threshold does not exist if the cost of
49
training were suciently low that promoting a worker without any additional information about their type
is still protable:
c < πCP (C) ≡ cmin
In this case, the rm would promote every worker. Furthermore, if the cost of training were suciently high
such that setting the threshold at the value of y that maximizes P (C|y), does not generate a high enough
probability that workers producing this signal are of Type C, then promoting no one, regardless of their
signal, maximizes rm prot. This occurs if
c > πCP (C|y),∀y
g(y−R1α(0))
or, equivalently, if πCc −1 < ∗ for all promotion standards y. The right-hand side of this inequalityg(y−h1(θC ;y))
attains its minimum at the threshold ymax such that h∗(θ ; ymax1 C ) = y
max. This is true when
ymax = √A +R1α(0)
σ 2π
Note that when y < h∗ ∗1(θC ; y), h1(θC ; y) is increasing in y, when y > h
∗
1(θC ; y), h
∗
1(θC ; y) is decreasing in
y, and when y = h∗1(θC ; y), h
∗
1(θC ; y) is the maximum of the Type C worker's best response function, as a
function of the promotion standard, y. This maximum number of hours from a Type C worker is attained
by ymax. In other words, if the rm chooses the threshold that maximizes the the number of hours worked
by the Type C worker, thereby maximizing the dierence between the means of the Type B and Type C
signals, and promoting a worker who meets this threshold is still not protable, there is no threshold that
the rm can set to distinguish protable promotions from unprotable ones. No worker will be promoted for
( ( )πCc > cmax = √ )
1 + g Aσ √ σ 2π
P (A)+P (B)
ε 2π P (C)
| ( g(y−
∗
)h1(θC ;y))P (C)The quantity P (C y) = ∗ is increasing in y for the feasible range ofg(y−h1(θC ;y))P (C)+g(y−R1α(0))(P (B)+P (A))
the equilibrium threshold y∗, ymin, ymax , a range guaranteed by rm costs, c (c 30min, cmax). πCP (C|ymin) <
c and πCP (C|ymax) > c. Therefore, the equilibrium y∗ exists for c [cmin, cmax].
( ) ( )
g y−h∗(θ ;y)
30 ( )This is because 1 C− is increasing in y for all y y
min, ymax and decr[easing in] y for all y 0, ymin , for someg(y R1α(0))
ymin < R1α(0). If c > cmin, πCP (C) < c, and therefore, πCP (C|y) < c for all y 0, ymin as well, so it must be that the
equilibrium threshold satises y∗ > ymin. c < c ∗max guarantees y < ymax. In other words, y∗will not be less than ymin nor
greater than ymax because we have specied a problem with rm costs such that the signal is informative to the rm's decision.
50
Figure 9: Type C Worker's Best Response to Promotion Standard y
Figure 10: (a) Solution to First-Order Condition for Figure 11: (b) Solution to First-Order Condition for
y < ymax y > ymax
Figure 12: Response of Distribution of ObservedWorker Productivity to Promotion Standard
Shift from y0 to ymax
( )
Uniqueness comes from the fact that πCP (C|y) is strictly increasing in y for all y ymin, ymax .
51
8 Appendix B: Robustness Checks
In this Appendix, I present a series of robustness checks to the results described in section 3. First, I show
that the control groups are unaected by the policy change; in control states, women under the age of 40
hired after the mandate face similar likelihood of promotion as those hired before. Furthermore, while women
under the age of 40 face a lower likelihood of promotion when hired after the enactment of the FMLA, women
over the age of 40 are unaected in comparison to men. Second, I show that the change in the likelihood of
promotion is not driven by the eect of the mandate on a given type of worker. Table 13 shows results for the
change in the probability of promotion when we restrict to observations from survey years after the FMLA,
still exploiting the variation in whether the respondent was hired before or after its enactment. Table 14
and Figure 14 demonstrate that the results are not driven by the inclusion of state- or demographic-specic
time xed eects. To conrm that dierential year trends do not drive the results, I show in Tables 15 -
17 results including state-specic, demographic-specic, and state-demographic-specic trends in the year of
hire, separately for each control group. While men under the age of 40 do witness a slight time trend in that
women under the age of 40 have an increasing likelihood of being promoted over time relative to men under
40, the enactment of the FMLA still retards that relative growth for women under the age of 40. Tables 18
and Figure 15 address the concern about the impact of the FMLA on a given type and demonstrate that
there appears to be little eect when restricting the sample to those who were hired before the FMLA.
