Sex Discrimination in the Promotion Process

Author(s): Craig A. Olson and Brian E. Becker

Source: ILR Review, Vol. 36, No. 4 (Jul., 1983), pp. 624-641
Published by: Sage Publications, Inc.
Stable URL: [Link]
Accessed: 21-10-2024 21:39 UTC

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@[Link].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
[Link]

Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to
ILR Review

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
 SEX DISCRIMINATION IN THE
PROMOTION PROCESS

CRAIG A. OLSON and BRIAN E. BECKER*

Previous research has suggested that intra-occupational earnings

differences are the principal source of the long-standing earnings gap
between men and women. Following that line of research, this study
examines the extent of gender differences in the incidence of and returns to
promotions. Drawing on the Quality of Employment Panel, the authors
compare the earnings and promotion experience of men and women over
the period 1973 - 1977. After controlling for unmeasured differences in job
level and constant individual ability in a fixed-effect model, they conclude
that the returns to promotion are comparable for men and women. Further
analysis indicates, however, that women are held to higher promotion
standards than men and therefore receive fewer promotions than men with
equal measured abilities. Although the female/male wage ratio in this
sample increased by nearly 6 percent over four years, the increase could
have been as much as 9.2 percent if women and men had been held to the
same promotion standards.

HE well-known earnings disparity be- principal empirical challenge of that re-

tween men and women has been of search has been to model the wage
policy and research interest for many years. determination process such that male-
Theoretical work has emphasized the female differences in productivity can be
discriminatory tastes of employers, employ- distinguished from employer discrimina-
ees, and consumers; occupational crowd- tion. Following this latter line of research,
ing; differential investments in human this paper examines an aspect of the wage
capital based on expected lifetime labor- determination process often suggested, but
force participation; and greater uncer- never tested, as an important source of
tainty about expected performance.' The male-female earnings differences: the in-
fluence of promotions on the male-female
*The authors are Assistant Professors of Industrial wage gap.
Relations and Human Resources at the School of Although the results of prior research
Management, the State University of New York at
Buffalo. An earlier version of this paper, "Sex,
Promotion, and Wages," was presented at the 1981 Race and Sex," Eastern Economic Journal, Vol. 1
Winter Meetings of the Econometric Society in (April/July 1974), pp. 103 - 10; Jacob Mincer and
Washington, D.C. Valuable comments on an early Solomon Polochek, "Family Investments in Human
draft were provided by Bert Barnow and Sharon Smith. Capital: Earnings of Women," Journal of Political
Economy, Vol. 82, No. 2 (March/April 1974, Part II),
'See, for example, Gary Becker, The Economics of pp. S76 - S108; and DennisJ. Aigner and Glen G. Cain,
Discrimination, (Chicago: University of Chicago Press, "Statistical Theories of Discrimination in Labor
1957); Barbara Bergman, "Occupational Segregation, Markets," Industrial and Labor Relations Review, Vol.
Wages and Profits When Employers Discriminate by 30, No. 2 (January 1977), pp. 175 - 87.

Industrial and Labor Relations Review, Vol. 36, No. 4 (July 1983). ?) 1983 by Cornell University
0019-7939/83/3604 $01.00

624

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
 SEX DISCRIMINATION 625
have varied widely, there is growing agree- for example, that while women earn
ment that while gender differences in the slightly greater returns to job specific
occupational distribution are substantial, training, their access to jobs imparting such
the more important influence on the male- training is substantially lower.6 This sup-
female wage gap is within-occupational ports the earlier work by Malkiel and
earnings differences.2 A number of ex- Malkiel in which earnings differences
planations for these differences have been within one professional organization were
offered, though most focus on average found to be largely attributable to male-
differences in human capital investments, female differences in levels of responsibility
particularly work experience. Landes ob- (jobs), after controlling for individual
serves, for example, that nearly two-thirds characteristics. Only half of this difference
of the within-occupational wage gap is in responsibility levels could be explained
explained by male-female differences in by male-female differences in measured
firm-specific human capital and in productivity characteristics.7
turnover.3 Mincer and Polachek attribute a This study extends the research on male-
substantial part of the wage gap for female earnings differences along the lines
married women to division of labor within suggested by this earlier work. Taking
the family, based on expected lifetime advantage of a somewhat richer data set
differences in the labor-force participation than has been employed to answer these
of women.4 On the other hand, using a data questions in prior research, we attempt to
set with very precise measures of actual further unravel the process by which
work experience, Corcoran and Duncan within-occupational earnings differences
find that the wage returns to experience are are generated. Our focus is the extent and
similar for men and women and that consequences of any differences in promo-
differences in qualifications explain only tion policies that affect men and women.
about half of the male-female wage gap.5 Specifically, the discussion is organized
A slightly different line of research around the following research questions:
examines the quality of work experience (1) To what extent do differential promo-
available to men and women within the tion rates and payoffs contribute to male-
firm. Particularly important are male- female earnings differences? (2) To what
female differences in job placement and extent do different promotion processes for
promotions. Duncan and Hoffman find, men and women contribute to differential
promotion rates and payoffs?
2Among others, see Alan S. Blinder, "Wage Dis-
crimination: Reduced Form and Structural Estimates,"
Journal of Human Resources, Vol. 8, No. 4 (Fall 1973),
Promotions and Wage Differentials
pp. 436-55; and Randall S. Brown, Marilyn Moon, Although men and women may receive
and Barbara S. Zoloth, "Incorporating Occupational different pay for the same or similar jobs
Attainment in Studies of Male-Female Earnings
Differentials," Journal of Human Resources, Vol. 15,
across firms, male-female pay differences
No. 1 (Winter 1980), pp. 3 -28. While the literature for the same job in the same firm do not
reflects the view that the major source of male-female appear to be the problem they once were.8
wage differences occurs within occupations, the broad Within a firm, that is, male-female earn-
occupational categories often used do not allow for an
ings differences are today largely attribut-
entirely unambiguous conclusion. In fact, pay dif-
ferentials within such categories are no doubt due in able to differences in job levels. It is
part to sex differences in the distribution of employees
among quite heterogeneous detailed categories. 6See GregJ. Duncan and Saul Hoffman, "On the Job
3Elisabeth M. Landes, "Sex Differences in Wages and Training and Earnings Differences by Race and Sex,"
Employment: A Test of the Specific Capital Hypothe- Review of Economics and Statistics, Vol. 61, No. 4
sis," Economic Inquiry, Vol. 15, No. 4 (October 1977), (November 1979), pp. 594-603.
pp. 523 - 38. 7Burton G. Malkiel and Judith A. Malkiel, "Male-
4Mincer and Polochek, "Family Investments in Female Pay Differentials in Professional Employment,"
Human Capital," p. S103. AmericanEconomicReview, Vol. 63, No. 4 (September
5Mary Corcoran and GregJ. Duncan, "Work History, 1973), pp. 703.
Labor Force Attachment and Earnings Differences 8See, for example, Francine D. Blau, Equal Pay in
Between the Races and Sexes," Journal of Human the Office, (Lexington, Mass.: Lexington Books,
Resources, Vol. 14, No. 1 (Winter 1979), pp. 3- 19. 1977), pp. 98 - 104.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
626 INDUSTRIAL AND LABOR RELATIONS REVIEW

