Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 23
1
Salaries, Salary Growth, and Promotions of
Men and Women in a Large, Private Firm
BARRY A. GERHART and GEORGE T. MILKOVICH
Studies of differences in earnings between
men and women have sought to adjust the
raw differential for gencler-related cliffer-
ences in factors thought to reflect individual
differences in productivity. In practice, these
studies have examined differences in pay
between men and women controlling for
(1) supply-side factors, such as investment
in human capital (e.g., general and firm-
specific work experience, education) and (2)
demand-side factors intencled to represent
the relative scarcity of specific types of labor
in specific markets (e.g., occupation, in-
clustry, geographic region). In most in-
stances, this research has used national sur-
vey data on inclivi(luals working in many
different (and unknown) firms.
In reviewing this research, Treiman and
Hartmann (1981:19) concluded that "worker
characteristics account for very little of the
difference in earnings (between men and
women]." Gen(ler-related differences in de-
mand-sicle factors, mainly occupational dis-
tribution, were found to explain a larger part
of the earnings gap. Nevertheless, after ad-
justing for both supply- and demand-sicle
factors, a substantial portion of the earnings
differential between men ancT women re-
23
mains. This remaining portion is taken as
evidence of labor market discrimination against
women and as being indicative of researchers'
inability to identify, measure, and control for
all aspects of worker productivity.
At least two problems exist with pay
discrimination research of the type de-
scribed above. First, some variables fre-
quently included on the right-hand side of
the wage equation may themselves be en-
dogenous to the process of discrimination
(Blinder, 1973; Cain, 1986; Oaxaca, 1973~.
Thus, for example, women may have less
experience in the labor market than men
because their work opportunities in the
market have been less favorable than tho.s'?
available to men having similar productive
abilities. As another example, it may be
that gender-basec] occupational (and firm)
segregation (see Reskin and Hartmann, 1985)
reflects not only different aspirations an
work values of men ant! women, but also
Differences that exist between men and women in
terms of preferences for different types of occupations
may also reflect discrimination, whether its source is
differential socialization in families, schools, and other
premarket institutions, or is market based.
OCR for page 24
24
unequal access to firms, occupations, and
jobs. Fuchs (1971), in fact, contended that
virtually the entire earnings gap between
men and women could be explained if an
occupational classification scheme having
sufficient detail were used, but he noted
that the question of why men and women
had `different occupational distributions then
had to be resolved (see also Sanborn, 1964~.
Although it is clear that access to jobs and
occupations is an important determinant of
earnings, little research has examiner] the
question of why such attainments slider ac-
cor(ling to gen(ler.2
A second problem with marketwicle stud-
ies of gender-based discrimination is that
they provide no direct information on work-
er productivity at the level of the firm. As
one consequence, previous research has had
to rely on indirect measures of productivity
(e.g., experience, tenure, and education).
Some evidence indicates, however, that such
measures may converge poorly with mea-
sures of productivity at the level of the firm
(e.g., Brown, 1982; Mecloff and Abraham,
1981~.
All firms attempt to assess the produc-
tivity of their workers in one way or another.
Many firms formalize this process by, for
example, regularly conducting performance
appraisals of employees. In these firms,
compensation and internal-staffing decisions
(e.g., promotion) are often explicitly based
on such productivity measures.3 These mea-
sures have a number of desirable charac-
teristics. First, they are designed to assess
worker productivity in a specific job in a
specific firm. Second, such measures are
2This issue of differential attainment due to possible
unequal access to firms, occupations, and jobs is some-
times referred to as employment discrimination.
3Based on a survey of personnel and industrial
relations executives, the Bureau of National Affairs
(1983) concluded that performance appraisal results
are used by 86 percent of firms for making salary in-
crease decisions and by 79 percent of firms for making
promotion decisions concerning their white-collar
workers.
PAY EQUITY: EMPIRICAL INQUIRIES
influenced by human capital characteristics
only insofar as the latter are useful in per-
forming the particular job in question. In
other words, firm-level productivity mea-
sures potentially carry information on the
quality of the match between the worker's
abilities and what the job requires. Thirdly
these job-specific productivity measures re-
flect differences in worker motivation, which,
together with the abilitybob requirements
match, affect actual inclividual performance
levels.
Given the discussion above, three ave-
nues of research on labor market discrim-
ination would appear to be especially useful.
