This is a repository copy of Gender wage discrimination in Mexico: A distributional

approach.

White Rose Research Online URL for this paper:

http://eprints.whiterose.ac.uk/9980/

Monograph:
Popli, G. (2008) Gender wage discrimination in Mexico: A distributional approach. Working
Paper. Department of Economics, University of Sheffield ISSN 1749-8368

Sheffield Economic Research Paper Series 2008006

Reuse
Unless indicated otherwise, fulltext items are protected by copyright with all rights reserved. The copyright
exception in section 29 of the Copyright, Designs and Patents Act 1988 allows the making of a single copy
solely for the purpose of non-commercial research or private study within the limits of fair dealing. The
publisher or other rights-holder may allow further reproduction and re-use of this version - refer to the White
Rose Research Online record for this item. Where records identify the publisher as the copyright holder,
users can verify any specific terms of use on the publisher’s website.

Takedown
If you consider content in White Rose Research Online to be in breach of UK law, please notify us by
emailing [email protected] including the URL of the record and the reason for the withdrawal request.

[email protected]
https://eprints.whiterose.ac.uk/
Sheffield Economic Research Paper Series

SERP Number: 2008006

ISSN 1749-8368

Gurleen Popli

Gender wage discrimination in Mexico: A distributional approach

April 2008

Department of Economics
University of Sheffield
9 Mappin Street
Sheffield
S1 4DT
United Kingdom
www.shef.ac.uk/economics
 Abstract:

This paper examines the observed wage differentials by gender in Mexico over the last
two decades (1984 to 2002). To estimate and understand the wage gap the paper uses a
nonparametric-distributional approach, and compares the results with the other
parametric approaches. The paper finds evidence of labour market discrimination
against women. The average discrimination in the first decade fell considerably, but
then started to increase again. The distribution of discrimination also changed over
time. The average fall in discrimination over time has resulted largely from a fall in the
estimated discrimination at the lower tail of the wage distribution, with little to no
change at the upper tail.

Key Words:. gender wage gap, nonparametric-distributions, Mexico

JEL Classification: J16, C14, O15
Acknowledgements: The research was supported by a grant from Social Sciences
Devolved Fund, University of Sheffield. I would like to thank Sarah Brown and Andy
Dickerson for helpful comments, and Syed Hamid for research assistance. All errors are
mine.

2
 Gender wage discrimination in Mexico: A
distributional approach
Gurleen K. Popli∗
Department of Economics
University of Sheffield

ABSTRACT

This paper examines the observed wage differentials by gender in Mexico

over the last two decades (1984 to 2002). To estimate and understand the
wage gap the paper uses a nonparametric-distributional approach, and com-
pares the results with the other parametric approaches. The paper Þnds
evidence of labor market discrimination against women. The average dis-
crimination in the Þrst decade fell considerably, but then started to increase
again. The distribution of discrimination also changed over time. The
average fall in discrimination over time has resulted largely from a fall in
the estimated discrimination at the lower tail of the wage distribution, with
little to no change at the upper tail.

JEL classiÞcation code: J16, C14, O15

Keywords: gender wage gap, nonparametric-distributions, Mexico
April 2008

Acknowledgements: The research was supported by a grant from Social

Sciences Devolved Fund, University of Sheffield. I would like to thank
Sarah Brown and Andy Dickerson for helpful comments, and Syed Hamid
for research assistance. All errors are mine.

∗
9 Mappin Street, Sheffield, UK, S1 4DT. Email: g.popli@sheffield.ac.uk; Tele.
+44(0)114-2223485; Fax 44(0)114-2223458.
1 Introduction

This paper examines the observed wage differentials, by gender, in Mex-

ico over the last two decades (1984 to 2002). To estimate and under-
stand the gender wage gap the paper uses a nonparametric-distributional
approach. Analyses of labor market discrimination largely rely on Blinder-
Oaxaca (BO henceforth) kind of decompositions (Blinder, 1973 and Oaxaca,
1973). Though a good starting point and a very good tool, BO decompo-
sition has certain limitations. First, it is a summary measure, it gives us
the behavior around the mean. To illustrate the problem associated with
focussing only on the mean, consider the following example:

“Suppose two sets of sample survey data about working women’s

wages relative to men’s are available. Each reveals an average
wage differential attributed to discrimination of 10%, but in the
Þrst survey all working women are underpaid by 10%, while in
the second half are underpaid by 20% and half are equally paid.”
(pp.82, Jenkins, 1994.)

Looking at the mean alone also misses out other useful information. For
e.g. there is evidence of a relationship between the wage gap and the wage
levels (Wood et al., 1993), which will not be captured by just looking at the
behavior around the mean. Knowing which segment of the distribution is
affected the most can be important for policy.
The second limitation of the BO decomposition is that it requires para-
metric speciÞcation of the conditional expectation function linking earnings
with the individual characteristics. The mis-speciÞcation of this conditional
function however can give misleading results.1
1
The two limitations mentioned here are not the only criticisms of the BO approach.
For a discussion of the other, related, criticisms of the BO approach see Neumark (2004,
pages 9-10).

1
 There have been attempts in the literature to correct for these two lim-
itations. Jenkins (1994) and DiNardo et al. (1996) give distributional
alternatives to the BO decomposition. Barsky et al. (2002) and Racine
et al. (2004) propose nonparametric alternatives to the BO method, which
do not require parametric speciÞcation of the conditional expectation func-
tion.2 The nonparametric-distributional approach used in this paper is in
the spirit of the work done by these authors.3
Previous research on gender wage gap in Mexico largely relies on the
BO kind of decompositions. Brown et al. (1999) use the National Urban
Employment Survey (ENEU) for 1987-1993, their Þndings suggest that most
of the male-female difference can be explained by the differences in the
human capital endowments. Sanchez et al. (2001) focus on the earnings gap
between the female and male-owned micro-enterprises, and reach the same
conclusion as the Brown et al. (1999) paper. Pagan and Ullibari (2000)
in their paper use the ENEU data for 1995 and the measure proposed by
Jenkins (1994) to analyze the gender wage gap in Mexico.4 They conclude
that the unexplained wage gap is largest for those with very low or very high
level of education.
The period since 1984 has been one of trade liberalization and increased
foreign direct investment for Mexico (Lustig 1998 and 2001). The effects
of free trade on women are not clear. One argument is, as competition in-
creases, returns to employer discrimination fall, reducing the discrimination
faced by women — both in terms of increasing their previously depressed
wages and by feminization of high paid jobs, particularly in the industrial
2
Another approach is the use of quantile regressions; while this does allow us to look
at more points in the earnings distribution it still does not give a complete distributional
experience and requires a parametric speciÞcation of the conditional expectation function
(see Gardeazabal et al., 2005, for an application).
3
A similar approach is used by Breunig and Rospabe (2005) to study the male-female
wage gap in France.
4
While using Jenkins’ measure takes care of the Þrst criticism mentioned above the
second limitation still applies.