52
Figure 13: Eect of Mandate on Gender Dierences in Promotion Rates
87 8 9 0 1 29 98 98 99 99 99 99
3 949 99
5 96 97 98 9 0 1 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1 1 1 1 1 1 1 1 1 19 19 19 19
9
20
0 002 19
8 981 19
8
19
9 99 99 9 9 9 9 9 9 9 01 1 19 19 19 19 19 19 19 20 20
0
Year of Hire Year of Hire
Parameter estimate 95% Confident Limit Parameter estimate 95% Confident Limit
(a) Women Under 40 Compared to Men Only (b) Women Over 40 Compared to Men
87 88 89 90 91 92 93 4 5 6 7 8 9 0 1 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
19 19 19 19 19 19 19 19
9 99 99 99 99 99 00 00 81 1 1 1 1 2 2 19 19
8 8 9 9 9 9 9 9 9 9 9 9 0 0
19 19 19 19 19 19 19 19 19 19 19 20 20
Year of Hire Year of Hire
Parameter estimate 95% Confident Limit Parameter estimate 95% Confident Limit
(c) Treatment States (d) Control States
Notes: The coec∑ients βτ,4 and 95% condence intervals reported are obtained from estimating
promoted 2001i,j,s,t = t=1988 βt,1(SurveyY eartt ·∑FemaleUnder40i,t · TreatmentState∑ ∑ s
)
+β2(FemaleUnder40i,t · TreatmentStates) + 2001t=1988 βt,3(SurveyY eartt · TreatmentStates)
+ 2001τ=1987 βτ,4(HiredY earτi,j · FemaleUnder40i,t) +
2001
τ=1987 βτ,5HiredY earτi,j
+β6FemaleUnder40i,t+β8Xi,t+β9Zj+λt+cs+fi+δi,t,s+εi,j,s,t, where the sample in Panel (c) is restricted to treatment states
only, and the sample in panel (d) is restricted to control states only. Standard errors are clustered at the state level. The sample
is restricted to respondents hired in 1987 or later. The regression controls for sex, age, age squared, tenure, tenure squared,
level of education attained, marital status, marital status interacted with sex, marital status interacted with FemaleUnder40,
and a dummy for nonwhite. The regression also controls for job characteristics (private rm, union), 24 standard occupational
categories, and 12 standard industry categories from the census occupational classication system.