therefore important to understand the in Table 1. Using data from the Quality of
process by which men and women come to Employment Panel, 1973- 77, Table 1
be employed in different jobs. Occupa- describes the wage gains of 57 men and 27
tional choice, of course, plays an important women who received a promotion between
part in job differentiation, but given the 1973 and 1977. The promotion measure is
relatively broad occupational categories based on a self-reported evaluation of job
available in most micro-data sets, it is changes by those individuals who did not
within-occupational differences in job turn over.' The exact question answered
levels that pose the more interesting re- was, "Compared to the job you had when
search questions. The importance of in- we interviewed you before (1973), is your
ternal mobility has been discussed by a present job at a higher level position
number of researchers, most of whom (1977)?" Based on this definition, approxi-
suggest that men and women receive mately 20 percent of the "intact" sample
differential access to promotions. Hall was promoted during this period.
observes, for example, that the "whole Several results are worth noting. First,
notion of a career with steady advancement the female-male wage ratio increased from
is relevant only for white males."9 Duncan 59 percent to 62 percent over this period.
and Hoffman agree, concluding that their Second, although the gross percentage
results support the proposition that promo- wage increase associated with a promotion
tion practices vary by race and sex.'0 The was quite dramatic for both sexes, the
Malkiel and Malkiel results are also consist- relative benefits of promotion were greater
ent with this view, though it is not clear to for men. Promoted men averaged a 104
what extent job levels observed at one percent increase in wages; men who were
particular time are attributable to promo- not promoted averaged a 42 percent
tion decisions or initial placement. Simi- increase. For women, the average hourly
larly, Ragan and Smith conclude that wage increase was 88 percent for those
"occupational and job level discrimination promoted and 52 percent for those not
appear to be the important contributors to promoted. Among women, that is, the
pay differences between men and average percentage wage increase was 69
women."" Finally, using data from three percent greater for those promoted than for
banking establishments, Cabrel, Ferber, those not promoted; among men, the
and Green find, after controlling for difference was 147 percent. Finally, it is
observed qualifications, that women re- worth observing that the percentages of
ceive lower initial job placements and are women and men receiving a promotion
less likely to be promoted after having been were almost identical (.19 and .20).
hired. Moreover, the differential promotion Because wage returns from promotion
rates occur even among men and women are apparently not equally shared by men
with the same initial placement level.'2 and women, the question becomes one of
The potential economic significance of determining whether this is attributable to
the wage returns to a promotion is reflected ability differences or discrimination. Sex
discrimination in the promotion process is
9Robert E. Hall, "Wages, Income and Hours of Work likely to take two forms: wage discrimina-
in the U.S. Labor Force," in Glen Cain and Harold tion and employment discrimination. In an
Watts, ed., Income Maintenance and Labor Supply example of wage discrimination, men and
(Chicago: Rand McNally, 1973), pp. 393- 94. women are promoted to the same job level,
'"Duncan and Hoffman, "On the Job Training,"
p. 601.
but are not paid equally on the new job. In
"See James E Ragan and Sharon P. Smith, "The
Impact of Differences in Turnover Rates on Male/ '3This table is based on individuals who reported the
Female Pay Differentials," Journal of Human Re- necessary data and did not turn over between 1973 and
sources, Vol. 16, No. 3 (Summer 1981), p. 346. 1977. Since inclusion in the initial 1973 wave required
'2Robert Cabral, Marianne A. Ferber, and Carole A. that respondents be working 20 or more hours per
Green, "Men and Women in Fiduciary Institutions: A week, the results in the table summarize the wage and
Study of Sex Differences in Career Development," promotion history over four years for a sample of
Review of Economics and Statistics, Vol. 63, No. 4 relatively stable, full-time workers. The sample selec-
(November 1981), pp. 573-80. tion process is described in detail later in the text.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
 SEX DISCRIMINATION 627
Table 1. Average Hourly Wage in 1973 and 1977 by Sex and Promotion Status.
(standard deviations in parentheses)

Men Women
W73 $ 5.45 $(2.63)
3.22
W77 7.93 (3.86)
4.92
Overall Means:

(1.48)
(2.12)
1973
1977
W I promotion:
5.15 3.16
9.22 (4.77)
5.72
(2.29) (1.10)
(1.83)
1973
1977
W No promotion:
5.53 3.23
7.61 (3.53)
4.73
(2.71) (1.56)
(2.14)
E(W77 - W73)/W73] lPromotion 1.04 .88
(1.46) (.50)
E(W77 - W73)/W73]INo Promotion .42 .52
(.42) (.50)
Total N 283
Promoted
Sample Size:

N 57125
24
Source: Robert P. Quinn and Graham Staines, Quality of Employment Survey, 1973-1977 Panel, 1st ed.
(Ann Arbor, Mich.: Inter-University Consortium for Political and Social Research, 1979).

employment discrimination, men and be fewer promotions, promotions to jobs

women of equal ability are promoted to requiring less training, or lower pay on the
different job levels. Both practices cause new job in an attempt to require women to
men and women of equal ability to receive pay a greater percentage of the training
different wage returns from a promotion. costs associated with the new job. Each of
The preceding explanations suggest that these policies would result in differential
differences in the wage returns from a wage returns from promotion.
promotion reflect discrimination, but an
alternative explanation is that men have a
greater average stock of human capital, Promotion Model
including greater work experience and A simple model of the promotion process
fewer interruptions in their past work is a useful tool in evaluating these alterna-
experience. Equally important may be tive explanations and obtaining an esti-
employers' predictions of male-female dif- mate of the likely effects of discrimination
ferences in future human capital invest-
ments: if employers assume that women ferences in Worker Quitting," Review of Economics and
have a higher turnover rate, they will Statistics, Vol. 62, No. 3 (August 1980), pp. 388-98;
attempt to invest less firm-specific training and Francine D. Blau and Lawrence M. Kahn, "Race
and Sex Differences in Quit Rates by Young Workers,"
in their female employees. 4 The result could
Industrial and Labor Relations Review, Vol. 34, No. 4
(July 1981), pp.563 - 77. These results suggest that firms
'4Recently reported results using micro data show unfairly discriminate against women if they deny
that women do not have higher quit rates than men, promotions to women or offer lower returns because of
other things equal. See W. Kip Viscusi, "Sex Dif- expected differences in turnover rates.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
628 INDUSTRIAL AND LABOR RELATIONS REVIEW

