NORC at the University of Chicago

The University of Chicago

Unequal Pay or Unequal Employment? A Cross‐Country Analysis of Gender Gaps

Author(s): Claudia Olivetti and Barbara Petrongolo
Source: Journal of Labor Economics, Vol. 26, No. 4 (October 2008), pp. 621-654
Published by: The University of Chicago Press on behalf of the Society of Labor Economists and
the NORC at the University of Chicago
Stable URL: http://www.jstor.org/stable/10.1086/589458 .
Accessed: 28/11/2014 11:01

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

The University of Chicago Press, Society of Labor Economists, NORC at the University of Chicago, The
University of Chicago are collaborating with JSTOR to digitize, preserve and extend access to Journal of
Labor Economics.

http://www.jstor.org

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
 Unequal Pay or Unequal Employment?
A Cross-Country Analysis
of Gender Gaps
Claudia Olivetti, Boston University

Barbara Petrongolo, London School of Economics, CEP,

CEPR, and IZA

We analyze gender wage gaps correcting for sample selection induced

by nonemployment. We recover wages for the nonemployed using
alternative imputation techniques, simply requiring assumptions on
the position of imputed wages with respect to the median. We obtain
higher median wage gaps on imputed rather than actual wage distri-
butions for several OECD countries. However, this difference is small
in the United States, the United Kingdom, and most central and north-
ern EU countries and becomes sizable in southern EU countries, where
gender employment gaps are high. Selection correction explains nearly
half of the observed negative correlation between wage and employ-
ment gaps.

I. Introduction
There is substantial international variation in gender pay gaps, from
around 30 log points in the United States and the United Kingdom to

We wish to thank Ivan Fernandez-Val, Larry Katz, Kevin Lang, Alan Manning,
and Steve Pischke for their very helpful suggestions. We also acknowledge comments
from seminars at several institutions, as well as from presentations at the Bank of
Portugal Annual Conference 2005, the SOLE/EALE Conference 2005, the Conference
in Honor of Reuben Gronau 2005, and the National Bureau of Economic Research
Summer Institute 2006. Olivetti aknowledges the Radcliffe Institute for Advanced

[ Journal of Labor Economics, 2008, vol. 26, no. 4]

䉷 2008 by The University of Chicago Press. All rights reserved.
0734-306X/2008/2604-0006$10.00

621

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
622 Olivetti/Petrongolo

between 10 and 25 log points in a number of central and northern Eu-

ropean countries, down to an average below 10 log points in southern
Europe. International differences in overall wage dispersion are typically
found to play a role in explaining the variation in gender pay gaps (Blau
and Kahn 1996, 2003). The idea is that a given level of dissimilarities
between the characteristics of working men and women translates into a
higher gender wage gap the higher the overall level of wage inequality.
However, the Organization for Economic Cooperation and Development
(OECD 2002, chart 2.7) shows that, while differences in the wage struc-
ture do explain an important portion of the international variation in
gender wage gaps, the inequality-adjusted wage gap in southern Europe
remains substantially lower than such gaps in the rest of Europe and the
United States.
In this article we argue that, in addition to differences in wage inequality,
and therefore in the returns associated to characteristics of working men
and women, a significant portion of the international variation in gender
wage gaps may be explained by differences in characteristics themselves,
whether observed or unobserved. This idea is supported by the striking
international variation in employment gaps, ranging from 10 percentage
points in the United States, the United Kingdom, and Scandinavian coun-
tries to 15–25 points in northern and central Europe, up to 30–40 points
in southern Europe and Ireland (see fig. 1). If selection into employment
is nonrandom, it makes sense to worry about the way in which selection
may affect the resulting gender wage gap. In particular, if women who
are employed tend to have relatively high-wage characteristics, low female
employment rates may become consistent with low gender wage gaps
simply because low-wage women would not feature in the observed wage
distribution. This idea could thus be well suited to explain the negative
correlation between gender wage and employment gaps that we observe
in the data.
Different patterns of employment selection across countries may in
turn stem from a number of factors. First, there may be international
differences in labor supply behavior and in particular in the role of house-
hold composition and/or social norms in affecting participation. Second,
labor demand mechanisms, including social attitudes toward female em-
ployment and their potential effects on employer choices, may be at work,
affecting both the arrival rate and the level of wage offers of the two
genders. Finally, institutional differences in labor markets regarding
unionization and minimum wages may truncate the wage distribution at
different points in different countries, affecting both the composition of

Studies for financial support during the early stages of the project. Petrongolo ak-
nowledges the ESRC for financial support to the Centre for Economic Performance.
Contact the corresponding author, Barbara Petrongolo, at [email protected].

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 623

Fig. 1.—Gender gaps in mean (log) hourly wages and in employment, 1999. Coefficient
of correlation: ⫺0.474.

employment and the observed wage distribution. In this article we will

be agnostic as regards the separate role of these factors in shaping gender
gaps, and aim at recovering alternative measures of selection-corrected
gender wage gaps.
Although there exist substantial literatures on gender wage gaps, on
the one hand, and gender employment, unemployment and participation
gaps, on the other hand,1 to our knowledge the variation in both quantities
and prices in the labor market has not been simultaneously exploited to
understand important differences in gender gaps across countries. In this
article we claim that the international variation in gender employment
gaps can indeed shed some light on well-known cross-country differences
in gender wage gaps. We will explore this view by estimating selection-
corrected wage gaps.
To analyze gender wage gaps across countries, allowing for sample
selection induced by nonemployment, we recover information on wages
for those not working in a given year using alternative imputation tech-

1
See Altonji and Blank (1999) for an overall survey on both employment and gender
gaps for the United States, Blau and Kahn (2003) for international comparisons of
gender wage gaps, and Azmat, Güell, and Manning (2006) for international compar-
isons of unemployment gaps.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
624 Olivetti/Petrongolo

niques.2 Our approach is closely related to that of Johnson, Kitamura,

and Neal (2000) and Neal (2004), and it simply requires assumptions on
the position of the imputed wage observations with respect to the median.
Importantly, it does not require assumptions on the actual level of missing
wages, as is typically required in the matching approach, nor does it
require the arbitrary exclusion restrictions often invoked in two-stage
Heckman sample selection correction models.
We then estimate median wage gaps on the sample of employed workers
and on a sample enlarged with wage imputation for the nonemployed.
The impact of selection into work on estimated wage gaps is assessed by
comparing estimates obtained under alternative sample inclusion rules.
The attractive feature of median regressions is that the results are only
affected by the position of wage observations with respect to the median
and not by specific values of imputed wages. If missing wage observations
are correctly imputed on the side of the median where they belong, then
median regressions retrieve the true parameters of interest.
Imputation can be performed in several ways, and our alternative im-
putation methods will address slightly different economic mechanisms of
selection. First, we use panel data and, for all those not in work in some
base year, we search backward and forward to recover wage observations
from the nearest wave in the sample. This implicitly assumes that an
individual’s position with respect to the base-year median can be signaled
by her wage from the nearest wave. As imputation is simply driven by
wages observed in other waves, we are in practice allowing for selection
on unobservables. Estimates based on this procedure tell what level of
the gender wage gap we would observe if the nonemployed earned “sim-
ilar” wages to those earned when they were employed, where “similar”
here means on the same side of the base-year median.
While this imputation method arguably uses the minimum set of po-
tentially arbitrary assumptions, it cannot provide wage information on
individuals who never work during the sample period. In order to recover
wages also for those never observed in work, we use observable char-
acteristics of the nonemployed to make educated guesses concerning their
position with respect to the median. In this case we are allowing for
selection on observable characteristics only, assuming that the nonem-
ployed would earn wages “similar” to the wages of the employed with
matching characteristics, where again “similar” means on the same side
of the base-year median. Having done this, earlier or later wage obser-
2
We do not attempt to provide a structural model of wage determination that would
in principle characterize general equilibrium effects of sample selection but would do
so at the cost of making assumptions on production technologies involving male and
female work. We are simply trying to estimate the gender wage gap correcting for
sample selection.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 625

vations for those with imputed wages in the base year can shed light on
the goodness of our imputation methods.
We next use probability models for assigning individuals on either side
of the median of the wage distribution. To do this we fit a probit model
for the probability of an employed worker’s belonging above their gender-
specific median, based on education and labor market experience, and
obtain predicted probabilities for the nonemployed. We then construct
an imputed sample using such predicted probabilities as sampling weights.
We complete our set of results by estimating bounds to the distribution
of wages (see Manski 1994), using either the actual or the imputed wage
distribution in turn. Bounds computed using the observed wage distri-
bution are interesting because they show that all our wage gap estimates
based on imputation do fall within these bounds. When the imputed wage
distribution is used, the increase in the proportion of individuals with a
wage (actual or imputed) allows us to tighten the bounds, as predicted
by the theory.
In our study we use panel data sets that are as comparable as possible
across countries, namely, the Panel Study of Income Dynamics (PSID)
for the United States and the European Community Household Panel
Survey (ECHPS) for Europe. We consider the period 1994–2001, which
is the longest time span for which data are available for all countries. Our
estimates on these data deliver higher median wage gaps on imputed rather
than actual wage distributions for most countries and across alternative
imputation methods. This implies, as one would have expected, that
women tend on average to be more positively selected into work than
men. However, the difference between actual and potential wage gaps is
small in the United States, the United Kingdom, and most central and
northern European countries, and it becomes sizable in southern Europe,
where the gender employment gap is highest. Under our most conser-
vative correction, sample selection into employment explains nearly half
of the observed negative correlation between gender wage and employ-
ment gaps. In particular, in Spain, Italy, Portugal, and Greece, the median
wage gap on the imputed wage distribution reaches levels closely com-
parable to those of the United States and of other central and northern
European countries.
Our results thus show that, while the raw wage gap is much higher in
Anglo Saxon countries than in southern Europe, the reason is probably
not to be found in more equal pay treatment for women in the latter
group of countries but mainly in a different process of selection into
employment. Female participation rates in Catholic countries and Greece
are low and and are concentrated among high-wage women. Having cor-
rected for lower participation rates, the wage gap there widens to levels
similar to those of other European countries and the United States.
The article is organized as follows. Section II discusses the related lit-

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
626 Olivetti/Petrongolo

erature. Section III describes the data sets used and presents descriptive
evidence on gender gaps. Section IV describes our imputation method-
ologies. Section V estimates raw median gender wage gaps on actual and
imputed wage distributions, to illustrate how alternative sample selection
rules affect the estimated gaps. Conclusions are brought together in Sec-
tion VI.

