You are on page 1of 7

The Gender Earnings Gap: Learning from International Comparisons

Author(s): Francine D. Blau and Lawrence M. Kahn


Source: The American Economic Review , May, 1992, Vol. 82, No. 2, Papers and
Proceedings of the Hundred and Fourth Annual Meeting of the American Economic
Association (May, 1992), pp. 533-538
Published by: American Economic Association

Stable URL: http://www.jstor.com/stable/2117457

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 support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

is collaborating with JSTOR to digitize, preserve and extend access to The American Economic
Review

This content downloaded from


179.6.208.243 on Thu, 27 Aug 2020 23:01:24 UTC
All use subject to https://about.jstor.org/terms
The Gender Earnings Gap:
Learning from International Comparisons

By FRANCINE D. BLAU AND LAWRENCE M. KAHN*

Despite some dramatic reductions in the policies to raise the relative pay of low-wage
male-female pay gap since the 1950's, gen- workers (regardless of gender) may indi-
der differentials persist in all industrialized rectly reduce the gender pay gap. U.S. pay-
nations. However, the size of these differ- setting is far less centralized than that in
entials varies considerably. This paper uses the other countries in this study, with the
micro data to analyze international differ- possible exception of Switzerland, a factor
ences in the gender pay gap among a sam- contributing to gender pay differences in
ple of eight industrialized nations (see Table the United States.1
1 for included countries). A major theme in We adapt a framework used by Chinhui
our analysis is that intercountry differences Juhn et al. (1989) in their analysis of
in the gender gap are affected by two pro- black-white wage trends in the United
cesses. First are gender-specific factors in- States to estimate the contribution of gen-
cluding differences in the qualifications of der-specific factors versus wage structure in
men and women and differences in the explaining international differences in the
treatment of equally qualified women (i.e., gender gap. The striking finding of our study
labor-market discrimination). Second is is that the higher level of wage inequality in
wage inequality, the prices the labor market the United States works to increase the
of each country places on various labor- gender differential in the United States rel-
market skills, both observed and unob- ative to all the other countries in our sam-
served. For example, suppose that in two ple and fully accounts for the lower gender
countries women have lower levels of skills earnings ratio in the United States com-
than men but that the differences in skills pared to the Scandinavian countries and
(somehow measured) is the same in the two Australia (the countries with the smallest
countries. If the return to skill is higher in gaps).
one country, then that nation will have a This approach helps us understand the
larger gender pay gap. seemingly paradoxical position of U.S.
Skill prices can be affected by relative women compared to women elsewhere. U.S.
supplies, by technology (e.g., high-tech in- women compare favorably with women in
dustries place a premium on highly trained other countries on several measures of skills
workers) or by the wage-setting institutions relative to men. Moreover, the United States
of each country. Specifically, centralized has had a longer commitment to policies of
wage-setting institutions which tend to re- equal pay and equal employment opportu-
duce interfirm and interindustry wage varia- nity for women than have the other coun-
tion and are often associated with conscious tries in our sample. Yet the United States
does not rank among the countries with the
smallest pay gaps; indeed the gender pay
differential in the United States is above
* Blau: Institute of Labor and Industrial Relations,
University of Illinois, Champaign, IL 61820, and NBER; average for our sample of eight countries.
Kahn: Institute of Labor and Industrial Relations, Uni- An important part of the explanation of this
versity of Illinois. We thank participants at the NBER
Conference on "Differences and Changes in Wage
Structures" for helpful comments. We are grateful to
David Blanchflower for making the International So- 'See our longer paper (Blau and Kahn, 1991) for a
cial Survey Programme (ISSP) data available to us. description of differences across these countries in
This research was supported by a grant from the Ford wage-setting institutions, gender-specific policies, and
Foundation to the NBER. gender differences in qualifications.

533

This content downloaded from


179.6.208.243 on Thu, 27 Aug 2020 23:01:24 UTC
All use subject to https://about.jstor.org/terms
534 ALA PAPERS AND PROCEEDINGS MAY 1992

pattern is that the U.S. labor market places TABLE 1-FEMALE/MALE EARNINGS RATIOS
CORRECTED FOR HOURS
a much larger penalty on those with lower
levels of labor-market skills (both measured Country
and unmeasured). (earnings All Married Single
measure) workers workers workers