First, firm-level research is neeclecl that
examines differences in men and women's
salaries adjusted for possible differences in
job-specific productivity measures. Second,
research on the determinants of occupa-
tional and job attainments would also be
helpful. Some evidence indicates that sub-
stantial gender-based, within-firm job seg-
regation may exist (BielLy and Baron, 1986),
which suggests that job level may be an
important determinant of gender-based sal-
ary differences within firms. Although with-
in-firm analyses have been rare in the lit-
erature, they have indeed found the effect
of job level on earnings differences to be
quite large, especially relative to withinjob
differences (Halaby, 1979; Malkiel and Mal-
kiel, 1973; Rosenbaum, 19851. Third, lon-
gitudinal data would permit stu(ly of changes
in salary. Further, such data would aid in
assessing the extent to which firms' pro-
motion practices contribute to job-level dif-
ferences. Again, it would also be useful to
know what role job-specific performance
measures play in decisions on promotion
and salary increases.
In this paper we examine possible gencler-
based differences in attainment in a large,
private firm. Specifically, we focus on three
general issues. First, controlling for job
level, performance ratings, and individual
characteristics, is there a salary disadvantage
for women? Second, given longitudinal data,
OCR for page 25
SALARIES, SALARY GROWTH, AND PROMOTIONS
we can move beyond] cross-sectional anal-
yses and examine models of salary change
over time, addressing questions of the fol-
lowing type: Do women's salaries grow at
the same rate as men's? Do men and women
receive equal salary increases for a given
level of performance? Do men and women
receive equal salary returns from promo-
tions within the firm? Third, controlling for
initial job level, performance, and individual
characteristics, do men and women receive
different numbers of promotions over time?
Unlike much previous research, we measure
not only whether a promotion has occurred,
but also the number of promotions over
time. In other words, are women, for ex-
ample, held to higher promotion standards
(Olson and Becker, 1983~?
THE FIRM
The data for this analysis are from the
personnel information system of a firm that
produces a highly diversified set of industrial
and consumer products. The firm has rough-
ly 100 manufacturing operations in more
than 30 states. It enjoyed general financial
success cluring the period of the study, as
evidences! by revenues that grew faster than
the Consumer Price Index and a return on
equity that comfortably exceeded the me-
clian for the Fortune 500.
Exempt jobs are the focus of this study.
Professional, managerial, sales, and tech-
nical jobs are the major broac] categories.
Examples of some of the most common job
titles include engineer, senior engineer,
sales representative, area sales manager,
administrative assistant, technologist, su-
pervisor, and maintenance supervisor.
Compensation policies and practices of
the firm are typical of those in the Fortune
500. The firm participates, for example, in
over six annual salary surveys for jobs in-
clucled in this study. The focus of these
surveys varies some focus on a selected
group of perhaps 10 to 15 product market
competitors and others focus on labor mar-
25
Let competitors that employ persons with
similar skills or have similar occupations.
Statistical methods are used to combine the
results of the surveys. Consiclerable jucig-
ment is also exercised, however, because
different degrees of confidence are placecl
on the results of the various surveys. This
is consistent with Rynes ant] Milkovich's
(1986) argument that ad hoc judgments are
typically made throughout the process.
Strategy also plays an important role in
determining pay level. During the time
period of the study, a policy of "paying with
the leaders" was follower! for the jobs we
examined. In practice, this policy meant
establishing the pay policy line at the fiftieth
percentile of the group of pay leaclers.
The pay structure of the jobs included in
this study was maintained with the help of
a single, national job evaluation system. The
structure is defined by 15 job classes, or
levels, each with minimum and maximum
rates of pay roughly 20 percent below and
above the midpoint. A Conference Boarc]
(1984) survey of 557 major U. S. firms found
that the median number of levels in exempt
pay structures for all industries was 19. The
compensable factors used in the job eval-
uation system are education/knowledge re-
quired, experience required, complexity of
duties, working conditions, and responsi-
bility. Of the 491 firms using formal job
evaluation in the Conference Board study,
90 percent used a formal plan. Of this group,
approximately 20 percent used a plan of the
general type used by the firm we studied.