II. Related Work

The importance of selectivity biases in making wage comparisons has
long been recognized since seminal work by Gronau (1974) and Heckman
(1974, 1979, 1980). The current literature contains a number of country-
level studies that estimate selection-corrected wage gaps across genders
or ethnic groups, based on a variety of correction methodologies. Among
studies that are more closely related to our article, Neal (2004) estimates
the gap in potential earnings between black and white women in the
United States by fitting median regressions on imputed wage distributions,
using alternative methods of wage imputation for women nonemployed
in 1990. He finds that the gap between potential earnings of white and
black women is at least 60% higher than the gap in actual earnings, thus
revealing that black women are more positively selected into work. Using
both wage imputation and matching techniques, Chandra (2003) finds
that the wage gap between black and white U.S. males is also understated,
due to selective withdrawal of black men from the labor force during the
1970s and 1980s.3
Turning to gender wage gaps, Blau and Kahn (2006) study changes in
the U.S. gender wage gap between 1979 and 1998 and find that sample
selection implies that the 1980s gains in women’s relative wage offers were
overstated and that selection may also explain part of the slowdown in
convergence between male and female wages in the 1990s. Their approach
is based on wage imputation for those not in work, along the lines of
Neal (2004). Mulligan and Rubinstein (2008) also argue that the narrowing
of the gender wage gap in the United States during 1975–2001 may be a
direct impact of progressive selection into employment of high-wage
women, in turn attracted by widening within-gender wage dispersion.
Correction for selection into work is implemented here using a two-stage
Heckman (1979) selection model. The authors show that, while in the
1970s the gender selection bias was negative, that is, nonemployed women
had higher earnings potential than working women, it became positive in
the mid 1980s.4
3
See also Blau and Beller (1992) and Juhn (2003) for earlier use of matching tech-
niques in the study of selection-corrected race gaps.
4
Earlier studies that discuss the importance of changing characteristics of the female
workforce in explaining the dynamics of the gender wage gap in the United States
include O’Neil (1985), Smith and Ward (1989), and Goldin (1990).

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 627

Related work on European countries includes Beblo et al. (2003), Al-

brecht, van Vuuren, and Vroman (2004), and Blundell et al. (2007). Blun-
dell et al. examine changes in the distribution of wages in the United
Kingdom during the period 1978–2000. They allow for the impact of
nonrandom selection into work by using bounds to the latent wage dis-
tribution according to the procedure proposed by Manski (1994). Bounds
are first constructed based on the worst-case scenario and then progres-
sively tightened using restrictions motivated by economic theory. Features
of the resulting wage distribution are then analyzed, including overall
wage inequality, returns to education, and gender wage gaps. Albrecht et
al. estimate gender wage gaps in the Netherlands having corrected for
selection of women into market work according to the Buchinsky’s (1998)
semiparametric method for quantile regressions, and they find evidence
of strong positive selection into full-time employment. Finally, Beblo et
al. show selection-corrected wage gaps for Germany using both the Heck-
man (1979) and the Lewbel (2007) two-stage selection models. They find
that correction for selection has an ambiguous impact on gender wage
gaps in Germany, depending on the method used.
Interestingly, most studies find that correction for selection has im-
portant consequences for our assessment of gender wage gaps. At the
same time, none of these studies use data for southern European countries,
where employment rates of women are lowest and thus the selection issue
should be most relevant. In this article we use data for the United States
and for a representative group of European countries to investigate how
nonrandom selection into work may affect international comparisons of
gender wage gaps.

III. Data
A. The PSID
Our analysis for the United States is based on the Michigan Panel Study
of Income Dynamics (PSID). This is a longitudinal survey of a represen-
tative sample of U.S. individuals and their households. It has been ongoing
since 1968. The data were collected annually through 1997 and every
other year after 1997. In order to ensure consistency with European data,
we use six waves from the PSID, from 1994 to 2001. We restrict our
analysis to individuals aged 25–54, having excluded the self-employed,
full-time students, and individuals in the armed forces.5
5
The exclusion of self-employed individuals may require some justification in so
far as the incidence of self-employment varies importantly across genders and coun-
tries, as well as the associated earnings gap. However, the available definition of income
for the self-employed is not comparable to the one we are using for the employees,
and the number of observations for the self-employed is very limited for European
countries. Both these factors prevent us from including the self-employed in our
analysis.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
628 Olivetti/Petrongolo

The wage concept that we use throughout the analysis is the gross
hourly wage. This is given by annual labor income divided by annual
hours worked in the calendar year before the interview date. Employed
workers are defined as those with positive hours worked in the previous
year.
The characteristics that we exploit for wage imputation for the nonem-
ployed are human capital variables, spouse income, and nonemployment
status, that is, unemployed versus out of the labor force. Human capital
is proxied by education and work experience controls. Ethnic origin is
not included here as information on ethnicity is not available for the
European sample. We consider three broad educational categories: less
than high school, high school completed, and college completed. They
include individuals who have completed less than 12 years of schooling,
between 12 and 15 years of schooling, and at least 16 years of schooling,
respectively. This categorization of the years of schooling variable is cho-
sen for consistency with the definition of education in the ECHPS, which
does not provide information on completed years of schooling but only
on recognized qualifications.
Information on work experience refers to years of actual labor market
experience (either full- or part-time) since the age of 18. When individuals
first join the PSID sample as a head or a wife (or cohabitor), they are
asked how many years they worked since age 18 and how many of these
years involved full-time work. These two questions are also asked ret-
rospectively in 1974 and 1985, irrespective of the year in which respon-
dents had joined the sample. The answers to these questions are used to
construct a measure for actual work experience, following the procedure
of Blau and Kahn (2006). Given the initial values reported, we update
work experience information for the years of interest using the longitu-
dinal work history file from the PSID. For example, in order to construct
the years of actual experience in 1994 for an individual who was in the
survey in 1985, we add to the number of years of experience reported in
1985 the number of years between 1985 and 1994 during which they
worked a positive number of hours.6 This procedure allows us to construct
the full work experience in each year until 1997. As the survey became
biannual after 1997, there is no information on the number of hours
worked by individuals between 1997 and 1998 and between 1999 and
2000. We fill missing work experience information for 1998 following
again Blau and Kahn (2006). In particular, we use the 1999 sample to
estimate logit models for positive hours in the previous year and in the
year preceding the 1997 survey, separately for males and females. The

6
The measure of actual experience used here includes both full-time and part-time
work experience, as this is better comparable to the measure of experience available
from the ECHPS.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 629

explanatory variables are race, schooling, experience, a marital status in-

dicator, and variables for the number of children aged 0–2, 3–5, 6–10, and
11–15 who are living in the household at the time of the interview. Work
experience in the missing year is obtained as the average of the predicted
values in the 1999 logit and the 1997 logit. We repeat the same steps for
filling missing work experience information in 2000.
Spouse income is constructed as the sum of total labor and business
income in unincorporated enterprises both for spouses and cohabitors of
respondents. Finally, the reason for nonemployment, that is, unemploy-
ment versus inactivity, is obtained from self-reported information on em-
ployment status.

B. The ECHPS
Data for European countries are drawn from the European Community
Household Panel Survey (ECHPS).7 This is an unbalanced household-
based panel survey, containing annual information on a few thousand
households per country during the period 1994–2001.8 The ECHPS has
the advantage that it asks a consistent set of questions across the 15
member states of the preenlargement European Union. The employment
section of the survey contains information on the jobs held by members
of selected households, including wages and hours of work. The house-
hold section allows us to obtain information on the family composition
of respondents. We exclude Sweden and Luxembourg from our country
set as wage information is unavailable for Sweden in all waves and un-
available for Luxembourg after 1996.
As for the United States, we restrict our analysis of wages to individuals
aged 25–54 as of the survey date, and we exclude the self-employed, those
in full-time education, and those in the military. The definitions of var-
iables used replicates quite closely those used for the United States.
Hourly wages are computed as gross weekly wages divided by weekly
usual working hours. The education categories used are less than upper
secondary high school, upper secondary school completed, and higher
education. These correspond to ISCED 0–2, 3, and 5–7, respectively.
Unfortunately, no information on actual experience is available in the

7
Previous work using ECHPS data for international comparisons of gender gaps
include the OECD (2002) survey and Arulampalam, Booth, and Bryan (2007), who
study the variation in gender pay gaps across quantiles of the wage distribution in 11
EU countries.
8
The initial sample sizes are as follows: Austria: 3,380; Belgium: 3,490; Denmark:
3,482; Finland: 4,139; France: 7,344; Germany: 11,175; Greece: 5,523; Ireland: 4,048;
Italy: 7,115; Luxembourg: 1,011; Netherlands: 5,187; Portugal: 4,881; Spain: 7,206;
Sweden: 5,891; United Kingdom: 10,905. These figures are the number of households
included in the first wave for each country, which corresponds to 1995 for Austria,
1996 for Finland, 1997 for Sweden, and 1994 for all other countries.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
630 Olivetti/Petrongolo

ECHPS, and we use a measure of potential work experience, obtained as

the current age of respondents minus the age at which they started their
working life. Spouse income is computed as the sum of labor and nonlabor
annual income for spouses or cohabitors of respondents. Finally, unem-
ployment status is determined using self-reported information on the main
activity status. (Descriptive statistics for both the U.S. and the EU samples
are reported in table A1 in the online appendix.)

