Está en la página 1de 49

ISSN 2539-102X (En Línea)

Vol. 20 No. 4 Documento de Trabajo en Economía

A theory of the gender pay gap. Evidence


from Colombia and the United States
Edgar Villa
Luz Karime Abadía
Ernesto A. Cárdenas

4
mayo de 2020

Facultad de Ciencias Económicas y


Administrativas Departamento de Economía
Serie: Documentos de Trabajo en Economía
Periodicidad continua

P. Arturo Sosa Abascal, S.


J. Gran Canciller

P. Carlos Eduardo Correa, S.


J. Vice-Gran Canciller

P. Jorge Humberto Peláez Piedrahita, S.


J. Rector de la Universidad

Luis David Prieto Martínez


Vicerrector Académico

P. Luis Alfonso Castellanos Ramírez, S.


J. Vicerrector del Medio Universitario

Catalina Martínez De Rozo


Vicerrectora Administrativa

P. Luis Fernando Álvarez Londoño, S. J.


Vicerrector de Extensión y Relaciones Interinstitucionales

Luis Miguel Renjifo


Vicerrector de Investigación

Jairo Cifuentes Madrid


Secretario General

Andrés Rosas Wulfers


Decano de Facultad Ciencias Económicas y Administrativas

Andrés Felipe Giraldo


Director Departamento de Economía

Silvia Gómez Soler


Directora Carrera de Economía

Luz Karime Abadía Alvarado


Directora de Posgrados en Economía

Gabriel Lubo Cárdenas


Secretario de Facultad

La serie de Documentos de Trabajo Vniversitas Económica es publicada para difusión y discusión. Estos trabajos no han sido
sometidos a evaluación por pares para su publicación en esta serie. Los contenidos son responsabilidad exclusiva de los autores y
no comprometen a la Facultad de Ciencias Económicas y Administrativas ni a la Pontificia Universidad Javeriana.
A theory of the gender pay gap. Evidence from
Colombia and the United States
Edgar Villa, Luz Karime Abadíayand Ernesto A. Cárdenasz; x

Abstract
We construct a competitive model that explains the gender pay gap per hour by a sta-
tistical discrimination mechanism based on unobserved e¤ort provided by men and women
in accordance to an intrahousehold division of labor in which only women engage in child
rearing activities. For unskilled labor an additional physical endowment e¤ect arises and
increases the pay gap against women. The model explains several empirical regularities
in the literature of regression analysis. We show empirical evidence for Colombia and the
United States that is consistent with the predictions of the model. Further, we study an
equal parental leave and equal pay for equal work policies and …nd conditions for these
policies to be e¤ective at reducing the pay gap.
Key Words: Gender Pay Gap, Statistical Discrimination, Self Selection, Equal pay for
Equal work.
JEL Code: J13, J16, J24, J71

1 Introduction
The gender pay gap is a relevant topic within and beyond the …eld of economics because
female workers have equalled almost in all features men and still fail to get the same income
as comparable male workers. In developed economies there has been a convergence across
genders in terms of human capital investment, employment prospects and outcomes relative
to men (Goldin, 2006). However, di¤erences in pay and employment levels that men and
women perform in the labor market remain (Blau and Kahn, 2006, 2007, 2017; Goldin, 2006,
Adjunct Researcher, Escuela Internacional de Ciencias Económicas y Administrativas, Universidad de
La Sabana. Address: Campus Puente del Común, Chía, Colombia. Corresponding author. Email:
edgarvipe@unisabana.edu.co. This research did not receive any speci…c grant from funding agencies in the
public, commercial, or non-pro…t sectors.
y
Associate Professor of the Department of Economics. Ponti…cia Universidad Javeriana, Bogota-Colombia.
Email: labadia@javeriana.edu.co.
z
Researcher, School of Economics. Universidad Sergio Arboleda. Email: ernesto.cardenas@usa.edu.co
x
We wish to thank the participants to the workshop seminar at Ponti…cia Universidad Javeriana in Cali and
Universidad del Rosario. We also would like to thank Kevin Lang for helpful comments on a previous draft.
Evidently in spite of our best e¤orts any remaining errors are our own.

1
2014; Olivetti and Petrongolo, 2016). The gender pay gap has been persistent in industrialized
developed economies which seems puzzling up to a certain degree given that these economies
have enacted policies to reduce the gender pay gap more so than other less developed economies
(Polachek 2014). For less developed economies things seem to be worse in terms of gender gaps
in human capital investment, employment and wages.
Even after enacting policies that are aimed to reduce the gender pay gap, like equal pay for
equal work type of policies, why is the gender pay gap still persistent in developed economies?
One standard explanation is the Beckerian taste based discrimination of employers against
women (Becker 1971). Still in competitive environments …rms that use this type of discrimi-
nation would be outperformed in the long run by …rms that do not as Becker (1971) showed.
Hence we ask if gender pay gaps could still persist in competitive environments, even after taste
based discrimination is eliminated in society, say by enacting laws that prevent women from
su¤ering this type of discrimination.
To answer this question we develop a model of a standard competitive economy that has
skilled individuals where men and women are considered perfect substitutes in skilled labor jobs
i.e. have the same disutility (in the absence of childrearing activities) of labor once they invest
in education, while they are assumed to have a di¤erent disutility only for unskilled labor jobs,
greater disutility for women than for men due to initial physical endowments. We assume that
women have the role of childrearing and housework in the economy which generates an extra
cost in e¤ort that reduces their supply of e¤ort in the workplace relative to men. We assume
the existence of a statistical discrimination mechanism against women in the workplace that
does not allow single women to credibly signal to employers that they would not have children
while they work in their jobs. This forces employers to use group information across genders
to determine wages based on average productivity. This generates a statistical discrimination
mechanism based on e¤ort, not on innate ability, which explains gender pay gaps across skilled
and unskilled occupations. The heterogeneity of physical endowments widens the gender pay
gap against women only for unskilled labor.
We establish the conditions for the existence of a rational expectations competitive equi-
librium that can rationalize the emergence of gender wage gaps across skilled and unskilled
occupations, under a statistical discrimination mechanism and physical heterogeneity on initial
endowments across genders. Our model is able to explain the following regularities: i) Neg-
ative relationship between fertility and female education, ii) Motherhood penalty (Anderson

2
et al., 2002; Goldin, 2014; Sigle-Rushton and Waldfogel, 2007; Hardoy et al., 2015 ) in which
women with dependent children face an income loss relative to women without dependent chil-
dren regardless of occupation; iii) Fatherhood premium (Francois, 1998, Goldin, 2014) in which
married men earn a higher wage per hour than comparable male workers, iv) Reverse trend
in education (Becker, Hubbard, and Murphy, 2010; Blau and Kahn, 2017) in which women in
more developed economies tend to invest more in education than men; v) ambiguous relation-
ship between the gender pay gap and education attainment (De la Rica et al., 2008, De la Rica
et al. 2015), vi) positive statistical relationship between wage inequality and gender pay gaps
(Blau and Kahn 2001).
A key contribution of our approach is that we are able to deduce an econometric population
regression equation implied by the model and rationalize several of the …ndings of the linear
regression approach used in the literature. We …nd empirical evidence that is consistent with
the theoretical predictions for a developed economy like the United States as well as for a less
developed country like Colombia.
A precise and detailed understanding of the emergence of a gender pay gap seems relevant
for assessing public policies aimed at closing the gap.1 Normatively we study two policies that
are thought to reduce the gender pay gap: an equal parental leave policy and an equal pay
for equal work policy. We …nd that each of these policies can be ine¤ective in reducing gender
pay gaps if they are not accompanied by a change in the role of women within a household.
We …nd that the …rst policy can reduce the gender pay gap for skilled labor by reducing the
inequality in the allocation of e¤ort/energy exertion in child-rearing activities among men and
women as well as changing the rational expectations about the stereotype belief for employers.
Some evidence of this e¤ect is found for within-couple managers by Hardoy et al., 2107.
For equal pay for equal work legislation, which mandates that skilled women and men are to
be paid by law the same wage rate for the same job, we show that if the role of women within
the household changes with this policy then the policy could potentially reduce the gender pay
gaps across occupations. Nonetheless if it does not change the female role within a household
then it could induce employers to implement a screening mechanism. This is because women
would be less productive than men on average which makes employers substitute them for men.
If the fertility rate is su¢ ciently large then the policy would generate the same e¤ective labor
levels and equilibrium relative wages across occupations that correspond to a REE by replacing
1
The importance of theoretical models supporting causal relationships among variables is a key imput for
the design and implementation of public policies. See Deaton (2009).

3
the statistical discrimination mechanism with a screening mechanism in which they o¤er two
contracts: one that pays a low wage rate for part time hours and a high wage rate for full time
hours. Workers self select optimally where to apply and we …nd that women with dependent
children would choose the part time contract while women without dependent children and men
choose the full time contract. Hence a separating allocation among women emerges under the
policy that would not reduce the gender pay gap in the aggregate and would actually deepen
the motherhood penalty.
The following sections of the article are organized in the following way. In the second section
we layout our conceptual framework in which e¤ort is conceptualized as the key driver of the
gender pay gaps, the third section presents a brief literature review in which our approach …ts in,
the fourth section presents the theoretical model, the …fth section presents the theoretical results
we …nd, the sixth section presents the reduced form econometric regressions that emerges from
the theoretical model as well as the empirical results, the eight section presents the analysis of
public policies that are designed to reduce the gender pay gaps while the last section concludes.

2 Conceptual Framework
Under perfectly competitive labor markets if men and women are identical then one should
expect to observe wage convergence across genders, in the absence of taste based discrimina-
tion. The empirical fact that a gender pay gap exists in all economies would necessarily imply
that either of two things arise: either labor markets are not perfectly competitive or men and
women are not identical in labor markets. In this article we focus on the second case where
we retain the idealized competitive environment in labor markets and study the gender wage
gap that comes from di¤erences across genders from two sources of heterogeneity: physical
(strength) endowments and a stereotype role in which women are responsible for child rearing
and housework activities within a household. The …rst heterogeneity is an ability di¤erence
based on physical attributes that generates a productivity di¤erence for unskilled jobs across
genders in occupations in which these endowments are relevant for employers. The second het-
erogeneity generates a productivity di¤erence across genders given that women …nd themselves
in the role of child rearing and housework activities which a¤ects the amount of e¤ort/energy
exertion they can supply in the workplace. The novelty of our approach is that it develops
a full microeconomic model that explains the gender pay gap that arises for both developed
(low fertility rate) as well as underdeveloped (high fertility rate) economies, something that is

4
absent in the literature which has focused more on developed economies.
The child rearing and house work heterogeneity is based on the social expected role that
women have of being responsible for these activities within a household. This generates a
statistical discrimination mechanism because employers cannot condition wage contracts for
women depending on whether they would have children or not. To illustrate the statistical
discrimination mechanism on e¤ort consider two identical skilled workers, a woman and a
man, that enter the labor market (in a particular …rm) with the same wage (for now) and job
responsibilities (including the same hours required of work) when hired and suppose that both
have the same number of dependent children with their respective (working) spouses. After a
certain period of time there is an evaluation of their performance based on their productivity
record, which is used to assess their merits for a wage increase. By productivity we understand
the amount of observable production accomplishments (say projects completed) by each worker
given that the e¤ort/energy exertion is not perfectly observable and veri…able by employers.
Assume that the cost of child rearing activities and housework falls on the respective woman in
each household and can be understood as a cost in terms of energy exertion in the e¤ort women
supply at work that comes from o¤ work hours activities. This cost a¤ects the e¤ort/energy
with which the woman that is hired performs her job but does not a¤ect her male coworker in
accordance with the role women have in their domestic chores, rational expectations by employer
would necessarily be consistent with this female role. Thus child rearing and housework costs
generate an external e¤ect for the woman which eventually would make the man outperform
her and obtain a wage increase relative to the otherwise identical female coworker.2 Hence
identical skilled workers get paid di¤erently based on productivity di¤erences that arise on the
job performance due to the external costs o¤ the job activities that a¤ects the supply of e¤ort.
If one wishes to model this situation in a static competitive equilibrium model then rational
expectations for employers would make them expect a productivity gap that would unfold in
the future in favor of men. Hence in equilibrium at the hiring moment a wage pay gap arises
against women due to this statistical discrimination mechanism.
To be clear we argue that the intensive margin of e¤ort/energy exertion in the labor life
of men and women is the key driver of the gender pay gap for skilled labor in the statistical
discrimination mechanism under rational expectations and not the extensive margin (number
2
In reality things could be quite more di¢ cult for women since the duties of the job pro…le and the child
rearing costs could make them either quit the job or make them take a part time job with a lower pay and lower
e¤ort/energy exertion.

5
of hours) since we are considering the gender pay gap in per hour terms.3 Women that foresee
not having children in their life would be a¤ected by this statistical discrimination mechanism
since they cannot credibly signal to the market that they would not have children during their
lifetime. The explicit role in the gender pay gap between extensive and intensive margins is a
contribution of our model to the existing theoretical literature.
The source of heterogeneity based on physical strength generates an additional source of the
gender pay gap for unskilled labor beyond that of the statistical discrimination mechanism. This
is because physical heterogeneity across men and women for unskilled jobs has an additional
e¤ect on the gender pay gap for these occupations. To illustrate consider two unskilled workers,
a woman and a man, where the latter has a greater physical endowment than the former which
manifests as lower disutility for physical unskilled work. This generates a higher sustained e¤ort
for men compared to women, within unskilled jobs, which then manifests as an additional gender
pay gap against women.
In the empirical literature the gender pay gap is measured as the di¤erence in per hour pay
between men and women. We argue that the hourly gap must re‡ect a per hour average pro-
ductivity di¤erence in competitive markets that remunerate work in accordance to productivity.
Two skilled workers can work in the same job and actually work an equal amount of hours to
perform their job but could end up having di¤erent productivity levels. If skilled workers have
the same ability (something we assume) and work the same amount of hours then the produc-
tivity di¤erence would necessarily come from e¤ort/energy exertion di¤erences. We therefore
consider e¤ective labor as the key ingredient for productivity di¤erences which is de…ned as ef-
fort (understood as energy exertion) multiplied by the number of hours worked. E¤ort/energy
exertion is assumed to be unobservable or unveri…able by employers. This assumption is at the
core of the statistical discrimination mechanism we study since it does not allow employers to
have certainty over the amount of e¤ort/energy a worker would be able and willing to supply
in the workplace given that the number of hours observed do not reveal any information in this
regard. This would force employers to take decisions to evaluate individual labor characteristics
based on say the use of aggregate group information such as group averages of e¤ort supplied
to the market by each gender group in each occupation. The statistical discrimination that we
study is di¤erent than the typical statistical discrimination mechanism used in the literature
that is based on types of workers and not on e¤ort that is actually chosen by individuals. We
3
The additional extensive margin e¤ect would increase furthermore the pay gap against women on top of
the intensive margin on the e¤ort/energy exertion.