I. Earnings Ratios in the Micro Data


Germany 0.6880 0.5730 1.0270
(monthly)
The following countries and time periods United Kingdom 0.6337 0.5966 0.9489
were used in this study: Austria [1985-1987], (annual)
United States 0.6849 0.5944 0.9552
West Germany [1985-1988], the United
(annual)
Kingdom [1985-1988], the United States Austria 0.7256 0.6558 0.9703
[1985-1988], and Switzerland [1987] (Inter- (monthly)
national Social Survey Programme [ISSP] Switzerland 0.6174 0.5768 0.9449
data); Sweden [1980] and Norway [1982] (monthly)
Sweden 0.7673 0.7242 0.9350
(Class Structure and Class Consciousness
(annual)
[CSCC] data); and Australia [1986] (Income Norway 0.7308 0.7160 0.9158
Distribution Survey [IDS] data).2 The sam- (annual)
ple was restricted to workers aged 18-65. Australia 0.7489 0.6909 0.9144
(annual)
To maximize sample size, we pooled years
of data for those countries in the ISSP Notes: The earnings ratios were evaluated at 40 hours.
surveyed more than once. The earnings ratios for married workers are for mar-
Table 1 gives estimated female/male ried workers with one person other than the spouse in
the household (for Sweden, Norway, and Australia,
earnings ratios by marital status for log
one child); those for single workers are for nonmarried
earnings corrected for hours. To obtain people with no other persons in the household.
these estimates, the following regression was
run separately by sex for each country:3

(1) ln(EARN) ent; OTHERS is number of people other


than the spouse in the household (ISSP
= bo + bjPART+ b2HRPART
countries), or number of children in the
+ b3HRFULL+ b4MARSP household (Scandinavia and Australia); and
e is an error term. The coefficients from (1)
+ b5OTHERS + e were used to compute a gender ratio for
each country, assuming a 40-hour work
where ln(EARN) is the natural log of earn- week.
ings; PART is a dummy variable for part- The first column in Table 1 indicates the
time employment; HRPART and HRFULL hours-corrected gender earnings ratio for
are interactions of weekly work hours with all workers. These estimates are obtained
part- and full-time status; MARSP is a by evaluating the family-composition vari-
dummy variable for married, spouse pres- ables at their mean values for each sex in
each country. The ratio is highest in the
Scandinavian countries, Australia, and Aus-
2The ISSP data are described in David Blanch-
tria, ranging from 73 to 78 percent, and
flower and Richard Freeman (1992). See Rachel lowest for Switzerland and the United King-
Rosenfeld and Arne Kalleberg (1990) for a description dom, ranging from 62 to 63 percent.4 The
of the CSCC data, originally compiled by Erik Wright;
and see McKinley Blackburn and David Bloom (1991)
for a description of the IDS data.
3For countries with more than one year of data, the 4While our estimates for Australia and the Scandi-
log-earnings variable was obtained by transforming each navian countries are below those in published sources
observation into its 1988 (or end year) equivalent on (ILO, various issues), they are consistent with other
the basis of regressions including only gender and year estimates from micro data (Blau and Kahn, 1991).
dummy variables. Published ratios for these countries are most likely

This content downloaded from


179.6.208.243 on Thu, 27 Aug 2020 23:01:24 UTC
All use subject to https://about.jstor.org/terms
VOL. 82 NO. 2 GENDER AND LABOR-MARKET OUTCOMES 535