C. Descriptive Evidence on Gender Gaps

Figure 1 plots raw gender gaps in log gross hourly wages and em-
ployment rates for all countries in our sample. All estimates refer to 1999,
which will be the base year in our analysis. At the risk of some oversim-
plification, one can classify countries into three broad categories according
to their levels of gender wage gaps. In the United States and the United
Kingdom, men’s hourly wages are between 27 and 33 log points higher
than women’s hourly wages. Next, in northern and central Europe, the
gender wage gap in hourly wages is between 11 and 25 log points, from
a minimum of 11 log points in Belgium to a maximum of 25 log points
in the Netherlands. Finally, in southern European countries, the gender
wage gap is on average below 9 log points, from 5 in Italy to 11 in Spain.
Such gaps in hourly wages display a roughly negative correlation with
gaps in employment to population ratios. Employment gaps range from
less than 13 percentage points, in the United States, the United Kingdom,
and Scandinavia,9 to 17–27 points in northern and central Europe, to 34–49
percentage points in southern Europe. The coefficient of correlation be-
tween the two series is ⫺0.474 and is significant at the 10% level.
Such negative correlation between wage and employment gaps may
reveal significant sample selection effects in observed wage distributions.
If the probability of an individual being at work is positively affected by
the level of her potential wage offers, and this mechanism is stronger for
women than for men, then high gender employment gaps become con-
sistent with relatively low gender wage gaps simply because low-wage
women are relatively less likely than men to feature in observed wage
distributions.
A simple and intuitive way to illustrate the role of sample selection
consists of making alternative conjectures about the potential wages of
the nonemployed, as a fraction of observed wages for the employed, as
suggested by Smith and Welch (1986, 123). For this purpose we divide
the population into three education groups: low, middle, and high, as
defined in Section III. True wages for each gender g (p male, female) can
be expressed as Wg p 冘 j djgWjg, where djg is the population share of edu-

9
Similar to other Scandinavian countries, the employment gap in Sweden over the
same sample period is 5.2 percentage points.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 631

cation group j for gender g and Wjg is the associated true wage. In turn,
Wjg is a weighted average of actual wages for the employed and potential
wages for the nonemployed. Assuming that the nonemployed would earn
a wage that is equal to a proportion g of the wage of the employed,
Wjg can then be expressed as
˜ [g ⫹ n (1 ⫺ g)],
Wjg p W (1)
jg jg

where njg is the employment rate of education group j and gender g and
W̃jg is their observed average wage. The reason for first computing (1) by
education and then aggregating over education groups is that gender em-
ployment gaps vary widely by education. Specifically, they everywhere
decline with educational levels, but if anything they do so more strongly
in southern Europe than elsewhere (see Olivetti and Petrongolo 2006,
table 1A).
The parameter g represents the type and extent of sample selection into
employment. In particular, values of g ! 1 (respectively 1 1) indicate pos-
itive (respectively negative) sample selection. For a given g, the role of
selection is magnified by a lower employment rate, njg . Denoting by w
the log of potential wages, the gender wage gap for education group j is
wjmale ⫺ wjfemale. This decreases with g if women have lower employment
rates than men, and it increases with the gender employment gap if there
is positive sample selection (g ! 1).
We can now assess the difference between observed and potential wage
gaps across alternative values of g after aggregating (1) across education
groups. This is shown in table 1 for g p 0.7, 0.5, and 0.3. Column 1
reports for reference the mean wage gap on the 1999 employed sample,
as also pictured in figure 1, together with its correlation with the em-
ployment gap and its coefficient of variation.10 Columns 2–4 report the
mean wage gap, having corrected for sample selection using (1). Gender
wage gaps increase everywhere with lower values of g, and, as expected,
they do more so in countries with high gender employment gaps. In other
words, the higher the gender employment gap, the stronger the impact
of a certain degree of positive sample selection. Selection correction gets
rid of the negative correlation between gender wage and employment gaps
and reduces the coefficient of variation in wage gaps. It is interesting to
10
The coefficient of correlation is better suited here to assess cross-country variation
than the simple standard deviation as the level of the wage gap is also systematically
affected by wage imputation. In col. 1, following Krueger and Summers (1988), we
adjust the standard deviation of estimated gender gaps across countries to account for
the upward bias induced by the least-squares sampling error, i.e., SD p (Var (bˆ c) ⫺
冘14cp1jˆ c2/14)1/2, where bˆ c is the estimated wage gap in country c, jˆ c is the corresponding
standard error, and 14 is the number of countries. To obtain the coefficient of variation,
we divide SD by the cross-country mean of the estimated b̂c’s. The same adjustment
applies to all coefficients of variation reported in tables 2–4.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
632 Olivetti/Petrongolo

Table 1
Mean Wage Gaps under Alternative Values of g
Base Sample g p .7 g p .5 g p .3

Country:
United States .325 .409 .434 .460
United Kingdom .269 .309 .336 .365
Finland .181 .239 .260 .283
Denmark .128 .167 .178 .189
Germany .226 .295 .333 .373
Netherlands .248 .334 .385 .440
Belgium .113 .202 .246 .292
Austria .233 .306 .359 .416
Ireland .178 .311 .367 .430
France .124 .207 .260 .318
Italy .052 .223 .311 .414
Spain .109 .296 .387 .494
Portugal .097 .218 .264 .313
Greece .089 .353 .471 .612
Correlation ⫺.474 .302 .616 .806
Coefficient of variation .462 .247 .245 .273
Sources.—Michigan Panel Study of Income Dynamics and European Community Household Panel
Survey.
Note.—The symbol g represents the ratio between the potential wages of the nonemployed and the
observed wages of the employed. Figures reported in rows 1–14 are gender differences in mean log wages.
Wages for each gender are obtained as weighted averages across three education groups. Wages for each
education group depend on g, as illustrated in eq. (1). Figures in the last two rows provide the cross-
country correlation between gender and employment gaps and the coefficient of variation for the gender
wage gap, respectively. Sample description: aged 25–54, excluding the self-employed, the military, and
those in full-time education, 1999.

note that such correlation becomes positive because selection correction

raises the resulting wage gap disproportionately more in countries with
very high employment gaps, most notably southern Europe.
Of course, values of g used here for the relative wages of the nonem-
ployed are hypothetical, and thus they only illustrate the mapping be-
tween the extent of sample selection and wage gaps. The rest of this article
seeks to retrieve evidence on the wages of the nonemployed. As will
become clear in the next section, the identifying assumptions needed to
do this are much weaker when one estimates median, rather than mean,
wage gaps. The focus in the rest of this article will thus be on median
gender pay gaps.

IV. Methodology
Let w denote the natural logarithm of hourly wages and F(wFg) the
cumulative log wage distribution for each gender, where g p 1 denotes
males and g p 0 denotes females. In what follows, our variable of interest
is the difference between (log) male and female median wages:

D p m(wFg p 1) ⫺ m(wFg p 0), (2)

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 633

where m() is the median function. The (log) wage distribution for each
gender is defined by
F(wFg) p F(wFg, I p 1) Pr (I p 1Fg)
⫹ F(wFg, I p 0)[1 ⫺ Pr (I p 1Fg)], (3)
where I p 1 for the employed and I p 0 for the nonemployed.
Estimated moments of the observed wage distribution are based on the
F(wFg, I p 1) term alone. If there are systematic differences between
F(wFg, I p 1) and F(wFg, I p 0), cross-country variation in Pr (I p 1Fg)
may translate into misleading inferences concerning the international var-
iation in the distribution of potential wage offers. This problem typically
affects estimates of female wage offer distributions; this is even more
the case when one is interested in cross-country comparisons of gender
wage gaps, given the cross-country variation in Pr (I p 1Fg p male) ⫺
Pr (I p 1Fg p female), measuring the gender employment gap. But
F(wFg), the term of interest, is not identified, because data provide in-
formation on F(wFg, I p 1) and Pr (I p 1Fg) but clearly not on
F(wFg, I p 0), as wages are only observed for those who are in work.
In particular, using (3), the median log wage for each gender, m, is
defined by
1
F(mFg, I p 1) Pr (I p 1Fg) ⫹ F(mFg, I p 0)[1 ⫺ Pr (I p 1Fg)] p . (4)
2
Our goal is to retrieve gender gaps in median (potential) wages, as illus-
trated in equation (2), with gender medians defined in equation (4). To
do this we need to retrieve information on F(mFg, I p 0), representing
the probability that nonemployed individuals have potential wages below
the median.
It can be shown that knowledge of F(mFg, I p 0) allows us to identify
the median wage gap in potential wages using median wage regressions,
as a simpler alternative to numerically solving (4). Let us consider the
linear wage equation
wi p b0 ⫹ b1 gi ⫹ ␧i , (5)
where wi denotes (log) potential wages, b0 is a constant term, b1 is the
parameter of interest, and ␧i is an error term such that m(␧iFgi ) p 0. De-
note by b̂ the hypothetical least absolute deviations (LAD) regression
estimator based on potential wages, that is, b̂ { arg min b 冘Nip1 Fwi ⫺
b0 ⫺ b1 giF, where b { [b0 b1 ]
.
However, wages wi are only observed for the employed; they are missing
for the nonemployed. Consider an example in which missing wages fall
completely below the median regression line, that is, wi ! wˆ i p bˆ 0 ⫹ bˆ 1 gi
for the nonemployed (Ii p 0), or equivalently, F(mFg, I p 0) p 1 . One can

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
634 Olivetti/Petrongolo

then define a transformed dependent variable yi that is equal to wi for

Ii p 1 and to some arbitrarily low imputed value w ˆ i)
— (such that —
wi ! w
for Ii p 0, and the following result holds (see Bloomfield and Steiger
[1983], sec. 2.3, for detail and formal proof):

b̂imputed p arg min

b
冘
N

ip1
Fyi ⫺ b0 ⫺ b1 giF

冘
N

p bˆ { arg min Fwi ⫺ b0 ⫺ b1 giF. (6)

b ip1

Condition (6) states that the LAD estimator is not affected by impu-
tation. In other words, obtaining b̂ using the transformed dependent var-
iable yi gives the same estimate that one would obtain if potential wages
were available for the whole population. Now consider an alternative
example in which missing wages fall completely above the median re-
gression line, that is, wi 1 w ˆ i for Ii p 0, or equivalently, F(mFg, I p
0) p 0. The result in (6) still holds, having set yi equal to some arbitrarily
high imputed value w ¯ (such that w ¯i1w ˆ i ) for the nonemployed. More in
general, the LAD estimator is not affected by imputation when the missing
wage observations are imputed “so as to maintain the same sign of the
residual” (Bloomfield and Steiger 1983, 52). That is, (6) is valid whenever
missing wage observations are imputed on the “correct” side of the median.
As a further example, suppose that the potential wages of the nonemployed
could be classified into two groups, L and U, such that wi ! w ˆ i for i 苸 L
and wi 1 w ˆ i for i 苸 U. One can define yi as a transformed variable such
that yi p wi for Ii p 1, yi p w — for Ii p 0 and i 苸 L, and yi p w ¯ for
Ii p 0 and i 苸 U, and LAD inference is still valid.
Using this result, one can estimate median wage gaps, based on wage
imputation for the nonemployed that simply requires assumptions on the
position of the imputed wage observations with respect to the median of
the wage distribution, as done in Johnson et al. (2000) and Neal (2004).
The attractive feature of median regressions is that results are only affected
by the position of imputed wage observations with respect to the median
and not by specific values of imputed wages as it would be in the matching
approach. In this article, we will estimate median wage gaps under alter-
native imputation rules, that is, under alternative conjectures over
F(mFg, I p 0). These imputation rules are described in detail below.