An explicit pay-for-performance policy ex-
ists for the determination of individual pay
increases. The policy is implemented through
the use of annual merit increase guides (see
Milkovich and Newman, 1987, for exam-
ples). These gui(les are designed to control
the cost of annual pay increases, as well as
to encourage a distribution of increases to
employees. According to the Conference
Board survey, 83 percent of firms used such
guides. In the firm we studied, recom-
mencled salary increase ranges were a func-
OCR for page 26
26
tion of (1) merit rating on one axis ant! (2)
current position in the salary range on the
other axis. Higher ratings and less pene-
tration into the salary range at a particular
level were associated with both larger and
more frequent salary increases.4
Promotions are based on performance as
well, and acIditional consideration is given
to years of experience with the firm. A
salary increase goes with promotions. Again,
the corporate compensation department is-
sues yearly guidelines that specify the size
of promotional increases. The importance
of examining the firm's promotion system
is increased by the fact that the firm engages
in a fairly strong practice of promotion from
within. Thus, women's access to higher level
jobs is most often governed by decisions
made while they are current employees.
Direct access from the external labor market
is limited.
Performance is assesses] through a formal,
annual performance appraisal process.s The
immediate supervisor rates each employee
on a 4-point scale, with 4 being the highest
performance level. Raters are instructed to
consider not only how well job requirements
were satisfied, but also the difficulty of the
job requirements and the appropriateness
of methods used to satisfy the job require-
ments. The numerical rating given is sup-
plemented by a written description of the
subordinate's performance during the year.
The complete appraisal is typically reviewed
by a higher level manager.
According to The Hay Group, the actual median
increase nationally for 1982 ranged from 9 percent for
"average performers" to 14.3 percent for "outstanding
performers." By 1986, the corresponding figures were
approximately 6 percent and 9 percent, respectively
(The Hay Group, 1986~. The corresponding figures for
the salary guides used in the firm were similar.
Main (1986) has argued that supervisory ratings of
performance are not "admissible" because they "might
reflect discrimination." The empirical evidence does
not support this hypothesis, however, despite the fact
that a large amount of both laboratory and field research
has been devoted to this question (see Dipboye, 1985,
for a review).
PAY EQUITY: EMPIRICAL INQUIRIES
The metho(1 of appraisal, as well as its
central role in making promotion and salary
increase decisions, is consistent with the
way firms typically operate with respect to
their exempt employees. The review system
is also a common feature of performance
appraisal plans.6
In regard to equal employment oppor-
tunity (EEO), the firm's standard training
for its managers includes materials on EEO
compliance with respect to staffing, access
to training, compensation, and performance
appraisal. The inclusion of EEO issues in
training programs is typical of large firms.7
Corporate personnel monitor managers' ac-
tions in these areas and encourage improve-
ment in those displaying subpar perfor-
mance.
Several external events may have also had
an elect on the firm's human resource prac-
tices. First, as with many large firms, it
faced EEO litigation during the 1970s. The
major case involvecI a cIass-action suit filed
by female hourly employees in a very small
number of the approximately 100 plants.
This suit was eventually settIe(1 out of court
for a substantial sum of money. As part of
the settlement, the firm stated that it would
also develop a plan to enhance the hiring
ancI promotion of female salaried employ-
ees. The litigation, however, dicI not involve
female salaried employees. Further, no for-
mal or numerical goals for salaried women
came about as a result of the litigation
activity.
A second external influence was the reces-
sion of the early 1980s. Again, as with many
other large firms, the firm we stuclied re-
6See the survey results reported in footnote 3 on
promotion and salary increases. The same survey also
indicates that rating scales and essays are the most
commonly used methods of appraisal. Further, 97
percent of firms review appraisals at a higher level.
iA Bureau of National Affairs (1985) survey found
that among firms with over 1,000 employees, over 60
percent included EEO issues in their manager training
programs. Further, EEO was the fourth (of 19) most
commonly included issue in such programs.
OCR for page 27
SALARIES, SALARY GROWTH, AND PROMOTIONS
duced the size of its white-colIar work force
during this period. Much of this reduction
came about through early retirement. As a
result, the number of men in exempt jobs
actually declined slightly over the course of
the stucly. In contrast, the number of women
in exempt jobs grew by over 50 percent
cluring the same period.
METHOD
Sample ant! Measures
This study involves two samples: exempt
employees in job levels 1 through 7 (1)
present in 1986 (the cross-sectional sample)
and (2) present in both 1980 an(l 1986 (the Analyses
longitudinal sample). The majority of ex-
empt employees are employee] in these lev-
els (approximately 84 percent of men, 97
percent of women).8 The cross-sectional
sample comprises 2,412 women and 9,647
men. The longitudinal sample comprises
840 women and 5,550 men.