on promotions and wage levels. Assume performance standard than men, such
that promotion to the job j in an organiza- that:
tion is based on expected performance in (3) P (Prom)m = f [P (u... > R- bX,
that job or the group of jobs to which it
for men, and
leads. Expected performance in turn will
reflect the stock of general and firm- (4) P (Prom). = f [P (U.
specific human capital of each worker, > (Rj + D) - bXw)],
including actual past performance in other
jobs held in the firm. An individual will be for women. A woman is held to a higher
promoted if his or her expected perform- promotion standard if D > 0, with the
ance is above the minimum requirement of result that she is less likely to receive a
job j and also exceeds the expected promotion to job j.
performance of other available workers. If In the preceding example, D represents
Rj equals the minimum level of perform- discrimination against women because the
ance required to be promoted to the job j, man and woman were assumed to have
then the probability of promotion for equal ability. Since u1 is unobserved, the
individual i is: empirical results would show that men and
women of equal measured ability have
(1) P (Prom,) = P[E(VMP1) > RJ different promotion probabilities to the job
* P [E (VMP,) - E (VMPN) j, though not necessarily lower promotion
rates to any job. Nevertheless, if men and
> 0 1 E ( TIMPi) > RJ],
women do not have equal unmeasured
where E (VMPN) is the distribution of the abilities, holding Xi constant, then D will
expected performance of other available represent male-female differences in mean
workers. Clearly, P [E (VMP1) > R.] and unmeasured abilities as well.
P [E (VMPi) - E (VMPN) > 0] increase In this case, the extent to which dis-
as individual i becomes more qualified. crimination contributes to unequal promo-
Expected future performance, E tion rates to the job j between men and
(VMPi), is a function of general and firm- women of equal observed ability cannot be
specific skills, as well as the effort and determined. If the average difference in
intensity with which those skills are ap- unmeasured abilities between the sexes (u)
plied. In practice, skills and effort are not has not been altered by employer practices,
completely observed, so that E (VMPS) can then the differential promotion rates are
be replaced by: not the result of discrimination by the
current employer. Alternatively, part of this
(2) E (VMPS.) = b1Xi + u1, differential can be attributed to dis-
where X is a vector of observable indicators crimination if firms hold women to higher
of skills and ability and u is a normally promotion standards (D > 0) or if they fail
distributed error term that consists of two to place women in jobs that allow firm and
components. One component of u com- individual investments in specific and
prises the constant, unmeasured abilities general skills, or both. This is the familiar
(V) an individual brings to his or her question raised in all studies of wage
current employer, and the second compo- discrimination -to determine whether the
nent comprises the general and firm- residual difference reflects discrimination
specific skills (t) a worker acquires with his or average differences in unmeasured
or her current employer that are not abilities between the groups that are not the
directly measured in the X vector. For a result of the job-placement practices and
man and a woman of equal ability training investments of firms.
(measured and unmeasured) and subject to To summarize, the probability that
no promotion discrimination, Equation 2 particular individuals will be promoted is
applies to both sexes, and the chances for a based on an assessment of their future
promotion will be the same for both. performance, for which complete measures
Promotion discrimination against women are unlikely to be available to the re-
means that women are held to a higher searcher. Sex differences in promotion,

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
 SEX DISCRIMINATION 629
controlling for Xi, may be attributable marginal product. If workers are to have an
either to differences in these performance opportunity to invest in firm-specific skills
characteristics or to discrimination. To the as well as to reap the returns from these
extent that differences in promotion are investments, they must be placed in jobs
based on Xi differences that are not that permit human capital investments by
attributable to firm discrimination, there both the firm and the worker. Although
can be some assessment of the degree of both the training and the reaping of
bias. But as the unmeasured characteristic subsequent rewards often occur within the
reflected in ui becomes more important same job, the existence of internal labor
(holding X1 constant), it is more difficult to markets with well-established job hierar-
distinguish between the hypothesis of dis- chies suggests that the returns to training
crimination and that of promotion based frequently take the form of promotions to
on unmeasured ability. If E (us) equals zero higher-level jobs. If for some reason, such
for both sexes, Equations 3 and 4 imply as sex discrimination, an employer is
that, conditional on X, a man will more unwilling to invest in training an individ-
likely be promoted than a woman, because ual, that individual is more likely than
of sex discrimination. Nevertheless, if men others to be placed in a job that does not
have a greater stock of unmeasured attri- offer the opportunity to acquire firm-
butes such that E (UM) - E (uw) > 0, the specific skills. Such a decision by an
observation that men are more likely to employer reduces an individual's chance for
receive a promotion (conditional on X) can a promotion, since the acquisition of such
be explained by different promotion stan- skills is an important determinant of
dards, average differences in unmeasured expected future performance in a higher
abilities acquired before their having level job.
joined the firm, average differences in skills Moreover, if employers hold a group of
acquired after having joined the firm, or workers, such as women, to a higher
some combination of these influences. promotion standard than others have to
meet (D > 0), this group will be less likely
to invest even in general skills, the entire
A Wage Model with Promotions cost of which is borne by the worker. In
To the extent that Equations 3 and 4 Equation 5, NProm can be interpreted as a
reflect actual employer practices, these proxy for the acquisition of human capital
practices would be expected to have an that leads to higher productivity and wages
important influence on the relative wages in the future, and a3, in turn, includes the
of men and women. An interpretation of wage returns to these investments.
the relative wage effects of promotions may In addition to measuring the returns to
be ambiguous, however; and it is therefore training investments, a3 estimates two other
instructive to examine the wage returns to a effects. The promotion model developed
promotion for a single group of workers. above assumes that firms promote on the
We begin with the following simple model: basis of ability, including unmeasured
(5) lnWi = ao + a1X1 abilities (V) captured in u, such as
intelligence and general motivational
+ a2JLi + a3NProm, + e,, levels, that are not the result of training
where JL represents the job level of an investments (t) by the firm and worker.
individual when hired. e - N(O, oe) consists Thus, ei and NProm will be positively
of two parts: the constant individual effect correlated if, as we might expect, some of
defined above ( Vi) and a random error these same unmeasured abilities influence
term. NProm equals the number of promo- earnings. The variable a3 also reflects the
tions a person has received at the firm. fact that promoted individuals are likely to
In a simple model of wage determina- face greater stress and risk in their new jobs.
tion, the real wage received by the average If these factors outweigh the nonpecuniary
worker will increase, because earlier invest- benefits of a promotion, such as more status
ments in general and specific human and a key to the executive washroom, then
capital increase the value of the individual's a3 would be positive even if abilities

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
630 INDUSTRIAL AND LABOR RELATIONS REVIEW