A. Imputation on Wages from Other Waves

We first exploit the panel nature of our data sets and, for all those not
in work in some base year t, we recover (the real value of) hourly wage
observations from the nearest wave in the sample, t
, and we use them as
imputed wages (yi) for estimating (6). The underlying identifying as-

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 635

sumption is that, for a given individual i, the latent wage position with
respect to her predicted (gender-specific) median when she is nonem-
ployed can be proxied by her wage in the nearest wave in which she is
employed. As the position with respect to the median is determined using
alternative information on wages, as opposed to measured characteristics,
we are allowing for selection on unobservables.
Formally, we will assume
F(mFgi , Iit p 0) p F(mFgi , Iit
p 1), (7)
where t is our base year and t
is the wave nearest to t in which we have
a nonmissing wage observation. In practice, we impute yit p wit
for
Iit p 0. This procedure of imputation makes sense if an individual’s po-
sition in the wage distribution stays on the same side of the median when
the individual is switching employment status. As we estimate median
wage gaps, we do not need an assumption of stable rank throughout the
whole wage distribution but only with respect to the median. Should the
position of individuals in the wage distribution change with employment
status, movements that happen within either side of the median do not
invalidate this method.
While imputation based on this procedure arguably uses the minimum
set of potentially arbitrary assumptions, it has the disadvantage of not
providing any wage information on individuals who never worked during
the sample period. It is therefore important to understand in which di-
rection this problem may distort, if at all, the resulting median wage gaps.
If women are on average less attached to the labor market than men, and
if attachment increases with potential wages, then the difference between
the median gender wage gap on the imputed and the actual wage distri-
bution tends to be higher the higher the proportion of imputed wage
observations in total nonemployment in the base year. Consider, for ex-
ample, a country with a very persistent female employment status: women
who do not work in the base year and are therefore less attached are also
unlikely to work at all in the whole sample period. In this case, low wage
observations for less-attached women are unlikely to be recovered and
the estimated wage gap is relatively low. Proportions of imputed wage
observations over the total nonemployed population in 1999 (our base
year) are reported in table A2 of the online appendix: the differential
between male and female proportions tends to be higher in Germany,
Austria, France, and southern Europe than elsewhere. Under reasonable
assumptions we should therefore expect the difference between the median
wage gap on the imputed and the actual wage distribution to be biased
downward relatively more in this set of countries. This, in turn, means
that we are being relatively more conservative in assessing the effect of
nonrandom employment selection in these countries than elsewhere.
Even so, it would, of course, be preferable to recover wage observations

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
636 Olivetti/Petrongolo

also for those never observed in work during the whole sample period.
To do this, we rely on the observed characteristics of the nonemployed.

B. Imputation on Observables
We use observable characteristics for wage imputation with two meth-
ods. With the first method, we make assumptions on the position of
missing wages with respect to their gender-specific median, based on a
small number of characteristics, summarized into the vector Xi. We can
illustrate this with a very simple example. Suppose that Xi only includes
years of completed education. This implies that we are using information
on education for someone who is nonemployed to place them above or
below their gender-specific median. We can define a threshold for Xi, x—
(say, 11 years of schooling), below which nonemployed individuals would
earn below-median wages, and another threshold x̄ (say, 16 years), above
which individuals would earn above-median wages.
More formally we assume that
F(mFgi , Ii p 0, Xi ≤ x)
— p 1; F(mFgi , Ii p 0, Xi ≥ x)
¯ p 0, (8)

where —x and x¯ are low and high values of Xi, respectively.11 In this case,
the imputed dependent variable yi is set equal to w— for i such that Ii p
0 and Xi ≤ x— and is set equal to w¯ for i such that Ii p 0 and Xi ≥ x¯ .
This method for placing individuals with respect to the median follows
an educated guess, based on their observable characteristics. However, we
can use wage information from other waves in the panel to assess the
goodness of such guess, as will be illustrated in Section V.B.
With the second method, we use probability models for imputing miss-
ing wage observations. In this case our imputation rule assumes that

F(mFgi , Ii p 0, Xi ) p Pˆ i , (9)
where P̂i is the predicted probability to belong below the median, based
on probit estimates.
We implement this imputation method in two steps. In the first step,
we estimate the probability of an individual’s wage belonging below the
median of the wage distribution, based on a set of observable character-
istics. On the employed sample, we define Mi p 1 for individuals earning
less than their gender-specific median and Mi p 0 for the others. We
estimate a probit model for Mi for each gender, with explanatory variables
Xi. Using the probit estimates, we obtain predicted probabilities of having
a latent wage below the median, Pˆ i p F(gXˆ i ) p Pr (Mi p 1FXi), for the
nonemployed subset, where F is the cumulative distribution function of
the standardized normal distribution and ĝ is the estimated parameter
11
All variables in (8) refer to the (same) base year, so time subscripts are omitted.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 637

vector from the probit regression. Predicted probabilities P̂i are then used
in the second step as sampling weights for the nonemployed. That is, we
construct an imputed sample in which the employed feature with their
observed wage and the nonemployed feature with a wage below median
with a weight Pˆ i and a wage above median with a weight 1 ⫺ Pˆ i. The
statistics of interest is the gender wage gap estimated on the imputed
sample. The associated variance is obtained by bootstrap to correct for
the fact that the weights used are based on probit estimates.
Note that in the first step we need a reference median wage in order
to define Mi. The readily available candidate would be the median ob-
served wage, but precisely due to selection this may be quite different
from the latent median wage, thus potentially delivering biased estimates.
In order to attenuate this problem, we also perform repeated imputation
on an expanded sample, augmented with wage observations from adjacent
waves. This allows us to get a better estimate of the potential median in
the first step of our procedure, thereby generating more appropriate es-
timates of the median wage gap on the final, imputed sample.

C. Discussion on Imputation Methodology

To ease the interpretation of the results presented in the next section,
we discuss here the main differences between alternative imputation meth-
ods. The three methods described differ in terms of the underlying iden-
tifying assumptions and the resulting imputed samples. The first method,
where missing wages are imputed using wage information from other
waves, implicitly assumes that an individual’s position with respect to the
median can be proxied by his or her wage in the nearest wave in the
panel. With this procedure, one can recover, at best, individuals who
worked at least once during the 8-year sample period. We thus emphasize
that this is a fairly conservative imputation procedure, one in which we
impute wages for individuals who are relatively weakly attached to the
labor market but not for those who are completely unattached and thus
have never been observed in work. This procedure has the advantage of
restricting imputation to a relatively “realistic” set of potential workers,
and thus it is the one we mostly rely upon to make quantitative statements.
In the second and third imputation methods, we assume instead that
an individual’s position with respect to the median can be proxied by
some of his or her observable characteristics. In the second method, we
use characteristics to take educated guesses regarding the position of miss-
ing wages. Clearly this procedure is more accurate for values of the ob-
servables in the tails than in the middle of the distribution. For example,
guessing the position with respect to the median for individuals with either
college or no education at all is safer than doing it for secondary school
graduates, who are thus best left without an imputed wage. In doing this,

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
638 Olivetti/Petrongolo

our imputed sample is typically larger than the one obtained with the
first method, although it is still substantially smaller than the existing
population. Finally, with the third method, we estimate the probability
of belonging above the median for the whole range of our vector of
characteristics, thus recovering predicted probabilities and imputed wages
for the whole existing population.
Different imputed samples will have an impact on our estimated median
wage gaps. In so far as women tend to be more positively selected into
employment than men, the larger the imputed sample with respect to the
actual sample of employed workers, the larger the estimated correction
for selection.
Having said this, it is important to stress that, with all three imputation
methods used, we never impose positive selection ex ante (except in a
benchmark example), and thus there is nothing that would tell a priori
which way correction for selection is going to affect the results. This is
ultimately determined by the wages that the nonemployed earned when
they were previously (or later) employed and by their observable char-
acteristics, depending on methods.
Before moving on to the discussion of our estimates, it is worthwhile
to motivate our choice of selection correction methodology and to frame
it in the context of the existing literature on sample selection. A number
of approaches can be used to correct for nonrandom sample selection in
wage equations and/or recover the distribution in potential wages. The
seminal approach suggested by Heckman (1974, 1979) consists in allowing
for selection on unobservables, that is, on variables that do not feature
in the wage equation but that are observed in the data.12 Heckman’s two-
stage parametric specifications have been used extensively in the literature
in order to correct for selectivity bias in female wage equations. More
recently, these have been criticized for lack of robustness and distribu-
tional assumptions (see Manski 1989). Approaches that circumvent most
of the criticism include semiparametric selection correction models that
have appeared in the literature since the early 1980s (see Vella [1998] for
an extensive survey of both parametric and nonparametric sample selec-
tion models). Two-stage nonparametric methods allow us in principle to
approximate the bias term by a series expansion of propensity scores from
12
In this framework, wages of employed and nonemployed would be recovered
as
E(wFZw, I p 1) p Zwdw ⫹ E(␧wF␧I 1 ⫺ZIdI)
E(wFZw, I p 0) p Zwdw ⫹ E(␧wF␧I ! ⫺ZIdI),
respectively, where Zw and ZI are the set of covariates included in the wage and
selection equations, respectively, with associated parameters dw and dI , and ␧w and
␧I are the respective error terms.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 639

the selection equation, with the qualification that the term of order zero
in the polynomial is not separately identified from the constant term in
the wage equation unless some additional information is available (see
Buchinsky 1998). Usually the constant term in the wage regression is
identified from a subset of workers for which the probability of work is
close to one, but in our case this route is not feasible since for no type
of women is the probability of working close to one in all countries.
Selection on observed characteristics is instead exploited in the matching
approach, which consists in imputing wages for the nonemployed by
assigning them the observed wages of the employed with matching char-
acteristics (see Blau and Beller 1992; and Juhn 1992, 2003). The approach
of this article is also based on some form of wage imputation for the
nonemployed, but it simply requires assumptions on the position of the
imputed wage observations with respect to the median of the wage dis-
tribution. Importantly, it does not require assumptions on the actual level
of missing wages, as is typically required in the matching approach, nor
it requires the arbitrary exclusion restrictions that are often invoked in
the two-stage Heckman sample selection correction models.