53
Regression Coefficients Regression Coefficients
  
−.2 −.15 −.1 −.05 0 .05 −.4 −.2 0 .2
Regression Coefficients Regression Coefficients
  
−.4 −.2 0 .2 .4 −.4 −.2 0 .2 .4
Table 11: Probability of Promotion Since Hire, Compared to Men Under 40
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Hired After*Female Under 40* -0.08*** -0.11*** -0.10*** -0.10*** -0.10***
Treatment State [0.023] [0.025] [0.025] [0.025] [0.025]
Female Under 40* 0.04** 0.05*** 0.05*** 0.05*** 0.05***
Treatment State [0.014] [0.017] [0.016] [0.017] [0.017]
Hired After*Treatment State 0.07*** 0.08*** 0.08*** 0.08*** 0.08***
[0.024] [0.027] [0.026] [0.025] [0.025]
Hired After*Female Under 40 0.01 0.04 0.04* 0.03* 0.04*
[0.019] [0.021] [0.021] [0.021] [0.020]
Female Under 40 -0.03** -0.05** -0.05*** -0.05** -0.05***
[0.016] [0.018] [0.017] [0.017] [0.017]
Hired After -0.00 -0.01 -0.01 -0.01 -0.01
[0.023] [0.027] [0.026] [0.026] [0.025]
Treatment State 0.03 0.06 0.02 0.02 0.02
[0.042] [0.071] [0.060] [0.059] [0.060]
Tenure 0.05*** 0.06*** 0.05*** 0.05*** 0.05***
[0.002] [0.003] [0.003] [0.003] [0.003]
Tenure Squared -0.00*** -0.00*** -0.00*** -0.00*** -0.00***
[0.000] [0.000] [0.000] [0.000] [0.000]
Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 29,558 25,353 25,295 25,295 25,295
Individual Fixed Eects 7,218 6,775 6,770 6,770 6,753
Mean 0.06 0.06 0.06 0.06 0.06
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes:The sample here is restricted to individuals under the age of 40 whose employment spell began in 1987 or later. All
regressions control for age, age squared, tenure, tenure squared, education level, marital status, marital status interacted with
sex, marital status interaction with a dummy for female under 40, and a dummy for nonwhite. The excluded education level is
high school graduate. All regressions also include year, state, and individual-xed eects. The dependent variable is whether
the respondent has been promoted since the time of hire. Column (2) includes controls for whether the respondent was a
member of a union and whether the employer was private. Occupations were divided into a set of 12 standard categories in
Column (3) and 12 categories reecting variation in human capital depreciation in Column (4). Column (5) includes both 24
standard occupation divisions and 8 standard industry divisions.
54
Table 12: Probability of Promotion Since Hire, Compared to Women Over 40
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Hired After*Female Under 40* -0.08*** -0.11*** -0.10*** -0.10*** -0.10***
Treatment State [0.023] [0.025] [0.025] [0.025] [0.025]
Female Under 40* 0.04** 0.05*** 0.05*** 0.05*** 0.05***
Treatment State [0.014] [0.017] [0.016] [0.017] [0.017]
Hired After*Treatment State 0.07*** 0.08*** 0.08*** 0.08*** 0.08***
[0.024] [0.027] [0.026] [0.025] [0.025]
Hired After*Female Under 40 0.01 0.04 0.04* 0.03* 0.04*
[0.019] [0.021] [0.021] [0.021] [0.020]
Female Under 40 -0.03** -0.05** -0.05*** -0.05** -0.05***
[0.016] [0.018] [0.017] [0.017] [0.017]
Hired After -0.00 -0.01 -0.01 -0.01 -0.01
[0.023] [0.027] [0.026] [0.026] [0.025]
Treatment State 0.03 0.06 0.02 0.02 0.02
[0.042] [0.071] [0.060] [0.059] [0.060]
Tenure 0.05*** 0.06*** 0.05*** 0.05*** 0.05***
[0.002] [0.003] [0.003] [0.003] [0.003]
Tenure Squared -0.00*** -0.00*** -0.00*** -0.00*** -0.00***
[0.000] [0.000] [0.000] [0.000] [0.000]
Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 29,558 25,353 25,295 25,295 25,295
Individual Fixed Eects 7,218 6,775 6,770 6,770 6,753
Mean 0.06 0.06 0.06 0.06 0.06
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: The sample here is restricted to females whose employment spell began in 1987 or later. All regressions control for age,
age squared, tenure, tenure squared, education level, marital status, marital status interacted with sex, marital status interaction
with a dummy for female under 40, and a dummy for nonwhite. The excluded education level is high school graduate. All
regressions also include year, state, and individual-xed eects. The dependent variable is whether the respondent has been
promoted since the time of hire. Column (2) includes controls for whether the respondent was a member of a union and
whether the employer was private. Occupations were divided into a set of 12 standard categories in Column (3) and 12
categories reecting variation in human capital depreciation in Column (4). Column (5) includes both 24 standard occupation
divisions and 8 standard industry divisions.