acquired before and after employment in promotion decisions. If the promotion

the firm could be controlled."5 An ordinary process is properly specified, a4 is the
least squares (OLS) estimate of the returns relationship between the unmeasured abili-
to a promotion (a3) would therefore be ties (u) used in promotion decisions and
expected to reflect both changing job unmeasured abilities (V) affecting wages.
characteristics and the wage gains asso- From the discussion above, this relationship
ciated with the higher abilities (V and t) of represents the wage returns from promo-
those promoted. tions attributable solely to the ability
differences (V) between those promoted
Estimation Issues and those not promoted. a3 is the return to
training investments made between t and
Two separate but related methods could
t - 1, as well as the nonpecuniary
be used to estimate the impact of a
characteristics of the higher level jobs. This
promotion while controlling for constant,
procedure assumes that in the entire
unmeasured abilities, V One is to replace
sample, the error terms in the promotion
NProm by NPfom, where NPiom is
model (u) and wage model (e) are distribu-
uncorrelated with ei and obtained by ted bivariate normal and independent of
regressing NProm on a set of instrumental
the exogenous variables in the respective
variables. Unfortunately, this method is not
equations.
feasible, because all the ability measures
Following the literature on fixed-effects,
influencing NProm should also be included
if constant individual effects are correlated
in the wage equation. A second method is
with the X vector in the wage equation, the
to redefine NPromi as a dummy variable
individual effects are also likely to be
(Prom) equal to one if a promotion
correlated with the exogenous variables in
occurred between time t - 1 and t and
the promotion model,"7 resulting in biased
define JL as the individual's job level in
estimates of the promotion coefficients and
period t - 1. Equation 5 then becomes:
incorrect estimates of X. In Equation 7,
(6) lnWtI{ = ao + aX1 therefore, the error term may still be
+ a2JLt l, + a3Promi + ei. correlated with the X vectors and Prom,
even when the selection-bias variable has
The correlation between Prom and e can been included.
now be controlled by including in Equation An alternative method that controls for
6 an estimate of E (utProm) using Heck- constant, unmeasured, individual effects
man's procedure for correcting for selectiv- (V) is a fixed-effect model based on
ity bias.16 Equation 6 then becomes: Equation 5. Employing the two periods of
wage data available in this study, the fixed-
(7) lnWt, = ao + aXZ + a2JLt l,
effect model is equivalent to the following
+ a3Promi + a4Xi + ei, first-difference specification, where e t , is
where X is an estimate of E (u|Prom)
obtained from the probit estimates of
Statistical Models of Truncation, Sample Selection and
Limited Dependent Variables and a Simple Estimator
"5The wage returns from a promotion may also reflect of Such Models," Annals of Economic and Social
the "prize" awarded to individuals who successfully Measurement, Vol. 5 (1976), pp. 475-92; James J.
compete in a promotion lottery designed to motivate Heckman, "Sample Selection Bias as a Specification
employees. Such a scheme is efficient in managerial Error," Econometrica, Vol. 47, No. 1 (January 1979),
positions for which the costs to the firm of measuring pp.153-61; and James J. Heckman, "Addendum to
individual performance are high. See Edward P. Lazear 'Sample Selection Bias on a Specification Error'," in
and Sherwin Rosen, "Rank Order Tournaments As Ernst W. Stromsdorfer and George Farkas, eds.,
Optimum Labor Contracts," Journal of Political Evaluation Studies Review Annual: Vol. 5 (Beverly
Economy, Vol. 89, No. 5 (October 1981), pp. 841 - 64. Hills: Sage, 1980), pp. 70 - 74.
Since our sample includes workers whose performance '7See James J. Heckman, "Heterogeneity and State
can be monitored by the firm at relatively low cost Dependence in Dynamic Models of Labor Supply,"
(namely, nonmanagerial and nonprofessional employ- mimeo no. 7837 (Chicago: University of Chicago, May
ees), it is unclear how much a, is influenced by 1978); and Jerry Hausman, "Specification Tests in
promotion lotteries. Econometrics," Econometrica, Vol. 46, No. 6 (Novem-
'6SeeJamesJ. Heckman, "The Common Structure of ber 1978), pp. 1251 - 71.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
 SEX DISCRIMINATION 631
the random error component from Equa- (9) lnWi,77 = ao + alxi +
tion 5 that is independent of V:
a2JL73, + a2(JL77 - JL73)i + e2 .
(8) InW77 - lnW73, = (ao,77 - a0,73)
If unmeasured individual abilities (V)
+ al(X77i - X73,) + a3Promi could be controlled, a2 would correspond to
+ (e7'7,i- e7-3,i) the wage return for a single unit change in
Since Vi andJLi are constant across the two job level. Promotion discrimination could
time periods, each cancels out and the error then be decomposed into different values of
term consists of e 3-e . This error term is
a2 for the sexes and differences in the
independent of wages and the X vector by average job-level change for each sex.
In our analysis, we do not observe either
assumption. In other words, ordinary least
squares applied to Equation 8 provides an JL73 orJL77. Rather, we only know whether
unbiased estimate of the wage returns of the change in job level was positive over this
changing job characteristics and the re- period. If the difference is positive, Prom
turns from the training investments (t). equals one. Asuming no demotions, this
relationship implies that:

(10) JL77 =JL73 + blProm,

The Problem of Job Level
where b1 is the average job-level change for
The preceding discussion also applies to
those promoted and Prom is the dummy
estimates of the promotion returns for men
and women. A comparison of the returns
variable defined above. Since JL73 is
unobserved and therefore included in the
from a promotion for men and women,
even after controlling for constant, un- error, we are left with the following
equation after substituting Equation 10
measured ability, is complicated, however,
into Equation 9:
by average unobserved differences between
the sexes in the "size" of the promotion. (11) lnWi,77 = ao + aX2
Recall that the use of higher promotion
standards for women means that they are + a2bPromi + (ei + a2JL73,i).
less likely to receive a promotion and more Consequently, in Equation 11, the wage
likely to receive a promotion that will move returns from a promotion reflect both the
them a shorter distance up the firm average size of the promotion and the
hierarchy than equally qualified men. It is returns from a unit increase in job level.
important to recognize this effect in viewing Lower estimated promotion returns for
this analysis, because we do not observe the women may therefore represent lower
"size" of the promotion received by the returns to a "unit" job level increase, or
promoted subsample. The different pro- differences in the size of the job level
motion returns in Table 1 may therefore changes between promoted men and
simply reflect measured and unmeasured women, or both.
individual characteristics or the fact that The arguments in this section can be
women may receive "smaller" promotions summarized in the following way. OLS
and lower wage returns because of dis- estimates of promotion returns reflect the
crimination, or both. characteristics of the higher level jobs as
The impact of promotion size on dif- well as the initial abilities ( V) and returns to
ferences in promotion returns between men the training investments (t) of those promo-
and women can be illustrated as follows. ted. To distinguish among these returns,
Assume thatJL73 indexes a person's job level two strategies are proposed. Both control
in the firm at the time of the 1973 interview. for the correlation between promotions and
In 1977, that person is asked if his or her constant unmeasured ability (V). The
current job is at a higher level - that is, is sample-selection bias technique assumes,
JL77 less than JL73? If JL77 and JL73 were however, that the error terms in the wage
measured, the dummy variable Prom and promotion model for the entire sample
would be replaced by JL77 - JL73 and are independent of the exogenous variables
Equation 6 becomes: in both the wage and promotion model.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
632 INDUSTRIAL AND LABOR RELATIONS REVIEW