D. Bounds
As discussed above, each imputation method is based on identifying
assumptions that are largely untested. In order to illustrate that results
delivered by our imputation methods are reasonable, we also provide
“worst case” bounds to the gap in potential median wages that do not
require any identifying assumption, as shown by Manski (1994) and Blun-
dell et al. (2007). We will then check that our estimated wage gaps on
imputed wage distributions fall into these bounds.
Manski notes that substituting the inequality 0 ≤ F(wFg, I p 0) p 1
into (3) gives bounds for the true cumulative distribution

F(wFg, I p 1) Pr (I p 1Fg) ≤ F(wFg)

≤ F(wFg, I p 1) Pr (I p 1Fg) ⫹ [1 ⫺ Pr (I p 1Fg)]. (10)

If one is interested in the median of F(), denoted by m, (10) implies

that

1
F(mFg, I p 1) Pr (I p 1Fg) ≤
2
≤ F(mFg, I p 1) Pr (I p 1Fg) ⫹ [1 ⫺ Pr (I p 1Fg)]. (11)

These bounds on F() deliver the following worst case bounds on the
gender-specific median

m l(wFg) ≤ m(wFg) ≤ mu (wFg) (12)

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
640 Olivetti/Petrongolo

such that
1
p F(m lFg, I p 1) Pr (I p 1Fg) ⫹ [1 ⫺ Pr (I p 1Fg)], (13)
2

1
p F(muFg, I p 1) Pr (I p 1Fg). (14)
2
Bounds on the gender-specific median can be obtained by solving (13)
and (14) using data on the observed wage distribution and employment
rates. Note that conditions (13) and (14) imply that one can only identify
bounds for the median if Pr (I p 1Fg) p 1/2. Hence, we will not be able
to obtain such bounds for the female median wage (and, therefore, for
the gender wage gap) in countries where less than 50% of the women
have a wage observation.
Having said this, the bounds for the median gender wage gap D defined
in (2) are obtained as follows:
m l(wFg p male) ⫺ mu (wFg p female)
≤ D ≤ mu (wFg p male) ⫺ m l(wFg p female). (15)

V. Results
A. Imputation on Wages from Adjacent Waves
In our first set of estimates, an individual’s position with respect to the
median of the wage distribution in the base year is proxied by the position
of his or her wage obtained from the nearest available wave. The kind of
imputation made here requires that individuals stay on the same side of
their gender median across different waves in the panel (see eq. [7]).
Results obtained with this method are reported in table 2.
Column 1 reports the actual wage gap for reference: this is the median
wage gap for individuals with an hourly wage in 1999, which is our base
year. The wage gaps of column 1 replicate very closely those plotted in
figure 1, with the only difference being that figure 1 plotted mean as opposed
to median wage gaps.13 As in figure 1, the United States and the United
Kingdom stand out as the countries with the highest wage gaps, followed
by central Europe, and finally by Scandinavia and Southern Europe.
In column 2, missing wage observations in 1999 are replaced with the
real value of the nearest wage observation in a 2-year window, while in
column 3, they are replaced with the real value of the nearest wage obser-
13
The absence of any important difference between mean and median wage
gaps on the observed wage distribution is good news for our approach, based on
the recovery of selection-corrected median wage gaps.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 641

Table 2
Median Wage Gaps under Alternative Sample Inclusion Rules:
Wage Imputation Based on Wage Observations from Adjacent Waves
1 2 3 4 5

Country:
United States .339 .352 .361 .354 .363
United Kingdom .256 .277 .284 .291 .301
Finland .160 .194 .197 .203 .206
Denmark .086 .100 .100 .093 .093
Germany .191 .223 .214 .234 .234
Netherlands .178 .193 .199 .195 .199
Belgium .078 .099 .112 .098 .112
Austria .192 .224 .234 .220 .229
Ireland .232 .273 .284 .292 .300
France .095 .133 .152 .140 .164
Italy .059 .062 .075 .070 .079
Spain .097 .153 .168 .143 .157
Portugal .150 .168 .186 .169 .185
Greece .111 .148 .184 .148 .185
Correlation ⫺.329 ⫺.263 ⫺.181 ⫺.269 ⫺.199
Coefficient of variation .484 .416 .382 .427 .392
Sources.—Michigan Panel Study of Income Dynamics and European Community Household Panel
Survey.
Note.—All wage gaps are significant at the 1% level. Figures in the last two rows display the cross-
country correlation between the reported gender wage gap and the gender employment gap and the
coefficient of variation of the gender wage gap. Sample description: aged 25–54, excluding the self-
employed, the military, and those in full-time education, 1999. Sample inclusion rules by columns: (1)
Employed at time of survey in 1999. (2) Wage imputed from other waves when nonemployed (⫺2 , ⫹
2 window). (3) Wage imputed from other waves when nonemployed (⫺5 , ⫹2 window). (4) Wage imputed
from other waves when nonemployed (⫺5 , ⫹2 window), adjusted for real wage growth by gender and
country. (5) Wage imputed from other waves when nonemployed (⫺5 , ⫹2 window), adjusted for real
wage growth by gender and country.

vation in the whole sample period, meaning a maximum window of [⫺5,

⫹2] years. Moving across from column 1 to column 3, gender wage gaps
tend to increase as more wage observations are included into the imputed
sample. This is indicative of positive sample selection, or, in other words,
estimated wage gaps on the observed wage distribution are downward bi-
ased due to nonrandom sample selection into employment because low-
wage women are less likely to feature in the observed wage distribution.
But there is important cross-country variation in the role of selection.
In particular, one can see that the median wage gap remains substantially
unaffected or marginally affected in the United States, the United King-
dom, Scandinavia, Germany, and the Netherlands; it increases by around
25% in Austria, Ireland, Italy, and Portugal; by 40% in Belgium; and by
more than 60% in France, Spain, and Greece. As expected, gender wage
gaps tend to respond more strongly to selection correction in countries
with high employment gaps. This can be clearly grasped by looking at
the cross-country correlation between employment and wage gaps. In
column 1, such correlation is ⫺0.329, and it falls by 45% in column 3.
Employment selection thus explains nearly a half of the observed cor-
relation between wage and employment gaps. Another indicative cross-

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
642 Olivetti/Petrongolo

country statistic is the coefficient of variation in the gender wage gap:

this falls by 31% between column 1 and column 3; thus selection explains
almost one-third of the observed cross-country variation of wage gaps.
For each sample inclusion rule in columns 1–3, one can compute the
adjusted employment rate for each gender, that is, the proportion of the
adult population that is either working or has an imputed wage. These
proportions are reported later in columns 1–3 of table 5. When moving
from column 1 to column 3, the fraction of women included increases
substantially in most countries, including some countries where the es-
timated wage gap is not greatly affected by the sample inclusion rules.
Moreover, the fraction of men included in the sample also increases, albeit
less so than for women. It is thus not simply the lower female employment
rate in several countries that drives our findings but also the fact that in
some countries selection into work seems to be less correlated to wage
characteristics than in others.
The estimates of columns 2 and 3 of table 2 do not control for aggregate
wage growth over time. If aggregate wage growth were homogeneous
across genders and countries, the estimated wage gaps based on wage
observations for other waves in the panel would not be affected. But, if
there is a gender differential in wage growth, and if such differential varies
across countries, then simply using earlier (later) wage observations would
deliver a higher (lower) median wage gap in countries where wage growth
for women is lower than for men.14 We thus estimate real wage growth
by regressing log real hourly wages for each country and gender on a
linear trend.15 The resulting coefficients are reported in table A3 in the
online appendix. These are then used to adjust real wage observations
outside the base year and to re-estimate median wage gaps. The resulting
median wage gaps on the imputed wage distribution are reported in col-
umns 4 and 5 of table 2. Despite some differences in real wage growth
rates across genders and countries, adjusting estimated median wage gaps
does not produce any appreciable change in the results reported in col-
umns 2 and 3, which do not control for real wage growth.
Note finally that, in table 2, we are only recovering wages for a quarter,
on average, of nonemployed women in the four southern European coun-
tries, as opposed to more than a half in the rest of countries (see table
A2 in the online appendix). For men, cross-country differences are less
marked, as respective proportions are 57% and 63%. Such differences

14
Note, however, that, even if real wage growth were homogeneous across
genders, imputation based on wage observations from adjacent waves would not
be affected only if the proportion of men and women in the sample remained
unchanged after imputation.
15
Of course, for our estimated rates of wage growth to be unbiased, this pro-
cedure requires that participation into employment be unaffected by wage growth,
which may not be the case.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 643

happen because (non)employment status tends to be relatively more per-

sistent in southern Europe than elsewhere, and this is much more so for
women than for men. As noted in Section IV, given that we recover
relatively fewer less-attached women in southern Europe, we are being
relatively more conservative in assessing the effect of nonrandom em-
ployment selection in southern Europe as compared to elsewhere. For
this reason it is important to try to recover wage observations also for
those never observed in work in any wave of the sample period, as ex-
plained in the next subsection.