55
Table 13: Probability of Promotion Since Hire, Restricted to Survey Dates August 1993 -2001
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Hired After*Female Under 40* -0.07*** -0.09*** -0.09*** -0.09*** -0.09***
Treatment State [0.026] [0.026] [0.026] [0.026] [0.026]
Female Under 40* 0.02 0.02 0.02 0.02 0.02
Treatment State [0.017] [0.020] [0.020] [0.020] [0.020]
Hired After*Treatment State 0.02* 0.03** 0.03** 0.03** 0.03**
[0.012] [0.013] [0.014] [0.014] [0.013]
Hired After*Female Under 40 0.03 0.06** 0.06*** 0.06*** 0.06**
[0.022] [0.022] [0.021] [0.021] [0.021]
Female Under 40 -0.01 -0.00 -0.00 -0.00 -0.00
[0.020] [0.022] [0.023] [0.022] [0.022]
Hired After 0.00 -0.01 -0.01 -0.01 -0.01
[0.010] [0.011] [0.011] [0.011] [0.011]
Treatment State -0.05 -0.01 0.05 0.01 -0.00
[0.043] [0.053] [0.056] [0.059] [0.061]
Tenure 0.04*** 0.05*** 0.05*** 0.05*** 0.05***
[0.002] [0.002] [0.002] [0.002] [0.003]
Tenure Squared -0.00*** -0.00*** -0.00*** -0.00*** -0.00***
[0.000] [0.000] [0.000] [0.000] [0.000]
Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Year Hired Fixed Eects Yes Yes Yes Yes Yes
State*Year Hired Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 34,202 29,088 28,992 28,992 28,798
Individual Fixed Eects 11,122 10,249 10,228 10,228 10,197
Mean 0.09 0.09 0.09 0.09 0.09
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: The coecients reported are the estimated coecients from the following equation:
promotedi,j,s,t = β1(HiredAfteri,j · FemaleUnder40i,t · TreatmentStates)
+ β2(FemaleUnder40i,t · TreatmentStates) + β3(HiredAfteri,j · TreatmentStates)
+ β4(HiredAfteri,j · FemaleUnder40i,t) + β5HiredAfteri,j + β6FemaleUnder40i,t
+ β7TreatmentStates + β8Xi,t + β9Zj + λt + cs + fi + εi,j,s,t
The sample is restricted to observations from August of 1993 or later and respondents whose employment spell began in 1987
or later. All regressions control for age, age squared, tenure, tenure squared, education level, marital status, marital status
interacted with sex, marital status interaction with a dummy5f6or female under 40, and a dummy for nonwhite. The excluded
education level is high school graduate. All regressions also include year, state, and individual-xed eects. The dependent
variable is whether the respondent has been promoted since the time of hire. Column (2) includes controls for whether the
respondent was a member of a union and whether the employer was private. Occupations were divided into a set of 12 standard
categories in Column (3) and 12 categories reecting variation in human capital depreciation in Column (4). Column (5)
includes both 24 standard occupation divisions and 8 standard industry divisions.