2
sector. The potential negative impacts of free trade include: ‘masculin-
ization’ of typical female jobs, as seen in maquiladoras in Mexico and a
decrease in the prices of commodities produced by women, particularly in
the agricultural sector (Artecona et al., 2002). The actual effect, however,
is an open empirical question.
There have been attempts in the literature to look directly at the gender
speciÞc impacts of structural adjustments in Mexico. Results are mixed:
Alarcon-Gonzalez et al. (1999) and Artecona et al. (2002) Þnd trade is
detrimental to women. Aguayo-Tellez et al. (2006) Þnd no impact of
trade on the gender wage gap. Anderson and Dimon (1995) use data from
1988 and focus on two female dominated industries (export processing and
tourism) in two Mexican cities (Tijuana and Torrean), their Þndings suggests
that the overall wage gap decreased as the export processing activity (as a
result of trade liberalization) increased.
This paper is different from the previous literature both in terms of the
methodology used and the longer horizon covered. All the studies mentioned
above cover the period till the mid-1990s. Most of them, like this study,
note that the unexplained gender wage gap in Mexico fell from the early
1980s to the mid 1990s, what they fail to notice is the subsequent upward
trend in the unexplained wage gap. Aguayo-Tellez et al. (2006) is the most
recent contribution and covers the same time horizon as this paper, but the
focus of their study is mainly on the employment changes faced by women.
The paper is structured as follows: the second section explains the dif-
ferent measures of discrimination used — the BO decomposition, the Jenkins
(1994) method and the nonparametric-distributional approach. The third
section describes the data used in the study and the empirical Þndings from
all three approaches. The last section concludes the paper.

3
2 Measures of Discrimination

The earnings equation is estimated separately for men (M ) and women (W ):

log yi = Xi β m + εm
i , for all i ∈ M, (1)

log yi = Xi β w + εw
i , for all i ∈ W,

where, yi is the wage of individual i; M and W are the sets of men and
women, respectively; Xi is the vector of individual characteristics, e.g. hu-
man capital; β m and β w are the vectors of unknown rates of returns to
individual characteristics, to be estimated for men and women, respectively;
εm w
i and εi are independent random residual terms for men and women,

respectively.

2.1 Blinder-Oaxaca Decomposition

The average wage gap between men and women can be decomposed as:

m w m w m w m w
log y − log y = (X b + X (β
− X )β b −β
b ), (2)

where the bar over the variable indicates the mean of the variable. The
Þrst term on the right hand side of equation (2) is referred to as the ex-
plained wage gap, the wage differential arising due to the differences in the
endowments. The second term is the part of the wage gap due to dif-
ferences in returns to similar endowments, and is an indirect measure of
discrimination.5 The summary measure of discrimination can be written
as:
5
There is a debate in the literature, about which wage structure — male or female —
should be used as the non-discriminatory one. All the measures presented here assume
that the male wage structure is the non-discriminatory wage structure. Another assump-
tion inherent in this analysis is that the distribution of X’s is not impacted by the β’s.
For discussion of the issues raised see Neumark (1988).

4
 DBO = 100[exp(S) − 1], (3)

w m w
b
where S = X (β b ).
−β DBO can be interpreted as the wage increase
women would have if there was no direct discrimination in the labor market.

2.2 Jenkins Measure

To estimate a distributional experience of discrimination, I start by con-

structing the Generalized Lorenz Curve (GLC) for ybiw = exp(log ybiw ), where
log ybiw is the estimated wage for women obtained from the estimated wage
equation in (1). Next, using the same ranking as for the GLC a General-
ized Concentration Centre (GCC) is obtained for rbiw = exp(log rbiw ), where
m
b .
log rbiw = Xiw β ri is the reference wage, the wage that women would
receive if their attributes are rewarded at rates β m ; this is the wage that
women would receive if there was no direct discrimination. As long as
ybiw < rbiw for all i ∈ W , the GCC must lie above the GLC. The further apart
the two curves are, the higher is the degree of discrimination. The area
between the two curves gives us the average discrimination index:

X
DJ = [1 + 1/(2nw )](rw − y w )/y w − (1/nw )2 ri − ybi )/y w ,
i(b (4)
i∈w

where the Þrst term is the difference between the means (hence compara-
ble to DBO above), and the second term is the weighted sum of deßated
wage gaps where the weights are a woman’s rank in the predicted earnings
distribution.
While DJ takes care of the limitation of looking only at the mean, i.e.
with this measure we are able to see the discrimination across the whole
distribution, it still has the problem of potential mis-speciÞcation of the
conditional expectation function (given by equation (1)). To correct for

5
both issues raised in the introduction I look at the nonparametric measure.6

2.3 Nonparametric Measure

The distribution of log earnings, h(log y|m) for men and h(log y|w) women,
respectively, can be written as:

Z
h(log y|m) = f (log y|m, x)g(x|m)dx, (5)
Z
h(log y|w) = f (log y|w, x)g(x|w)dx,

where f (log y|m, x) is the conditional log wage distribution given individual
characteristics (X) and gender; and g(x|m) is the distribution of individual
characteristics, given gender. The empirical counterpart of (5) is:

n
X µ ¶
b θilog y − log yi
h(log y|m) = K , for all i ∈ M, (6)
nh h
i=1
Xn µ ¶
b θi log y − log yi
h(log y|w) = K , for all i ∈ W,
nh h
i=1

where θ are the sample weights; K(.) is the kernel function; and h is the
window width.
BO decomposition and the Jenkins measure focus on the counterfactual
mean log wage, log riw = Xiw β m , the corresponding counterfactual distribu-
tion would be:
6
One advantage of the Jenkins measure, not central to the analysis done here but worth
noting, is that it satisÞes a set of axioms (normative properties) often used in the poverty
and deprivation literature (del Rio, Gradin and Canto, 2006).

6
 Z
hw (log y|m) ≡ f (log y|m, x)g(x|w)dx (7)
Z
= ω(x)f (log y|m, x)g(x|m)dx,

where ω(x) is the re-weighting function deÞned as ω(x) ≡ g(x|w)/g(x|m).

hw (log y|m) is the re-weighted distribution of men, such that the distribution
of individual characteristics is as that of women, but they are paid as men
would be. This is in the spirit of the work done by DiNardo et al. (1996).
To estimate the counterfactual distribution we need an estimate of the re-
weighting function, which using Baye’s rule can be written as:

Pr(w|x)/ Pr(w)
ω(x) = . (8)
Pr(m|x)/ Pr(m)

The conditional probabilities in the re-weighting function can be estimated

using a probit model. Once the re-weighting function has been estimated
the empirical counterpart of equation (7) is:

n
X µ ¶
b θi log y − log yi
hw (log y|m) = ω
b (x)K , for all i ∈ M. (9)
nh h
i=1

The difference between the two distributions, hw (log y|m) and h(log y|w),
gives us the extent of discrimination faced by women. To make the estimates
obtained here comparable with the summary measures above, I deÞne:

DN P = 100[exp(Sd ) − 1], (10)

R
where S d = Ew [log y|m] − E[log y|w]; E[log y|w] = log yh(log y|w)d log y
R
and Ew [log y|m] = log yhw (log y|m)d log y.