6
turn to the literature review.

3 Related literature review


Several theories of discrimination have attempted to explain why two groups of individuals
are paid di¤erent average wages. They have attributed di¤erences to either employers’preju-
dices, i.e. taste discrimination like in Becker (1971), or imperfect information, i.e. statistical
discrimination like in Phelps (1972) and Arrow (1973). However, all of these theories in their
earliest and simplest version are not supported by empirical evidence. Models built on taste
discrimination predict the elimination of the male/female di¤erential in the long run, whereas
their persistency has been observed for decades in most OECD countries. Models based on
statistical discrimination have proved deceptive since they have been unable to explain a gen-
der/racial gap in mean wages. Nonetheless in contrast to taste-based theories of discrimination,
statistical discrimination theories derive group inequality without having to assume preference
bias against members of a minority group. In this sense models of statistical discrimination are
more attractive since they suggest discrimination may be a rational response by employers to
imperfect information on individual productivity.
For example Oettinger (1996) uses a statistical discrimination model that incorporates the
notion of job-matching. In particular, the productivity of any job match is imperfectly observed
ex ante by employers, that the initial productivity signal is noisier for the “minority” group
(i.e. women) and that additional information about the productivity of the match is revealed
with tenure on the job. In this framework productivity is something that is not chosen by an
individual but that is an innate ability of the individual that is to be revealed to the employer,
more di¢ cult for women than for men. Altonji and Blank (1999) argue that the costlessness
of verifying worker quali…cation constitutes the major drawback to this type of reasoning since
transcripts, resumes and references are easily available and can be veri…ed, and ultimately
workers can also be submitted to tests in order to verify their productivity. Hence the noisy
assumption for the minority group seems unrealistic unless it is justi…ed by another mechanism
which is what we study in this article. There is another dimension to productivity that is
related to the sustained e¤ort that is chosen by an individual and not as an innate ability that
an individual is endowed with when working.
Our approach di¤ers from the traditional statistical discrimination approach since we assume
that what is unobservable is the e¤ort provided by individuals which di¤ers across genders since

7
only women have the social role of childrearing and housework activities. This mechanism
generates productivity di¤erences among men and women due to the di¤erent roles they play
in the household. Since women (during their age for childbearing) cannot signal credibly to
employers that they will not have children then employers have the rational belief that women
eventually will end up having dependent children sooner or later. Rational expectations then
has to include this eventuality which is re‡ected in the wages o¤ered to women and men. More
importantly the Altonji and Blank (1999) critique would not apply here to our approach since
even if an employer learns the productivity of an individual there is no guarantee that it will
be the same productivity level supplied since individuals choose the amount of e¤ort to provide
in the workplace, which is restricted by the amount of o¤ the job activities that also require
e¤ort.
Machado (2008) uses a statistical discrimination model as an explanation for gender dis-
crimination. She argues that fertility intentions of women are treated as the key unobservable
characteristic for employers since those that intend to have children may become less productive
in the future if they plan to allocate less e¤ort to employment after childbearing. A compar-
ative advantage of women in child raising activities relative to men cause women to specialize
in these activities within the household making them less productive at work over the longer
term. Equilibrium is characterized by population fertility rates that exactly match employer’s
beliefs about fertility intentions. Since motherhood is believed to decrease women’s productiv-
ity, women’s wages are a decreasing function of expected future fertility rates. The gender wage
gap is explained in the model for equally skilled individuals that are paid di¤erently according
to gender, with the gap being proportional to their expected fertility rates. The model predicts
a negative correlation between skill and fertility choices and a smaller rate of skill investment
among women.
An alternative model for a statistical discrimination mechanism has been provided by Al-
banesi and Olivetti (2009) where employer’s uncertainty is related to household division of labor
and hours in home production. Allocation of home hours depend on spouses’relative earning,
while labor contracts in turn depend on expected hours at home production. A self-ful…lling
equilibrium in which women allocate more time to home production and earn lower wages
arises. Another theoretical contribution is that of Chichilnisky (2009). This author develops a
non-cooperative game theoretic model that explains the gender pay gap as a Nash equilibrium
of a game with incomplete information about women’s work at home and in the marketplace.

8
Expectations about women’s lower wages leads to over utilization of women in the household
which in turn leads to lower productivity and lower wages for women in the marketplace, simi-
lar to a prisoner’s dilemma situation that is Pareto inferior. According to Chichilnisky (2009):
"Inequity at home breeds inequity in the marketplace and reciprocally, leading to a persistent
gender gap". This author argues that "with learning by doing, at high levels of skill there is a
Pareto superior equilibrium where men and women share e¤orts equally at home and receive
the same pay in the marketplace, …rms enhance their pro…ts, and there is additionally more
welfare at home". The author proposes that Family Law and appropriate contracts can help
resolve this Pareto inferior situation as well as increase productivity and economic growth in
the economy as a whole.
Francois (1998) provides also a theoretical strategic explanation for the gender pay gap in
competitive labor markets but unlike other explanations the theory developed arises from the
interaction between women and men within the household where if one member of a couple
alone has an e¢ ciency wage job, both can bene…t by a trade which increases the household work
of the low wage member (women in equilibrium) in return for a monetary transfer from the high
to the low wage member. To a person in an e¢ ciency wage employment, this bene…t from trade
acts as a bene…t in kind occurring in addition to the e¢ ciency wage payment. Employers can
lower e¢ ciency wage payments while still satisfying a worker’s incentive compatibility constraint
because threat of job loss entails loss of gains from trade in addition to loss of the wage premium.
This is where the desire of a …rm to discriminate comes from. If all other …rms discriminate
and allocate e¢ ciency wage jobs to one gender only, then a single …rm can ensure its employees
receive this bene…t in kind. Thus discrimination by other …rms makes it strictly better for any
one …rm to also discriminate. The author argues that if labor market discrimination itself causes
gender di¤erences then altering a discriminatory outcome may require direct policy responses
such as a¢ rmative action.
In summary from Machado (2008) we take the point that children play a key role in the
restrictions household women face in the workplace given the role women with dependent chil-
dren play in childrearing and housework activities. From Albanesi and Olivetti (2009) we take
the importance of the division of labor within a household and its e¤ect on the e¤ort and hours
women o¤er in the labor market. From Chichilnisky (2009) we take the key assumption that
labor contracts cannot be based directly on unobservables out of work activities of employees
but that still can a¤ect the e¤ort women supply in the workplace which in turn a¤ects the way

9
they may be treated in the labor market. From Francois (1998) we take the point that e¤ort
plays a key role in generating the gender pay gap.

4 The model
Consider an economy that produces a consumption good C, which we normalize its price
to unity, with physical capital, skilled and unskilled labor. Skilled labor is non-physically
intense and requires education investment by an individual while unskilled labor is a physically
intense e¤ort job that does not. Real wages per hour for the two type of jobs are denoted
!n (! n;w ; ! n;m ) and ! H ! H;w ; ! H;m that can vary across genders, where n denotes
unskilled labor, H denotes skilled labor, w denotes women and m denotes men. These wages
are determined in competitive markets where employers have rational expectations about the
average e¤ort levels individuals provide in the work place based on aggregate information across
men and women related to childrearing activities and housework.
Labor has both extensive and intensive margins. The former is the amount of hours that
workers are willing to o¤er in labor markets like in any standard model while the later is the
amount of e¤ort per unit of time that workers exert in the work place. E¤ective labor is
de…ned to be the multiplication of the intensive and extensive margins which is the amount of
labor that employers are interested in obtaining when they hire workers. Our key assumption
is that individuals not only choose the amount of hours to o¤er but also the amount of e¤ort
they exert. We conceptualize e¤ort as energy that is expended when individuals work and it
is assumed to be imperfectly observed by employers. Women and men di¤er in their o¤ work
activities, namely due to child rearing activities and housework, which generates a cost in terms
of e¤ort/energy exertion for individuals that are in charge of these activities.
The second key assumption is that there is a social norm belief in the economy that only
women are the ones that stay at home to take care of children and do housework (e.g. cooking,
laundry etc).4 There are also individuals of both genders that do not incur in child-rearing
costs either because they do not want to have children or if they do they already have their
children grown up and therefore are not dependent.
4
We follow Becker (1985) in assuming that men and women within a household have a division of labor such
that women are in charge of childrearing and housework. The role of women in the economy is taken as a social
convention or norm that is exogenous. This does not have to be necessarily the case since it can be considered
to be an endogenous convention that depends on how men and women in households negotiate and share these
o¤ the work activities (Francois 1998). Even so we consider this social convention as being exogenous due to
historical reasons that make it quite persistence the role that women have in terms of these activities.

10
Utility for an individual comes from the consumption of a good while disutility comes from
e¤ective labor that increases marginally due to child rearing and housework activities. This
can be represented by the following quasilinear utility function

I;M l
U = ln c (1)
2
where c denotes the consumption good, which is the source of energy that an individual can
acquire through the energy production function5 = c; l represents the number hours which
i;I;M
is the extensive margin of labor. E¤ective labor is de…ned to be l, > 0 is the disutility
weight on e¤ect labor for an individual de…ned as

i;I;M
Di;I + S (1 M ) h + M (nc + 2h)

i
where Di;I (1 I) +I represents the disutility of labor for i = w; m which depends
on the occupation the individual chooses I = 0; 1 where I = 0 denotes low or no education
i
(unskilled labor) and I = 1 denotes high education (skilled labor). The parameter denotes
the disutility weight of unskilled labor that di¤ers across genders i = w; m while denotes the
disutility weight for skilled labor which is identical across genders since education is assumed
to level the skills for men and women. We assume that women incur in a higher disutility than
m w
men for unskilled jobs i.e. < which are based on physical e¤ort and that skilled labor
m w
jobs generate a lower disutility of e¤ort i.e. 1 < < . The variable M is an indicator
variable that takes the value one if a woman is in a couple and potentially would become a
"mother" and zero otherwise where nc is the number of children a woman bears and rears. The
intuition behind the utility function speci…cation that we use is that marginal disutility of l
increases in absolute value with e¤ort per unit of time and is ampli…ed by having a greater
disutility parameter DI and a greater number of children nc and housework when M = 1. S
5
The conceptual framework that e¤ective labor is the multiplication of time (hours) by energy/e¤ort in the
utility speci…cation is inspired by Hamermesh and Rees (1988) pgs 57-58. Key insight is that labor economics in
general treats supply of labor as hourse supplied and does not take into account the supply of e¤ort, assuming
that the intensity of e¤ort is always the same per hour. "Each hour at work has been treated as if it measured
the same amount of e¤ort being o¤ered and thus the same e¤ective supply of labor. In reality this may not be
the case. If workers are erractic in their attendance at the workplace, their productivity per hour on the job
will be less than that of otherwise identical workers and their cost to their employers may be too great."pg 57.
"Hours away from work are easy to identify. Yet at least as important but far harder to measure is e¤ort while
at work. The pace of work should be faster where hours are short than where they are long, since a worker
cannot sustain an intense pace for long periods. There should also be an increase, within some range, in the
intensity of work as real wages rise. Where real wages are very low, a wage increase may call forth a greater
supply of e¤ort simply because better-paid workers will have a more adequate diet and better medical care, and
will therefore, for essentially biological reasons, be able to work harder" pg. 58.