B. Imputation on Observables
In table 3, we exploit some observable characteristics of the nonem-
ployed for assigning them on one or the other side of their gender median
(eq. [8] gives the formal identifying assumption). Column 1 reports for
reference the median wage gap on the base sample, which is the same as
the one reported in column 1 of table 2. In column 2, we assume that all
those not in work in 1999 would have wage offers below the median for
their gender.16 This is an extreme assumption, and it is the only case in
which we impose ex ante positive sample selection. This assumption is
clearly violated for countries like Italy, Spain, and Greece, in which more
than half of the female sample is not in work in 1999, and thus estimates
are not reported for these three countries. However, also for other coun-
tries, there are reasons to believe that not all nonemployed individuals
would have wage offers below their gender mean. Having said this, es-
timated median wage gaps increase substantially for most countries, except
for Denmark and Finland. The correlation with employment gaps turns
positive and quite strong because the wage gap in high employment-gap
countries increases disproportionately relative to other countries.
Of course, one cannot know exactly what wages these individuals
would have received had they worked in 1999. But we can form an idea
of the goodness of this assumption by looking again at wage observations
(if any) for these individuals in all other waves in the panel. This allows
us to see whether an imputed observation had a wage that was indeed
below their predicted gender median at the time he or she was observed
working. Specifically, we take all imputed observations in 1999. Among
these, we select those who ever worked at some time in the sample period.
Out of this subset, we compute the proportion of observations who had
wages below the predicted gender median. Such proportions are computed
for men and women and are reported in columns headed “M” and “F,”

16
In practice, whenever we assign someone a wage below the median, we pick
wi p ⫺5, this value being lower than the minimum observed (log) wage for all
countries and thus lower than the median. Similarly, whenever we assign someone
a wage above the median, we pick w̄i p 20.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
 Table 3
Median Wage Gaps under Alternative Imputation Rules: Wage Imputation Based on Observables—Educated Guesses
1 2 3 4 5 6
Goodness Goodness Goodness Goodness
Imputation Imputation Imputation Imputation
(%) (%) (%) (%)
Wage Wage Wage Wage Wage Wage
Gap Gap M F Gap M F Gap M F Gap M F Gap

Country:
United States .339 .432 .90 .74 .335 1.00 .89 .348 .67 .78 .348 .82 .88 .367
United Kingdom .256 .375 .80 .63 .238 .83 .81 .246 .71 .56 .253 .87 .89 .287
Finland .160 .191 .86 .73 .179 .87 .86 .161 .40 .43 .160 .80 .83 .188
Denmark .086 .122 .77 .77 .100 .85 .81 .076 .63 .55 .092 1.00 .60 .100
Germany .191 .368 .87 .53 .214 .88 .74 .192 .53 .54 .196 .94 .86 .215
Netherlands .178 .353 .67 .50 .246 .67 .62 .208 .67 .78 .189 .79 .78 .221
Belgium .078 .228 .68 .70 .113 .70 .82 .082 .80 .52 .084 .69 1.00 .116
Austria .192 .360 .89 .60 .210 .91 .72 .192 .00 .43 .208 .88 .80 .234
Ireland .232 .591 .83 .33 .213 .87 .94 .237 .75 .79 .241 .73 .92 .291
France .095 .357 .81 .56 .134 .84 .82 .114 .75 .79 .102 .83 .86 .161
Italy .059 ... .75 ... .092 .78 .69 .200 .93 .74 .088 .83 .86 .186
Spain .097 ... .71 ... .182 .74 .65 .199 .70 .78 .109 .72 .96 .235
Portugal .150 .357 .65 .52 .188 .61 .69 .264 .67 .71 .162 .69 .69 .258
Greece .111 ... .78 ... .175 .81 .73 .349 .67 .70 .160 .73 .60 .362
Correlation ⫺.329 .625 ⫺.120 .435 ⫺.200 .395
Coefficient of variation .484 .364 .329 .393 .430 .336

All use subject to JSTOR Terms and Conditions

Sources.—Michigan Panel Study of Income Dynamics and European Community Household Panel Survey.
Note.—All wage gaps are significant at the 1% level. In specification 2 no results are reported for Italy, Spain, and Greece as more than 50% of women in the sample
are nonemployed. The goodness of imputation is measured using wage observations (if any) from other waves in the panel, and specifically computing what proportion of
these is on the same side of the median that we assumed in imputation. Figures in the last two rows display the cross-country correlation between the reported gender
wage gap and the gender employment gap, and the coefficient of variation of the gender wage gap. Sample: aged 25–54, excluding the self-employed, the military and those
in full-time education, 1999. Sample inclusion rules by columns: (1) Employed at time of survey in 1999. (2) Impute wage ! median when nonemployed. (3) Impute wage

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

! median when unemployed. (4) Impute wage ! median when nonemployed and education ! upper secondary and experience ! 10 years; impute wage 1 median when
nonemployed and education ≥ higher education and experience ≥ 10 years. (5) Impute wage ! median when nonemployed and spouse income in bottom quartile. (6) Wage
imputed from other waves when nonemployed (⫺5, ⫹2 window) and sample inclusion rule 4.
A Cross-Country Analysis of Gender Gaps 645

respectively. They are fairly high for men, but they are sensibly lower for
women, which makes the estimates based on this extreme imputation case
a benchmark rather than a plausible measure for the gender wage gap.
In column 3, we impute a wage below the median to all those who are
unemployed (as opposed to nonparticipants) in 1999. The unemployed,
by definition, are receiving wage offers (if any) below their reservation
wage, while the employed have received at least one wage offer above
their reservation wage. At constant reservation wages, the unemployed
have lower potential wages than the observed wages of the employed and
are thus assigned an imputed wage value below the median. This impu-
tation leaves the median wage gap roughly unchanged with respect to the
base sample in the United States, the United Kingdom, Scandinavia, Ger-
many, Austria, and Ireland and raises it substantially elsewhere, especially
in southern Europe. Also, the proportion of “correctly” imputed obser-
vations, computed as for the previous imputation case, is now much
higher. Those who do not work because they are unemployed are thus
relatively more likely to be overrepresented in the lower half of the wage
distribution. Selection now explains 64% of the correlation between wage
and employment gaps and 32% of the cross-country variation.
In column 4, we follow standard human capital theory and assume that
all those with less than upper secondary education and fewer than 10
years of labor market experience have wage observations below the me-
dian for their gender. Those with at least higher education and at least 10
years of labor market experience are instead placed above the median. In
the four southern European countries, the gender wage gap increases
enormously with respect to the actual wage gap of column 1 and, as a
consequence, the correlation with employment gaps turns positive. It is
interesting that the proportion of correctly imputed observations is high
in Ireland, France, and southern Europe but that is not so much so in
other countries, where imputation based on unemployment works better
than imputation based on human capital components.
The imputation method of column 5 is implicitly based on the as-
sumption of assortative mating along wage attributes and consists of as-
signing wages below the median to those whose partners have total income
in the bottom quartile of the gender-specific distribution. The assumption
is that individuals married to low-productivity spouses also have low
productivity, and thus the spouse’s wage is taken as a proxy for an in-
dividual’s potential wage offer. The results are qualitatively similar to those
of column 3: the wage gap is mostly affected in southern Europe, but on
average it is less affected than in other imputation examples. It would be
natural to perform a similar exercise at the top of the distribution by
assigning a wage above the median to those whose partner earns income
in the top quartile. However, in this case, the proportion of correctly
imputed observations was too low to rely on the assumption used for

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
646 Olivetti/Petrongolo

imputation. We have also considered imputation based on low spouse

education, obtaining very similar results as with low spouse income. Fi-
nally, we considered imputation based on high spouse education, and,
similar to imputation based on high spouse income, the goodness of im-
putation turned out worse.
We also combined imputation methods by using, first, wage observa-
tions available from other waves and then imputing the remaining missing
ones using education and experience information, as done in column 4.
The results, reported in column 6, show again a much higher gender gap
in France and southern Europe and not much of a change elsewhere with
respect to the base sample of column 1.
Similar to the previous imputation method, we report in columns 4–8
of table 5 the proportion of men and women included in our imputed
samples. As expected, we are now able to recover wage information for
a higher fraction of the adult population. In column 4, such proportions
are generally not equal to 100% because we did not impute wages to
those who are employed but have missing information on hourly wages
due to nonresponse, as the selection mechanism driving nonresponse is
clearly different from that driving nonemployment.
Finally, we report estimates based on a probabilistic, two-step impu-
tation technique, summarized in equation (9). In the first step, we use the
1999 base sample to estimate a probit model for the probability of be-
longing above the gender-specific median, controlling for education (up-
per secondary and higher education), experience, and its square.17 The
estimated coefficients for the first-stage probit regression (not reported)
conform to standard economic theory: individuals with higher levels of
educational attainment and/or of labor market experience are more likely
to feature in the top half of the wage distribution. These estimates are
used as sampling weights in the second step to construct an imputed
sample, on which we estimate the median gender wage gap and the cor-
responding bootstrapped standard error (with 200 replications).
The results of this exercise are summarized in table 4. Column 1 reports
the median wage gap for the base sample, which is the same as the one
reported in column 1 of tables 2 and 3. Column 2 reports the estimated
median wage gap obtained from the probabilistic model described, having
used the observed 1999 median as the reference median for our probit
estimates. As fewer than 50% of women are employed in Italy, Spain,
and Greece, we cannot credibly estimate a probit model where Mi p 1
17
We also estimated a more general specification that also controls for marital
status, the number of children of different ages, and the position of the spouse
in their gender-specific distribution of total income. Since the results of the exercise
do not vary in any meaningful way across specifications, we only report findings
for the human capital specification.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 647

Table 4
Median Wage Gaps under Alternative Imputation Rules:
Wage Imputation Based on Observables—Probabilistic Model
Weighted
Base Sample Imputation
(1) (2) (3)

Country:
United States .339 .359 .371
United Kingdom .256 .264 .292
Finland .160 .179 .199
Denmark .086 .100 .100
Germany .191 .200 .232
Netherlands .178 .229 .235
Belgium .078 .117 .154
Austria .192 .205 .236
Ireland .232 .319 .341
France .095 .182 .186
Italy .059 ... .229
Spain .097 ... .333
Portugal .150 .272 .272
Greece .111 ... .593
Correlation ⫺.329 .291 .686
Coefficient of variation .484 .339 .427
Sources.—Michigan Panel Study of Income Dynamics and European Community
Household Panel Survey.
Note.—All wage gaps are significant at the 1% level. In specification 2 no results
are reported for Italy, Spain, and Greece as more than 50% of women in the sample
are nonemployed. Figures in the last two rows display the cross-country correlation
between the reported gender wage gap and the gender employment gap and the co-
efficient of variation of the gender wage gap. Sample description: aged 25–54, excluding
the self-employed, the military, and those in full-time education, 1999. Sample inclusion
rules by columns: (1) Employed at time of survey in 1999. (2) Impute wage ! (resp. 1)
median with probability Pˆ i (resp. 1 ⫺ Pˆ i) if nonemployed. Pˆ i is the predicted probability
of having a wage below the gender-specific base sample median, as estimated from a probit
model including two education dummies, experience, and its square. (3) Impute wage !
(resp. 1) median with probability Pˆ i (resp. 1 ⫺ Pˆ i) if nonemployed. Pˆ i as above, having
enlarged the base sample with wage observation from adjacent waves.