The fixed-effect model does not rely on this First, individuals were entirely excluded
assumption. In both cases, the returns from from the study if in 1973 they were
a promotion will include both the wage employed in the construction industry, were
change associated with a "unit" change in self-employed, or did not answer questions
job level and the number of job-level units necessary to construct the variables in-
the worker rises as a result of the promo- cluded in a 1973 earnings function and an
tion. attrition equation. This reduced the sam-
ple to 979. Those self-employed and those
Model Specification employed in construction were excluded
and Estimation because they had no real opportunity for
In the preceding sections, we argued that promotion. The sample reduction caused
to understand the impact of promotion by a respondent's failure to answer all of the
decisions, we must model both wages and questions was assumed to be independent
promotions. An issue that must be consid- of 1973 earnings and the probability of
ered as well is the selection process that promotion between 1973 and 1977. The
reduced the original sample of 1,455 two other reasons for the reduced sample
individuals to the 408 observations size reflect sample attrition between 1973
described in Table 1. In particular, we and 1977. First, individuals who left their
believe the attrition process is related to 1973 employer, either voluntarily or invol-
promotions and wages and must be expli- untarily, were excluded from the analysis
citly considered in the analysis. As noted, because our definition of promotion in-
our study draws on data collected in the cludes only internal upward mobility, that
Quality of Employment Panel, 19 7 3 - 19 7 7. is, promotion within a firm. Second, the
Based on a national probability sample of figures reported in Table 1 required infor-
persons aged 16 years or older and working mation on 1977 wages and promotions
more than 20 hours per week, the full data between 1973 and 1977. If these data were
set includes 1,455 respondents. There are incomplete for a respondent or if the data
several advantages to these data: they offer required to construct a 1977 earnings
longitudinal observations (respondents equation were missing for a respondent,
were interviewed in both 1973 and 1977); that individual was excluded from the
their measure of labor-force experience calculations reported in Table 1. All told,
captures actual rather than potential years this selection process left a remaining
in the labor force; and they provide an sample of 408 individuals.
opportunity to measure the process of If attrition were independent of wages
upward internal job mobility, namely, and promotion decisions, we could ignore
promotions. attrition in the wage and promotion
The large nonrandom reduction in the analysis and observe only the "intact"
sample size required to obtain complete individuals. The assumption of the inde-
data on promotion decisions and their wage pendence of attrition, promotions, and
consequences causes a potential problem wages from one another is likely to be
that must be carefully considered. As others untenable, however. Our earlier discussion
have demonstrated, ignoring sample attri- suggested that employers may attempt to
tion will lead to biased estimates if attrition minimize the cost associated with high
is related to the earnings process. 18 The turnover rates among women by adopting
strong possibility that attrition, earnings, policies that would reduce the incidence of
and the promotion process are not indepen- promotions among women or reduce their
dent from one another is apparent from the wage returns from promotion, or both."9 In
three principal sources of sample attrition. other words, employment stability is con-
sidered to be an additional productivity
characteristic. In this sample, there are
'8Heckman, "Sample Selection Bias," and Jerry
Hausman and David Wise, "Attrition Bias in Experi-
mental and Panel Data: The Gary Income Main- 'It may also be the case that some women turn over
tenance Experiment," Econometrica, Vol. 47, No. 2 because of the promotion discrimination they experi-
(March 1979), pp. 455-73. ence in the firm.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
 SEX DISCRIMINATION 633
gender differences in the rate of attrition, Although the three equations could be
with 55 percent of the men and 65 percent estimated using the maximum-likelihood
of the women having left the sample procedure outlined by Hausman and Wise,
between 1973 and 1977. we use for this analysis the two-step
To handle those problems, a three- consistent procedure developed by Heck-
equation model has been adopted, in man.' The attrition equation is estimated
which the endogenous variables are wages, on the "complete" sample (979 observa-
the promotion decision, and sample attri- tions) using probit analysis. The promotion
tion. Three alternative specifications of the and wage equations are then estimated
wage equation are used. In one specifica- using the sample of individuals who worked
tion, the dependent variable is the natural for the same firm in both 1973 and 1977
log of the hourly wage in 1977. Except for (408 observations). The selection-bias vari-
the addition of a dummy variable indicat- ables obtained from both the attrition and
ing a promotion between 1973 and 1977, the promotion equations are included in
the model is specified as a conventional the OLS and ANOVA wage equations to
earnings model and corresponds to Equa- account for the possible correlation be-
tion 11.20 The second model corresponds to tween promotions, attrition, and the error
the fixed-effect specification of Equation 8. term in the equations.22 When the fixed
The third specification is a simple ANOVA effect wage model is estimated, only the
model that regresses change in the natural Mills ratio from the attrition equation is
log of wages between 1977 and 1973 on the included, because the correlation between
promotion dummy variable. This last promotions and constant individual effects
specification is identical to the fixed-effect was controlled by the first differencing.
model when none of the exogenous vari- Separate estimates were calculated for the
ables changes over the four-year period. male and female samples.
Equation 11 and the ANOVA model are
also estimated with and without the Results and Discussion
selection-bias variables obtained from the Promotion effects. Before turning to the
promotion and attrition probit estimates. results from the wage models, we will
The specification of the promotion summarize the estimates yielded by the
equation includes the same 1973 human promotion and attrition probit models.23
capital measures as the wage model; The attrition models are significant at
proxies for promotion standards, R (union conventional levels (ax = .05) for both sexes,
membership, occupation, industry, and suggesting that attrition is a nonrandom
firm size); one dummy variable denoting process. As might be expected, the proba-
the gender of the employee's immediate bility of attrition for each sex is negatively
supervisor; and a second dummy variable related to tenure with the firm, with the
denoting that the person had no immediate strongest effect among individuals having
supervisor. The measures excluded from five to ten years of service. Generally, the
the promotion equation, but included in probability of attrition also declines with
the wage equation, are the region and labor-force experience, union member-
urban variables. These variables are ex- ship, pension coverage, and the number of
cluded because they reflect cost-of-living school-age children in the family, though
differences that affect wages but not
promotion probabilities. Finally, the attri-
2'Hausman and Wise, "Attrition Bias," and Heck-
tion equation includes all of the variables in man, "Sample Selection Bias."
the OLS wage equation (1973 values) as 22The promotion model was estimated with and
well as other family income (OTHINC), the without the correction for attrition. The results
number of children in the household reported here do not include attrition in the promotion
model because the attrition variable was insignificant,
(KIDSO-5, KIDS6-15), age, and pension and for the most part, the coefficients in the promotion
coverage. model changed very little when the variable was
included.
20The variable definitions and sample means are 23The complete results of the promotion and attrition
available from the authors. models are available from the authors.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
634 INDUSTRIAL AND LABOR RELATIONS REVIEW

these relationships are not always statisti- tion differences on the female/male wage
cally significant for both sexes. differential is examined in the next section
In contrast, the promotion models for of the paper.
each sex are not very stable. For males, the Wage effects. The wage effects of promo-
promotion equation is significant at the .01 tion are summarized for each sex and
level [x2 (.01, 23) = 75.2 > 41.6], although model in Tables 2 and 3. The OLS results
most of the individual coefficients are show that a promotion between 1973 and
insignificant. The results for many of the 1977 yielded about 13 percent higher
individual variables are also insignificant in average hourly earnings for men and 22
the female model, but the overall equation percent for women.27 Using a one-tailed
is significant at the .05 level [X2 (. 05, 19) = test, we conclude that promotion returns
31.8 > 30.1].24 are significantly different from zero at the
To determine whether the promotion .05 level for women and at the .10 level for
process varies significantly by gender, we men. The difference in promotion returns
combined the samples and constrained all between the sexes is not significant at the
coefficients, including the constant, to be .05 level, but the absolute size of these
the same for men and women. In com- estimates indicates that, as expected, pro-
paring the constrained and unconstrained motions have a substantial influence on
estimates, we are unable to reject at the .10 wages.
level the hypothesis that the coefficients are As we pointed out earlier, the O LS results
equal.25 Nevertheless, when comparing this include the payoffs from changing job
constrained version of the promotion model characteristics, constant unmeasured abil-
to estimates that allow the intercept to vary ity differences, and training investments
by gender, we can reject at the .05 level the not captured in X. In an effort to unravel
hypothesis that the intercept is the same for the contribution of each of these factors to
men and women.26 Consequently, in this promotion wage gains, we estimated the
small sample, the impact of measured Xs fixed-effect models reported in line 5 of
on the promotion process is the same for Tables 2 and 3. These estimates represent
men and women. The coefficient on the wage gains from promotions, after constant
gender dummy variable, however, is individual effects have been controlled.
-.50075, indicating that, for some un- Presumably, these returns will largely
measured reason, a woman is significantly represent changing job characteristics and
less likely than a man to receive a returns from training. For both men and
promotion, other things equal. One ex- women, the estimated returns from a
planation of this finding is that women are promotion increased and the standard
held to a higher promotion standard than errors declined vis-a-vis the OLS results.
men (D > 0). Another is that employers For women, the change from 22 to 27
predict that men will have a higher level of percent corresponds to a 23 percent in-
performance, based on employers' observa- crease in the estimated returns from a
tions of attributes that are not measured by promotion. For men, the estimated promo-
the estimates. The impact of these promo- tion return of 30 percent in the fixed-effect
model is 115 percent greater than in the
OLS estimates. Similar changes in the
24The insignificance of many of the individual
coefficients may simply reflect the fact that many of the
promotion returns also obtain in the simple
continuous variables were collapsed into sets of dummy ANOVA model where changes in the log of
variables in the original data set. Correct significance wages was regressed on the promotion
tests for a construct thus require joint tests of dummy. Like the OLS results, the estima-
significance.
ted promotion returns for men and women
25The value of the log-likelihood function was
- 149.75 for the unconstrained model and -164.32 for in the fixed-effect and ANOVA models are
the constrained model. This yields a chi square value of not significantly different at the .05 level.
29.13 < 31.41 = X2a5, 2 The relatively larger wage returns from
26The product of - 2 times the difference in the
values of the log-likelihood function is 4.88 > 3.84 =
2
X .05, 1- 27These percentages equal exp (B) - 1.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
 SEX DISCRIMINATION 635
Table 2. Promotion and Selection-Bias Coefficients in the Male Hourly Wage
Model (N = 283).
(standard errors in parentheses)