Table 14: Probability of Promotion Since Hire, Without State- or Demographic-Specic Time Fixed Eects
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Hired After*Female Under 40* -0.03* -0.05** -0.05** -0.05** -0.05**
Treatment State [0.019] [0.021] [0.021] [0.021] [0.021]
Female Under 40* 0.02** 0.02** 0.02** 0.02** 0.02**
Treatment State [0.008] [0.009] [0.009] [0.009] [0.009]
Hired After*Treatment State 0.02** 0.03** 0.03** 0.03** 0.03**
[0.011] [0.013] [0.013] [0.013] [0.013]
Hired After*Female Under 40 0.02 0.04** 0.04*** 0.04*** 0.04**
[0.013] [0.014] [0.014] [0.014] [0.014]
Female Under 40 -0.04** -0.04** -0.04** -0.04** -0.04**
[0.016] [0.018] [0.017] [0.017] [0.017]
Hired After -0.01 -0.02 -0.02 -0.02 -0.02
[0.010] [0.011] [0.012] [0.012] [0.011]
Treatment State 0.01 0.08 0.02 0.02 0.03
[0.031] [0.052] [0.032] [0.033] [0.034]
Tenure 0.05*** 0.05*** 0.05*** 0.05*** 0.05***
[0.002] [0.002] [0.002] [0.002] [0.002]
Tenure Squared -0.00*** -0.00*** -0.00*** -0.00*** -0.00***
[0.000] [0.000] [0.000] [0.000] [0.000]
Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 57,269 49,108 48,987 48,987 48,637
Individual Fixed Eects 13,913 12,904 12,889 12,889 12,857
Mean 0.07 0.07 0.07 0.07 0.07
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: The coecients reported are the estimated coecients from the following equation:
promotedi,j,s,t = β1(HiredAfteri,j · FemaleUnder40i,t · TreatmentStates)
+ β2(FemaleUnder40i,t · TreatmentStates) + β3(HiredAfteri,j · TreatmentStates)
+ β4(HiredAfteri,j · FemaleUnder40i,t) + β5HiredAfteri,j + β6FemaleUnder40i,t
+ β7TreatmentStates + β8Xi,t + β9Zj + λt + cs + fi + εi,j,s,t
The sample is restricted to respondents hired in 1987 or later. All regressions control for age, age squared, tenure, tenure
squared, education level, marital status, marital status interacted with sex, marital status interaction with a dummy for female
under 40, and a dummy for nonwhite. The excluded education level is high school graduate. All regressions also include
year, state, and individual-xed eects. The dependent variab5l7e is whether the respondent has been promoted since the time
of hire. Column (2) includes controls for whether the respondent was a member of a union and whether the employer was
private. Occupations were divided into a set of 12 standard categories in Column (3) and 12 categories reecting variation in
human capital depreciation in Column (4). Column (5) includes both 24 standard occupation divisions and 8 standard industry
divisions.
Figure 14: Eect of Mandate on Promotion Rate of Women Under 40
Without State- or Demographic-Specic Time Fixed Eects
87 88 89 0 1 2 3 4 5 6 7 8 9 0 1
19 19 19 19
9 99 99 99 99 9 9 9 9 9 0 01 1 1 1 19 19 19 19 19 20 20
Year of Hire
Parameter estimate 95% Confident Limit
58
Regression Coefficients
 
−.3 −.2 −.1 0 .1
Table 15: Probability of Promotion Since Hire, With Time Trends in the Year of Hire
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Hired After*Female Under 40* -0.06* -0.09*** -0.09*** -0.09*** -0.08***
Treatment State [0.030] [0.032] [0.032] [0.032] [0.031]
Female Under 40* 0.01 0.01 0.01 0.01 0.01
Treatment State*Year Hired [0.004] [0.004] [0.004] [0.004] [0.004]
Female Under 40* -0.00 -0.00 -0.00 -0.00 -0.00
Year Hired [0.004] [0.004] [0.004] [0.004] [0.004]
Female* 0.01*** 0.01*** 0.01*** 0.01*** 0.01***
Year Hired [0.002] [0.002] [0.002] [0.002] [0.002]
Year Hired -0.03*** -0.03*** -0.03*** -0.03*** -0.03***
[0.005] [0.005] [0.005] [0.005] [0.005]
Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Year Hired Fixed Eects Yes Yes Yes Yes Yes
State*Year Hired Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 57,269 49,108 48,987 48,987 48,637
Individual Fixed Eects 13,913 12,904 12,889 12,889 12,857
Mean 0.07 0.07 0.07 0.07 0.07
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: This sample is restricted to respondents hired in 1987 or later. All regressions control for age, age squared, tenure,
tenure squared, education level, marital status, marital status interacted with sex, marital status interaction with a dummy
for female under 40, and a dummy for nonwhite. The excluded education level is high school graduate. All regressions also
include year, state, and individual-xed eects. The dependent variable is whether the respondent has been promoted since the
time of hire. Column (2) includes controls for whether the respondent was a member of a union and whether the employer was
private. Occupations were divided into a set of 12 standard categories in Column (3) and 12 categories reecting variation in
human capital depreciation in Column (4). Column (5) includes both 24 standard occupation divisions and 8 standard industry
divisions.