United States and Germany comprise an where Yij is the log of wages; Xi is a vector
intermediate category.
of explanatory variables; Bi is a vector of
In the next two columns, we use equation coefficients; Oij is a standardized residual
(1) to estimate the earnings ratio separately (i.e., with mean zero and variance 1 for each
for married workers with one nonspouse country); and oj is the country's residual
"other" (or child) in the household5 and for standard deviation of wages (i.e., its unex-
single (i.e., nonmarried) people with no oth- plained level of male wage inequality).
ers (children) in the household. The pay Then, the male-female wage gap for
ratio is uniformly very high among single country j is
workers, ranging from 0.91 to 1.03. In con-
trast, the pay gap is much larger for married (3) Dj-Ymj -Yfj = 8 Xj Bj + 0 y01
workers, with the rankings for this group
fairly similar to the overall rankings. (The where the m and f subscripts refer to male
most notable differences are that Austria and female averages, respectively; and a 8
falls into the intermediate group and Ger- prefix signifies the average male-female
many falls into the lowest group.) These difference for the variable immediately fol-
findings suggest that intercountry differ- lowing. Equation (3) states that the country's
ences in the earnings ratios of married pay gap can be decomposed into differences
workers drive the international pattern of in measured qualifications (5X1), and dif-
gender pay gaps and that in order to explain ferences in the standardized residual (60j)
this pattern, we need to concentrate on multiplied by the log money value per unit
married workers.6 difference in the standardized residual (o).
The pay-gap difference between two
II. Analysis of the Intercountry Gender Gap countries j and k can then be decomposed
as follows:
Juhn et al. (1989) have devised a method
that allows us to decompose the cross-coun- (4) Dj - Dk
try differences in the gender gap into a
portion due to gender-specific factors and a
portion due to differences in the overall - (Xj - Xk)Bk + 5Xj(Bj -Bk)
level of wage inequality. Following their no-
tation, suppose that we have for male worker
+ (89 - 50k )o'k + 50j(o,i - Ck )
i and country j a male wage equation:
The first term in (4) reflects the contribu-
(2) Yij = XijBj + o-jOij tion of intercountry differences in observed
labor-market qualifications (X) to the gen-
der gap. The second term reflects the im-
pact of different measured prices across
overstated in that the Scandinavian data are limited to countries for observed labor-market quali-
manufacturing workers, and the Australian data ex- fications. The third term measures the ef-
clude supervisory personnel.
5The average number of nonspouse others in the fect of cross-country differences in the rela-
household is roughly 1. tive residual wage positions of men and
6The very small gaps for single workers may be due women (i.e., whether women rank higher or
in part to their being disproportionately young (the pay lower within the male residual wage distri-
gap -is lower for young workers); in addition, single
bution). Such differences in rankings may
males are less productive than married males (see
Sanders Korenman and David Neumark, 1991), reflect gender differences in unmeasured
whereas the opposite is likely to be the case for women. characteristics or the impact of labor-market
In our longer paper (Blau and Kahn, 1991), we show discrimination against women. We label this
that differences in the composition of the labor force
term the "gap" effect. Finally, the fourth
with respect to marital status do not explain differences
in the pay gap across countries, again suggesting that
term of (4) reflects intercountry differences
we must concentrate on differences among married in residual inequality. Suppose that unmea-
workers. sured deficits in female relative skills or

This content downloaded from


179.6.208.243 on Thu, 27 Aug 2020 23:01:24 UTC
All use subject to https://about.jstor.org/terms
536 ALA PAPERS AND PROCEEDINGS MA4Y 1992

discrimination lower women's position in (4). With labor-market discrimination, the


the male distribution of wage residuals as is
last term in part reflects the interaction
likely. The larger the penalty a country between country j's level of discrimination
places on being below average in wages, the (pushing women down the distribution of
larger will be its pay gap. In the empiricalwages) and intercountry differences in the
work below, we label this the effect of "un-overall level of inequality which determine
observed prices." how large the penalty is for that lower posi-
Following Juhn et al. (1989), the third tion in the distribution (Juhn et al., 1989).
term, (80Q - 0k)dok, can be obtained empir-The observed price effect may also reflect
ically by assigning each woman a percentile discrimination if, for example, women
number corresponding to her position in "crowd" into certain sectors lowering rela-
her country's distribution of male wage tive earnings there even for men (Barbara
residuals. We can then find what the coun- Bergmann, 1974).
try-j mean female wage residual would have In light of the evidence presented above
been given the percentile rankings in coun- that intercountry differences in the pay gap
try j and the distribution of male wage are driven by differentials for married work-
residuals in country k (note that the mean ers, we implement this decomposition using
male residual is always zero). The difference as the dependent variable YMCH, hours-
between this imputed wage residual for corrected earnings simulated for married
country j and the actual female mean wage people with one nonspouse other (or child)
residual for country k is used to find the in the household using equation (1).7 An
estimate of (80 - 0k)o-k. That is, we find alternative to using wages for married work-
the contribution to the cross-country dif- ers (with one other or child) as the depen-
ference in the gender gap that would result dent variable in equation (2) would be to
if two countries had the same levels of control for marital status and number of
residual male wage inequality and differed children (others) on the right-hand side.
only in their percentile rankings of the fe- However, such an approach is problematic
male wage residuals. The fourth term of (4), in that these variables may measure higher
0j(oj - ok), may be obtained analogously skills for men and lower skills for women.
and measures the contribution to the cross- The decomposition of YMCH is shown in
country difference in the gender gap that Table 2. The underlying regressions include
would result if two countries had the same the traditional human capital variables
percentile rankings of the female wage (education, potential experience, and its
residuals and differed only in the extent of square) and a vector of dummy variables for
male residual wage inequality. union status, one-digit industry, and one-
According to (4), the full impact of gen- digit occupation.8 The structural variables
der-specific factors is reflected in the first may reflect both worker skills and rents
and third terms, the effect of gender differ- received by workers with these characteris-
ences in qualifications and of gender dif- tics. Unfortunately, the data sets available
ferences in wage rankings at a given level of
measured characteristics. Labor-market
7For each worker i, we have
structure is reflected in the second and
fourth terms, the impact of intercountry
YMCHi = ln(EARN)i - bjPART1 - b2HRPART1
differences in returns to measured and un-
measured characteristics. In a traditional
-b3(HRFULLi -40)- b4(MARSPi -1)
decomposition, the sum of the third and
fourth terms represents the impact of inter- - b5 (OTHERSi - 1) .
country differences in the "unexplained"
differential, which is commonly taken as an 8For Australia, union status is unavailable, and the
estimate of discrimination. occupation and industry codes differ from those in the
ISSP or CSCC data. For U.S-Australia comparisons,
The possibility of discrimination compli- we estimated a U.S. equation that had the same vari-
cates the interpretation of the last term of ables as Australia's.