Model and Independent Variables

Dependent Variables Promotion Xproniotion Xattrition Other Controls R

(1) in (W77) - in (W73) .282*** None .065

ANOVA:

(.064)
(2) in (W77) - in (W73) .499*** -.161** -.101 None .080
(.125) (.083) (.082)
(3) in (W77) .123 a .230
OLS:

(4) In (W77) .047 .069 -.267 a .241

(.083)

(.236) (.143) (.168)

(5) in (W77) - in (W73) .262*** b .138C
Fixed Effect:

(.075)
(6) in (W77) - in (W73) .268*** -.085 b .145C
(.076) (.097)
aEducation, tenure, labor-force experience, union, marital status, occupation, region, urban, and firm size.
Education, tenure, labor-force experience, union, marital status, occupation.
CThis is the variance in In (W77) - In (W73) explained by the regression and does not include the variance in wages
over the two time periods explained by the individual constants that are implied by first differencing the two years of
data.
*Significant at the .10 level in a two-tailed test.
**Significant at the .05 level in a two-tailed test.
***Significant at the .01 level in a two-tailed test.

Table 3. Promotion and Selection-Bias Coefficients in the Female Hourly Wage

Model (N = 1.25).t
(standard errors in parentheses)

Dependent Variables 2
Model and Independent Variables
Promotion Xprornotion Xattrition Other Controls R

(1) In (W77) - In (W73) .240*** None .068

ANOVA:

(.080)
(2) In (W77) - In (W73) .304* -.052 -.012 None .071
(.171) (.110) (.079)
(3) In (W77) .200** a .649
OLS:

(.102)

(4) In (W77) .777***

(.251) .378**
(.151) -.166***
(.023) a a .672
(5) in (W77) - in (W73) .240*** b .429c
Fixed Effect:

(.078)
(6) in (W77) - In (W73) .201** -.117 b .440c
(.103) (.120)
tFootnotes a, b, and c and levels of significance denoted by asterisks are the same as in Table 2.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
636 INDUSTRIAL AND LABOR RELATIONS REVIEW
promotions in the fixed-effect model were differences between the OLS and fixed-
unexpected. Earlier, we had argued that effect models were greater for men, we can
OLS estimates of promotion returns reflect conclude that promoted men came from
the greater responsibilities of the higher- relatively lower-level positions, within their
level job, returns from constant individual respective job hierarchy, than did promo-
abilities, and training investments. If that is ted women. These within-group compari-
true, the promotion coefficient in the fixed- sons, of course, say nothing about the
effect model would be expected to drop if absolute job levels held by either sex.
firms promote those individuals with the Nonetheless, this is the pattern we might
highest levels of constant unmeasured expect to observe if women, on average,
abilities. We believe there are two explana- hold lower-level jobs than men. Policies
tions for the unexpected increase in this designed to promote women from within
coefficient. First is the unlikely possibility would draw on those women who held the
that firms promote less-able individuals. A highest positions within the relatively lower
more plausible explanation is that the OLS class of jobs to which women had been
wage estimate of promotion returns is confined in the past.
picking up the effects of both unmeasured Finally, promotion and attrition
ability and initial job level in 1973, as selection-bias variables are included in
reflected in the error term in Equation 11. both the OLS and ANOVA models to
Since JL73 is unobserved, the wage return control for the covariance between the
from a promotion will be confounded ifJL73 error term in the wage equation and the
is correlated with wages and promotions. errors in the promotion and attrition
JL73 will be correlated with Prom if the models. The results for these variables are
chances for a promotion vary with an also reported in Tables 2 and 3.28 The
employee's position in the firm's hierarchy. inclusion of these two variables in the
We argue that this correlation is negative, ANOVA model led to larger estimated
simply because there are more opportuni- promotion returns for both sexes. Only X
ties for promotion from lower-level posi- promotion, however, appears to be signifi-
tions than from higher-level positions. The cant (at the .05 level).
correlation between JL73 and wages is The results obtained when the selection-
expected to be positive. Independent of bias variables are added to the OLS and
unchanging abilities, those employees on fixed-effect equations are not entirely
higher-level jobs receive higher wages as consistent with those in the ANOVA model.
compensation for their investments in firm- As expected, the attrition parameter was
specific skills as well as for the greater negative in each equation, suggesting that
responsibilities, risks, and longer hours less-productive workers were more likely to
associated with a higher-level job. As a leave the sample. The results for the Mills
result, the positive correlation betweenJL73 ratio representing promotion decisions in
and wages and the negative correlation the OLS equation are different from the
between JL73 and Prom will cause the
estimated promotion wage returns to in-
28The standard errors reported in Tables 2 and 3 for
crease whenJL73 is controlled in the fixed- the equations that include the selection-bias variables
effect model. are those obtained from the OLS regressions and are
For both men and women, the promo- biased. With the exception of the coefficient on the
tion returns estimated in the OLS model selection-bias correction for promotions in the ANOVA
model, none of the incorrect t-values approach
suggest that in 1973, promoted individuals
conventional levels of significance. The reported
held relatively lower-level positions than standard errors would thus have to be biased signifi-
nonpromoted individuals. Since the pro- cantly upward in order to change the conclusion in the
motion returns for both sexes increased in text. This seems unlikely, and for the reasons discussed
in the next paragraph, we have chosen to discount the
the fixed-effect model over those in the
selection-bias estimates for other reasons. See William
OLS model, the chances of a promotion H. Green, "Sample Selection Bias as a Specification
declined as individuals moved up the Error: Comment," Econometrica, Vol. 49, No. 3 (May
hierarchy of jobs held by their respective 1981), pp. 795 - 98; and Heckman, "Addendum to
sex. Furthermore, since the coefficient Sample Selection Bias."