59
Table 16: Probability of Promotion Since Hire, With Time Trends in the Year of Hire
Among Women Only
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Hired After*Female Under 40* -0.09** -0.12*** -0.12*** -0.12*** -0.12***
Treatment State [0.042] [0.040] [0.041] [0.040] [0.040]
Female Under 40* -0.00 -0.00 -0.00 -0.00 -0.00
Treatment State*Year Hired [0.006] [0.006] [0.006] [0.006] [0.006]
Female Under 40* -0.01 -0.01 -0.02 -0.02 -0.02
Year Hired [0.005] [0.005] [0.005] [0.005] [0.005]
Year Hired -0.01 -0.01* -0.02* -0.02** -0.02*
[0.007] [0.008] [0.008] [0.008] [0.008]
Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Year Hired Fixed Eects Yes Yes Yes Yes Yes
State*Year Hired Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 29,558 25,353 25,295 25,295 25,118
Individual Fixed Eects 7,218 6,775 6,770 6,770 6,753
Mean 0.06 0.06 0.06 0.06 0.06
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: This sample is restricted to women hired in 1987 or later. All regressions control for age, age squared, tenure, tenure
squared, education level, marital status, marital status interacted with sex, marital status interaction with a dummy for female
under 40, and a dummy for nonwhite. The excluded education level is high school graduate. All regressions also include
year, state, and individual-xed eects. The dependent variable is whether the respondent has been promoted since the time
of hire. Column (2) includes controls for whether the respondent was a member of a union and whether the employer was
private. Occupations were divided into a set of 12 standard categories in Column (3) and 12 categories reecting variation in
human capital depreciation in Column (4). Column (5) includes both 24 standard occupation divisions and 8 standard industry
divisions.
60
Table 17: Probability of Promotion Since Hire, With Time Trends in the Year of Hire
Among Men and Women Under 40 Only
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Hired After*Female Under 40* -0.05* -0.08** -0.08** -0.08** -0.07**
Treatment State [0.030] [0.033] [0.032] [0.033] [0.032]
Female Under 40* 0.00 0.00 0.00 0.00 0.00
Treatment State*Year Hired [0.005] [0.005] [0.005] [0.005] [0.005]
Female Under 40* -0.00 -0.00 -0.00 -0.00 -0.00
Year Hired [0.004] [0.004] [0.004] [0.004] [0.004]
Year Hired -0.03*** -0.03*** -0.03*** -0.03*** -0.03***
[0.006] [0.006] [0.006] [0.006] [0.006]
Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Year Hired Fixed Eects Yes Yes Yes Yes Yes
State*Year Hired Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 47,662 41,323 41,219 41,219 40,922
Individual Fixed Eects 12,095 11,251 11,234 11,234 11,202
Mean 0.08 0.08 0.08 0.08 0.08
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: This sample is restricted to respondents under the age of 40 hired in 1987 or later. All regressions control for age, age
squared, tenure, tenure squared, education level, marital status, marital status interacted with sex, marital status interaction
with a dummy for female under 40, and a dummy for nonwhite. The excluded education level is high school graduate. All
regressions also include year, state, and individual-xed eects. The dependent variable is whether the respondent has been
promoted since the time of hire. Column (2) includes controls for whether the respondent was a member of a union and
whether the employer was private. Occupations were divided into a set of 12 standard categories in Column (3) and 12
categories reecting variation in human capital depreciation in Column (4). Column (5) includes both 24 standard occupation
divisions and 8 standard industry divisions.