11
is a binary variable that takes the value of one if the individual is single (no children). Note
that mothers would have twice the amount of housework, denoted h, when M = 1 relative to
fathers since the latter would avoid doing housework because the corresponding mother incurs
this cost.6
We assume nc as the same for all women and exogenous in the model where we use it to
de…ne the development of an economy. If nc is su¢ ciently low (resp. relatively high) then
the economy is said to have completed (resp. not completed) a demographic transition that
makes it become a developed (resp. underdeveloped) economy. There is no theoretical value
that de…nes a threshold value for nc . Empirically a low nc is less than two while it is relatively
high for values greater than seven. For our results to follow we do not need a threshold value
but just a sense of low and high fertility levels.7 We abstract from housework initially to make
the model as simple as possible and then we incorporate it back to explain the marriage male
premium. The bene…t of doing this is that all results follow through and make the presentation
of the model much simpler.
We assume that e¤ort/energy exertion is unobservable by employers and a social norm exists
in the economy in which women have the responsibility of child rearing. This gives rise to
an e¤ort statistical discrimination mechanism in which …rms use rational expectations across
genders to evaluate a worker’s characteristic based on aggregate group information, such as
group averages in e¤ort supplied by genders. This implies that workers belonging to di¤erent
genders are treated di¤erently even if skilled men and women are considered to be perfect
substitutes from the point of view of the production side.
Human capital (H m ; H w ) is supplied by men and women who decide to invest in education
at proportional cost ei of per unit of time for i = w; m where perfect …nancial markets are
assumed to exist. This means an individual that takes the decision to study can do so without
…nancial constraints. In this sense the decision of an individual to study and get a skilled labor
job depends only on his or her disutility of skilled e¤ective labor and the net return per unit
of time on education ! i;H (1 ei ) where ! i;H is the gross skilled wage for individual of gender
i8 and assume ew em . Moreover assume that there exists nm men and nw women in the
6
Important to stress that in the term i;I;M of the utility speci…cation the units of measurement of h, nc are
in the same units as disutility Di;I .
7
If nc is endogenized then the level of technology, in the neoclassical production function, would be the
criterion to de…ne a developed or underdeveloped economy since it would determine the amount of children
per woman given that it determines the labor occupations of women and therefore the amount of children they
would bear and rear.
8
Note that ei varies across men and women since it represents the cost of education as a proportion of total

12
economy and denote by nh min fnm ; nw g the number of couples/parents in the economy.
The budget constraint for an individual of gender i = w; m in occupation I is given by

c W i;I l (2)

where we normalize the price of the consumption good to one and W i;I I! i;H (1 ei ) +
(1 I) ! i;n represents the real wage per hour that depends on the gender and occupation of an
individual. Here we conceptualize e¤ective labor as being remunerated by the real wage per
hour. Since e¤ort is assumed to be unobservable then the real wage is the true price of e¤ective
labor, not only of an hour of time, since an hour of time can be used to shirk at the workplace
if the employer does not have the capacity to detect it. What employers are interested in hiring
is the time and e¤ort of a worker, where the latter only arises through work time.
The production side of the economy produces the aggregate consumption good C through
a neoclassical constant return to scale production function with both types of e¤ective labor
H, Ln and physical capital K under constant returns to scale

C = F (A; K; H; Ln ) (3)
wage income earned by the individual of gender i. Since wages in the labor market end up being di¤erent in
equilibrium for men and women then the proportional cost of education relative to the income wage that an
individual earns must also be di¤erent for men and women even though the total cost of education would be
the same. We can illustrate how the proportional cost ei di¤ers across genders in a simple two period set up.
Suppose that an individual of gender i = w; m chooses to study in the …rst period and work as a skilled worker
in the second period. The present value of that choice is

! i;H li
P V H;i = h+
1+r
where r is the interest rate which we assume it to be equal to the discount rate for the individual, h is the cost
of education which is assumed to be the same for men and women and ! i;H is the gross real wage per unit of
time that the individual can obtain. In order to write the present value as ! i;H 1 ei assume that the cost
of education for individual i is a proportion of the gross wage rate the individual receives in the future under
perfect …nancial markets, namely h = i ! i;H for a given i 2 (0; 1) where h is the cost per unit of time of
schooling. Hence the present value can be written as

! i;H li
$i;H li 1 ei = i
! i;H li +
1+r
1
= ! i;H li i
1+r

This implies that


r
ei + i
1+r
which means that proportional cost ei varies across genders due to i
.

13
where A is the technology level, K is physical capital, H denotes skilled e¤ective labor or human
m;H w;H
capital de…ned as H lm;H H m + lw;H H w , H i represents the number of skilled laborers
of gender i hired (extensive margin) by the production sector, li;H represents the amount of
i;H
hours the …rm hires a skilled worker of gender i and represents the unobserved expected
e¤ort/energy from each laborer of gender i when hired (intensive margin). Similarly we de…ne
w;n n;m
Ln lw;n Lw;n + ln;m Ln;m as the e¤ective amount of unskilled labor where the term Li;n
represents the number of laborers (extensive margin) hired of gender i where li;n represents the
n;i
amount of hours the …rm hires worker of gender i in the unskilled occupation n; the term
represents the unobserved expected e¤ort/energy exerted by an unskilled worker in occupation
n and gender i.9

4.1 Optimal decisions for consumers


Consider the problem of maximizing utility for an individual according to equation (1) subject
to the budget constraint in equation (2) and function of e¤ort/energy
I;M l
max ln c (4)
c;l T 2
s:t: : c W i;I l, = c.

Since utility is increasing in c then the budget constraint has to bind then …rst order condition
is
1 i;I;M
W i;I l = 0
l
which yields the individual supply of time and e¤ort as well as the demand for the consumption
good
s
W i;I
ci;I;M
d = i;I;M
s = i;I;M
for i = w; m (5)

i;I;M i;I;M 1 1
s ls = i;I;M
; lsi;I;M = p i;I;M
W i;I
9
The set up simpli…es into a standard analysis if one assumes that men and women supply the same amount
of e¤ort as in a perfectly competitive model in which e¤ort is observable by employers. To see this note that
the a neoclassical production function, say a Cobb-Douglas with constant returns to scale, can be written under
the assumption that men and women do not di¤er in the average e¤orts they supply in both type of jobs
C = AK 1 2 H lH H (2 n ln Ln ) or equivalently C = BK 1 H (Ln ) which shows that the average
e¤ort levels are imbeded in the technological coe¢ cient B A 2 H lH (2 n ln ) associated with skilled and
unskilled labor in the production function. In this representation H and Ln would denote the number of workers
and not e¤ective labor. This shows that e¤ort supplied by an individual is key to the production side of the
economy since without it (say B = 0 would arise under zero e¤ort) the production of the good would be zero.

14
where the only restriction is that lsi;I;M cannot exceed the endowment of time T . The indirect
utility function is
1 W i;I
U i;I;M = ln i;I;M
1 for i = w; m (6)
2
Importantly from (5) note that the number of hours lsi;I;M supplied is a decreasing function of
i;I;M
real wages W i;I and of the number of children nc for mothers (M = 1) while that of e¤ort s
i;I;M i;I;M
is increasing such that the amount of e¤ective labor s ls is independent of wages but
does vary across gender, occupations and motherhood which is a key insight of our approach.10
To make matters simple assume h = 0 which we then relax.

4.2 Occupation choices


Using the de…nitions of W i;I , Di;I and the indirect utility function in equation (6) we can
determine the occupation choices of each type of individual across genders. An individual
without dependent children (M = 0) would prefer I = 1 relative to I = 0 if and only if
U i;1;0 U i;0;0 or equivalently

! i;H
i for i = w; m. (7)
! i;n (1 ei )
w m ! i;H
Since by construction > then m > w which means that as the relative wage ! i;n
decreases men would be the …rst ones that would switch to unskilled jobs relative to women
since they would require a lower relative wage for this type of job to be appealing. Note that
men with or without dependent children are identical in the work place.
10
Even though we take the number of children as exogenous and equal for all women for simplicity, it can be
endogenized in the following way for h = 0. Suppose the utility function is
i i
Di;I + M nc l
U i = ln ci + M ln (nc + 1)
2
where 2 0; 12 is the utility weight on having children. The utility payo¤ M ln (nc + 1) arises only if M = 1
and nc > 0. First women solve utility maximization problem taking nc as given. The indirect utility function
i;I
is then U i;I;M = 21 ln Di;IW+M nc + M ln (nc + 1) 12 . This function can be maximized with respect to nc
i;I
by using the …rst order condition for M = 1 which yields nc +1 2(Di;I1 +nc ) = 0 and therefore nI; c = 2 1D 2 1
for I = 0; 1which is the optimal amount of children (integer solution evidently). Note that the solution is
independent of the wages and that it is an increasing function of.Di;I which means that women working in
unskilled jobs have optimally more children than skilled women, a prediction that is consistent with empirical
evidence. Finally this solution is valid if the second derivative
p
of the objective function evaluated at the solution,
i;I
1 I; 2 D 1
2 + 2 is strictly negative or n c < p : This is satis…ed for I = 0; 1 if the parameter
(nI;
c +1) 2(D i;I +nI;
c )
1 2

is su¢ ciently greater than one for a given level of 2 0; 12 since that guarantees that Di;I > 0 is su¢ ciently
large for a positive solution. Assuming then that nc is exogenous is without loss of generality.

15
A woman with dependent children (M = 1) prefers to work in a skilled e¤ort job (I = 1)
relative to an unskilled one (I = 0) if and only if U w;1;1 U w;0;1 or equivalently

! w;H + nc
w
! w;n ( + nc ) (1 ew )
+nc 11
Note that w
+nc
> w which means that mothers would …nd unskilled jobs appealing for a
lower relative wage than non mothers. Table 1 summarizes the occupation choices for both
types of women among the two jobs.

n o M =0 M =1
! w;H +nc
! w;n
> max w ;
(1 ew ) ( w +nc )(1 ew )
I=1 I=1
+nc ! w;H
w w I=1 I=0
( nc )(1 ew )
+n ! w;n (1 oew )
! w;H
min ( w +n+n c
w ;
c )(1 e )
w >
(1 ew )
I=0 I=0
! w;n
Table 1: Occupation Choices for Women

4.3 Optimal decisions for production


Constant returns to scale in the technology implies that pro…ts are zero and therefore consumers
determine the amount to be produced of the consumption good in the economy, the amount of
time to supply in the workplace, the occupation in which they would choose to work in and the
ensuing e¤ort they would supply. The production side determines then the relative real wages
for each occupation, the amount of laborers to work in the two labor occupations as well as the
amount of physical capital required in order to produce a certain amount of the consumption
good to minimize cost. The cost minimization problem for a given required level to be produced
of C is given by

min ! m;H lm;H H m + ! w;H lw;H H w (8)


K;H i ;Li;n
w;n w;n
+! l Lw;n + ! m;n lm;n Lm;n + rK
s:t : F (A; K; H; Ln ) C for C > 0
m;H w;H w;n m;n
H lm;H H m + lw;H H w ; Ln lw;n Lw;n + lm;n Lm;n
w;H
11
When the number of children is endogenous then U w;1;1 U w;0;1 is equivalent to !!w;n
0; 2 1; 2
(nc +1) ( +nc ) 0; 1; (n0;
c +1) ( +n1;
c )
2 0; . Since n c > nc then the more likely case is 2 > w for a given
(nc +1) ( w +nc )(1 e )
1; i
(nc +1) ( w +n0;
1;
c )

value of ei which gives rise to the same conclusions that we get in the model from the stronger assumption that
the number of children is exogenous.

16
First order conditions with respect to K, Ln;i and H i for i = m; w are

i;H
r = Fk ; ! i;H = FH (9)
i;n
! i;n = FLn for i = w; m

Hence relative wages across genders for skilled and unskilled jobs satisfy
w;j
! w;j
= m;j for j = H; n (10)
! m;j
which re‡ects that relative wages across occupations are determined in terms of relative e¤ort
levels supplied by skilled men and women re‡ecting productivity di¤erences.

4.4 Rational expectations and group averages


Since e¤ort/energy exertion is assumed to be unobserved by employers then rational expecta-
tions allows them to take decisions to evaluate individual labor characteristics based on the
use of aggregate group information, such as group averages of e¤ort supplied to the market by
each gender group in each occupation. Hence rational expectations on the supply of e¤ort by
men and women in skilled and unskilled labor markets must equal the group averages across
occupations

i;n nh i;0;0 nh i;0;1


1 s + s ; (11)
nw nw
i;H nh i;1;0 nh i;1;1
1 s + s , for i = w; m
nw nw
i;I;M
where e¤ort supply functions s for i = w; m come from equation (5) and are to be deter-
mined in equilibrium depending on the occupation choices of individuals. Since in this simple
set up nc is assumed to be the same across occupations then the group averages use the same
nh
weight nw
across occupations.12

4.5 Rational expectations competitive equilibrium


De…ne a rational expectations competitive equilibrium (REE) as a set of positive wages ! i;j > 0
for i = w; m and j = H; n and an allocation of workers across occupations (Lw;n; , Lm;n; , H m; , H w; )
12 nh
The weight nw on women with children can vary across occupations if nc would be endogenized in the
model.

17
where consumers maximize utility subject to their budget constraint, the productive sector min-
imizes costs and attain zero pro…ts, labor markets balance and expected e¤orts across genders
are equal to group averages under rational expectations.
Proposition 1. There exists a REE.
Proof: See Appendix.

5 Theoretical results
In this section we study the properties of a REE with respect to di¤erences among men and
women.
Proposition 2. a) In a REE women earn less per hour than men within occupations; b)
gender pay gaps across occupations decrease with lower fertility rates in the economy.
Proof: See appendix.
This proposition explains the main empirical regularity of the existence of a gender pay gap
in a per hour basis among workers due to the statistical discrimination mechanism. Unskilled
labor have a greater gender gap due to the di¤erence in physical endowments across genders.
Moreover, these gaps arise because women o¤er on average less amount of e¤ective e¤ort/energy
relative to men in equilibrium consistent with rational expectations from employers. The gender
pay gaps within the two type of labor occupations in a REE are
" s !#2
w;H w
! 1 e nh
= 1 1 (12)
! m;H 1 em nw + nc

" s !#2
m w
! w;n nh
= w 1 1 w : (13)
! m;n nw + nc
Note that these pay gaps do not change with the level of technology. Moreover, the …rst
of these gender pay gaps, which is for skilled workers, depends on the number of children that
women bear, the fraction of women with dependent children and education costs. As the av-
erage number of children decreases then the pay gaps decrease correspondingly. Alternatively,
if we keep constant the number of children per woman and lower the fraction of women with
dependent children then the pay gap also would decrease. Hence both the intensive and ex-
tensive margins of childbearing and rearing determines the gender pay gap for skilled workers.
The same reasoning applies for unskilled labor. Note that if nc goes to zero then the wage gap
m
ew
for unskilled workers would still remain since w < 1 as well for skilled labor since em
1.