7
3 Empirical Findings

3.1 Data and Descriptive Statistics

The data used for the study is from Encuesta National de Ingresos y Gastos
de los Hogares (ENIGH). ENIGH is the national household survey, which
started in 1984, continued in 1989, 1992 and every two years thereafter.
The analysis in this paper focuses on four years — 1984, 1994, 1996 and
2002. To look at the period before the trade reforms were introduced I
analyze the data for 1984. The majority of reforms were implemented by
1994, hence comparison of the results for 1984 with 1994 give us some idea
about the ‘immediate’ impact of the changing macroeconomic environment
in the country. December 1994, for Mexico, was marked by a major currency
crisis followed by an economic recession.7 To see what impact this crisis
had on gender wage differentials, I analyze the data for 1996. Since 1996,
the Mexican economy has seen relatively stable growth, analysis of the 2002
data allows us to capture the impact of stabilization and growth on the
gender wage gaps.
Sample Selection: The sample is only of wage earners, over the age
of 16 years. All those individuals who have more than one job are excluded;
this is done to insure that workers can be correctly classiÞed as wage earners
only, it will rule out individuals who are wage earners, but also help out in
the family business and receive income from that. Students, retired people,
and those who give their reason for not working as ‘inability to work due to
disability’ are also dropped from the sample. ENIGH interviews a different
number of households each year, the sample selected here ranges from 30%
to 35% of the total ENIGH sample.
ENIGH employs a ‘stratiÞed sampling’ technique making the data na-
tionally representative. Sample weights made available by ENIGH are used
7
In 1995 the GDP growth rate for Mexico was -6.2%. Source: Table 7. The Mexican
Economy 1999, Banco de Mexico.

8
in the analysis below. The earnings measure is hourly real wage (net of
taxes), in 2002 Mexican Pesos, calculated from last months income and
usual hours worked last week.
Descriptive Statistics: The labor force participation of women has
increased over time in Mexico. In 1984 women were 28% of the sample,
this number increased steadily to 36% by 2002.8 Descriptive statistics for
the sample used here are reported in Tables 1a and 1b, for women and men
respectively.
Real wages for men and women increased between 1984 and 1994; the
currency crisis of 1994 and the recession following that are associated with
a decrease in real wages, which recovered only by year 2002.9 The average
age of the wage earners has increased over time; women however continue
to be on average two years younger then men. Hours worked per week are
also lower for women, by an average of seven hours per week. More men
tend to be heads of their households, relative to women. On average the
unionization rates among women tend to be higher than the rates for men,
though the rates of unionization have declined for both over time. Wage
earning women, on average, tend to have higher educational qualiÞcations
than men, and this gap has widened over time.10
Table 1c reports the occupational distribution by gender across the years.
The occupations are ranked from the lowest to the highest paying, the rank-
ing is based on the 1984 average wages for the whole sample in each oc-
8
Though the labor force participation (LFP) of women in Mexico has increased over the
last two decades, it is still much lower than what is prevalent in the developed countries.
According to OECD statistics (accessed on 31/10/2007) in 2002 LFP of women in Mexico
was 40% and for US and UK it was 66%.
Since the majority of the work force in Mexico is male, it is not unreasonable to assume
that the male wage structure is closest to the productivity related wages, and hence the
non-discriminatory wage structure. (Refer back to footnote 5)
9
The fall in the real wages, during the period of the currency crisis, is true not only for
the sample considered here but for the whole economy. Following the 1994 crisis, the real
wages in the manufacturing sector fell by 12.5%, and the real minimum wage (nationwide
average) fell by 12.3% (Source: Banco de Mexico).
10
Working women in Mexico are more educated than working men - this Þnding is
consistent with the Þndings of Pagan and Sanchez (2000) and Anderson and Dimon (1995).

9
cupational category; the ranking has not changed much across the years.
There is some evidence of occupational segregation, which is persistent over
the years. Women are largely employed as personal service workers (do-
mestic and establishment), education workers or as administrators in the
industrial sector. Men are largely employed as agricultural workers, and
as operators and less-skilled workers in the industrial sector. In most of
the occupations, that are female or male dominated, men on an average
earn more than women. Three occupations where women, across the years,
on average earned more than men are - vendors, peddlers with no business
representation, workers in arts, entertainment and sports, and police and
armed forces. However, a small proportion of the workforce, both men and
women, work in these three occupations.

3.2 Discrimination

Overall trend: The average raw log wage differential11 between men and
women is not very high. On average, over the entire period under consid-
eration female wages were about 99% of the male wages. This difference is
not statistically signiÞcant. The average conditional log wage difference12
however is higher and statistically signiÞcant (see Table 3). In 1984, female
wages were 21% lower than the male wages, this proportion decreased to
17% by 2002.
An earnings equation is estimated for both men and women separately13 ,
results are reported in Table 2. There is debate on what should/should not
be included as the control variable. Some argue that the control variables
11
The average raw log wage differential is simply the difference between the average log
wages for men and women.
12
The average conditional log wage differential is obtained by running an OLS regression
on the pooled sample of men and women. The dependent variable in the regression is
log wages and the independent variables are: age and age squared, dummy variables for
gender, education, union status, region of residence and occupation. The conditional log
wage difference is the coefficient on the ‘gender’ dummy.
13
A Chow test rejects the null hypothesis of a single regression for the pooled male and
female sample.

10
(vector X) should not include factors that themselves can be affected by
discrimination - like industry and occupation (Blau and Ferber, 1987). Re-
sults presented here are for the speciÞcation including the following control
variables: age, age squared, dummy variables for union status, education
(no education is the base category), region of residence (south is the base
category), and occupation (agricultural workers are the base category). Al-
ternate speciÞcations were also estimated to check the robustness of the
results, which are commented on later.
All coefficients are jointly signiÞcant. Age has a signiÞcant and a pos-
itive effect, at a decreasing rate, on log wages. Education at all levels has
a positive impact on the wage, with premiums increasing as the level of ed-
ucation increases. Returns to education are higher for women relative to
men, particularly at mid levels of education (more than primary, but less
than college); though the gap has narrowed over time. The union dummy
is signiÞcant, with the effect being larger for women than for men.
Regional dummies are signiÞcant and positive, with signiÞcant premiums
for those (both men and women) in the capital region or in the north (closer
to US) of the country. Returns in almost all the occupations are higher for
men relative to women, with a few exceptions, notable exceptions being the
three occupations where women on average earn more than men.
Table 3 reports the summary discrimination measures from all the three
approaches. Whichever summary measure we look at (DBO , DJ , DNP ),
discrimination was the highest in 1984, declined in the mid 1990s, since then
it has been rising again.
In 1984, if there were no discrimination, wages of women, depending on
the measure used, would have been 21% to 25% higher than their observed
wages. Given that to begin with average female wages were almost the same
as those of men, this would mean that the female wages, on average, would
be 21% to 25% higher than the male wages. Over the next decade, the