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
 SEX DISCRIMINATION 637
ANOVA results, however. This is particu- Such a decomposition is misleading
larly true in the men's sample, where the however. Although the unadjusted proba-
sign is positive though insignificant. The bilities of promotion are nearly the same for
results for the women's sample are in the men and women, the estimates of the
expected direction, though the magnitude promotion model indicate that, other
of the change in the OLS promotion things equal, women are less likely to
coefficient is surprising. receive a promotion. To examine the
These results could be attributed to impact of this difference on the female/
several factors. First, the selection- male wage differential, we decompose this
adjustment procedure has the desirable difference in wages, using the wage results
property of consistency for large samples. and the average number of promotions
In this analysis, however, the total sample of women (men) would have received had they
women is small, and the number of women been treated like men (women) in the
promoted is even smaller. As a result, the promotion process. A three-step procedure
selection-adjustment procedure may yield is used to estimate the average promotion
unstable results. A second possible problem probabilities women would have experi-
with the OLS selection-bias results is the enced had they been promoted under the
high collinearity between the selection- men's standard.30 First, SEX is set equal to
adjustment variable and the exogenous zero and each woman's X values are
variables in the OLS wage model, making inserted into the promotion model where
it difficult to identify the selection process. each coefficient, except the constant, is
This collinearity occurs because the exoge- constrained to be the same for each sex.
nous variables in the wage and promotion This yielded Zk for each observation. Using
models are nearly identical. We are in- the cumulative distribution function for a
clined to discount the selection-bias esti- standard normal variable, F(Z), we then
mates not only for these reasons, but also calculate the promotion probability for
because the fixed-effect model controls for each individual. Next, by calculating F(Z)
constant unmeasured characteristics we obtain the average promotion probabil-
without having to depend on the promotion ity that women could have expected had
probit estimates for a selection- adjustment they been treated like men (Pr6mJ). The
variable. procedure to obtain Pr6mm for men is the
same, except the gender variable is set
Decomposition of Average equal to one.
Wage Differences
The negative coefficient on the gender
The estimated promotion returns re- variable in the promotion model suggests
ported in Tables 2 and 3 illustrate the that firms do not use the same promotion
significant impact of promotions on wages standards for men and women and that this
for both sexes. With the exception of the difference works to the disadvantage of
OLS selection-bias estimates, however, women. If women in this sample had been
none of the estimated returns from a promoted according to the men's stan-
promotion was statistically different at the dards, nearly 32 percent of the women
.05 level for the two sexes. The implication would have been promoted, instead of only
here is that men and women have equal 19 percent, as in the observations. Al-
opportunities for promotion ternatively, if the men in this sample had
(Promm, Promw)
30This is similar to the method used in Francine Blau
and similar percentage wage gains. Thus, if
and Lawrence M. Kahn, "Causes and Consequences of
we were to "decompose" the earnings Layoffs," Economzc Inquzry, Vol. 19, No. 2 (April
difference between the sexes based on these 1981), pp. 270- 96. The procedure was adopted
results, we would conclude that promotions because, unlike ordinary least squares, the nonlinearity

did not have an important influence on the of the probit model does not allow prediction of the
mean promotion probability in the sample using the
average female-male wage difference.29 mean values of the independent variables. This
procedure yields an unbiased estimate of the mean
2>For example, see Blinder, "Wage Discrimination." promotion probability in the population.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
638 INDUSTRIAL AND LABOR RELATIONS REVIEW

been held to the women's standard, only for in W73 and the ANOVA model of
10.8 percent would have been promoted promotion wage effects for (in W77 -
instead of the 20.1 percent observed. W-7 3 ) .33 The decomposition of the mean
The wage implications of this apparent difference in the log of 1977 wages for men
discrimination against women can be and women then becomes either:
demonstrated by examining how the (14) 1nW77w - nW77,M = (aOW- aoJ

BC
female/male wage differential for this
A
sample would have narrowed between 1973
and 1977 if women (men) had been + (X73- XM) al73 + X73 (al7W-al7M3
promoted according to the men's (women's)
promotion model. In 1973, the difference + (bow - boM) +
between men's and women's wages can be D
decomposed in either of the following ways
(Promw - PromM)bpro7.,M
using OLS wage estimates:
E
(12) lnW73W - lnW73,M = (aow - aOM)
+ Promw (bpronzW - bPronzM)
+ (Xw - XM ) alM + Xw (aw - alM) F

or or

(13) InW73W - InW73,M = (aOw - aOM) (15) lnW77,W - lnW77,M = (aO7W -a l

BC
A
+ (Xw - XM ) a, + XM (aw - alM) .
The first and the last terms in each + (X73-Xm) a1 + Xm (a1w- alm)
equation represent the portion of the mean
wage difference unexplained by different + (bow - boM) +
average levels of X for men and women. D

Assuming the wage model is correctly (Promw - PromM)bPro.,w

specified, these two terms represent direct E
discrimination. When the decomposition
+ PromM (bPronzW - bProtnM).
employs results from a standard earnings
F
model with the male (female) coefficients
as the base, we obtain the results shown in The subscripts M and W refer to the
the first three lines of Table 4 (Table 5).32 results for men and women, respectively.
Depending on whether the male or female The terms A, B, and C come from the 1973
wage coefficients are used as a base, 68 wage estimates and reflect the amount of
percent (.37/.54) or 87 percent (.46/.54) of the difference between men's and women's
the gross wage differential is unexplained 1977 wages that is attributable to 1973
and attributed to discrimination. endowments and discrimination up to
A similar decomposition of 1977 wages 1973. Terms D, E, and F come from the
could also be done using the regression ANOVA promotion results and represent
the portion of the average wage difference
results from a 1977 wage model. Since we
in 1977 attributable to changes in wages
are primarily interested in the wage effects
of promotions, however, 1977 earnings between 1973 and 1977. If the promotion
differences can be analyzed using the process was not a possible source of
equation,
discrimination, then D + F represents the
only portion of the 1977 differential attrib-
lnW77 = lnW73 + (lnW77 - lnW73), utable to post-1973 discrimination. As the
and substituting the 1973 wage estimates previous results suggest, however, women

330ne of the other promotion wage models could also

3Of course, discrimination may also account for be used. The OLS promotion estimates were rejected,
differences in the levels of X. This is not of immediate however, because the fixed-effect results indicate
concern, however, because of our interest in the impact significant individual and job-level effects. The
of promotions on wages. ANOVA results were used instead of the fixed-effect
32The wage equations used in this analysis are results because of the similarities in promotion returns
available from the authors upon request. in the two models.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
 SEX DISCRIMINATION 639
Table 4. Decomposition of 1973 and 1977 Mean Average In Wage Differences
for Males and Females.

Equation 14 Values of Equation 14

(1) InW73,w- InW73,m -.5391
(2) alm w- X3) -m.1661
(3) (aw - aonm) + XWZ3 (aiw - aim) -.3730
(4) lnW77,w - lnW77,m -.4794
(5) Change in the earning gap from 1973 to 1977 (4-1) +.0597
(6) (bow - bom) + Promw (bPromw- bpromm) (.0705 -.0081) = + .0624
(7) (Promw - Promm) bprom -.0027
(8) (bow - bom) + Pr6mw (bpromw - bpromm) .0567
(9) (Pr6mw - Promm) bpromm .0355

Table 5. Decomposition of 1973 and 1977 Mean Average In Wage Differences

for Males and Females.