The depenclent variables are annual sal-
ary, salary growth, and promotions.9 The
first set of independent variables, referred
8The firm did not provide data on former employees
or on active employees in levels 8 through 15.
9The number of promotions is defined using salary
increase codes, which indicate the reason for an in-
crease (e. g., merit or promotion). The number of
promotional increases between 1980 and 1986 is used
as the definition. A promotion can occur without a
change in job level. Similarly, job level can change
in the absence of a promotion. Thus, the correlation
between changes in job levels and the number of
promotions is high but not perfect (r = . 731. Although
some evidence suggests that women receive more
within-level promotions (Flanders and Anderson, 1973;
Stewart and GudyLunst, 1982), this correlation did not
differ by gender, which suggests that changes in job
level and the number of promotions, as defined by
the firm, were related in a similar manner for both
men and women. Moreover, the female/male ratio of
change in job levels was virtually identical to the
female/male ratio of the number of promotions (1.33
and 1.38, respectively). Thus, we chose to use the
firm's definition of a promotion rather than attempt
to draw inferences from changes in job level.
27
to as human capital (HC) variables in this
study, are firm tenure, it job tenure (years
at a particular job level), potential experi-
ence (age minus years of schooling minus
six), and education dummies for highest
degree. Squarect terms for job tenure, firm
tenure, and potential experience are also
includecI. In the cross-sectional analyses,
the most recent performance rating prior
to the most recent salary change is used.
In the longitudinal analyses, the average
performance rating over the 1980-1986 time
frame is used. The variables used are sum-
marizec] in Appendix A.
The following salary equation is estimated
separately for the years 1980 and 1986:
In(Si~) = XilBt + eat (1)
where In(Si`) is a vector of the natural log-
arithm of salaries for i persons cluring time
period t; Xi~ is a matrix of observations on
the exogenous variables contained in Ap-
pen(lix A; B is a coefficient vector; and e is
a disturbance term composed of all un-
measured causes of salaries.
Given the availability of longitudinal data,
we also estimate the following salary growth
equation:
In(Si~986/Si~98o) = Xi~980B + ei~g80 (2)
Thus, salary growth is definecl as the
natural logarithm of the ratio of 1986 salary
divided by 1980 salary.
Finally, we estimate a similar equation
for the number of promotions received clur-
ing the 1980-1986 period:
PROM = Xi~g80B + ei~g80. (3)
infirm tenure is based on the date used for cal-
culating benefits. It is important to note that this date
can differ from the original hiring date. Therefore,
this measure should give an accurate indication of the
amount of actual time spent with the firm even for
persons not continuously employed with the firm.
OCR for page 28
28
The salary growth and promotion equa-
tions provide a unique opportunity to study
the attainment of men and women over
time. These analyses may help explain the
process by which men and women reach
the differential levels of attainment so wide-
ly observed in cross-sectional research.
Estimates of the three equations were
obtained separately for men and women and
used to decompose salary, salary growth,
ant] promotion differences into two com-
ponents (Blinder, 1973; [ones, 1983~: (1)
differences in mean levels of endowments
and (2) differences in coefficients or prices
received for those endowments. Because
the result of a decomposition varies as a
function of which group is used as the
standard (Cain, 1986), we report decom-
positions using both the advantaged and
disadvantaged group as the stanciard. In
a(l(lition, we report corresponding "a(ljusted
ratios" (Cain, 1986:746~.
As ~liscussed by Blinder (1973), Cain (1986),
Oaxaca (1973), and others, such decompo-
sitions assume that the variables on the
right side of the equation are exogenous to
tithe inclusion of individual-specif~c intercepts in
Equations (2) and (3) could be used to eliminate bias
due to any lack of independence between X time-
invariant components of e (Mundlak, 1978~. Given data
at two points in time, the use of individual-specif~c
intercepts is equivalent to a first-di~erencing model.
There are at least two problems with this model,
however. First, variables that do not change over time
(e.g., firm tenure) must be excluded. Second, the
model exacerbates any unreliability problems. As a
result, differences in parameter estimates may stem
from unreliability rather than elimination of the effects
of nonindependence of X and e. In this analysis, for
example, performance rating is a key variable. King
et al. (1980) have estimated the upper-bound reliability
of supervisory ratings to be approximately .60. In our
analysis, the correlation between performance ratings
in 1980 and 1986 is approximately .20. Using a formula
given by Guilford (1954:394), the resulting reliability
of the change in performance rating would then be
approximately .50. Across adjacent years, the corre-
lation between performance ratings is closer to .40,
which results in a difference score reliability of .33.