This content downloaded from


179.6.208.243 on Thu, 27 Aug 2020 23:01:24 UTC
All use subject to https://about.jstor.org/terms
VOL. 82 NO. 2 GENDER AND LABOR-MARKET OUTCOMES 537

TABLE 2-ANALYSIS OF LOG WAGES (YMCH) that these differences in rankings raise the
Mean
differential relative to the United States,
percentile often substantially (the unweighted average
of female
residuals effect is 0.2374). 10 The column headed
Gender in male "Unobserved prices" shows that the lower
Country differential distribution Di-DUSA
level of residual wage inequality in each of
Germany 0.5569 15.78 0.0367 the other countries has a negative effect,
United Kingdom 0.5165 20.98 -0.0037
United States 0.5202 30.44 - often quite considerable, on its gap relative
Austria 0.4220 21.61 -0.0982
Switzerland 0.5502 25.37 0.0300
to that in the United States (the unweighted
Sweden 0.3227 30.91 - 0.1975 average effect is - 0.2791).
Norway 0.3341 27.88 -0.1861
Australia 0.3698 31.45 -0.1504
Table 2 also provides estimates of the
impact of measured skills and their prices
Decomposition of D,-DUSA on intercountry differences in the pay gap.
Observed Observed Unobserved The "observed X's" effect is positive with
Country X's prices Gap prices
the exception of Sweden, indicating that
Germany 0.0240 -0.1038 0.5739 -0.4574 U.S. women have relatively favorable levels
United Kingdom 0.0170 -0.0302 0.3465 - 0.3370
Austria 0.0633 -0.1475 0.3412 -0.3552
of the measured characteristics (the un-
Switzerland 0.0889 - 0.0397 0.2384 -0.2576 weighted average effect is 0.0335). The "ob-
Sweden -0.0218 -0.0394 0.0145 -0.1508
Norway 0.0229 -0.0948 0.1865 -0.3007 served prices" effect is negative in every
Australia 0.0399 -0.0561 - 0.0389 - 0.0953 case, indicating that the male returns to
Notes: Log earnings were estimated at 40 hours per week for measured characteristics decrease the pay
married individuals with one nonspouse other (or child). The re- gap in other countries relative to the United
gression includes controls for education, potential experience and
its square, union status, and occupation and industry dummy vari- States (the unweighted average effect is
ables.
- 0.0731).
The results in Table 2 suggest that U.S.
women fare well with respect to gender-
to us lack information on actual labor- specific factors (as measured by the contri-
market experience or weeks worked, which bution of the observed X's and of gap) but
remain important omitted variables in these are adversely affected by the high level of
analyses.9 wage inequality in the United States (as
The column headed Di - DUSA shows the measured by the contribution of observed
difference between each country's gender and unobserved prices)." Indeed, the higher
differential and that of the United States. level of wage inequality in the United States
The U.S. gender gap is higher than in the fully accounts for the lower gender ratio in
Scandinavian countries, Australia, and Aus- the United States in comparison to the
tria, about the same as in the United King- Scandinavian countries and Australia. How-
dom, and lower than in Germany and ever, as noted earlier, what we have labeled
Switzerland. The mean percentile of the wage inequality could also reflect the im-
female residual in the male distribution pact of the interaction of differences in wage
ranges from 16 percent in Germany to 30-31 structure and the country-i (non-U.S.) level
percent in Australia, Sweden, and the of discrimination.
United States. It is noteworthy that U.S.
women place near the top of the list.
The column headed "Gap" shows the t0Note that although the percentile ranking of
contribution of each country's female place- women in Sweden is slightly higher than in the United
States, the gap effect is small and positive. This is
ment in the male residual wage distribution
because the gap effect takes into account not simply
to its relative pay gap. The figure is positive the rankings, but also the associated wage rewards
for all countries except Australia, indicating (penalties).
1"Further indication of U.S. women's high relative
labor-market qualifications compared to women in
9To some degree, our controls for education, hours, other countries is the far greater incidence of full-time
industry, and occupation may pick up the effects of work among U.S. employed women (relative to men)
such omissions. than for those elsewhere (Blau and Kahn, 1991).