18
For unskilled labor the pay gap is a combination of both the statistical mechanism and the
di¤erences in physical endowments among men and women.
Fertility rate determines the distribution of women across occupations in a REE. The fol-
lowing corollary that is proven within the proof of proposition 1 illustrates this result.
Corollary 1. In a REE the amount of women that invest in education is a decreasing
function of the fertility rate for women.
The corollary can be illustrated by table 2 and equation (??). When fertility in the economy
w;H
n o
! w;H
is su¢ ciently large such that !!w;n satis…es min ( w +n+n
c
c
)(1 ew) ; w
(1 ew) > ! w;n
then all
women in a REE would self select into unskilled labor and would not invest in education. As
! w;H
fertility decreases the equilibrium relative wage ! w;n
would end up increasing eventually
+nc ! w;H
up to the point in which ( w
+nc )(1 ew ) ! w;n w
(1 ew )
where women without dependent
children invest in education while women with dependent children (mothers) would work as
w;H
n o
unskilled workers. Finally if nc is low enough such that !!w;n > max w (1 ew ) ; ( w +n+n c
w
c )(1 e )

then all women would invest in education and work as skilled workers.
Table 2: Distribution of women across occupations n o
! w;H +nc
H w; = nw , Lw;n; = 0 if ! w;n
> max w ;
(1 ew ) ( w +nc )(1 ew )
+nc ! w;H
H w; = nw nh , Lw;n; = nh if ( w
+nc )(1 ew ) ! w;n w
ew )
n o (1
+nc ! w;H
H w; = 0, Lw;n; = nw if min ( w
+nc )(1 ew )
; w
(1 ew )
> ! w;n

Proposition 3. In a REE the e¤ective labor supplied on average by women in the workplace
is strictly less than that by men within occupations, which widens with a greater amount of
children per woman.
Proof: See appendix.
This proposition shows that in terms of e¤ective labor women o¤er less than men for both
occupations. Therefore the gender pay gaps in each occupation arise due to the lower average
e¤ort/energy that women o¤er relative to men consistent with what employers would expect.
Proposition 4. In a REE the gender pay gap may not decrease with female education
m
(1 em )
attainment if fertility among women is su¢ ciently low given that w
(1 ew )
< 1. Otherwise, the
gender pay gap decreases with female education attainment.
Proof: See appendix.
This proposition can rationalize the empirical …ndings of Albrecht et al. (2003) and De la
Rica et. al (2008), De la Rica et al. (2015). The latter authors …nd for countries like Spain
that the gender pay gap is higher for unskilled than for skilled workers while the …rst authors

19
…nd the opposite for countries like Denmark, Sweden and United Kingdom. Hence, female
education attainment can be associated empirically with lower or wider gender pay gaps which
m
(1 em )
in our model depends on whether w
(1 ew )
is less or greater than one. This result explains the
apparent puzzle since female education attainment is believed to reduce the gender pay gap.
Our results suggests that this is not always the case.
An empirical regularity observed across economies is what has been called reversal trend in
education where women in more developed economies tend to invest more in education than
men. There is empirical evidence that shows that more women obtain tertiary degrees in most
European countries (OECD 2013) relative to men. The following result shows that our model
can explain this regularity.
Proposition 5. In a REE the reversal trend in education arises in economies in which
nm = nw and the fertility rate is su¢ ciently low.
Proof: See appendix.
The intuition of this comes from the fact that as fertility decreases the equilibrium relative
wage among skilled and unskilled women tends to decrease in accordance with table 2. Hence if
w;H
n o
fertility is su¢ ciently low, such that !!w;n > max w (1 ew ) ; ( w +n+n c
w
c )(1 e )
, then all women
end up being skilled and only men would work in unskilled labor jobs, which is an essential
type of job in the economy. Therefore the reversal trend in education arises.
Another empirical regularity that has been observed as reported by Belley et al. (2012) and
Goldin (2014) for the United States is that skilled men and women have the same wage at the
start of their career but female wages grow at a slower rate creating a gender wage gap which
then is reversed somewhat when women are older. This is called the motherhood penalty. The
model can rationalize the emergence of this regularity.
Proposition 6. In a REE women with dependent children experience an income loss (moth-
erhood penalty) regardless of the occupation. The income loss is larger for economies with higher
fertility rates.
Proof: See appendix.
This proposition is straightforward since labor income is a decreasing function of the amount
of children women bear and rear. Hence women without dependent children would have a higher
income regardless of the occupation. Once children grow up and become independent women
end up getting back higher income levels. There is empirical evidence in line with this result
(Grimshaw et al 2015) which has found a larger motherhood penalty in less developed economies

20
(higher fertility rate economies).
Francois (1998) and Goldin (2014) among others have noted the existence of a wage pre-
mium for married men which has been a robust empirical …ndings in the social sciences. Our
theoretical model can be extended to include housework in order to explain the e¤ect of mar-
riage on income for male workers. Under the role that women are in charge of both housework
and childrearing activities then married men would save themselves both of these costs while
men and women without a spouse or partner would not be able to save the cost of housework.
The female spouse on the other hand endures the housework for herself as well as that of her
spouse. We can use the utility function in (1) with h > 0. The results proved above hold for
h = 0 so when h > 0 and is su¢ ciently close to zero then by continuity of all the relevant
functions the results would still hold. The main result is that the model generates now wider
gender pay gaps compared to when h = 0 since now we would have
h q i2
nh +h
! w;H
1 w
e 1 nw
1 +2h+nc
= h q i2 (14)
! m;H 1 em nh +h
1 nw
1

h q w
i2
nh +h
! w;n m
+h 1 nw
1 w
+2h+nc
= w h q i2 : (15)
! m;n +h nh m
+h
1 nw
1 m

Moreover a fatherhood premium across occupations arises in terms of total income since now
income for male workers across occupations are
s r
! m;H; ! m;H;
W m;1; lsm;1;1; = ; W m;0; lsm;0;1; = m
+ Sh + Sh

Non-fathers (single men) have S = 1 while fathers have S = 0. Hence total income increases
with fatherhood in equilibrium.

6 Econometrics of gender pay gaps


Any theory of the gender pay gap should be consistent with the econometric evidence that
has been found using regression analysis of log earnings, the typical dependent variable in the
literature, on independent variables like log of reported hours worked, female dummy, education
attainment dummy (or continuous variable version of it), marital status dummy (in our setting
having dependent children is the variable related to marital status) and other variables like

21
occupation and industry dummies. The regression results for di¤erent countries, developed as
well as underdeveloped, have shown that a di¤erence in earnings persists even when controlling
for reported hours worked, industry of work and human capital characteristics of workers,
leading scholars to suggest that this provides evidence of wage discrimination against women.
In order to test whether our model is consistent with several of the empirical …ndings in this
empirical literature note that total wage income in a REE is
s
W i;I;
W i;I; lsi;I;M; = i;I;M
for i = w; m, (16)

I = 0; 1, M = 0; 1, S = 0; 1

while the optimal supply of e¤ective labor supply from equation (5) is

i;I;M; 1
s lsi;I;M; = i;I;M
for i = w; m, (17)
I = 0; 1, M = 0; 1, S = 0; 1

With these functions we can deduce the data generating process for total wage income in the
economy according to the model using three binary random variables. Let E take the value one
if an individual is skilled, zero otherwise, let F take the value one if the individual is female, zero
otherwise and …nally let G take the value one if an individual has dependent children (marital
status of being married in our model), zero otherwise. As shown in the appendix by using
the log versions of equations (16) and (17) we can obtain the following semielastic regression
equation

ln We = 0 + 1 ln l + 2F + 3E + 4G + 5 (F G) (18)
+ 6 (F E) + 7 (E G) + 8 (F E G) + "

where We denotes total wage earnings (say per month) across genders, occupations and mari-
tal/dependent children status, l denotes total amount of hours reported (again say per month),
" is a zero mean error term that depends on unobserved e¤ort levels supplied in the labor
market across genders, occupations and marital/dependent children status. The reduced form

22
regression parameters in terms of the regression are
s
m
$w;n; ( + h) $m;H; (1 em ) ( m + h)
1 b, 2 ln m;n; w , 3 ln
$ ( + h) $m;n; ( + h)
r s
m m p
+h
4 ln m , 5 ln w ln $w;n; ,
+ nc + 2h) ( m + h)
(
s s s
$m;n; ( w + h) $w;H; (1 ew ) m
( + h)
6 ln + ln , 7 ln ,
$w;n; ( m + h) $m;H; (1 em ) ( m + h)
s
( m + h) ( w + nc + 2h) p
8 ln + ln $w;n;
m
( + nc + 2h) ( w + h)2

It is shown in the appendix that under the assumptions of the theoretical model the esti-
mation of equation (18) by OLS delivers a negatively biased estimator of 1 while for the other
OLS
variables the estimators e , i = 2; ::; 8 are unbiased.
i

6.1 Signs of parameters


The coe¢ cient 1 is an elasticity while the other parameters are semielasticities that can be
interpreted as percent changes after multiplying by 100. The regression parameters that have
unambiguous signs according to theoretical model are the following

1 > 0, 2 < 0, 3 > 0, 4 > 0, 5 < 0, 7 >0

while parameters 6 and 8 have ambiguous signs. The reference population in the regression
equation arises when F = E = G = 0 which refers to unskilled men without dependent
children. Coe¢ cient 1 = b is the elasticity of total income on hours reported and measures
the responsiveness of labor income to increase in hours worked. The coe¢ cient 2 < 0 is
the coe¢ cient associated with F which represents the gender pay gap for unskilled women
relative to the reference population and which has been veri…ed extensively in the literature as
being negative13 . 3 > 0 is the coe¢ cient associated with E which measures relative earnings
13
A taste-based discrimination approach can explain the sign of this coe¢ cient but cannot explain the other
regression coe¢ cients. For example taste based discrimination, constant across occupations, implies that equi-
! w;H
librium relative wages from the …rst order condidtions of cost minimization conditions satisfy ! m;H
=
! w;n
! m;n = d for a given value d < 1 that measures discrimination against women. Under the auxiliary assump-
tions m = w = , h = 0 and ew = em then one would have 2 ln d < 0 but at the same time would imply
4 = 6 = 7 = 0. These last coe¢ cient values are rejected in the empirical literature which shows evidence not
consistent with a taste based discrimination approach that assumes a constant discrimination against women
across occupations.

23
of skilled men without dependent children with respect to unskilled men without dependent
children, where this positive sign has also been veri…ed empirically since it measures the positive
return of education for men. Coe¢ cients 4 > 0 and 7 > 0 measure the marital status of
unskilled (G) and skilled men (E G) respectively which just says that married men with
dependent children across occupations earn more relative to unmarried men without dependent
children, the so called wage premium for married men in the empirical literature, which has
also been extensively veri…ed in the literature.14 The coe¢ cient 5 < 0 associated with F G
re‡ects the motherhood penalty relative to the reference population, which re‡ects proposition
6 and has been veri…ed also empirically in the literature.
The ambiguity of the sign of 6 associated with E F shows that skilled women do not
always close the gender pay gap with education attainment relative to the reference population.
This re‡ects the result of proposition 4 since 6 is the ratio of the equilibrium gender pay gap
for skilled labor relative to the equilibrium gender pay gap for unskilled labor. Depending
on whether this ratio is greater or less than one then education attainment for women with
dependent children can decrease or increase the gender pay gap for them. The coe¢ cient 8

associated with E F G measures how much skilled women with dependent children earn
relative to unmarried men without dependent children which according to the theoretical model
is ambiguous. Recall that under a REE the equilibrium unskilled wage for women $w;n; is a
decreasing function with respect to nc . The …rst term that comprises coe¢ cient 8 is negative
while the second is positive. This re‡ects the motherhood penalty which is more severe for
economies with lower unskilled wages for women. In economies in which the fertility rate is
relatively large then 8 is negative for a given value of $w;n; which just says that skilled mothers
would not close the gender gap with respect to the reference population. Nonetheless when the
fertility rate is relatively low then 8 could become positive (not necessarily though) for higher
values of $w;n; which means that skilled mothers can close the gender wage gap relative to
the reference population. The fertility rate in the economy plays a key role in how education
attainment can decrease the gender pay gap for skilled mothers in accordance with proposition
4.
The model generates additional predictions related to the fertility rates since it has an e¤ect
on some of these regression coe¢ cients. Recall that a developed economy is one in which the
fertility rate is relatively small while an underdeveloped economy is one in which the fertility
14
This empirical result has been di¢ cult to explain by other theoretical approaches. To our knowledge only
Francois (1998) developes another theoretical approach that can explain this empirical result beside ours.

24
rate is relatively large. We have established that in a REE the relative equilibrium gender
pay gaps are decreasing strictly with respect to the fertility rate of the economy as well as the
equilibrium wages for women. Hence 2 would be lower in absolute value in an underdeveloped
economy than in a developed one as well as coe¢ cient 5 which means that for less developed
economies the gender pay gap is larger for unskilled women as well as the motherhood penalty.
Coe¢ cient 6 could be negative for a less developed economy while positive for developed
economies (not necessarily though). This means that education attainment for women in less
developed countries would not close the gender pay gap while it could close it more likely in
a more developed economy. Evidently depending on the values of the other parameters the
opposite conclusion can arise. Polachek (2014) and Polachek and Xiang (2014) …nd evidence
that gender pay gaps decrease in both type of economies with fertility rates and education
attainment.
In a REE wage levels (not relative equilibrium wages) are increasing in the technology
level in the economy. Economies with higher levels of technology would have e¤ects on two
regression parameters in (18), namely 5 and 8. Higher levels of technology generate a greater
motherhood penalty (lower 5) for unskilled women while it reduces this e¤ect for skilled women
15
(higher 8 ).

6.2 Empirical evidence for Colombia and the United States


In this section we report regressions as in equation (18) for a less developed economy like
Colombia in table 2 and a developed economy like United States in table 3. In the appendix
we report descriptive statistics on the variables used in these regressions.
15
This conclusion has a caveat. When the number of children is endogenized, as shown in a footnote above
where we showed that skilled women have less amount of children than unskilled women, then a greater level
of technology would generate a greater demand of skilled labor which then reallocates labor from unskilled
towards skilled labor. This would then lower the fertility rate in the economy since more skilled women would be
employed and therefore a lower amount of children per woman would arise in the economy. Hence a greater level
of technology would have an additional e¤ect through the fertility rate in the economy when nc is endogenous
besides what we have discussed for parameters 5 and 8 .