for workers earning less than the median for these countries. In column
3, we use as the reference median the one obtained on a wage distribution
enlarged with wage imputation from all other waves, and in this case the
fraction of missing wages is below 50% for men and women in all coun-
tries. If wage imputation is correct, this procedure delivers a reference
median that is closer to the latent median than the observed median.
Comparing column 1 to columns 2 and 3 shows that the median wage
gap on imputed wage distributions increases mildly in most countries
down to Austria but rises substantially in Ireland, France, and Portugal
and enormously in Italy, Spain, and Greece, which are the countries with
the highest employment gaps.
To broadly summarize our findings (a summary is provided in table
5), one could note that, whether one corrects for selection on unobserv-
ables (table 2) or observables (tables 3 and 4), our results are qualitatively
consistent in identifying a clear role of sample selection in countries with

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
 Table 5
Percentage of Adult Population in Samples for Tables 2–4
No. observa-
tions, 1999 1 2 3 4 5 6 7 8 9
M F M F M F M F M F M F M F M F M F M F

United States 2,835 3,551 95.9 83.3 98.2 91.4 98.5 92.7 100.0 100.0 96.4 84.1 96.9 88.9 96.7 86.3 99.0 95.3 99.7 99.3
United Kingdom 1,998 2,458 87.2 74.3 92.4 84.9 93.3 88.0 96.9 96.8 91.2 76.2 90.5 81.0 89.9 77.1 95.3 91.1 98.8 98.2
Finland 1,464 1,697 92.6 82.5 96.9 92.9 97.2 93.6 99.4 98.8 98.8 91.2 93.2 86.7 93.2 83.3 97.4 95.3 100.0 99.6
Denmark 996 1,054 94.0 88.1 99.5 97.4 99.5 97.7 98.3 98.1 97.5 94.1 95.0 90.5 94.6 88.9 99.7 98.0 99.9 99.6
Germany 2,822 3,064 91.2 68.8 96.7 83.1 98.1 87.0 98.9 93.4 97.6 75.2 92.2 72.0 92.6 70.7 98.4 88.4 99.5 97.0
Netherlands 2,322 2,750 94.1 70.4 96.7 81.5 97.1 84.3 99.8 99.1 96.7 82.0 95.8 77.8 95.6 72.9 98.1 89.1 99.8 99.6
Belgium 1,170 1,382 90.3 70.2 94.1 77.9 94.9 81.5 98.7 98.2 95.8 80.2 91.6 77.1 92.8 73.5 95.7 86.6 99.0 97.5
Austria 1,296 1,378 95.8 69.3 98.5 78.7 98.7 81.6 99.8 97.8 99.2 72.2 95.8 70.9 96.6 71.4 98.7 82.7 99.8 95.8
Ireland 1,053 1,345 84.9 58.1 89.8 70.8 90.7 73.8 99.6 99.0 92.9 61.0 88.3 64.0 88.9 62.2 93.0 78.3 99.8 99.2
France 2,601 2,923 73.1 55.7 92.8 75.6 94.4 80.1 85.5 90.2 79.8 64.6 75.3 63.7 74.8 58.0 95.7 84.9 98.4 97.0
Italy 3,113 3,744 79.3 45.2 89.9 54.9 91.3 57.9 94.9 96.8 91.9 56.1 83.2 64.0 81.5 49.7 93.8 74.9 98.6 96.6
Spain 2,680 3,065 83.7 45.5 91.2 59.4 92.6 62.5 99.7 99.6 93.8 56.1 87.0 60.1 86.5 47.9 94.7 73.8 99.8 99.6
Portugal 2,019 2,251 91.5 67.2 95.1 75.5 95.9 78.1 99.7 99.3 95.0 73.5 94.6 82.0 92.6 70.1 98.4 90.2 99.7 99.1
Greece 1,379 1,938 87.1 38.4 94.5 49.0 95.1 52.1 99.9 99.4 95.9 47.1 89.7 60.6 88.5 42.1 96.4 71.1 100.0 99.3
Sources.—Michigan Panel Study of Income Dynamics and European Community Household Panel Survey.
Note.—Figures in cols. 1–9 represent the proportions of males and females included in the sample across imputation rules of tables 2–4. Sample description: aged 25–54,

All use subject to JSTOR Terms and Conditions

excluding the self-employed, the military, and those in full-time education. Sample inclusion rules by column: (1) Employed at time of survey in 1999. (2) Wage imputed from
other waves when nonemployed (⫺2 , ⫹2 window). (3) Wage imputed from other waves when nonemployed (⫺5 , ⫹2 window). (4) Impute wage ! median when nonemployed.
(5) Impute wage ! median when unemployed. (6) Impute wage ! median when nonemployed and education ! upper secondary and experience ! 10 years; impute wage ! median
when nonemployed and education ≤ higher education and experience ≥ 10 . (7) Impute wage ! median when nonemployed and spouse income in bottom quartile. (8) Sample
inclusion rules 3 and 6 above. (9) Sample inclusion rule 3 above and wage imputed using probabilistic model (see note to table 4).

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

A Cross-Country Analysis of Gender Gaps 649

high employment gaps, especially in France and southern Europe. Quan-

titatively, the correction for sample selection is smallest when wage im-
putation is performed using wage observation from other waves in the
panel, and it increases when it is instead performed using observed char-
acteristics of the nonemployed. As argued above, this is mainly due to
different sizes of the imputed samples. While only individuals with some
degree of labor market attachment feature in the imputed wage distri-
bution in the first case, the use of observed characteristics may in principle
allow wage imputation for the whole population, thus including individ-
uals with no labor market attachment at all. Interestingly, the fact that
controlling for unobservables does not greatly change the picture obtained
when controlling for a small number of observables alone (education,
experience, and spouse income) implies that most of the selection role
can indeed be captured by a set of observable individual characteristics.18

C. Bounds
Each imputation rule requires assumptions about the position of the
nonemployed relative to the median of the potential wage distribution.
In order to show that we obtain reasonable estimates for the median wage
gap under each specification, we compute bounds following the procedure
discussed in Section IV. Table 6 reports “worst case” bounds to the po-
tential distribution for the base sample and for a subset of wage imputation
rules.
Column 1 reports bounds using the actual wage distribution to obtain
the F() terms in conditions (13) and (14). All estimates for the median
wage gap obtained with alternative imputation methods and reported in
tables 2–4 lie within the bounds to the potential distribution reported in
column 1. Note that, mechanically, the bounds for the gender-specific
18
We have performed a number of robustness tests and more disaggregate anal-
yses on the results reported in tables 2–4. First, we repeated all estimates using
a common set of age weights (obtained from the U.S. 1999 sample) for all countries.
Results using such weights were virtually identical to those obtained without
weights, and thus variation in the age structure across countries does not seem
to explain much of the observed variation in gender pay gaps. Second, for the
imputation rules reported in tables 2 and 3, we have repeated our estimates sep-
arately for three education groups (less than upper secondary education, upper
secondary education, and higher education), and we found that most of the se-
lection occurs between rather than within groups, as median wage gaps disag-
gregated by education are much less affected by sample inclusion rules than in
the aggregate. Finally, we have repeated our estimates separately for three de-
mographic groups: single individuals without children in the household, married
or cohabiting without children, and married or cohabiting with children. We found
evidence of a strong selection effect in France and southern Europe among those
who are married or cohabiting, especially when they have children, and much
less evidence of selection among single individuals without children.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
650 Olivetti/Petrongolo

Table 6
“Worst Case” Bounds to Median Wage Gaps under Alternative
Imputation Rules
1 2 3 4
Lower Upper Lower Upper Lower Upper Lower Upper
Bound Bound Bound Bound Bound Bound Bound Bound

United States .153 .513 .268 .397 .164 .504 .218 .447
United Kingdom ⫺.042 .562 .126 .373 .005 .505 .045 .460
Finland .059 .255 .147 .234 .131 .236 .073 .236
Denmark .035 .155 .074 .099 .064 .115 .040 .150
Germany ⫺.065 .521 .110 .297 .013 .399 ⫺.025 .462
Netherlands ⫺.028 .406 .088 .274 .072 .289 .040 .322
Belgium ⫺.184 .321 ⫺.061 .203 ⫺.070 .203 ⫺.110 .246
Austria ⫺.030 .417 .087 .288 .011 .365 ⫺.017 .402
Ireland ⫺.533 .815 ⫺.037 .494 ⫺.374 .680 ⫺.311 .639
France ⫺.668 .830 ⫺.089 .265 ⫺.387 .565 ⫺.430 .623
Italy ... ... ⫺.419 .377 ⫺.496 .485 ⫺.282 .360
Spain ... ... ⫺.510 .374 ⫺.718 .715 ⫺.577 .640
Portugal ⫺.396 .470 ⫺.111 .323 ⫺.208 .376 ⫺.066 .315
Greece ... ... ⫺.940 .924 ... ... ⫺.496 .566
Sources.—Michigan Panel Study of Income Dynamics and European Community Household Panel
Survey.
Note.—In specification 2 no results are reported for Italy, Spain, and Greece as more than 50% of
women in the sample are nonemployed. Similarly, this is the case in specification 3 for Greece. Sample
description: aged 25–54, excluding the self-employed, the military, and those in full-time education, 1999.
Sample inclusion rules by columns: (1) Employed at time of survey in 1999. (2) Wage imputed from
other waves when nonemployed (⫺5 , ⫹2 window). (3) Impute wage ! median when unemployed. (4)
Impute wage ! median when nonemployed and education ! upper secondary education and experience
! 10 years; impute wage 1 median when nonemployed and education ≥ higher education and experience
≥ 10 years.