By averaging performance ratings and counting pro-
motions over time, the reliability problem is reduced.
PAY EQUlf Y: EMPIRICAL INQUIRIES
gender. If not, additional equations for these
right-sirle variables can, in theory, be a(lde(1
to the model. In the present context, for
example, an equation for job level might
be warrantecl. As Blinder argued, however,
implementing this latter strategy is often
clifficult in practice because of identification
nrohlems. Consistent with this point, we
found the determination of job level ant]
salary to be so closely intertwined that iden-
tification of a two-equation mode] was not
possible.
As an alternative, we follow the Blinder
and Oaxaca approach of estimating a series
of equations, introducing variables of more
questionable enclogeneity in steps to the X
vector. Thus, in the cross-sectional salary
analyses, for example, we introduce job
level last. This strategy permits an exami-
nation of how the decomposition results
change in response to different model spec-
ifications.
RESULTS
Table 1-1 reports mean salaries for all
employees active in 1986 and for employees
active in both 1980 an(l 1986. The female/
mate salary ratio is somewhat higher based
on employees active in both 1980 and 1986.
One explanation may be the fact that this
latter group does not include new entrants
to the jobs. New entrants are more likely
to begin in lower job levels and have less
time accumulated at each level. Both factors
contribute to Tower pay relative to higher
tenure employees. The larger growth of
women's employment in the firm's exempt
jobs relative to that of men suggests that
most new entrants were women.
Table 1-2 reports mean 1980 and 1986
salaries and their ratio as a function of gender
and 1980 job level. These results allow
comparisons between men and women start-
ing at the same level in 1980 (but not
necessarily at the same level as of 1986~.
Overall, the ratio of women's salary to men's
salary is .84 in 1980 and rises to .88 in 1986,
OCR for page 29
SALARIES, SALARY GROWTH, AND PROMOTIONS
TABLE 1-1 1986 Salaries of Men ancI Women, by 1986 lob Level
1986 Cross-Sectional Sample
29
1986 Longitudinal Sample
1986 Women Men Women Men
Job Mean Mean Mean Mean
Level N Salary N Salary W/Ma N Salary N Salary W/Ma
All 2,412 $35,503 9,647 $42,049 .84 840 $40,004 5,550 $45,620 .88
1 859 29,451 1,777 31,875 .92 114 31,092 386 32,519 .96
2 412 32,870 896 35,222 .93 156 34,675 325 35,869 .97
3 521 36,209 2,151 37,795 .96 174 37,697 1,032 38,789 .97
4 162 40,745 641 42,935 .95 88 41,589 468 43,584 .95
5 286 43,925 2,184 46,307 .95 182 44,274 1,617 46,819 .95
6 158 50,568 1,859 53,410 .95 109 51,067 1,607 53,557 .95
7 58,968
14 55,415 138 59,002 .94 17 55,805 114 .95
aW/M = women/men ratio.
thus eliminating 25 percent of the salary
differential. Consistent with this narrowing
differential, the mean salary for women in-
creased by a greater percentage (61 percent)
between 1980 and 1986 than did the mean
salary for men (54 percent).
Within 1980 job levels, similar trends
emerge, although the salary differentials are
much narrower. The ratios of women's mean
salaries to those of men range from .93 to
.95 in 1980. In all cases, these ratios in-
creased between 1980 and 1986. Again, the
decreasing salary differentials are consistent
with the greater salary growth of women
observed at each 1980 job level. A cursory
examination of the 1980 job levels having
sufficient numbers of both men an(l women
(1 through 3) reveals no obvious relationship
between job level and gencler differences
in salary or salary growth.
One possible concern with the numbers
in Tables 1-1 and 1-2 is that there may be
a selection process related to gentler and
salary because the longituclinal sample in-
cludes only employees still active in 1986.