25
Year 2002 2010 2018
Dep Variable: Log monthly reported income (1) (2) (3) (4) (5) (6)
Explanatory variables (theoretical variable)
log hours 0.454*** 0.354*** 0.583*** 0.337*** 0.648*** 0.230***
(0.0354) (0.0645) (0.0153) (0.0267) (0.0182) (0.0317)
Female (F) -0.174*** -0.116*** -0.202*** -0.132*** -0.143*** -0.112***
(0.0203) (0.0187) (0.00802) (0.00759) (0.00666) (0.00666)
Tertiary Education (E) 1.174*** 0.642*** 0.874*** 0.477*** 0.637*** 0.319***
(0.0632) (0.0550) (0.0193) (0.0170) (0.0144) (0.0119)
Married with Children <=12 years old (G) 0.111*** 0.0880*** 0.0926*** 0.0745*** 0.0518*** 0.0581***
(0.0201) (0.0183) (0.00805) (0.00753) (0.00699) (0.00659)
Female x Married with Children<=12 (FG) -0.0278 -0.0438* -0.0685*** -0.0526*** -0.0477*** -0.0350***
(0.0311) (0.0262) (0.0136) (0.0122) (0.0121) (0.0110)
Female x Tertiary Education (FE) 0.0322 -0.0184 -0.0325 -0.00525 0.0145 0.0392***
(0.0779) (0.0642) (0.0240) (0.0202) (0.0185) (0.0149)
Tertiary Education x Married with Children <=12 (EG) 0.167* 0.125* -0.0326 -0.00978 0.0452 0.0161
(0.0925) (0.0749) (0.0310) (0.0258) (0.0278) (0.0224)
Female x Tertiary Education x Married with Children<=12 (EFG) -0.199 -0.119 0.0425 0.0123 -0.0103 -0.00974
(0.135) (0.116) (0.0420) (0.0350) (0.0383) (0.0309)
Constant 10.96*** 11.31*** 10.42*** 11.56*** 10.20*** 12.35***
(0.191) (0.361) (0.0832) (0.150) (0.0986) (0.175)
Includes control variables? No Yes No Yes No Yes
Observations 15317 15317 68122 68122 73902 73902
R-squared 0.335 0.561 0.327 0.516 0.285 0.496
Robust standard errors in parentheses; *p<0.1, **p<0.05, ***p<0.01
Controls: age, part time dummy, 6 occupation dummies: professionals & technicians, managers and public supervisors, admin. staff, traders and sellers, service workers, agri. & forestry workers
economic sector dummies: agriculture & livestock, mines & quarries, manufacture industry, electricity & gas-water, construction, transport & communications, financial establishments;
Base Categories: Over-Time; Large Firm; Non-agricultural workers and operators; Personal, Social and community services; Antioquia .
Source: DANE Microdatos

Table 2: Results for Colombia

We report cross section regressions for three years 2002, 2010 and 2018 for both countries
with heterocedastic standard errors where we report with and without control variables that
include among others age, part time, state or department dummies as well as economic sector
dummies. These controls are relevant in the sense that the theoretical model assumes away
heterogeneity across industries and occupations as well as part time and full time job positions.
The empirical results are consistent with the predictions of the theoretical model in both
countries. Speci…cally, for log hours reported 1 = b estimates are positive and statistically
signi…cant at the 1% for both countries but it is much closer to one for the United States (once
controls are included) than for Colombia, which is quite far below one. This could be explained
by the fact that Colombia does not require employers to pay overtime at a higher rates relative
to regular hour rates while the U.S. would have labor market regulations that require employers
to pay time and a half for any extra hour supplied by an employee. As noted above the OLS
estimator of 1 is negatively biased which is consistent with the empirical results for both
countries given that when controls are included the point estimates tend to decrease in both
countries.
In both countries, relative to the base population of single unskilled men, the gender pay
gap for unskilled women is negative 2 < 0 and statistically signi…cant at the 1% level where
in absolute value the gap is of similar magnitude in Colombia relative to the United States.
Moreover, in both countries also relative to the base population, skilled labor for men has a

26
Year 2002 2010 2018
Dep Variable: Log monthly reported income (1) (2) (3) (4) (5) (6)
Explanatory variables (theoretical variable)
log hours 1.467*** 1.115*** 1.557*** 1.198*** 1.530*** 1.142***
(0.00919) (0.0134) (0.00875) (0.0127) (0.00918) (0.0130)
Female (F) -0.142*** -0.160*** -0.125*** -0.139*** -0.149*** -0.143***
(0.00476) (0.00490) (0.00506) (0.00508) (0.00514) (0.00524)
Tertiary Education (E) 0.469*** 0.243*** 0.458*** 0.249*** 0.449*** 0.246***
(0.00642) (0.00615) (0.00702) (0.00666) (0.00637) (0.00614)
Married with Children <=12 years old (G) 0.112*** 0.0969*** 0.0751*** 0.0721*** 0.0802*** 0.0736***
(0.00635) (0.00571) (0.00750) (0.00698) (0.00759) (0.00710)
Female x Married with Children<=12 (FG) -0.0747*** -0.0450*** -0.0287*** -0.000424 -0.0556*** -0.0343***
(0.00974) (0.00881) (0.0111) (0.0103) (0.0124) (0.0115)
Female x Tertiary Education (FE) -0.0455*** -0.0188** -0.0273*** -0.00351 -0.00479 0.00134
(0.00880) (0.00804) (0.00935) (0.00852) (0.00857) (0.00787)
Tertiary Education x Married with Children <=12 (EG) 0.0369*** 0.0408*** 0.0916*** 0.0874*** 0.0842*** 0.0737***
(0.0112) (0.0101) (0.0128) (0.0119) (0.0121) (0.0110)
Female x Tertiary Education x Married with Children<=12 (EFG) 0.0600*** 0.0216 0.0126 -0.0145 0.0368** 0.0193
(0.0168) (0.0152) (0.0186) (0.0172) (0.0186) (0.0172)
Constant 0.436*** 2.440*** -0.0566 1.918*** 0.113** 2.263***
(0.0473) (0.0707) (0.0450) (0.0665) (0.0473) (0.0686)
Includes control variables? No Yes No Yes No Yes
Observations 130847 130847 118011 118011 115128 115128
R-squared 0.412 0.527 0.412 0.513 0.401 0.501
Robust standard errors in parentheses; *p<0.1, **p<0.05, ***p<0.01
Controls: age, part time dummy, occupation dummies for 2002: professional specialty, techn. & related support, sales occupation, administrative support, private household occupations,
protective service occupations, service occupations, precision production, machine operators, transportaton & material, handlers & equip cleaners, farming-forestry & fishing,
sector dummies: agriculture, mining, construction, manufacturing durable goods, manufacturing non-durable goods, transportation, communications, utilities & sanitary services,
whole and retail trade, private households, business-auto & repair, personal services, entertainment & recreation, hospitals, medical services (excluding hospitals), educational services,
social services, other professional services, forestry and fisheries; occupation dummies for 2010-2018: professional and related occupations, service occupations, sales occupations,
office and admin. support occupations, construction and extraction occupations, installation-maintenance and repair occupations, production occupations, transportation
and material moving occupations; state dummies. Base Categories: Full-Time; Executive & Administrative(2002);Finance, Insurance & Real State(2002);
Management, Business & Financial Occupations(2010-2018); Financial Activities(2010-2018); California.
Source: Basic Current Population Survey

Figure 1: Table 3: Results for the United States

positive return 3 > 0 which is statistically signi…cant at the 1% level where in absolute value
the return is greater in Colombia than in the United States.
Furthermore, in both countries unskilled men with dependent children earn more than men
without dependent children 4 > 0 (married men premium) which is statistically signi…cant
at the 1% level. For skilled men 7 > 0 (married men premium) this also arises for both
countries but it is only statistically signi…cant at the 1% level for the United States for all years
while it is only signi…cant in Colombia for 2002. Motherhood penalty arises in both countries
since 5 < 0 which is statistically signi…cant at the 1% level. The motherhoold penalty does
not arise in both countries for skilled mothers since the estimate of 8 is ambiguous and not
statistically signi…cant once control variables are included. Finally, the estimates for parameter
6 are negative or insigni…cant for the United States while they are positive or insigni…cant
for Colombia. This is consistent with the ambiguos prediction of the theoretical model with
respect to this reduced form parameter.

27
7 Public policies
According to the theoretical model gender pay gaps decrease in developed economies as these
economies complete a demographic transition towards lower fertility rates. Nonetheless the
gaps would not disappear totally even if fertility rates would be su¢ ciently close to zero since
there would still be di¤erences in physical endowments across men and women for unskilled
labor as well as di¤erences in schooling costs that can persist across genders. This motivates to
consider policies that can push furthermore the reduction of the gender pay gaps. We go back
to the case where housework cost h = 0 and analyze three such type of policies in what follows.
Importantly each of these policies can be ine¤ective in reducing gender pay gaps if they are not
accompanied by a change in the role of women in childrearing costs within the household.

7.1 Equal parental leave


A policy intervention that can change, even though not necessarily, the role that only women
should do child-rearing activities is equal parental leave which assumes that the father of a
child is equally responsible for child care after birth, as the time in the womb of the mother
would be counted solely to be time of the mother. In such a policy, mothers would be expected
to consume leave of absence due to pregnancy, delivery and early lactation which would be
matched by the father’s time later in the life of the child. An equal parental leave policy would
contribute to gender income equality since it can change the belief that mothers are the sole
responsible parent for child-rearing to a situation in which such activities are more equally
shared by both parents. This policy can change the allocation of child-rearing time within the
household and therefore a¤ect positively the average e¤ort women with dependent children
would o¤er in labor markets while decreasing that of men with dependent children reducing the
gender pay gaps across occupations. Housework might not necessarily change with this type of
policy which would reduce the e¤ectiveness of it. There is empirical evidence in this direction
as in Patnaik (2019). The following proposition summarizes this conclusion where we assume.
Proposition 7. Assume nm = nw then the policy for equal parental leave would reduce
gender pay gaps across occupations if it is accompanied by childrearing cost sharing within the
household.
Proof: See the appendix.

28
7.2 Equal pay for equal work
Another type of policy is one that can be called equal pay for equal work (EPEW). This policy
is one in which by law employers would be required to pay the same wage rate for both men
and women for the same type of work. If the production sector adheres to this policy then
there are two cases that may arise which depend on whether the childrearing costs are shared
or not within the household. If fathers share childrearing costs then the EPEW policy would
increase the average e¤ort exerted by women and reduce that of men such that the …rst order
w;j
! w;j
conditions of cost minimization in equation (10) ! m;j
=1= m;j for j = H; n would hold given
that female e¤ort increases with wages according to equation (5).
What happens when childrearing sharing costs do not change within the household? In
this case note that the …rst order conditions of cost minimization in equation (10) would be
w;j
! w;j
violated ! m;j
= 1 6= m;j for j = H; n since women would still have the role of childrearing
and therefore would provide less e¤ective labor on average to the workplace than men. Hence
the …rst consequence of the policy would induce employers to try to reestablish the optimality
of cost minimization somehow. There are two restrictions for employers to this: i) e¤ort is
unobservable by employers, ii) there is a negative relationship between e¤ort and time supplied
in the workplace. Note that the e¤ort supplied by women, in accordance with (5), is an
increasing function of the real wage across occupations while the supply of hours is a decreasing
function of the real wage where the amount of e¤ective labor would be constant, in accordance
w;I;M w;I;M 1
with (5), s ls = Dw;I +M nc
for I = 0; 1, which is independent of the real wage. Hence in
order to lower production costs employers would be forced to reduce the amount of time that
would be required to be supplied by women while increasing the amount of e¤ort supplied by
them and keep the amount of e¤ective labor constant.
A screening mechanism that separates women types depending on the e¤ort they can supply
in the workplace could implement the same e¤ective labor values and relative equilibrium
wages as in a REE that was studied under a statistical discrimination mechanism. We study a
screening mechanism in which the e¤ective labor of a REE is implemented and where employers
o¤er a menu of contracts that are incentive compatible such that men and women would self
select in order to satisfy the cost minimization …rst order condition under an EPEW policy.
The menu of contracts would eliminate the gender pay gap per unit of time for a given type
of contract that is o¤ered. Nonetheless this mechanism would generate women with dependent
children to be hired as part time workers while women without dependent children, as well as

29
men, to be hired as full time workers across occupations. The policy back…res in the sense that
only women without dependent children would earn the same real wage than men per hour
while those with dependent children would work less amount of hours and earn less per hour
making them earn less total income. This increases unambiguously the motherhood penalty.
The following proposition shows that a menu of contracts exists that can implement, through
a screening mechanism, the same e¤ective labor levels and equilibrium relative wages across
genders consistent with a REE. This menu o¤ers men and women the same wage for the same
type of job that satisfy the …rst order cost minimization conditions and therefore satis…es the
EPEW policy.
h j j
i
Proposition 8. There exists a menu of contracts sj ; l ; sj ; l2 = [Fj ; Pj ] for j = H; n
under a screening mechanism that implements the same e¤ective labor levels and equilibrium
relative wages across genders in accordance to a REE and are consistent with the cost mini-
mization …rst order conditions.
Proof: See the appendix.
The following corollary reports two consequences of the screening mechanism that would
arise under a EPEW policy.
sj
Corollary 2. If nc > nc for some threshold nc > 0, sj
dj for some threshold dj >
1 for j = H; n then a) the screening mechanism generates a separating allocation in which
women without dependent children and men self select into the full time contracts while women
with dependent children self select into the part time contracts; b) the gender pay gaps across
occupations are not eliminated in the aggregate while the motherhood penalty is deepened.
The interesting point of corollary 2 is that an EPEW policy thought as a screening mech-
anism would not eliminate necessarily the gender pay gaps in the aggregate. To see this note
nh
that the average wage rate per unit of time for a woman in the population is sj 1 nw
+ sj nnwh
for j = H; n which is smaller than sj , the wage rate that men earn given that men do not share
n n
sj (1 n h )+sj n h
in childrearing costs. Hence w
s j
w
< 1 for j = H; n. Moreover the motherhood penalty
is deepened since women with dependent children are the ones that earn less per unit of time
and work less amount of hours compared to women without dependent children. In conclusion
we …nd that an EPEW policy that is not accompanied by a change in childrearing sharing costs
within a household would back…re since on the one hand it increases the motherhood penalty
and does not reduce the gender pay gaps in the aggregate. A caveat is in order. If the fertility
rate is su¢ ciently low then all women would self select as full time workers and therefore the

30
EPEW policy would eliminate the gender pay gaps as well as the motherhood penalty even in
the absence of a change in childrearing sharing costs within households.
Polachek (2014) …nds evidence that countries with EPEW legislation have not been able
to reduce the gender pay gaps. Moreover this author …nds that the gender pay gap is lowest
between single men and women while being the largest between married men and women
something that can be rationalized within our model.