This content downloaded from


179.6.208.243 on Thu, 27 Aug 2020 23:01:24 UTC
All use subject to https://about.jstor.org/terms
538 ALA PAPERS AND PROCEEDINGS MA Y 1992

What are we then to conclude about the that to understand changes in the gender
role of labor-market structure? From a pay gap fully, it would also be fruitful to
number of indirect indicators we conclude examine the impact of changes in wage
that it is important, even though it may not structure. Given growing wage inequality in
be possible to estimate its effect precisely. the United States in the 1970's and 1980's
First, U.S. wage-setting institutions are con- (Juhn et al., 1990), American women may
siderably less centralized than in other have been swimming upstream in a labor
countries, thus making a finding of the im- market increasingly unfavorable to low-wage
portance of wage structure plausible. Sec- workers.
ond, the U.S. commitment to policies of
equal pay and equal employment opportu- REFERENCES
nity compares positively to the other coun-
tries in our sample, although such policies Bergmann, Barbara R., "Occupational Segre-
may well be more effectively implemented gation, Wages and Profits When Employ-
under more centralized wage-determination ers Discriminate by Race or Sex," East-
systems.12 Further, U.S. women compare ern Economic Journal, April-July 1974, 1,
favorably to women in the other countries 103-10.
in terms of their qualifications relative to Blackburn, McKinley L. and Bloom, David E.,
men. Thus, it is credible that gender-specific "Changes in the Structure of Family In-
factors do not explain the relatively high come Inequality in the United States and
pay gap in the United States. Third, we Other Industrialized Nations During the
found that the wage and residual wage vari- 1980s," Working Paper, Columbia Uni-
ation of both men and women in the United versity, June 1991.
States considerably exceed those of the same Blanchflower, David and Freeman, Richard,
gender group in other countries (results not "Unionism in the United States and
shown), suggesting that the same set of fac- Other Advanced OECD Countries," In-
tors (measured and unmeasured prices and dustrial Relations, Winter 1992, 31, 56-79.
wage-setting institutions) affect the wages of Blau, Francine D. and Kahn, Lawrence M., "The
both men and women in each country in a Gender Earnings Gap: Some Interna-
similar way. Finally, and perhaps most im- tional Evidence," unpublished manuscript
portantly, even though the estimated wage presented at the NBER Conference on
inequality effect may include the impact of "Differences and Changes in Wages
gender discrimination as it interacts with Structures," London, 1991.
wage structure, our findings nonetheless Juhn, Chinhui, Murphy, Kevin M. and Pierce,
suggest an extremely important role for wage Brooks, "Accounting for the Slowdown in
inequality in affecting the gender ratio. Black-White Wage Convergence," un-
published manuscript, University of
III. Conclusion Chicago, October 1989.
_ and , "Wage In-
The major finding of this paper is the equality and the Rise in Returns to Skill,"
importance of wage structure, specifically unpublished manuscript, University of
the higher level of wage inequality in the Chicago, January 1990.
United States relative to other industrial- Korenman, Sanders and Neumark, David,
ized countries, in explaining international "Does Marriage Really Make Men More
differences in the gender pay gap. Much Productive?" Journal of Human Re-
attention has focused on women's growing sources, Spring 1991, 26, 282-307.
relative levels of skills and labor-force com- Rosenfeld, Rachel and Kalleberg, Arne, "A
mitment as causes of changes in the pay gap Cross-National Comparison of the Gen-
in the United States. Our research suggests der Gap in Income," American Journal of
Sociology, July 1990, 96, 69-106.
12In addition, the comparable-worth approach pur- ILO, Yearbook of Labour Statistics, Geneva:
sued in Australia might be expected to produce a International Labor Organization, various
larger immediate impact on wages. issues.

This content downloaded from


179.6.208.243 of:ffff:ffff on Thu, 01 Jan 1976 12:34:56 UTC
All use subject to https://about.jstor.org/terms

You might also like