11
levels of discrimination faced by women fell, with almost all of the decline
happening after the 1994 currency crisis. Subsequent growth and stabiliza-
tion of the economy are associated with increased levels of discrimination
faced by women.
Next we look at each measure individually and see what additional in-
formation they provide.
BO decomposition: The entire raw wage differential can be explained
by discrimination. Not surprisingly the component attributable to ‘endow-
ments’ is negative — women have better human capital endowments.
Jenkins Measure: For all years, the GCC lies above the GLC every-
where, indicating that discrimination is present across the whole distribution
(see Figures 1a to 1d). But how has the experience of discrimination across
the years changed? Discrimination dominance checks across the years are
reported in Table 4. The mean predicted reference (i.e. counterfactual)
wage for each year is higher than the mean predicted wage. Both the 1984
and 1994 distributions lie above the 1996 and 2002 distributions, indicating
that the 1984 and the 1994 distribution of wages had more discrimination.
Of interest is the comparison of the 1984 and the 1994 distributions. While
the summary statistics indicate the two distributions have a similar degree
of discrimination, the nature of this discrimination is different for the two
years. Below the median, the 1994 distribution has less discrimination than
the 1984 distribution, above it the discrimination was higher in 1994 relative
to 1984.
Nonparametric Measure: Figures 2a to 2d, give us the distributions
of men and women for all the four years.14 The dispersion of wages is higher
for women relative to men in all years. For the years 1994, 1996 and 2002,
the actual distributions for men and women look similar at the lower tail;
14
To estimate the nonparametric distributions, I used the Gaussian kernel and window
width, h = 1.06σn−1/5 , where σ is the standard deviation of the log wages and n is the
sample size.

12
differences are in the upper tail where there seems to be another, smaller,
peak for women. Imposed on these graphs are also the counterfactual
wage distributions for women (i.e. the distribution for men reweighted).
If women had the endowments that they do and were paid as men are,
then their distribution would be as men’s but slightly to the right (with the
exception of 1996). Though the peak on the upper tail still remains for the
later years.
Using the estimated actual and the counterfactual distributions, the dis-
tributional experience of discrimination can be summarized using the fol-
lowing equation:

log ypm − log ypw = (log ypm − log rpw ) + (log rpw − log ypw ), (11)
T ermA T ermB

where log ypm is the log wage at the pth percentile of the male wage distri-
log ypm
R
bution, such that b
h(log y|m) = p. Similarly log ypw and log rpw are the
0
log wages at the pth percentile of the female actual wage distribution and
the counterfactual distribution, respectively. The left hand side of equation
(11) is the difference between the male and female wages, at each percentile,
calculated from the estimated kernel distributions. This total difference
is decomposed into the difference explained by the ‘endowments’ (T ermA
of equation (11)) and the ‘unexplained’ (T ermB of equation (11)) differ-
ence. The unexplained difference is interpreted as the indirect measure of
discrimination. The results of this exercise are reported in Table 5.
Just like in the BO decomposition we get the explained difference as neg-
ative - better endowments for women should, if left to market forces, yield
higher returns; this is true not only on average but is observed at every
percentile of the distribution. Discrimination is greater at the lower tail
of the distribution, and decreases as one moves up the distribution though

13
not monotonically.15 Changes in the discrimination (both the decline in
the Þrst decade and the upward trend in the subsequent decade) come from
changes in the ‘unexplained’ differences at the lower tail of the wage distri-
bution, with the levels of discrimination at the upper tail remaining largely
unchanged. Just as the Jenkins measure indicated, 1994 is an interest-
ing case, while the discrimination at the lower tail of the distribution has
decreased it has increased at the upper tail, relative to 1984.

3.3 Robustness checks

As mentioned above there is a debate in the literature about what should

and should not be included in the control variables. Numerous speciÞcations
were tried to check the robustness of the Þndings. The results from this
exercise are summarized here, detail results are available from the author
on request.
One of the alternative speciÞcation that was checked and is of particular
interest is one with no occupation dummies. The results for the summary
statistics (DBO , DJ , DNP ) from this speciÞcation are qualitatively the same
as the ones reported here, but are quantitatively different, with the discrim-
ination in this speciÞcation half of what is reported above. This is contrary
to the Þndings for the US where evidence suggests that including occupa-
tional dummies reduces the estimated discrimination (Neumark, 1988; Blau
and Kahn, 2006).
Two other variables that deserve a mention are the marital status of the
individual and the number of young children in the family. Including both
the variables in either of the above two speciÞcations signiÞcantly reduces
the estimated discrimination, without changing the results qualitatively.16
15
This is contrary to the Þndings of Wood et al. (1993).
16
This Þnding is consistent with that for the US (Neumark, 1988). The marriage
premium is high and signiÞcant for married men and low and insigniÞcant for women. If
marriage is related to unobserved productive characteristics then it can explain away all
discrimination (Neumark, 2004, chapter 1).

14
But these results should be intrepreted with caution, as these two variables
are not measured accurately in the data. While we can identify how many
children there are in the household, we cannot attach a child to an adult
woman in the household. This is because of the complex joint family struc-
ture prevalent in Mexico. The question regarding marital status was asked
only from the 1996 survey on. Before that it is possible only to identify
whether or not the head of the household is married, and the spouse of the
head of the household.
Including industry dummies, in either of the above speciÞcations, does
not signiÞcantly alter the results - qualitatively or quantitatively.

4 Concluding discussion

Establishing direct causality between the impact of structural reforms on

gender wage differentials is not straightforward. What this paper does is to
look at how the wage differentials, by gender, have changed over time given
the changing macro conditions in the country over these two decades.
Average raw wage differentials between men and women in Mexico are
not signiÞcant amongst the wage earners, who are predominantly in the for-
mal sector of the economy. This Þnding in itself is surprising, especially
when one looks closely at the endowments and the returns to these endow-
ments for men and women. Wage earning women have higher educational
achievement and higher unionization rates relative to men, the returns to
these endowments are also higher for women relative to men. These two
factors combined together should make the average wages for women higher
than the average wages for men.
Factors that let women down, it would seem, are — their age, their region
of residence, and their occupation. In the absence of information on actual
labor market experience, age is used here as a proxy for experience. Though
the returns to age are marginally lower for women than for men, what drives