8 Conclusions
In this article we make the point that the gender pay gap per hour for skilled as well as unskilled
labor arises and can actually persist when labor markets are perfectly competitive given that
men and women are di¤erent in two fundamental aspects: physical strength and the role of
child-rearing and housework responsibilities for women. We show that in a rational expectations
equilibrium (REE) the gender pay gaps arises because of a statistical discrimination mechanism
based on a intrahousehold division of labor that employers rationally believe that women engage
in child-rearing and housework activities while men do not. This a¤ects negatively e¤ective
labor supplied at the workplace by women. For unskilled workers there is an additional source
that increases the gender pay gap against women which comes from di¤erences in physical
endowments that men and women are born with which decreases furthermore e¤ective labor
supplied at the workplace by women.
The model explains other empirical regularities observed for women besides the gender pay
gaps across occupations, namely the inverse relationship between education attainment and
the average fertility in an economy, the motherhood penalty in which women with dependent
children face an income loss relative to women without dependent children; the reverse trend in
education in which women in more developed economies tend to invest more in education than
men; the potential failure of education attainment to decrease the gender pay gap, marital wage
premium for men (fatherhood premium). We also …nd that in our model there is a negative
relationship between wage inequality and gender pay gaps under the mild assumption that
women earn less than the average wage in the economy across occupations.
Furthermore the model can rationalize the …ndings of regression analysis of wage earnings
per hour in the literature and explains how regression parameters are a¤ected by the fertility
rate of an economy consistent with the demographic transition of a high fertility underdeveloped
economy towards a developed economy with a low fertility rate. The empirical regularities and

31
regression analysis results that our model can explain seem to be more di¢ cult to explain with
other approaches like taste-based discrimination, women with weak bargaining skills among
other explanations. Our approach can be seen as a uni…ed approach to explain all these empir-
ical regularities.
We …nd that the gender pay gap decreases in developed economies as it completes a de-
mographic transition towards lower fertility rates. Nonetheless the gender pay gaps would not
disappear totally even if fertility would be su¢ ciently close to zero due to di¤erences in phys-
ical endowments across men and women for unskilled labor as well as possibly di¤erences in
schooling costs that can arise across genders for skilled labor. This then motivates policies that
can push towards lower levels of the gender pay gaps. Normatively we study two policies that
are thought to reduce the gender pay gap like an equal parental leave policy and an equal pay
for equal work (EPEW) policy.
We …nd that the …rst policy can attenuate and decrease the role that women ful…ll in an
economy since men become more responsible for child-rearing activities, even though housework
may not be necessarily a¤ected. This contributes to gender income equality since employers
would change their beliefs consistent with an increase of the average e¤ort women supply in
labor markets while a decrease in the average e¤ort men supply. We …nd that if a EPEW
policy is implemented without changing the role of women in childrearing and the fertility
is su¢ ciently large (for certain parameter values) then this policy could implement the same
e¤ective labor levels and relative equilibrium wage rates in a REE and would generate women
with dependent children to self select into being part time workers, while men (and women
without dependent children) would self select into being full time workers. This policy would
increase the motherhood penalty and would not reduce in the aggregate the gender pay gaps
across occupations even though women without dependent children would bene…t. But the issue
is that mothers would be negatively a¤ected since they would earn less per hour and would
also work less amount of time compared to non mothers. This would deepen the motherhood
penalty.

References
Albanesi, Stefania and Claudia Olivetti. 2009. Home production, market production and the
gender wage gap: Incentives and expectations. Review of Economic Dynamics 12 (1): 80-107.
Albrecht, James, Anders Björklund and Susan Vroman. 2003. Is there a glass ceiling in

32
Sweden? Journal of Labor Economics, 21 (1): 145-177
Altonji, Joseph and Rebecca Blank. 1999. Race and gender in the labor market. In
Handbook of Labor Economics, Vol 3. ed. O. Ashenfelter and D.Card. North-Holland.
Anderson Deborah, Melissa Binder and Kate Krause. 2002. The motherhood penalty:
Which mothers pay it and why? The American Economic Review, 92 (2): 354-358.
Arrow, Kenneth J. 1973. The Theory of Discrimination. In Discrimination in Labor Markets
ed. Ashenfelter and Rees.
Becker, Gary S. 1971. The Economics of Discrimination. Second Edition. The University
of Chicago Press. University of Chicago.
— — . 1985. Human capital, e¤ort, and the sexual division of labor. Journal of Labor
Economics 3 (1).
Becker, Gary S., William H. J. Hubbard, and Kevin M. Murphy. 2010. Explaining the
worldwide boom in higher education of women. Journal of Human Capital 4 (3): 203–41.
Belley, Phillipe, Nathalie Havet and Guy Lacroix. 2012. Wage growth and job mobility in
the early career: Testing a statistical discrimination model of the gender wage gap. Working
paper GATE 2012-29.
Blau, Francine D and Lawrence M. Kahn. 2001. Understanding International Di¤erences
in the Gender Pay Gap. Journal of Labor Economics, 21 (1): 106-144.
— — . 2006. The U.S. Gender Pay Gap in the 1990S: Slowing Convergence. ILR Review 60
(1): 45-66.
— — . 2007. The Gender Pay Gap. Have Women Gone as Far as They Can? Academy of
Management Perspectives. 21: 7-23.
— — . 2017. The Gender Wage Gap: Extent, Trends and Explanations. Journal of Economic
Literature. 55 (3): 789-865.
Chichilnisky, Graciela. 2009. Gender Pay Gap. International Journal of Green Economics
3 (2): 157-174.
Deaton, Angus S. 2009. Instruments of development: Randomization in the tropics, and
the search for the elusive keys to economic development (No. w14690). National Bureau of
Economic Research.
De la Rica, Sara, Juan J. Dolado and Vanesa Llorens. 2008. Ceilings or ‡oors? Gender
wage gaps by education in Spain. Journal of Population Economics 21(3): 751-776.
De la Rica, Sara, Juan J. Dolado and Raquel Vegas. 2015. Gender Gaps in Performance

33
Pay: New Evidence from Spain. Annals of Economics and Statistics. No 117/118: 41-59.
Francois, Patrick. 1998. Gender discrimination without gender di¤erence: theory and policy
responses. Journal of Public Economics 68: 1–32.
Goldin, Claudia. 2006. The Quiet Revolution That Transformed Women’s Employment,
Education, and Family. American Economic Review, 96 (2): 1-21.
________. 2014. A Grand Gender Convergence: Its Last Chapter. American Economic
Review 104(4):1091-1119.
Grimshaw, Damian and Jill Rubery. 2015. The motherhood pay gap: A review of the issues,
theory and international evidence. International Labor O¢ ce.
Hamermesh, Daniel S. and Albert Rees. 1988. The Economics of Work and Pay, Fourth
edition, Harper & Row, Publishers, Inc.
Hardoy, Inés, Pål Schøne and Kjersti Misje Østbakken. 2015. Children and the gender gap
in management. Labour Economics 47: 124-137.
Machado, Cecilia. 2008. Statistical Discrimination with Fertility: Understanding the Gen-
der Wage Gap and Other Labor Market Di¤erences. Working Paper. Columbia University.
Olivetti, Claudia. and Barbara Petrongolo. 2016. The Evolution of Gender Gaps in Indus-
trialized Countries. IZA DP No. 9659. Working paper.
Oettinger, Gerald S. 1996. Statistical discrimination and the early career evolution of the
black-white wage gap. Journal of Labor Economics 14(1), 52–78
Patnaik, Ankita. 2019. Reserving Time for Daddy: The Consequences of Fathers’Quotas,
Journal of Labor Economics, 37 (4).
Phelps, Edmund. 1972. The Statistical Theory of Racism and Sexism. American Economic
Review 62: 659-661.
Polachek, Solomon. 2014. Equal Pay Legislation and the Gender Wage Gap. IZA world of
labor,16.
Polachek, Solomon. and Jun Xiang. 2014. The Gender Pay Gap Across Countries: A
Human Capital Approach. Working Paper No. 8603.
Sigle-Rushton, Wendy and Jane Waldfogel. 2007. Motherhood and Women’s Earnings
in Anglo-American, Continental European, and Nordic Countries. Feminist Economics 13(2):
55–91.

34
Appendix
Proof of proposition 1. Since technology has constant returns to scale relative equilibrium
wages are determined by the production side of the economy while quantities are determined
by the demand side of the economy. Zero pro…ts determine the amount that is to be produced
by the economy that would satisfy the demand of the consumption good. Equilibrium relative
wages for skilled workers are determined by the …rst order conditions of the production side
in equation (10), rational expectations for skilled labor that equal expected e¤ort to group
averages as in equation (11) and the e¤ort/energy supply functions from equation (5) yields
q q
nh ! w;H (1 ew ) nh ! w;H (1 ew )
! w;H w;H 1 nw
+ nw +nc
m;H
= m;H = q q
! 1 nnmh ! m;H (1 em ) m;H
+ nnmh ! (1 e )
m

where the …rst equality arises due to …rst order conditions, the second equality arises due to
rational expectations when optimal e¤ort supply levels are included for each type of individual.
Therefore when the expression is simpli…ed the equilibrium relative wage between skilled men
and women is " s !#2
! w;H 1 ew nh
= 1 1 : (19)
! m;H 1 em nw + nc
Since nc > 0 and ew em then a gender pay gap arises against skilled women in equilibrium
w;H
i.e. !!m;H < 1. Note that as nc decreases (respectively nnwh decreases) the gender pay gap
diminishes ceteris paribus.
Equilibrium relative wage between unskilled men and women is determined by equation (10)
for j = n, rational expectations for this type of unskilled labor from equation (11) and e¤ort
supply functions from equation (5) which yields
q q
nh ! w;n nh ! w;n
! w;n w;n 1 nw w +
nw w
+nc
m;n
= m;n = q q (20)
! 1 nh ! m;n
+ nh !
m;n
m m
nm nm
" s !#2
m w
nh
= w 1 1 w
nw + nc
Since w > m and nc > 0 then a gender pay gap arises against unskilled women in equi-
w;n
librium i.e. !!m;n < 1. Again when nc or nnwh decreases the gender pay gap diminishes ceteris
paribus. Nonetheless even if nc or nnwh would be arbitrarily close to zero the gender pay gap
would remain since w > m which just re‡ects the gender di¤erences for the disutility in regards
to unskilled physical labor.
From …rst order conditions (9), rational group expectations (11) and e¤ort supply functions
from equation (5) the relative wage between skilled and unskilled women are
0 q 12
nh
! w;H
FH w;H
FH
2 w
(1 e ) @ w 1 nw
1 +nc
= w;n = q w A (21)
! w;n FLn FLn nh
1 1nw w
+nc

35
Similarly the relative wage between skilled and unskilled jobs for men are
m;H 2 m
! m;H FH FH (1 em )
m;n
= m;n = (22)
! FLn FLn
n
Equations (21) and (22) depend on LH through FFLHn which means they are not the equilibrium
relative wages. We look for an equilibrium in which men work in both type of jobs. Since men
are identical in labor markets according to the statistical discrimination mechanism given that
they o¤er always the same e¤ort and time then they must be indi¤erent between the two job
occupations. Hence according to equation (7) for i = m we have that the equilibrium relative
wage between skilled and unskilled labor for men is

! m;H
= m (23)
! m;n (1 em )
otherwise men would self select into only one of the jobs. Now we can use this previous equilib-
w;H m;H ! w;H w;H
rium relative wage and multiply it by !!w;H in order to get !!m;n ! w;H
= m (1 em ) !!w;H
which rearranged yields

! w;H ! w;H
= m (24)
! m;n (1 em ) ! m;H
" s !#2
w
(1 e ) nh
= m 2 1 1
(1 m
e ) nw + nc

w;H
where we replaced !!m;H from equation (19). We can divide the equilibrium relative wage
(24) by the equilibrium wage (20) to get
h q i2
! w;H nh
! w;H
! m;n
w
(1 e )w 1 nw
1 +nc
w;n
= !w;n = m 2 h q i2 . (25)
! ( (1 em )) n w
! m;n
1 nw 1
h
w
+nc

From table 1 depending on whether this previous value takes on then we can have an
equilibrium situation in which all women are skilled or unskilled or only single women are
skilled and mothers are unskilled. It well depend crucially on the values of fertility for skilled
and unskilled women. For levels of fertility su¢ ciently large then the equilibrium relative wage
among skilled and unskilled women in equation (25) decreases su¢ ciently that according to
w;H
table 1 all women would self select into unskilled jobs i.e. !!w;n < w (1 ew ) . As fertility rate
decreases then for a certain range, according to table 1, only single women would self select into
! w;H
skilled jobs while mothers would self select into unskilled jobs i.e. ( w +n+n c
w
c )(1 e ) ! w;n

w
(1 ew )
. Finally, when fertility decreases su¢ ciently then all women would self select into
! w;H +nc
skilled labor jobs i.e ! w;n
> ( w
+nc )(1 ew )
. The distribution of women across jobs depends
! w;H
then on the value of ! w;n
with respect to the threshold values in table 1.