For example, one scenario is that men were
observed to receive lower salary increases
because men experiencing larger salary
growth were more likely to have left the
firm. Alternatively, discrimination against
women could have resulted in all but the
"cream of the crop" quitting, leaving us to
TABLE 1-2 1980 and 1986 Salaries of Men and Women, by 1980 Job Level
1980 Women Men Women/Men
Job 1986a 1986a 1986
Level N 1980 1986 1980 N 1980 1986 1980 1980 1986 1980
All 840 $24,786 $40,004 1.61 5,550 $29,606 $45,620 1.54 .84 .88 1.14
1 348 21,954 35,762 1.63 872 23,516 36,479 1.55 .93 .98 1.14
2 177 24,014 38,862 1.62 774 25,171 40,009 1.59 .95 .97 1.05
3 195 26,635 43,048 1.62 1,515 28,220 44,229 1.57 .94 .97 1.09
4 48 28,881 46,760 1.62 485 30,848 47,661 1.55 .94 .98 1.14
5 58 31,894 50,042 1.57 1,307 33,801 51,346 1.52 .94 .97 1.10
6 14 35,717 52,768 1.48 571 37,705 55,225 1.46 .95 .96 1.03
7 0 26 36,840 54,321 1.47
al986 salary/1980 salary.
bWomen tsalary 1986/salary 1980] 1/men [salary 1986/salary 1980] - 1.
OCR for page 30
30
observe only the latter group, which re-
ceived relatively large increases.
Supplementary data suggest, however,
that these are not likely problems. First,
the voluntary quit rate averaged only about
2 percent per year over the course of the
study. Second, anti more important, we
examined salary data on all men and women
active in 1980 or 1984. The average salary
of men and women grew by 35 percent and
40 percent, respectively, during this period.
Restricting the sample to only employees
active in both 1980 and 1984 yielded growth
rates for men and women of 46 percent and
53 percent, respectively. The key point is
that the femaTe/male ratio of growth rates
was approximately 1.14 in both cases, which
suggests that focusing only on employees
active in both years does not influence the
observer] relative salary growth of men and
women. Finally, note that the ratio of 1.14
is identical to that found in the first row,
last column of Table 1-2.
Decomposition of Salary Differences
As Table 1-3 indicates, human capital
variables alone explain up to 32 percent of
the salary advantage of men in the cross
section. Adding job level and performance
rating to the mode} raises the explaine(l
percentage to between 68 and 77 percent,
which suggests that most of the gap is due
to the fact that men tend to hold higher
level jobs. The corresponding adjusted sal-
ary ratios (AM and AW) suggest that equal-
izing human capital would raise the salary
ratio to .86 to .90. Equalizing job level and
performance rating as well would raise it
to .95 to .97. 12
Note that adding performance rating ac-
tually results in a slight decrease in the
explained percentage of the salary cliffer-
i2Within levels, very little of the pay gap could be
explained. Because, however, the raw ratios were high
(see Tables 1-1 and 1-2), the adjusted ratios were also
high.
PAY EQUITY: EMPIRICAL INQUIRIES
ential. Although not shown here, this find-
ing was even stronger in analyses of specific
job levels. The reasons are twofold. First,
the mean performance rating of women is
slightly higher than that of men (2.59 versus
2.52 overall in 1980~. Second, although
women receive a slightly greater return for
a given performance rating, inclu(ling per-
formance rating in the mo(lel changes some
of the other coefficients. 13
In summary, the cross-sectional results
suggest that men, in general, receive greater
returns to explanatory variables. These re-
sults are consistent with previous cross-
sectional analyses of male-female salary clif-
ferentials (Cain, 1986~.
Decomposition of Salary Growth
Differences
Although women's salary levels fell short
of men's, women, as noted earlier, expe-
rienced greater salary growth in percentage
terms. Moreover, as the decompositions
reported in Table 1-4 (remonstrate, this
salary growth advantage cannot be entirely
explaine(1 by the mo(lels. The parentheses
in Table 1-4 in(licate instances in which
(differences in coefficients favor women. Ap-
pen(lix B reports the regression results.
In the overall analyses, 40 to 49 percent
of women's greater salary growth can be
explaine(l by differences in human capital.
Adding average performance rating raises
the explained part to 50 to 62 percent.
Finally, the inclusion of job level and num-
ber of promotions raises this figure to 72
to 74 percent. Within specific job levels,
there is a good clear of variance in the extent
to which human capital, average perfor-
mance rating, an(l promotions can explain
the salary growth advantage of women.
Table 1-5 shows the contribution of spe-
cific factors to the salary growth (lifferential
i3For example, adding performance rating to the
equation having human capital variables increased the
advantage realized by men in returns to job tenure.