15
the results is that the women in the sample are on average two years younger
than men (their actual experience could be even lower).
When we look at the discrimination measures, there does seem to be
evidence of discrimination against women — all the measures indicate that
women should be earning, on average, higher than what they are earning
currently. The Þndings are consistent across all the three measures used
in the paper; though each measure provides additional information. The
summary statistics often hide the distributional experience of discrimination.
For example we Þnd that while the average discrimination is similar in 1984
and 1994, the distributions are very different: with the 1994 distribution
showing lower discrimination at the lower tail and higher discrimination at
the upper tail, relative to 1984. On the other hand the summary statistics
for 2002 indicate much lower levels of discrimination relative to 1984, but
the distributional experiences are almost similar - with discrimination falling
as one moves up the distribution.
The average wage discrimination fell in the Þrst decade of the period con-
sidered here, with most of the decline coming right after the 1994 currency
crisis, in the second decade the discrimination started to increase. The dis-
tribution of discrimination has also changed over time. The average fall in
discrimination over time has resulted largely from the fall in the estimated
discrimination at the lower tail of the wage distribution, with little to no
change at the upper tail.
From a policy perspective the important things to note are the evidence
of a sticky ßoor and glass ceiling. There is evidence of a sticky ßoor with
unexplained wage differentials being higher at the lower tail, though these
have narrowed over time. The wage differentials at the upper tail, the glass
ceiling, however show no signs of relenting.
Two Þndings from this study need further exploration. First is the dra-
matic decrease in discrimination observed between 1994 and 1996. This

16
was the period of economic recession in Mexico. Recessions are often per-
ceived to increase the gender wage gap, as economic downturns are expected
to break the tenuous connection that women have with the labour market
(Langton and Konrad, 1998). The empirical evidence in support of this
hypothesis is however mixed with labour market structure being more im-
portant (Baden, 1993). For Mexico one possible explanation for the decrease
in the gender wage gap following the recession could be the differential im-
pact the recession had on the industrial and the services sector. Women
are predominantly in the service sector, this sector saw a lower downturn
compared with the industrial sector which mainly employs men.17
The second Þnding that needs further exploration is the impact of the
occupational structure on the wage gap. The Þndings here are contrary to
those in the literature.
The above results however carry some qualiÞcations. First, the study
looks only at the wage earners. There could be issues with selection bias,
selection into the formal wage earning sector cannot necessarily be assumed
random.18 Second, as the robustness checks carried out in this paper sug-
gest, there always is an issue of the control variables — what should and
should not go into X (the vector of control variables). It might be the
case that the Þnding that women do not earn higher than men, even if they
have higher education, is due to unobserved, productivity related, variables
which have not been controlled for. Third, there are issues with pre-labor
market discrimination which need to be explored.
17
Alaez Aller and Ullibarri Arce (2001) Þnd results supporting this explanation for
Spain.
18
For e.g., there is a higher concentration of women in the informal sector in Mexico,
may be women do not enter the formal sector due to other socioeconomic constraints
(e.g. care responsibilities within the family). Sample selection could also explain why
the results here are different from the Þndings of Brown et al. (1999) and Sanchez et al.
(2001).

17
References
1. Aguayo-Tellez E., J. Airola, and C. Juhn, 2006. “Did trade liberaliza-
tion help women? The case of Mexico in the 1990s,” Working Paper,
University of Houston.

2. Aláez Aller, R. and M. Ullṍbarri Arce, 2001. “Explaining the reduc-

tion in the gender wage gap during the 1990—94 economic recession in
Spain,” Atlantic Economic Journal, vol. 29, no. 1, pp. 63-74.

3. Alarcon-Gonzalez, D. and T. McKinley, 1999. “The adverse effects of

structural adjustments on working women in Mexico,” Latin American
Perspectives, vol. 26, no.3, pp. 103-117.

4. Anderson J.B. and D. Dimon, 1995. “The impact of opening markets

on Mexican male/female wage and occupational differentials,” The
Social Science Journal, vol. 32, no. 4, pp. 309-326.

5. Artecona, R. and W. Cunningham, 2002. “Effects of trade liberaliza-

tion on the gender wage gap in Mexico,” Policy Research Report on
Gender and Development, World Bank.

6. Baden, S., 1993. “The impact of recession and structural adjustment

on women’s work in selected developing countries,” Report No. 15,
BRIDGE (development — gender).

7. Banco de Mexico. The Mexican Economy 1999.

8. Barsky, R., J. Bound, K.K. Charles and J.P. Lupton, 2002. “Ac-
counting for the black-white wealth gap: A nonparametric approach,”
Journal of American Statistical Association, vol. 97, no. 459, pp.
663-673.

18
 9. Blau, F.D. and M. Ferber, 1987. “Discrimination: Empirical evidence
from the United States,” American Economic Review, vol. 77, no. 2,
pp. 316-20.

10. Blau, F.D. and L.M. Kahn, 2006. “The US gender pay gap in the
1990s: Slowing convergence,” Industrial and Labor Relations Review,
vol. 60, no.1, pp. 45-66.

11. Blinder, A.S., 1973. “Wage discrimination: Reduced form and struc-
tural estimates,” Journal of Human Resources, vol. 8, pp. 436-455.

12. Breunig, R. and S. Rospabe, 2005. “Parametric vs. semi-parametric

estimation of the male-female wage gap: An application to France,”
Social Science Research Network Paper 844565.

13. Brown C.J., J.A. Pagan and E. Rodriguez-Oreggia, 1999. “Occupa-

tional attainment and gender earnings differential in Mexico,” Indus-
trial and Labor Relations Review, vol. 53, no.1, pp. 123-135.

14. del Rio, C., C. Gradin and O. Canto, 2006. “The measurement of
gender wage discrimination: The distributional approach revisited,”
ECINEQ Working Paper 2006-25.

15. DiNardo, J., N.M. Fortin and T. Lemieux, 1996. “Labor market insti-
tutions and the distribution of wages, 1973-1993: A semi-parametric
approach,” Econometrica, vol. 64, pp. 1001-1044.

16. Gardeazabal, J. and A. Ugidos, 2005. “Gender wage discrimination

at quantiles,” Journal of Population Economics, vol. 18, no. 1, pp.
165-179.

17. Jenkins, S., 1994. “Earnings discrimination measurement: A distribu-

tional approach,” Journal of Econometrics, vol. 16, no.1, pp. 81-102.

19
18. Langton, N. and A.M. Konrad, 1998. “The impact of labor market
structure on sex differences in earnings,” Gender, Work and Organi-
zation, vol. 5, no. 2, pp. 83-101.

19. Lustig, N., 1998. Mexico: The Remaking of an Economy, Second

Edition (Brookings Institution Press, Washington DC).

20. Lustig, N., 2001. “Life is not easy: Mexico’s quest for stability and
growth.” Journal of Economic Perspectives, vol. 15, pp. 85-106.

21. Neumark, D., 1988. “Employers’ discriminatory behaviour and the

estimation of wage discrimination,” The Journal of Human Resourses,
vol. 23, no. 3, pp. 279-295.

22. Neumark, D., 2004. Sex Differences in Labor Markets, Routledge.

23. Oaxaca, R., 1973. “Male-female wage differential in urban labor mar-
kets,” International Economic Review, vol. 14, no. 3, pp. 693-709.