36
We can replace the equilibrium relative wage (23) in equation (22) to get

FH
= m (26)
FLn (1 em )
FH
which determines the relative amount of H and Ln in the economy since the ratio FLn
is a
n
function of LH
FH Ln
g (27)
FLn H
such that g is a strictly positive increasing function.16
The amount of hours demanded by the economy in the production of the consumption good
is determined by the average of individuals who optimally supply them. The average amount
of hours o¤ered across occupations and genders under the stereotype belief are the following

nh nh w;0;1 m;n
lw;n = 1 lsw;0;0 + l ; l = lsm;0;1 = lsm;0;0 (28)
nw nw s
nh nh w;1;1
lw;H = 1 lsw;1;0 + l ; lm;H = lsm;1;1 = lsm;1;0
nw nw s

We replace in these functions the optimal amount of hours according to (5) which yields
" s !#
w
1 n h 1
lw;n = p w 1 1 w ; lm;n = p m (29)
! w;n nw + nc ! m;n
" s !#
1 n h 1
lw;H = p 1 1 ; lm;H = p
(1 ew ) ! w;H nw + nc (1 em ) ! m;H

The amount of equilibrium e¤ective labor demanded per individual across occupations and
genders are determined by the supply of these where using (5), (11) and (29) we are able to get
" s !#2
w
w;n 1 nh
(lw;n ) = w 1 1 w (30)
nw + nc
" s !#2
w;H 1 nh
lw;H = 1 1
nw + nc
m;n 1 m;H 1
(lm;n ) = m; lm;H =

16
In the particular case in which the techonology is a Cobb-Douglas producation function

F (K; H; Ln ) = K 1 H (ALn )

such that + < 1 then


FH Ln
=
FLn H
Ln
which is a stictly increasing function of H .

37
From the de…nitions of e¤ective labor we have
m;H w;H
H lm;H H m + lw;H Hw (31)
n w;n w;n w;n m;n m;n m;n
L (l ) L + (l ) L

while from the amount of men and women in the economy we must have

nm H m + Lm;n , nw H w + Lw;n : (32)

We can now determine the distribution of men and women across the two occupations Lw;n ,
L , H m , H w . To do so we replace equations (27), (31) and (32) in equation (26) to get after
m;n

rearranging

(lw;n w;n ) Lw;n + (lm;n m;n ) Lm;n 1


m;H
=g m
lm;H (nm Lm;n ) + lw;H w;H (nw Lw;n ) (1 em )

where g 1 is the inverse function of g which exists since g is a monotonic function. This equation
can be written as
Lm;n = b1 c1 Lw;n (33)
where
m;H w;H
lm;H nm + lw;H nw
b1 m;H (lm;n m;n )
>0
lm;H + 1
g (
m (1 em ) )
w;n w;n
w;H w;H (l )
l + g 1
(
m (1 em ) )
c1 m;H A(l m;n m;n
)
>0
lm;H + g 1
( m (1 em ) )

which shows a negative linear relationship between Lm;n and Lw;n . From table 1 the distribution
w;H
of women across jobs depends on the value that !!w;n takes which yields

Table A1: Distribution of women across occupations


n o
w;H +nc
H w; = nw , Lw;n; = 0 if !!w;n > max w (1 ew )
;( w
+nc )(1 ew )
+nc ! w;H
H w; = nw nh , Lw;n; = nh if ( w
+nc )(1 ew ) ! w;n w
ew )
n o (1
+nc ! w;H
H w; = 0, Lw;n; = nw if min ( w
+nc )(1 ew )
; w
(1 ew )
> ! w;n

The distribution of men across occupations is then determined using (33) to get

Table A2: Distribution of men across occupations n o


! w;H +nc
H m; = nm Lm;n; , Lm;n; = b1 if ! w;n
> max w
(1 ew )
;( w
+nc )(1 ew )
m; m;n; m;n; +nc ! w;H
H = nm L ,L = b1 c1 nh if ( w +nc )(1 ew ) ! w;n w
ew )
n o (1
+nc ! w;H
H m; = nm Lm;n; , Lm;n; = b1 c1 nw if min ( w
+nc )(1 ew )
; w
(1 ew )
> ! w;n

38
Moreover, e¤ective labor in both type of jobs are determined to be
m;H w;H
H lm;H H m; + lw;H H w;
m;n w;n
Ln; (lm;n ) Lm;n; + (lw;n ) Lw;n; .

Replacing these equilibrium levels in the production function and using the …rst order condition
r = FK , where FK can be written as a strictly decreasing function of K for a given value of C
C
i.e. FK = k K , and k 0 > 0, in order to get

C
C =F 1
; H ; Ln;
k (r)
which determines implicitly the amount of the consumption good C as a …xed point problem.
This problem would have a solution under the Inada conditions which is a standard assumption
of a neoclassical production function. Replace this value C in the …rst order condition which
yields the aggregate amount of capital in the economy
C
K = 1
. (34)
k (r)
Finally equilibrium wages can be found by using (11), (9) and (22) to get the equilibrium wages
(1 em )
! m;H; = (FH (K ; H ; Ln; ))2 (35)

" s !#2
w
2 1 e nh
! w;H; = (FH (K ; H ; Ln; )) 1 1
nw + nc
1
! m;n; = (FLn (K ; H ; Ln; ))2 m (36)
" s !#2
n; 2 w
(FLn (K ; H ; L )) nh
! w;n; = w 1 1 w
nw + nc
which all are strictly positive.
Proof of proposition 2. a) This is a corollary of proposition 1. Note that equations (19)
and (20) show that across skilled and unskilled occupations there is a gender pay gap that
arises against women in a REE; b) gender pay gaps decrease with fertility rates as can be seen
in equations (19) and (20).
Proof of proposition 3. From (30) we can obtain the relative e¤ective labor supply
functions of women relative to men, or e¤ective labor gaps, within occupations in a REE
" s !#2
w;n w;n m
(l ) nh
m;n = w 1 1
(lm;n ) nw + nc
" s !#2
w;H
lw;H nh
m;H
= 1 1
lm;H nw + nc

39
where all are strictly less than one given that < m < w and nnwh 2 (0; 1). Moreover the
e¤ective labor gaps widen with the average number of children nc since the numerators of these
gaps are decreasing in nc .
w;H w;n
Proof of proposition 4. Note that !!m;H > !!m;n is equivalent to having
" s !#2 " s !#2
m w
1 ew nh nh
1 1 > w 1 1 w
1 em nw + nc nw + nc
h q w
i2
We can divide on both sides of the previous inequality by 1 nnwh 1 w
+nc
and re-
arrange to get
h q i2
1 nnwh 1 +nc
m
(1 em )
h q w i2 > w .
nh (1 ew )
1 nw 1 w
+nc

The left hand side is a strictly decreasing function in nc . Hence it follows that the previous
inequality can only hold for nc su¢ ciently low. As nc gets close to zero the left hand side of
m
em )
the previous equation tends to one and given that w (1(1 ew )
< 1 then the conclusion holds.
! w;H
Proof of proposition 5. From equation (25) we know that is a strictly decreasing
! w;n
n o
! w;H
function of nc . Hence if nc is su¢ ciently low such that !w;n > max w (1 ew ) ; ( w +n+n c
w
c )(1 e )
w;
according to table A1 then all women would eventually end up being skilled workers H =
nw . Nonetheless unskilled labor is essential in the economy since the production function is
neoclassical. This implies that for nc su¢ ciently low only men would be the ones performing
unskilled labor Lm;n; > 0, given that in a REE they are indi¤erent between the two type of
job. Since nm = nw by assumption then H m; = nm Lm;n; < nw = H w; holds.
Proof of proposition 6. Income for a woman is de…ned to be W w;I lsw;I;M for I = 0; 1
where the optimal amount of time supplied for an individual is given by equation (5) which
yields s
W w;I
W w;I lsw;I;M =
Dw;I + M nc
Since wages do not di¤er among women with or without dependent children according to the sta-
tistical discrimination mechanism then it must be the case that income is lower for women with
dependent children M = 1, given that nc > 0, relative to women without dependent children
M = 0, more so when women have a greater amount of children. Since equilibrium wages for
women decrease with nc then as nc is larger per woman then the motherhood penalty is higher
given that the numerator in the previous equation decreases slower than the denominator.
Deduction of the population regressions in the econometrics section
Recall that total wage income and optimal e¤ective labor supply in a REE for an individual
i = w; m are given by equations (16) and (17). Let E denote the random variable that takes
the value one if an individual reports being skilled and zero otherwise, let F denote the random
variable that takes the value one if the individual is a woman and zero otherwise and …nally
let G denote the random variable that takes the value one if an individual reports having

40
dependent children and zero otherwise. The income for each type of woman across occupations
and motherhood status allow us to de…ne the following where we …rst take logarithms on both
sides of equation (16) and then multiply appropriately on both sides to obtain the following
s
$w;H; (1 ew )
EF G ln W w;1; lsw;1;1; = EF G ln (37)
+ nc + 2h
s
$w;H; (1 ew )
EF (1 G) ln W w;1; lsw;1;0; = EF (1 G) ln
+h
r
$w;n;
(1 E) F G ln W w;0; lsw;0;1; = (1 E) F G ln w
+ 2h + nc
r
w;0; w;0;0; $w;n;
(1 E) F (1 G) ln W ls = (1 E) F (1 G) ln w
+h
Similarly for men we have
s
$m;H; (1 em )
E (1 F ) G ln W m;1; lsm;1;1; = E (1 F ) G ln (38)
s
$m;H; (1 em )
E (1 F ) (1 G) ln W m;1; lsm;1;0; = E (1 F ) (1 G) ln
+h
r
m;0; $m;n;
(1 E) (1 F ) G ln W lsm;0;1; = (1 E) (1 F ) G ln m
r
m;0; $m;n;
(1 E) (1 F ) (1 G) ln W lsm;0;0; = (1 E) (1
m F ) (1 G) ln
+h
The data generating processes for the random variable log of total wage earnings per month
is de…ned to be
ln We EF G ln W w;1; lsw;1;1; + EF (1 G) ln W w;1; lsw;1;0; (39)
w;0;
+ (1 E) F G ln W lsw;0;1;
w;0;
+ (1 E) F (1 G) ln W lsw;0;0;
+E (1 F ) G ln W m;1; lsm;1;1;
+E (1 F ) (1 G) ln W m;1; lsm;1;0;
+ (1 E) (1 F ) G ln W m;0; lsm;0;1;
+ (1 E) (1 F ) (1 G) ln W m;0; lsm;0;0;
The data generating process for the random variable log of hours ln lsF;E;G reported per month
across F , E and G is de…ned to be
ln lsF;E;G EF G ln lsw;1;1; + EF (1 G) ln lsw;1;0; (40)
+ (1 E) F G ln lsw;0;1; + (1 E) F (1 G) ln lsw;0;0;
+E (1 F ) G ln lsm;1;1; + E (1 F ) (1 G) ln lsm;1;0;
+ (1 E) (1 F ) G ln lsm;0;1;
+ (1 E) (1 F ) (1 G) ln lsm;0;0;

41
Add all of the terms from (37) and (38), use de…nition (39), add and substract b ln lsF;E;G on
the RHS, where b is a positive parameter that depends on the labor market, to get
s
$w;H; (1 ew )
ln We = b ln lsF;E;G + EF G ln (41)
+ nc + 2h
s r
$w;H; (1 ew ) $w;n;
+EF (1 G) ln + (1 E) F G ln w
+h + nc + 2h
r s
$w;n; $m;H; (1 em )
+ (1 E) F (1 G) ln w + E (1 F ) G ln
+h
r r
$w;n; $m;n;
+ (1 E) F (1 G) ln w + (1 E) (1 F ) G ln m
+h
s
$m;H; (1 em )
+E (1 F ) (1 G) ln
+h
r
$m;n;
+ (1 E) (1 F ) (1 G) ln m b ln lsF;E;G
+h
where
( w + nc + 2h) ( m + h) w
+h
ln lsF;E;G = ln w m EF G + ln m EF (42)
( + nc + 2h) ( + h) +h
( w + h) m m
( + h)
+ ln w m F G + ln EG
( + nc + 2h) ( + h) ( m + h)
m m m
+h +h +h
+ ln E + ln w F + ln m G
+h +h
1
+ ln m ln F;E;G
s v F;E;G ln F;E;Gs
+h

which comes from the constant e¤ective labor supply of individuals consistent with (5).
Equation (41) can be written as the following log linear semielastic population regression
equation where we destroy all the brackets of the right hand side variables and then group terms
in order to achieve the following reduced form equation that corresponds to equation (18)

ln We = + 1 ln lsF;E;G + 2 F + 3 E + 4 G +
0 5 (F G) (43)
+ 6 (E F ) + 7 (E G) + 8 (E F G) + "

where " E b v F;E;G ln F;E;G


s b v F;E;G ln F;E;Gs denotes the unobservable error
term such that E ( ) is the expected value operator. Note that E (") = 0 and V ar (") =
b2 V ar v F;E;G ln F;E;G
s is heterocedastic since it varies with gender, skill and marital/dependent
children status. The coe¢ cients are functions of structural parameters de…ned in the following

42
way
m
F;E;G F;E;G $w;n; ( + h)
0 E b v ln , 1 s b, 2 ln m;n; w ,
$ ( + h)
s r
$m;H; (1 em ) ( m + h) m
+h
3 ln , 4 ln m ,
$m;n; ( + h)
s
m p
5 ln w m ln $w;n; ,
( + nc + 2h) ( + h)
s s s
$m;n; ( w + h) $w;H; (1 ew ) m
( + h)
6 ln + ln , 7 ln ,
$w;n; ( m + h) $m;H; (1 em ) ( m + h)
s
( m + h) ( w + nc + 2h) p
8 ln m w 2 + ln $w;n; .
( + nc + 2h) ( + h)
On the other hand, from the de…nition of the error term the following mean conditional
independence condition holds
E "j ln lsF;E;G ; X = E "j ln lsF;E;G (44)
where X (F; E; G; F G; EF; EG; EF G). This means that once one conditions on ln ls the
conditional expected value of the error term is independent of X. This is useful since from
equation (43) and (44) we get
E ln We j ln lsF;E;G ; X = 0 + 1 ln ls + 2F + 3E + 4G + 5 (F G)
+ 6 (E F) + 7 (E G) + 8 (E F G) + E "j ln lsF;E;G

This shows that if equation (43) is estimated by OLS then OLS estimator f1 is biased while
the other ei for i = 2; ::; 8 are unbiased. The bias of f1 is negative since
F;E;G F;E;G
Cov b ln lsF;E;G ; " = Cov b ln lsF;E;G ; E b v F;E;G ln s b v F;E;G ln s
F;E;G
= Cov b ln lsF;E;G ; b v F;E;G ln s
= bCov ln lsF;E;G ; ln lsF;E;G = bV ar ln lsF;E;G < 0:
Proof of proposition 7. We prove the result for h = 0 since by continuity of the equi-
librium relative wage functions the conclusion would still hold for h > 0 su¢ ciently close to
zero. Suppose now that the policy a¤ects childrearing sharing costs within the household in the
following way. The utility function for a father would need to represent the cost for childrearing
(Di;I +M nc ) l
i.e. U m = ln c 2
where 2 (0; 1) represents the childrearing share cost for a father.
(Di;I +M (1 )nc ) l
The utility function for a mother is therefore U w = ln c 2
. The gender pay
gaps across occupations would end up being
h q w
i2
nh
w;n m 1 1
! nw w
+(1 )nc
= w h q i2 (45)
! m;n nh m
1 nw 1 m
+ nc

43
h q i2
nh
! w;H 1 ew 1 nw
1 +(1 )nc
= h q i2 (46)
! m;H 1 em nh
1 nw
1 + nc

given that nm = nw . As approaches a half the gender pay gaps decrease.