Equation 15 Values of Equation 15

(1) 1nW73,w- lnW73,m -.5391
(2) aiw (X73 - X73) -.0719
(3) (aow - aom) + X73 (a1w - aim) -.4672
(4) 1nW77,w - 1nW77,m - .4794
(5) Change in the earning gap from 1973 to 1977 (4-1) + .0597
(6) (bow - bom) + Promw (bprom - bprom) (.0705 -.0085) = +.0620
(7) (Promw - Promm) bprom -.0023
(8) (bom - bow) + Pr6mm (bProm - bpromr) .0660
(9) (Promw - Pr6mm) bpromw .0201

would receive more promotions if they were difference in the log of wages was .4794, or
subject to the same promotion standards as about a 6 percent decline in the female-
men. Thus, term E is also partially a result male gap from 1973 (line 5 in Table 4). As
of discrimination. line 6 in each table illustrates, firms
The wage impact of discrimination in apparently corrected some of their earlier
the promotion process can be calculated wage discrimination against females by
by substituting Pr6mw for Promw granting them relatively larger wage in-
in Equation 14 and Pr6mM for PromM in creases over this period (see, for example,
Equation 15, where Pr6mw (Pr6mM) is the (bow - boM) in Table 5). By itself, this
average promotion probability for women process would have narrowed the wage
(men) if they had been treated like men differential by about 7.3 percent. The
(women). The results of these calculations promotion process worked in favor of men,
are summarized in lines 3 through 9 in however, and reduced this relative gain to 6
Tables 4 and 5. In 1977, the male-female percent.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
640 INDUSTRIAL AND LABOR RELATIONS REVIEW

The unadjusted estimates in line 7 and our earlier results indicate that men and
the latter term in line 6 in Tables 4 and 5 women earn the same relative wage gains
show that men benefitted slightly from from promotions, the opportunities for
both greater promotion rates (-.0027) and such gains are not equally available.
greater returns from promotion (-.0081). The reader is reminded that some
The overall decline in the gap of .0597 obvious caveats apply to our results. First,
points can then be attributed to the the study utilizes a very small set of workers.
generally faster rise in female wages More stable estimates of the models could
(+.0705) and the greater benefits of the be expected with a larger sample. More-
promotion process enjoyed by men over, although the total sample size is
(-.0081-.0027). These estimates under- probably adequate for most analyses, the
state the potential influence of discrimina- actual numbers of men and women promo-
tion, however, since differences in ted during this period were 57 and 24,
promotion rates (Promw - PromM) may respectively. These results are therefore
also reflect discrimination. When Pr6mw vulnerable to the effects of one or two
and Pr6mM replace Promw and PromM in outliers. Second, the promotion model is a
Tables 4 and 5, respectively, we obtain very simple view of promotion decisions.
estimates of the wage difference that would An important aspect of the process that is
exist if men and women were held to the not considered here is that a worker is
same promotion standard. These results eligible for a promotion only to higher-level
are reported in lines 8 and 9 of the tables. If jobs. Thus, for any particular job opening,
women had been promoted using the men's the applicant pool is limited to individuals
promotion standard, the relative wage who have not already progressed past the
differential would have declined by .0355 open position. Correctly modeling the
points (line 9) because women would have promotion process is obviously important
received over 50 percent more promotions because these results are used to simulate
than men.34 The overall increase in the the promotion experience each sex would
female/male differential would have been have had if promotions had been made
.0922 points (lines 8 and 9), or over fifty using the promotion process of the other
percent larger than the observed rate. sex. We believe our results have indirectly
Alternatively, if men had been treated as captured this aspect of the promotion
women were in promotion decisions and process, but it was not explicitly included in
these results were used to decompose the our model.
wages using the female coefficients as the A third potential problem with the
base, the portion of the earnings gap that promotion results is the bias introduced by
would be eliminated was somewhat smaller. the sample heterogeneity that the fixed-
As Table 5 shows, the overall decline in the effect wage estimates show to be important
wage differential would have been .0861 in wage determination. This issue, how-
points (.0660 + .0201), or a 44 percent ever, cannot be examined using this sam-
increase over the actual change in the ple. Fourth, this study suffers from the
differential. same problem encountered in all wage-
Taken together, these results suggest that discrimination studies. Does "unexplained"
although the overall wage gap between the difference reflect discrimination or un-
sexes declined by 6 percent over this four- measured ability, and how much of the
year period, the decline actually would "explained" difference in wages reflects
have been between 8.6 and 9.2 percent if indirect employer discrimination? As in
men and women had been exposed to the other studies of this kind, our results do not
same promotion process. Thus, although provide an obvious choice between these
two competing explanations. Finally, it is
important to recognize that this study was
34The magnitude of the promotion rates for women
done using a relatively stable group of full-
when exposed to male standards would be expected to
approach the male level in the long run. This effect time, male and female workers. We think
may be a short-run phenomena reflecting the "over this is an advantage, because the two
qualified" status of many women in their current jobs. groups' unmeasured characteristics may be

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]
 SEX DISCRIMINATION 641
more similar to each other than those of a returns to promotion are comparable for
random sample drawn from all workers. men and women. A more systematic
Nevertheless, our sample does limit the analysis of differences in the promotion
population to which the results can be process for men and women also casts
generalized. doubt on the apparent equality of promo-
tion opportunities suggested in Table 1. For
example, our results indicate that women
Summary and Conclusions (men) would have received substantially
There were several motivations for this more (fewer) promotions had they been
research. There was a general interest in held to the same standards as men
modeling the promotion process of organi- (women). Although the female-male wage
zations and the consequent impact on gap narrowed by about 6 percent between
individual earnings. This general topic was 1973 and 1977, it would have narrowed
pursued in the context of an issue of even further-by 8.6 to 9.2 percent-if
concern to scholars, policy makers, and men and women had been promoted on the
workers: the gap between male and female basis of the same criteria.
earnings. The promotion process seemed Taken together, these results suggest that
particularly germane to the question of although the returns from promotion are
male-female earnings differences, since largely the same for men and women, the
prior research had demonstrated the im- promotion process is quite different for the
portant contribution of intra-occupational two groups. As shown in earlier work by
wage differences to the overall male-female Duncan and Hoffman, unequal access to
earnings gap. opportunities, rather than unequal re-
We argued that there are essentially two turns, constitutes the principal source of
ways in which firms might discriminate male-female differences in employment
against women in the promotion process outcomes.35 The weight of the evidence in
and in turn influence wages. First, firms this study indicates that the promotion
may promote men and women on the same process is an important influence on the
basis but compensate women less than men male-female earnings gap and one that is
for the promotion. This is the picture not operating in a direction that is consist-
reflected in the unadjusted descriptive ent with public policy.
comparison reported in Table 1. Subse-
quent analysis, however, does not support
this interpretation. The promotion coeffi- 35Duncan and Hoffman, "On the Job Training,"
cients in the wage equations show that the p. 601.

This content downloaded from [Link] on Mon, 21 Oct 2024 21:39:57 UTC
All use subject to [Link]