24. Pagan, J.A. and M. Ullibarri, 2000. “Group heterogeneity and the
gender earnings gap in Mexico,” Economia Mexicana Nueva Epoca,
vol. IX, num. 1, pp. 23-40.

25. Pagan, J.A. and S.M. Sanchez, 2000. “Gender Differences in Labor
Market Decisions: Evidence from Rural Mexico,” Economic Develop-
ment and Cultural Change, vol. 48, no. 3, pp. 619-637.

26. Racine, J.S. and C. Green, 2004. “Robust estimation of union and
gender wage gaps,” Manuscript, McMaster University.

27. Sanchez, S.M. and J.A.Pagan, 2001. “Explaining gender differences

in earnings in micro-enterprise sector,” in M. Correria and E. Katz
(eds) The Economics of Gender in Mexico: Work, Family, State and
Market (World Bank).

20
28. Wood, R.G., M.E. Corcoran and P. Courant, 1993. “Pay Differences
among the highly paid: The male-female earnings gap in lawyers,”
Journal of Labor Economics, vol. 11, no. 3, pp. 417-441.

21
 Table 1a: Average Sample Characteristics1 - Women
1984 1994 1996 2002
Personal Characteristics
Log real wage (hourly)2 2.63 (0.85) 2.81 (0.85) 2.43 (0.83) 2.66 (0.84)
Age (in years) 31.46 (11.51) 30.34 (10.78) 31.96 (10.75) 33.77 (11.73)
Hours worked per week 40.11 (12.96) 41.25 (13.63) 41.62 (13.51) 41.19 (13.80)
Union member (1=yes) 0.31 (0.46) 0.23 (0.42) 0.23 (0.42) 0.19 (0.39)
Head of the household (1=yes) 0.19 (0.39) 0.14 (0.35) 0.16 (0.36) 0.18 (0.38)
Education Level (proportion of the sample)
Primary incomplete 0.18 (0.38) 0.12 (0.32) 0.12 (0.33) 0.11 (0.31)
Primary complete 0.27 (0.44) 0.16 (0.37) 0.18 (0.38) 0.16 (0.37)
Junior high incomplete 0.04 (0.19) 0.04 (0.20) 0.03 (0.18) 0.04 (0.19)
Junior high complete 0.23 (0.42) 0.32 (0.47) 0.30 (0.46) 0.27 (0.45)
High school incomplete 0.03 (0.16) 0.04 (0.20) 0.04 (0.19) 0.06 (0.24)
High school complete 0.09 (0.28) 0.12 (0.33) 0.11 (0.31) 0.12 (0.33)
Some college 0.04 (0.19) 0.05 (0.22) 0.07 (0.25) 0.09 (0.28)
College complete 0.06 (0.23) 0.09 (0.29) 0.10 (0.31) 0.10 (0.30)
More than college - 0.01 (0.08) 0.00 (0.07) 0.01 (0.08)
Region3 of Residence (proportion of the sample)
North 0.28 (0.45) 0.29 (0.45) 0.28 (0.45) 0.28(0.45)
Center 0.37 (0.48) 0.40 (0.49) 0.29 (0.45) 0.43(0.50)
Capital 0.10 (0.29) 0.16 (0.37) 0.14 (0.35) 0.16(0.37)
South 0.26 (0.44) 0.14 (0.35) 0.18 (0.39) 0.13(0.33)
Observations 1136 3417 3482 5625
1. Standard deviation in parentheses. Sample weights are used in all calculations.
2. In 2002 Mexican Pesos.
3. 32 states of Mexico are divided into four regions. Northern states: Baja California, Baja California Sur, Chihuahua, Coahuila, Durango, Nayarit, Nuevo Leon, San Luis
Potosi, Sinaloa, Sonora, Tamaulipas, Zacatecas; Central states: Aquascalientes, Colima, Jalisco, Guanajuato, Hidalgo, Mexico, Michoacan, Morelos, Puebla, Queretaro,
Tlaxcala; Southern states: Campeche, Chiapas, Guerrero, Oaxaca, Quintana Roo, Tabasco, Veracruz, Yucatan; Capital: Federal District.
Source: Author’s calculations from ENIGH dataset for various years.

22
 Table 1b: Average Sample Characteristics - Men
1984 1994 1996 2002
Personal Characteristics
Log real wage (hourly) 2.65 (0.79) 2.79 (0.87) 2.43 (0.82) 2.66 (0.79)
Age (in years) 33.77 (12.81) 33.07 (12.73) 33.50 (12.00) 35.00 (13.16)
Hours worked per week 46.52 (11.67) 48.96 (13.33) 49.52 (13.63) 48.94 (13.26)
Union member (1=yes) 0.22 (0.41) 0.16 (0.36) 0.15 (0.36) 0.14 (0.35)
Head of the household (1=yes) 0.68 (0.47) 0.64 (0.48) 0.67 (0.47) 0.62 (0.49)
Education Level (proportion of the sample)
Primary incomplete 0.27 (0.45) 0.18 (0.39) 0.16 (0.37) 0.14 (0.35)
Primary complete 0.23 (0.42) 0.22 (0.41) 0.21 (0.40) 0.19 (0.39)
Junior high incomplete 0.06 (0.24) 0.07 (0.25) 0.05 (0.22) 0.05 (0.22)
Junior high complete 0.15 (0.36) 0.21 (0.41) 0.24 (0.43) 0.26 (0.44)
High school incomplete 0.03 (0.17) 0.04 (0.20) 0.06 (0.23) 0.06 (0.24)
High school complete 0.05 (0.21) 0.07 (0.25) 0.09 (0.29) 0.10 (0.30)
Some college 0.04 (0.20) 0.06 (0.23) 0.06 (0.22) 0.07 (0.25)
College complete 0.06 (0.24) 0.08 (0.26) 0.07 (0.26) 0.08 (0.26)
More than college 0.00 (0.06) 0.01 (0.07) 0.01 (0.09) 0.01 (0.10)
Region of Residence (proportion of the sample)
North 0.28 (0.45) 0.27 (0.44) 0.29 (0.45) 0.28 (0.45)
Center 0.40 (0.49) 0.43 (0.49) 0.28 (0.45) 0.41 (0.49)
Capital 0.13 (0.34) 0.19 (0.39) 0.16 (0.37) 0.20 (0.40)
South 0.18 (0.38) 0.12 (0.32) 0.14 (0.35) 0.11 (0.31)
Observations 3011 7770 8004 10394
For sources and definitions refer to notes at the end of Table 1a.