Proof of proposition 8. We prove again the result for h = 0 since by continuity of the
equilibrium relative wage functions the conclusion would still hold for h > 0 close to zero. Say
a¢ rmative action requires a > 0 skilled labor positions for women. Two cases to consider.
i) Suppose that the role of childrearing does not change in the economy. In this case all
the positions are …lled by women without dependent children then the only e¤ect is on the
distribution of women as shown in table A3

n o
! w;H +nc
H w; = nw , Lw;n; = 0 if ! w;n
> max w
(1
;
ew ) ( w
+nc )(1 ew )
w; w;n; +nc ! w;H
H = a + nw nh , L = nh a if ( w +nc )(1 ew )
= ! w;n
+nc ! w;H
H w; = nw nh , Lw;n; = nh if ( w
+nc )(1 ew )
> ! w;n w
(1 ew )
! w;H
H w; = a, Lw;n; = nw a if w
(1 ew )
= ! w;n
n o
w; w;n; +nc ! w;H
H = 0, L = nw if min w
(1 ew ) ( w +nc )(1 ew )
; > ! w;n
Table A3: Distribution of women across occupations

The corresponding distribution of men across occupations is determined using (33) as in


table A2 given the values in A3. Note that the policy would e¤ective in changing the distribution
w;H
of women and men across occupations only if the relative equilibrium wage !!w;n makes the
women self select voluntarily into the skilled positions such that they are indi¤erent between
being skilled or unskilled. Otherwise women that …ll the positions would drop out afterwards
given that they would prefer to be unskilled if the relative wage is not high enough. In this
case the gender pay gap is not changed by the policy.
ii) Suppose now that the role of childrearing does change in the economy when the spouses
of women with dependent children share on the childrearing costs. Suppose that a fraction of
the reserved a skilled positions are …lled by women with dependent children such that their
spouses share with proportion childrearing costs. In this case the distribution of women
across occupations would still be determined as in table A3 where the equilibrium relative wage
! w;H
! m;H
would be determined as in equation (46). In this case the gender pay gap for skilled
labor is reduced compared to the case in which the policy would not implemented.
Proof of proposition 9. Set the high wage in occupation j = H; n as the male REE
real wage rate sj = ! m;j; and the lower wage in the same occupation as the female REE
wage sj = ! w;j; which by construction satis…es sj > sj . Across contracts the relative wages
per unit of time correspond to the equilibrium gender pay gaps in a REE under a statistical
discrimination mechanism. Moreover the average e¤ective labor demanded for occupation j
j n
is de…ned as l where in each occupation it can be set to be that of a REE where l
1 H 1
m, l which correspond to that provided by men in a REE in accordance with (30),

44
h q w
i2 h q i2
n 1 nh H 1 nh
l w 1 nw
1 w
+nc
, l 1 nw
1 +nc
which correspond to that
n H
provided by women in accordance with (30). Evidently l > ln and l > lH under the
assumptions of the model.
The corresponding time and e¤ort supplied by an individualqshould satisfy the optimal
n n n H H
supply functions that come from (5). Hence l = p w1 n , = s
w and l = p1 H , =
s s
q
sH
. On the other hand the amount of time part time contract would satisfy half the amount
n H
of time supplied in the full time contract i.e . ln = , and lH = l2 . The corresponding l
q nh q w i2 2
n s nh
optimal e¤ort supplied for part time contracts would be = 2 w 1 nw 1 w
+nc
q h q i 2
H
and H = 2 s 1 nnwh 1 +nc
where
0 " s !#2 1
n 1 @ nh A>0
l ln w 1 1 1
nw + nc
0 " s !#2 1
H 1@ nh A > 0:
l lH 1 1 1
nw + nc

j
By construction lj < l for j = H; n which means that low e¤ective labor corresponds also with
j
part time low wage contracts where > j for j = H; n for nc is su¢ ciently large. The …rst
order conditions for cost minimization would be satis…ed trivially since the wages for the same
occupation and contract are equal. Given the menu of contracts we need to …nd the distribution
of workers across contracts for men and women.
For unskilled labor …rst note that men would prefer the high-wage full-time contract if
n n
U m (sn ; ln ) is satis…ed. This inequality is equivalent to satisfying ssn e [ l l ] .
n m n
U m
sn ; l
Second women without dependent children would prefer the high-wage full-time contracts if and
n n
e [ l l ] : Furthermore
n w n
only if U w sn ; l U w (sn ; ln ) which is equivalent to having ssn
women with dependent children prefer the high-wage full-time contracts if nc is su¢ ciently
low. To see this we use the direct utility function (1) for i = w evaluated at the two type of
n
contracts for unskilled jobs to see whether U w;M sn ; l U w;M (sn ; ln ) is satis…ed which is
n
e( +nc )[ l l ] :Hence if n is su¢ ciently low then women with depen-
n w n
equivalent to having s sn c
dent children would prefer the high wage part time contract. For skilled labor note that men
and women without dependent children are identical in the sense that their utility functions are
the same. Hence it su¢ ces to …nd when these individuals would preferhthe fulli time contract,
H
H H l lH
i.e. U m sH ; l U m sH ; lH which is satis…ed if and only if ssH e holds. On the
other hand, skilled women with dependent children (mothers) would prefer the high wage full
H
time contracts for nc su¢ ciently low i.e. when U w;M sH ; l U w;M sH ; lH is satis…ed which
h H i
sH ( +nc ) l lH
is equivalent to having e sH
. The preference for contracts across individuals is
summarized in tables A4 and A5.

45
Men Women (M = 0) Women (M = 1)
n
sn ( w
+nc )[ l ln ]
if sn
e Fn Pn Fn Pn Fn Pn
n n
+nc )[ l ln ]
e [ ln ]
w w
( sn l
if e > sn
Fn Pn Fn Pn Fn Pn
n n
if e [ l ]> e [ ]
w n m n
l sn l l
sn
Fn Pn Fn Pn Fn Pn
n
if e [ l ln ]
m n
>s sn
F n Pn F n Pn F n Pn
Table A4: Preference of contract Unskilled
h H i
Men Women (M = 0) Women (M = 1)
sH ( +nc ) l lH
if sH
e FH PH FH PH FH PH
h H i h H
i
( +nc ) l lH sH l lH
if e > sH
e FH PH FH PH FH PH
h H
i
H
l l sH
if e > sH
FH P H FH PH FH PH
Table A5: Preference of contract Skilled
h H
i
n
lH
e [ ln ]
w l
n l H
Proof of corollary 2. a) De…ne d > 1, d e > 1 and nc
1 sn w 1 sH
max n
ln
ln sn ; H ln sH then women with and without dependent children
l l lH
self select respectively into full and part time contracts for unskilled and skilled workers, i.e.
they are separated by the screening mechanism. This comes from the second case in tables A4
and A5.
b) Given these results then the distribution of men and women are such that mothers work
as part time workers L ew;n + H ew
part;M part;M = nh while the rest of women work as full time workers
e
L w;n e w
nh . Men on the other hand would work only as full time workers in both
f ull + Hf ull = nw
skilled and unskilled jobs i.e. L em;n + H e m = nm . The gender pay gaps in the aggregate across
f ull f ull
occupations are not eliminated in terms of per unit of time since the average wage rate per unit
of time for a woman in the economy is now sj 1 nnwh +sj nnwh for j = H; n which is less than sj ,
the wage rate than men earn given that mothers are the ones in charge of childrearing. Hence
n n
sj (1 n h )+sj n h
w
s j
w
< 1 for j = H; n. Moreover, the screening mechanism under the conditions above
generates that women choose di¤erently occupations relative to a competitive REE since now
mothers work in both type of occupations but as part time workers while non mothers work
in both type of occupations as full time workers. For nc > nc this reallocation of female labor
increases the motherhood penalty relative to women without dependent children since now they
earn less per unit of time and work less amount of time than what they would have in a REE
under the statistical discrimination mechanism. To see this recall that income for mothers in a
REE for skilled and unskilled labor are respectively
s r
w;H; w;H;M $w;H; w;n; w;n;M $w;n;
$ ls = ; $ ls = w
+ nc + nc

while in the screening mechanism under a EPEW policy the income of mothers are respectively
s r
H H n
s l 1 sH sn l 1 sn
= ; = w .
2 2 + nc 2 2 + nc

46
Since in the screening mechanism by construction we set sj = ! w;j; for j = H; n then we have
that for skilled labor
s s
H
sH l 1 sH sH
= < = $w;H; lsw;H;M
2 2 + nc + nc

which is always satis…ed. Similarly for unskilled labor


n r r
sn l 1 sn sn
= w < w = $w;n; lsw;n;M
2 2 + nc + nc
which is also satis…ed.
Descriptive Statistics
Tables A6 and A7 report the descriptive statistics of the data used for Colombia and the
United States respectively that come from o¢ cial sources like DANE which is the statistics
bureau in Colombia and the Current Population Survey for the United States.
2002 2010 2018
Female Men Female Men Female Men
Variable Avg Std dev Avg Std dev Avg Std dev Avg Std dev Avg Std dev Avg Std dev
Log wage per hour 8.082 0.765 8.237 0.725 8.293 0.729 8.427 0.701 8.463 0.645 8.523 0.61
Log monthly wage 13.37 0.746 13.6 0.701 13.59 0.737 13.8 0.69 13.74 0.664 13.87 0.605
Age (years) 32.96 10.02 33.5 10.38 34.66 10.43 34.62 10.75 35.71 11.06 35.44 11.43
Weekly hours reported 48.21 12.64 51.66 13.18 48.22 11.81 51.39 10.94 47.09 9.631 50.17 9.478
+Children <12 0.62 0.486 0.62 0.487 0.57 0.495 0.53 0.499 0.52 0.499 0.47 0.499
+Married or Cohabiting 0.43 0.495 0.61 0.488 0.45 0.497 0.59 0.493 0.48 0.499 0.58 0.494
+Part-time 0.11 0.31 0.07 0.249 0.09 0.286 0.04 0.185 0.08 0.265 0.03 0.173
+Full-time 0.55 0.498 0.51 0.5 0.61 0.487 0.58 0.493 0.73 0.446 0.68 0.465
N. Observations Unweighted 7640 7677 32991 35131 36124 37778
N. Observations Weighted 1051779 980240.25 1972774 2200713 2499359 2704381
Source: DANE Microdatos, + denotes a dummy variable

Table A6: Descriptive Statistics Colombia

2002 2010 2018


Female Men Female Men Female Men
Variable Avg Std dev Avg Std dev Avg Std dev Avg Std dev Avg Std dev Avg Std dev
Log wage per hour 2.844 0.58 3.068 0.621 2.862 0.622 3.042 0.661 2.914 0.615 3.092 0.638
Log monthly wage 7.908 0.671 8.241 0.699 7.916 0.72 8.195 0.751 7.983 0.709 8.253 0.713
Age (years) 38.49 11.92 38.37 11.72 40.03 12.52 39.79 12.21 40.1 12.94 40.09 12.7
Weekly hours reported 37.96 7.974 42.05 8.228 37.59 8.11 41.33 8.553 38.06 7.883 41.54 8.154
+Children <13 0.32 0.468 0.32 0.466 0.31 0.461 0.31 0.462 0.28 0.45 0.28 0.45
+Married or Cohabiting 0.53 0.499 0.60 0.49 0.52 0.5 0.59 0.492 0.49 0.5 0.55 0.497
+Part-time 0.19 0.388 0.06 0.245 0.20 0.402 0.09 0.288 0.18 0.383 0.07 0.262
+Full-time 0.82 0.388 0.94 0.245 0.80 0.402 0.91 0.288 0.82 0.383 0.93 0.262
N Observations Unweighted 61435 69412 56460 61551 54010 61118
N Observations Weighted 119167160 141056712 123583643 142164222 141629556 167509667
Source: Basic Current Population Survey

Table A7: Descriptive Statistics United States

47

También podría gustarte