median are tighter the higher the employment rate for that gender. Since
variation in male employment rates is low relative to variation in female
rates, cross-country differences in the tightness of the bounds mostly stem
from differences in women’s employment selection across countries.
Bounds for the median gender wage gap are thus much tighter for the
United States, the United Kingdom, and countries all the way down to
Austria than they are for Ireland, France, and Portugal—for which they
are so large as to be completely uninformative. Indeed, we cannot even
obtain bounds to the median wage gap for Italy, Spain, and Greece on
the base sample because less than 50% of women are employed.
A restriction typically used to tighten such bounds is that of stochastic
dominance (see Blundell et al. 2007), which assumes various forms of
positive selection into employment. As this is precisely something that
our article is assessing, we cannot use it as an identifying assumption. But
we can instead compute bounds after wage imputation, that is, using
imputed wage distributions to compute the F() terms in (13) and (14).
This procedure has the advantage of tightening the bounds without as-
suming positive sample selection ex ante. The estimated bounds are re-
ported in columns 2–4 of table 6. In column 2, the wage distribution used
is one in which missing wage observations are replaced by observed wages

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 651

in the nearest available wave (as in col. 2 of table 2). In column 3, missing
wage observations are imputed below the median if an individual is un-
employed (as in col. 3 of table 3). In column 4, they are imputed using
education and experience levels of the nonemployed (as in col. 4 of table
3). As employment rates are higher in columns 2–4 than in column 1,
bounds do become tighter. However, they still remain relatively large in
southern Europe, where employment rates remain relatively low even
after wage imputation.

VI. Conclusions
Gender wage gaps in the United States and the United Kingdom are
much higher than in other European countries, and especially so with
respect to France and southern Europe. Although at first glance this fact
may suggest evidence of a more equal pay treatment across genders in
the latter group of countries, appearances can be deceptive.
In this article, we note that gender wage gaps across countries are
negatively correlated with gender employment gaps, and we illustrate the
importance of nonrandom selection into work in understanding the ob-
served international variation in gender wage gaps. To do this, we perform
wage imputation for those not in work by simply making assumptions
on the position of the imputed wage observations with respect to the
median. Imputation is performed according to different methodologies
based on observable or unobservable characteristics of missing wage
observations.
We find higher median wage gaps on imputed rather than actual wage
distributions for most countries in the sample, meaning that, as one would
have expected, women tend on average to be more positively selected into
work than men. However, this difference is small in the United States,
the United Kingdom, and a number of central and northern European
countries, and it is sizable in France and southern Europe, that is, in
countries in which the gender employment gap is particularly high. Our
(most conservative) estimates suggest that correction for employment se-
lection explains about 45% of the observed negative correlation between
wage and employment gaps. In Italy, Spain, Portugal, and Greece, the
median wage gap on the imputed wage distribution ranges between 20
and 30 log points across specifications. These levels are closely comparable
to those of the United States and of other European countries.
Another interesting result is that we obtain qualitatively similar esti-
mates whether we impute missing wages using available wage information
from other waves in the panel or whether we use observable characteristics
of the nonemployed. This implies that employment selection mostly takes
place along a small number of measurable characteristics.
Our analysis identifies directions for future work. We argue that gender

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
652 Olivetti/Petrongolo

employment gaps are key to understanding cross-country differences in

gender wage gaps. Employment gaps may, in turn, be driven by supply
or demand forces, or both. In recent work, Fernández and Fogli (2005)
and Fortin (2005, 2006) emphasize the role of “soft variables,” such as
cultural beliefs about gender roles and family values and individual at-
titudes toward greed, ambition, and altruism, as important determinants
of women’s employment decisions as well as of gender wage differentials.
These “fuzzy” variables may also shape employers’ beliefs about women’s
labor force attachment and thus the demand for female labor. In addition,
labor market and financial institutions, as well as the sectoral composition
of the economy, may play an important role in the determination of gender
employment gaps. Disentangling supply and demand factors that drive
cross-country differences in female employment is thus the next step for
understanding existing variation in gender pay gaps.

References
Albrecht, James, Aico van Vuuren, and Susan Vroman. 2004. Decom-
posing the gender wage gap in the Netherlands with sample selection
adjustment. IZA Discussion Paper no. 1400, Institute for the Study of
Labor, Bonn.
Altonji, Joseph G., and Rebecca M. Blank. 1999. Race and gender in the
labor market. In Handbook of labor economics, vol. 3C, ed. Orley
Ashenfelter and David Card, 3141–3259. Amsterdam: North-Holland.
Arulampalam, Wiji, Alison L. Booth, and Mark L. Bryan. 2007. Is there
a glass ceiling over Europe? Exploring the gender pay gap across the
wage distribution. Industrial and Labor Relations Review 60, no. 2:
163–86.
Azmat, Ghazala, Maia Güell, and Alan Manning. 2006. Gender gaps in
unemployment rates in OECD countries. Journal of Labor Economics
24, no. 1:1–37.
Beblo, Miriam, Denis Beninger, Anja Heinze, and François Laisney. 2003.
Measuring selectivity-corrected gender wage gaps in the EU. ZEW
Discussion Paper no. 03-74, Centre for European Economic Research,
Mannheim.
Blau, Francine D., and Andrea H. Beller. 1992. Black-white earnings over
the 1970s and 1980s: Gender differences in trends. Review of Economics
and Statistics 74, no. 2:276–86.
Blau, Francine D., and Lawrence M. Kahn. 1996. Wage structure and
gender earnings differentials: An international comparison. Economica
63 (May): S29–S62.
———. 2003. Understanding international differences in the gender pay
gap. Journal of Labor Economics 21, no. 1:106–44.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
A Cross-Country Analysis of Gender Gaps 653

———. 2006. The US gender pay gap in the 1990s: Slowing convergence?
Industrial and Labor Relations Review 60, no. 1:45–66.
Bloomfield, Peter, and W. L. Steiger. 1983. Least absolute deviations: The-
ory, applications and algorithms. Boston: Birkhauser.
Blundell, Richard, Amanda Gosling, Hidehiko Ichimura, and Costas
Meghir. 2007. Changes in the distribution of male and female wages
accounting for employment composition using bounds. Econometrica
75, no. 2:323–63.
Buchinsky, Moshe. 1998. The dynamics of changes in the female wage
distribution in the USA: A quantile regression approach. Journal of
Applied Econometrics 13, no. 1:1–30.
Chandra, Amitabh. 2003. Is the convergence in the racial wage gap il-
lusory? NBER Working Paper no. 9476, National Bureau of Economic
Research, Cambridge, MA.
Fernández, Raquel, and Alessandra Fogli. 2005. Culture: An empirical
investigation of beliefs, work, and fertility. NBER Working Paper no.
11268, National Bureau of Economic Research, Cambridge, MA.
Fortin, Nicole M. 2005. Gender role attitudes and the labour market
outcomes of women across OECD countries. Oxford Review of Eco-
nomic Policy 21, no. 3:416–38.
———. 2006. Greed, altruism, and the gender wage gap. Unpublished
manuscript, Department of Economics, University of British Columbia.
Goldin, Claudia. 1990. Understanding the gender wage gap: An economic
history of American women. Oxford: Oxford University Press.
Gronau, Reuben. 1974. Wage comparison: A selectivity bias. Journal of
Political Economy 82, no. 6:1119–43.
Heckman, James J. 1974. Shadow prices, market wages, and labor supply.
Econometrica 42, no. 4:679–94.
———. 1979. Sample selection bias as a specification error. Econometrica
47, no. 1:153–63.
———. 1980. Addendum to sample selection bias as a specification error.
In Evaluation studies review annual, vol. 5, ed. Ernst W. Stromsdorfer
and George Farkas. Beverly Hills, CA: Sage.
Johnson, William R., Yuichi Kitamura, and Derek Neal. 2000. Evaluating
a simple method for estimating black-white gaps in median wages.
American Economic Review Papers and Proceedings 90:339–43.
Juhn, Chinhui. 1992. Decline of male labor market participation: The role
of declining market opportunities. Quarterly Journal of Economics 107,
no. 1:79–122.
———. 2003. Labor market dropouts and trends in the wages of black
and white men. Industrial and Labor Relations Review 56, no. 4:643–62.
Krueger, Alan B., and Lawrence H. Summers. 1988. Efficiency wages and
the inter-industry wage structure. Econometrica 56, no. 2:259–93.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions
654 Olivetti/Petrongolo

Lewbel, Arthur. 2007. Endogenous selection or treatment model esti-

mation. Journal of Econometrics 127, no. 2:777–806.
Manski, Charles F. 1989. Anatomy of the selection problem. Journal of
Human Resources 24, no. 3:343–60.
———. 1994. The selection problem. In Advances in econometrics, Sixth
World Congress, vol. 1, ed. Christopher A. Sims. Cambridge: Cam-
bridge University Press.
Mulligan, Casey B., and Yona Rubinstein. 2008. Selection, investment,
and women’s relative wages over time. Quarterly Journal of Economics
123, no. 3:1061–1110.
Neal, Derek. 2004. The measured black-white wage gap among women
is too small. Journal of Political Economy 112 (February): S1–S28.
OECD (Organization for Economic Cooperation and Development).
2002. Employment outlook. Paris: OECD.
Olivetti, Claudia, and Barbara Petrongolo. 2006. Unequal pay or unequal
employment? A cross-country analysis of gender gaps. CEPR Discus-
sion Paper 5506, Centre for Economic Policy Research, London.
O’Neill, June. 1985. The trend in the male-female wage gap in the United
States. Journal of Labor Economics 3, no. 1, pt. 2:S91–S116.
Smith, James P., and Michael Ward. 1989. Women in the labor market
and in the family. Journal of Economic Perspective 3 (Winter): 9–24.
Smith, James P., and Finis R. Welch. 1986. Closing the gap: Forty years
of economic progress for blacks. Santa Monica, CA: RAND Corpora-
tion, R-3330-DOL.
Vella, Francis. 1998. Estimating models with sample selection bias: A
survey. Journal of Human Resources 33, no. 1:127–69.

This content downloaded from 155.97.178.73 on Fri, 28 Nov 2014 11:01:15 AM

All use subject to JSTOR Terms and Conditions