23
 Table 1c: Occupation – proportion of the sample
1984 1994 1996 2002
Men Women Men Women Men Women Men Women
Personal service workers -domestic 0.01 (0.08) 0.11 (0.31) 0.01 (0.08) 0.15 (0.35) 0.01 (0.11) 0.13 (0.34) 0.01 (0.12) 0.14 (0.35)
Agricultural workers 0.19 (0.39) 0.05 (0.21) 0.13 (0.33) 0.02 (0.12) 0.11 (0.31) 0.02 (0.13) 0.11 (0.31) 0.01 (0.12)
Vendors, peddlers- with no business 0.00 (0.05) 0.01 (0.08) 0.01 (0.09) 0.01 (0.11) 0.01 (0.09) 0.01 (0.09) 0.01 (0.11) 0.02 (0.13)
representation
Less skilled workers- industrial 0.10 (0.30) 0.01 (0.10) 0.14 (0.35) 0.03 (0.17) 0.11 (0.32) 0.02 (0.15) 0.13 (0.33) 0.03 (0.17)
production
Police and armed forces 0.03 (0.16) 0.00 (0.03) 0.04 (0.19) 0.00 (0.06) 0.05 (0.23) 0.01 (0.08) 0.05 (0.21) 0.01 (0.08)
Personal service workers - establishments 0.05 (0.22) 0.13 (0.34) 0.07 (0.25) 0.07 (0.26) 0.07 (0.26) 0.08 (0.27) 0.07 (0.25) 0.11 (0.31)
Workers, operators – industrial 0.25 (0.43) 0.13 (0.33) 0.22 (0.41) 0.12 (0.33) 0.21 (0.41) 0.16 (0.37) 0.21 (0.41) 0.14 (0.35)
production
Salespersons 0.07 (0.25) 0.09 (0.28) 0.08 (0.27) 0.12 (0.32) 0.08 (0.26) 0.11 (0.31) 0.09 (0.29) 0.14 (0.35)
Transport workers 0.07 (0.25) - 0.08 (0.28) - 0.10 (0.30) - 0.09 (0.29) 0.00 (0.04)
Administrative workers –industrial 0.09 (0.29) 0.28 (0.45) 0.08 (0.27) 0.23 (0.42) 0.10 (0.29) 0.23 (0.42) 0.10 (0.29) 0.19 (0.40)
production
Technicians 0.03 (0.17) 0.07 (0.25) 0.04 (0.20) 0.10 (0.29) 0.04 (0.20) 0.06 (0.24) 0.03 (0.17) 0.05 (0.23)
Supervisors- industrial production 0.03 (0.18) 0.01 (0.09) 0.02 (0.15) 0.01 (0.11) 0.03 (0.17) 0.02 (0.12) 0.03 (0.17) 0.02 (0.12)
Workers in arts, entertainment and sports 0.01 (0.10) 0.00 (0.03) 0.01 (0.10) 0.01 (0.08) 0.01 (0.11) 0.01 (0.07) 0.01 (0.08) 0.00 (0.06)
Education workers 0.03 (0.16) 0.10 (0.30) 0.02 (0.16) 0.09 (0.29) 0.02 (0.15) 0.10 (0.30) 0.03 (0.16) 0.08 (0.27)
Professional 0.02 (0.14) 0.01 (0.12) 0.02 (0.15) 0.03 (0.16) 0.02 (0.15) 0.04 (0.20) 0.03 (0.17) 0.04 (0.19)
Senior directors, Administrators in public 0.03 (0.17) 0.01 (0.09) 0.03 (0.16) 0.02 (0.14) 0.03 (0.17) 0.02 (0.13) 0.02 (0.14) 0.02 (0.13)
and private sector
Occupations are ranked from lowest to highest paying. The ranking is based on the 1984 average wages for the whole sample in each occupational category. The ranking does
not change much across years.
Standard deviation in parentheses. Sample weights are used in all calculations.
Source: Author’s calculations from ENIGH dataset for various years.

24
 Table 2: Estimated Coefficients of Earnings Equation
Dependent Variable: Log real wage (hourly); Method of estimation: Least Squares
1984 1994 1996 2002
Independent Variables Men Women Men Women Men Women Men Women
Age 0.06* 0.05* 0.06* 0.06* 0.06* 0.04* 0.05* 0.05*
Age Square (x 10-3) -0.71* -0.55* -0.57* -0.55* -0.68* -0.44* -0.52* -0.59*
Union 0.24* 0.27* 0.10* 0.13* 0.19* 0.21* 0.19* 0.33*
Education dummies
Primary incomplete 0.27* 0.50* 0.16* 0.22* 0.12* 0.22* 0.17* 0.31*
Primary complete 0.46* 0.76* 0.30* 0.46* 0.30* 0.35* 0.26* 0.41*
Junior high incomplete 0.54* 0.94* 0.40* 0.52* 0.36* 0.53* 0.39* 0.50*
Junior high complete 0.56* 0.99* 0.44* 0.61* 0.44* 0.56* 0.38* 0.56*
High school incomplete 0.67* 0.86* 0.52* 0.72* 0.57* 0.64* 0.51* 0.67*
High school complete 0.77* 1.00* 0.68* 0.96* 0.65* 0.78* 0.55* 0.77*
Some college 0.79* 1.02* 0.86* 0.89* 0.90* 0.91* 0.82* 0.94*
College complete 0.96* 1.06* 1.22* 1.14* 1.08* 1.06* 1.07* 1.10*
More than college 1.41* - 1.51* 1.50* 1.35* 1.47* 1.39* 1.36*
Region dummies
North 0.13* 0.24* 0.25* 0.17* 0.18* 0.19* 0.37* 0.37*
Centre 0.09* 0.07 0.24* 0.22* 0.04** 0.05*** 0.27* 0.28*
Capital 0.21* 0.35* 0.34* 0.34* 0.19* 0.26* 0.25* 0.39*
Occupation dummies
Personal service workers -domestic 0.04 -0.12 0.39* 0.34* 0.11*** -0.05 0.26* 0.26*
Vendors, peddlers- with no business
representation 0.17 0.51** 0.41* 0.60* -0.03 0.1 0.12** 0.22*
Less skilled workers- industrial production 0.26* -0.12 0.30* 0.30* 0.14* -0.09 0.26* 0.12
Polish and armed forces 0.23* 0.24 0.31* 0.36** 0.28* 0.44* 0.32* 0.49*
Personal service workers - establishments 0.46* 0.35* 0.50* 0.34* 0.31* 0.08 0.31* 0.35*
Workers, operators – industrial production 0.39* 0.38* 0.53* 0.33* 0.38* -0.07 0.44* 0.22*
Salespersons 0.47* 0.18*** 0.54* 0.40* 0.36* -0.01 0.49* 0.33*
Transport workers 0.50* - 0.53* - 0.38* - 0.39* 0.35***
Administrative workers –industrial
production 0.69* 0.62* 0.81* 0.78* 0.66* 0.35* 0.67* 0.66*
Technicians 0.70* 0.69* 0.92* 0.84* 0.58* 0.43* 0.64* 0.73*
Supervisors- industrial production 0.80* 0.39** 0.77* 0.67* 0.73* 0.30* 0.78* 0.66*
Workers in arts, entertainment and sports 0.79* 0.86*** 0.98* 1.09* 0.91* 0.73* 0.77* 1.73*
Education workers 0.94* 0.95* 1.07* 1.10* 0.79* 0.72* 0.83* 0.89*