61
Table 18: Probability of Promotion Since Hire, Restricted to Those Hired Before Aug. 1993
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Post-Period*Female Under 40* 0.04** 0.04* 0.04* 0.04* 0.04*
Treatment State [0.017] [0.020] [0.020] [0.020] [0.020]
Female Under 40*Treatment State -0.00 0.00 0.01 0.01 0.01
[0.010] [0.012] [0.013] [0.013] [0.013]
Post-Period*Treatment State -0.04*** -0.05*** -0.05*** -0.05*** -0.05***
[0.015] [0.016] [0.017] [0.017] [0.017]
Post-Period*Female Under 40 -0.02 -0.03 -0.03 -0.03 -0.03
[0.016] [0.018] [0.018] [0.018] [0.019]
Female Under 40 -0.04* -0.04* -0.04 -0.04 -0.04*
[0.023] [0.025] [0.024] [0.025] [0.025]
Female 0.00 0.00 0.00 0.00 0.00
[0.000] [0.000] [0.000] [0.000] [0.000]
Post-Period 0.01 0.02 0.02 0.02 0.02
[0.013] [0.015] [0.015] [0.015] [0.015]
Treatment State 0.10** 0.12*** 0.23*** 0.23*** 0.13***
[0.041] [0.045] [0.061] [0.061] [0.044]
Tenure 0.04*** 0.05*** 0.05*** 0.05*** 0.05***
[0.003] [0.003] [0.003] [0.003] [0.003]
Tenure Squared -0.00*** -0.00*** -0.00*** -0.00*** -0.00***
[0.000] [0.000] [0.000] [0.000] [0.000]
Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 39,067 33,529 33,456 33,456 33,206
Individual Fixed Eects 10,194 9,284 9,273 9,273 9,249
Mean 0.11 0.11 0.11 0.11 0.11
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: The coecients reported are the estimated coecients from the following equation:
promotedi,j,s,t = β1(PostPeriodt · FemaleUnder40i,t · TreatmentStates)
+ β2(FemaleUnder40i,t · TreatmentStates) + β3(PostPeriodt · TreatmentStates)
+ β4(PostPeriodt · FemaleUnder40i,t) + β5PostPeriodt + β6FemaleUnder40i,t
+ β7TreatmentStates + β8Xi,t + β9Zj + λt + cs + fi + εi,s,t
This sample is restricted respondents hired between 1987 and August of 1993. "Post-period" is an indicator variable equal to
1 if the survey date took place after August of 1993. All re6gr2essions control for age, age squared, tenure, tenure squared,
education level, marital status, marital status interacted with sex, marital status interaction with a dummy for female under
40, and a dummy for nonwhite. The excluded education level is high school graduate. All regressions also include year, state,
and individual-xed eects. The dependent variable is whether the respondent has been promoted since the time of hire.
Column (2) includes controls for whether the respondent was a member of a union and whether the employer was private.
Occupations were divided into a set of 12 standard categories in Column (3) and 12 categories reecting variation in human
capital depreciation in Column (4). Column (5) includes both 24 standard occupation divisions and 8 standard industry
divisions.