OCR for page 31
SALARIES, SALARY GROWTH, AND PROMOTIONS
TABLE 1-3 Decomposition of Salary Differences
Decomposition Standard
R2 Men Women Differ'?ntin
31
Variables in Equation Women Men Coeff.a Endow.b Coeff.a Endow.b Raw ARC AWE
1980 Longitudinal Sample
Human capital (HC) .19 .29 .73 .27 .86 .14 .84 .88 .86
HC, perf. rating 80 (PA80) .21 .33 .76 .24 .88 .12 .88 .86
HC, job level .58 .71 .28 .72 .31 .69 .96 .95
HC, job level, PA80 .62 .73 .31 .69 .32 .68 .95 .95
N = 840 5,550
1986 Longitudinal Sample
HC .29 .21 .80 .20 1.00 .00 .88 .90 .88
HC, PA86 .32 .28 .82 .18 1.01 -.01 .90 .88
HC, job level .79 .80 .21 .79 .29 .71 .97 .97
HC, job level, PA86 .82 .83 .23 .77 .30 .70 .97 .96
N = 840 5,550
1986 Cross-Sectional Sample
HC .30 .31 .68 .32 .73 .27 .84 .89 .88
HC, PA86 .33 .37 .70 .30 .75 .25 .89 .88
HC, job level .81 .83 .25 .75 .22 .78 .96 .96
HC, job level, PA86 .83 .85 .27 .73 .24 .76 .96 .96
N = 2,412 9,647
aProportion of salary differential due to unequal coefficients or returns.
bProportion of salary differential due to unequal endowments.
CAdjusted salary ratio using men's coefficients as standard.
Adjusted salary ratio using women's coefficients as standard.
in the mode! containing only human capital
variables. An important factor accounting
for women's advantage is potential experi-
ence. Because potential experience has a
negative impact on salary increases, women
benefit from having lower levels of potential
experience and having a less negative coef-
ficient. 14 Similarly, with respect to job level,
women benef~tec] from having a lower mean
and a less negative return. The addition of
promotion to the mode} sharply reduces the
14The use of potential experience (age minus school-
ing minus six), rather than actual experience, is prob-
lematic for persons with intermittent labor force par-
ticipation. Thus, the role of potential experience in
explaining salary growth differences in our study may
be partly artifactual. As an indirect test, we restricted
the sample of women to unmarried women only. The
resulting decomposition (of the full model) actually
increased the importance of potential experience in
explaining women's faster salary growth.
importance of potential experience and job
level in accounting for the differential.
Another important factor is job tenure. 15
Women received larger percentage salary
increase returns to job tenure. Recall that
the firm's salary increase guicle recommends
smaller percentage increases as employees
progress within the salary range at a given
level. Because this position in the salary
range is not likely to be a perfect function
of job tenure, the latter may not completely
capture the effect of current position in the
salary range. If men tend to be higher in
the salary range than women, we would
expect men to receive smaller increases than
women.
Women's higher salary growth was also
MA more flexible functional forn1 (dummy variables
for each year of job tenure) did not change this con-
clusion.
OCR for page 32
32
PAY EQUITY: EMPIRICAL INQUIRIES
TABLE 1-4 Decomposition of Salary Growth Differences, by 1980 Job Level
Decomposition Standard
R2 Men Women Differential
Variables in Equation Women Men Coeff.a Endow.b Coeff.a Endow.b Raw AWC AMP
All Levels
Human capital (HC) .14 .21 (.51)e .49 ( 60) .40 1.14 1.07 1.06
HC, avg. perf.
rating (AVGPA) .23 .30 (.38) .62 (.50) .50 1.09 1.07
HC, job level .15 .21 (.46) .54 (.53) .47 1.08 1.07
HC, job level, AVGPA .24 .31 (.25) .75 (.42) .58 1.11 1.08
HC, job level, AVGPA,
promotion (PROM) .39 .43 ( 26) .74 (.28) .72 1.10 1.10
Level 1
HC .16 .29 (1.14) - .14 (1.02) - .02 1.14 .98 1.00
HC, AVGPA .23 .43 ( 51) .49 (.80) .20 1.07 1.03
HC, AVGPA, PROM .35 .57 (.67) .33 ( 77) .23 1.05 1.03
Level 2
HC .26 .30 (.50) .50 (1.10) - .10 1.05 1.03 1.00
HC, AVGPA .37 .34 - .13 1.13 ( 46) .54 1.06 1.03
HC, AVGPA, PROM .56 .45 - .28 1.28 ( 10) .90 1.06 1.05
Level 3
HC .14 .18 (.32) .68 (.52) .48 1.09 1.06 1.04
HC, AVGPA .31 .31 (.30) .70 (.58) .42 1.06 1.04
HC, AVGPA, PROM .53 .44 (.14) .86 ( 41) .59 1 OS 1 05
NOTE: N = 840 for women, 5,550 for men.
aProportion of salary di~erentiai due to unequal coefficients or returns.
bProportion of salary differential due to unequal endowments.