25
Table 2 (Continued)

Professional 0.79* 1.04* 0.95* 1.04* 0.66* 0.49* 0.93* 1.00*

Senior directors, Administrators in public and
private sector 1.18* 1.05* 1.42* 1.11* 1.21* 0.85* 1.27* 1.17*
Constant 0.36* 0.23 0.40* 0.20*** 0.22* 0.56* 0.48* 0.15
R-squared 0.54 0.60 0.56 0.57 0.50 0.53 0.51 0.54
Observations 3011 1136 7770 3417 8004 3482 10394 5625
For sources and definitions refer to notes at the end of Table 1a.
* significant at 1% level; ** significant at 5% level; *** significant at 10% level.
1. Nine education dummies are included in estimation, base category is – ‘no education’.
2. Three regional dummies are included in estimation, base category is – ‘south’
3. Fifteen occupational dummies are included in estimation, base category is – ‘agricultural workers’

26
 Table 3: Gender Wage Differentials
1984 1994 1996 2002
Raw log wage differential1 (standard error) 0.02 (0.03) -0.02 (0.02) -0.00 (0.02) 0.01 (0.01)
Conditional log wage differential2 (standard error) 0.19* (0.02) 0.20* (0.01) 0.16* (0.01) 0.16* (0.01)
Blinder-Oaxaca Decomposition
Component attributable to:
Endowments -0.17 -0.20 -0.14 -0.14
Discrimination (t-stat) 0.19 (7.41) 0.18 (10.12) 0.14 (9.02) 0.15 (12.47)
Discrimination index (DBO) 20.75 20.05 14.76 15.89
Jenkins Measure
Average discrimination index (DJ) 21.74 20.88 16.38 18.53
Nonparametric Measure
Discrimination index (DNP) 24.68 19.56 13.17 15.36
* significant at 1% level; ** significant at 5% level; *** significant at 10% level.
1. The raw log wage differential is simply the difference between the average log wages for men and women.
2. The conditional log wage differential is obtained by running an OLS regression on the pooled sample of men and women. The dependent variable in the regression is log
wages and the independent variables are: age and age squared, dummy variables for gender, education, union status, region of residence, and occupation. The conditional
log wage difference is the coefficient on the ‘gender’ dummy.
Source: Author’s calculations from ENIGH dataset for various years.

27
 Table 4: Jenkins Measure – summary statistics
1984 1994 1996 2002
Mean predicted reference wage for women ( r w ) 19.09 23.71 15.20 19.31
Mean predicted wage for women ( y w ) 16.36 19.84 13.38 17.11
r w – yw 2.73 3.87 1.82 2.20
Cumulative sample share (%) Generalized Lorenz ordinates for r w − y w
10 0.22 0.15 0.04 0.15
20 0.48 0.37 0.12 0.33
30 0.71 0.63 0.24 0.55
40 0.90 0.91 0.38 0.76
50 1.12 1.19 0.55 0.99
60 1.31 1.47 0.73 1.23
70 1.48 1.72 0.92 1.42
80 1.74 1.99 1.13 1.57
90 2.06 2.41 1.39 1.72
100 2.73 3.87 1.82 2.20
Source: Author’s calculations from ENIGH dataset for various years.

Table 5: Nonparametric Measure – summary statistics1

1984 1994 1996 2002
Term A Term B Term A Term B Term A Term B Term A Term B
Wage percentiles (%)
5 -0.30 0.62 -0.29 0.34 -0.13 0.23 -0.16 0.27
10 -0.30 0.49 -0.25 0.28 -0.10 0.16 -0.15 0.22
25 -0.25 0.27 -0.20 0.20 -0.09 0.12 -0.15 0.19
50 -0.20 0.14 -0.19 0.14 -0.14 0.12 -0.13 0.14
75 -0.21 0.11 -0.23 0.07 -0.19 0.09 -0.14 0.07
90 -0.09 0.08 -0.15 0.14 -0.12 0.09 -0.13 0.02
95 -0.03 0.08 -0.02 0.21 -0.03 0.09 -0.05 0.07
1. Term A is the wage difference explained by ‘endowments’; Term B is the unexplained wage difference, interpreted as the indirect measure of discrimination. Refer to
equation (11) in the main text on how these are calculated.
Source: Author’s calculations from ENIGH dataset for various years.

28
 Figure 1a. Joint distribution of actual (y) and
counterfactual (r) estimated wages, 1984
p x Mean wage amongst the lowest p
0 5 10 15 20 25

0 .2 .4 .6 .8 1
Population share, p (women ranked by their estimated wage, y)

Generalized Lorenz Curve for y

Generalized Concentration Curve for r
Source: Authors calculations from ENIGH 1984.

Figure 1b. Joint distribution of actual (y) and

counterfactual (r) estimated wages, 1994
p x Mean wage amongst the lowest p
0 5 10 15 20 25

0 .2 .4 .6 .8 1
Population share, p (women ranked by their estimated wage, y)

Generalized Lorenz Curve for y

Generalized Concentration Curve for r
Source: Authors calculations from ENIGH 1994.

29
 Figure 1c. Joint distribution of actual (y) and
counterfactual (r) estimated wages, 1996
p x Mean wage amongst the lowest p
0 5 10 15 20 25

0 .2 .4 .6 .8 1
Population share, p (women ranked by their estimated wage, y)

Generalized Lorenz Curve for y

Generalized Concentration Curve for r
Source: Authors calculations from ENIGH 1996.

Figure 1d. Joint distribution of actual (y) and

counterfactual (r) estimated wages, 2002
p x Mean wage amongst the lowest p
0 5 10 15 20 25

0 .2 .4 .6 .8 1
Population share, p (women ranked by their estimated wage, y)

Generalized Lorenz Curve for y

Generalized Concentration Curve for r
Source: Authors calculations from ENIGH 2002.

30
 Figure 2a. Density of log wages, 1984
.6
.4
Density
.2
0

-2 0 2 4 6 8
Log hourly wage

Women Men
Men - reweighted
Source: Authors calculations from ENIGH 1984.

Figure 2b. Density of log wages, 1994

.6
.4
Density
.2
0

-2 0 2 4 6 8
Log hourly wage

Women Men
Men - reweighted
Source: Authors calculations from ENIGH 1994.

31
 Figure 2c. Density of log wages, 1996
.6
.4
Density
.2
0

-2 0 2 4 6 8
Log hourly wage

Women Men
Men - reweighted
Source: Authors calculations from ENIGH 1996.

Figure 2d. Density of log wages, 2002

.6
.4
Density
.2
0

-2 0 2 4 6 8
Log hourly wage

Women Men
Men - reweighted
Source: Authors calculations from ENIGH 2002.

32