Table 19: Probability of Promotion Since Hire, Restricted to Those Hired Before Aug. 1993
With Year Trends
(1) (2) (3) (4) (5)
VARIABLES Promoted Promoted Promoted Promoted Promoted
Post-Period*Female Under 40* 0.01 0.01 0.01 0.01 0.01
Treatment State [0.020] [0.021] [0.020] [0.020] [0.021]
Female Under 40* 0.01** 0.01** 0.01** 0.01** 0.01**
Treatment State*Survey Year [0.003] [0.003] [0.003] [0.003] [0.003]
Female Under 40* -0.00 -0.00 -0.00 -0.01 -0.00
Survey Year [0.002] [0.002] [0.002] [0.002] [0.002]
Survey Year 0.02*** 0.02*** 0.02*** 0.02*** 0.02***
[0.004] [0.005] [0.005] [0.005] [0.005]
State-Specic Year Trends Yes Yes Yes Yes Yes
Survey Year Fixed Eects Yes Yes Yes Yes Yes
State Fixed Eects Yes Yes Yes Yes Yes
Firm Characteristics Yes Yes Yes Yes
Occupation Controls 12 12 24
Industry Controls 8
Observations 39,067 33,529 33,456 33,456 33,206
Individual Fixed Eects 10,194 9,284 9,273 9,273 9,249
Mean 0.11 0.11 0.11 0.11 0.11
Robust standard errors in brackets, clustered at the state level
*** p<0.01, ** p<0.05, * p<0.1
Notes: The coecients reported are the estimated coecients from the following equation:
promotedi,j,s,t = β1(PostPeriodt · FemaleUnder40i,t · TreatmentStates)
+ β2(FemaleUnder40i,t · TreatmentStates) + β3(PostPeriodt · TreatmentStates)
+ β4(PostPeriodt · FemaleUnder40i,t) + β5PostPeriodt + β6FemaleUnder40i,t
+ β7TreatmentStates + β8Xi,t + β9Zj + λt + cs + fi + εi,s,t
This sample is restricted respondents hired between 1987 and August of 1993. "Post-period" is an indicator variable equal
to 1 if the survey date took place after August of 1993. All regressions control for age, age squared, tenure, tenure squared,
education level, marital status, marital status interacted with sex, marital status interaction with a dummy for female under 40,
and a dummy for nonwhite. The excluded education level is high school graduate. All regressions also include year, state, and
individual-xed eects. The dependent variable is whether the respondent has been promoted since the time of hire. Column (2)
includes controls for whether the respondent was a member of a union and whether the employer was private. Occupations were
divided into a set of 12 standard categories in Column (3) and 12 categories reecting variation in human capital depreciation
in Column (4). Column (5) includes both 24 standard occupation divisions and 8 standard industry divisions.
63
Figure 15: Gender Dierences in Promotions in the Pre- and Post-Mandate Periods
Among Women Under 40 Hired Before Aug. 1993
8 9 0 1 2 3 4 5 6 7 8 9 0 1
19
8 98 99 99 99 99 99 99 99 99 9 9 0 01 1 1 1 1 1 1 1 1 19 19 20 20
Survey Year
Parameter estimate 95% Confident Limit
Notes: The coec∑ients βτ,1 and 95% condence intervals reported are obtained from estimating
promoted = 2001i,j,s,t t=1988 βt,1(SurveyY eartt ·∑FemaleUnder40i,t · TreatmentStates)
+β∑2(FemaleUnder40 2001i,t · TreatmentStates) + t=1988 βt,3(SurveyY eartt · TreatmentStates)
+ 2001t=1988 βt,4(SurveyY eartt · FemaleUnder40i,t) + β6FemaleUnder40i,t + β7TreatmentStates
+β8Xi,t + β9Zj + λt + cs + fi + εi,s,t.
Standard errors are clustered at the state level. The sample is restricted to respondents hired in August of 1993 or later. The
regression controls for sex, age, age squared, tenure, tenure squared, level of education attained, marital status, marital status
interacted with sex, marital status interacted with FemaleUnder40, and a dummy for nonwhite. The regression also controls
for job characteristics (private rm, union), 24 standard occupational categories, and 12 standard industry categories from the
census occupational classication system.
64
Regression Coefficients
 
−.1 0 .1 .2