CAdjusted salary ratio using women's coefficients as standard.
Adjusted salary ratio using men's coefficients as standard.
eParentheses indicate that coefficients favor women.
partly a function of their higher average
v ~
performance ratings. Nevertheless, the
coefficient on average performance rating
was smaller for women, which indicates they
received a smaller payoff for performance.
On the other hand, adding promotion to
the mocle] eliminated this disadvantage.
Decomposition of Promotion
Differences
Table 1-6 reports the mean number of
promotions received by men and women
between 1980 and 1986 overall an(1 as a
function of 1980 job level. As in the case
of salary increases, women had a (listinct
promotion advantage. Of further interest,
this advantage, like the women's salary growth
advantage, does not decline at higher job
levels. In fact, these simple descriptive sta-
tistics suggest that the advantage may ne
larger at higher job levels.
Table 1-7 presents the results of the de-
composition of the promotion cTi~erential
under (li~erent model specifications. (Ap-
pendix B presents the regression results.)
A key finding is that the conclusion re-
garding the final model depends on which
coefficients are used as the stan(lard. When
men's coefficients are used as the standard,
the promotion advantage of women is com-
pletely explained by the variables in the
final motlel. Table 1-8 presents results of
the decomposition into specific factors. In
the mo(le} containing human capital vari-
ables, the main advantage of women is again
OCR for page 33
33
o
Cot
C)
C:
an
Cal
o
o
Cat
Cat
V)
Cat
Cat
3
m
C)
a'
of
._
3
o
5-
C)
Cad
t1 ~
to
._
V,
o
CO)
a:
o
o
~ o
Pa C)
~ C
au ~
C)
c)
o
Cat
E-°
,- a;
a.' of
c 3
=~
~ JO
~ .
~ o
C)
Cut
o
EM
o
3
JO
~ o~
~ m~
~ - ;
C) ~
o
Cat
E-°
.=
c~ ~ oo cM o ~ u) o c~ 'o
cl
1 1
cM cM ~ o
;,
1
1
o 'o x o o oo
~ 1 ~ c~
o ~ ~ ~ x - c~ ~ c~
1 1
c~ cN o
~ 1 '~
_ 1
1
~ ~ ~ c~ ~
1 °° 1 _
o co
1 ~ ~' 1 ~ C~ c~
1
1
1
cO cs ~ o cO
- - - ~
oO
o x
1 ~
~ c~ o o c~ -
I=1 ~ 1 X
1 ~ 1 _
1
oo o oo
IO CD _
c~ 1 c~
1
_ C~ o
C~ _ o
1 _
u)
CO CO _ o
1 '~ ~
oo C~ o
~o ~ ~ 1 '~ '=
1
CM CD O
1 ~
Xo ~ ~ ~ I ~ 8
1 ~ _
bC
._
C~
~ _ Z ~ ° ~ ~ ~
V,
-
C~
~o
Ct
cr
o
s~ ~
C~ ~ i
C~ ~
V~ C
c:
o
._
c,, 3
o o
~ C~
o ~
v' ~
~ o
~ o
o c)
c: ~
c, ~
OCR for page 38
38
(1987), among others, have argued that the
adjusted female/maTe salary ratio might be
closer to unity if unmeasured differences
between men and women and relevant labor
markets could be better incorporated in
salary equations. We were able to move in
this direction by including, for example,
several experience measures, job level, and
performance rating in our moclels.
Our results, however, indicate that at
least in the case of performance rating, its
inclusion actually led to a slight reduction
in the adjusted female/male salary ratio. In
the promotion and salary growth equations,
in which the unadjusted ratios exceeded
unity, the inclusion of average performance
rating lee! to sometimes sizable increases in
adjusted ratios. Thus, the common thread
is that inclusion of performance rating does
not help explain the raw salary advantage
of men, but it does explain some of the raw
advantage of women with respect to salary
growth and promotions.
Although women had an overall advan-
tage in the salary increase and promotion
process, they receive(1 a smaller payoff for
their performance ratings. Based on the full
promotion equation, for example, men re-
ceived an average of .45 additional pro-
motions for each additional point in average
performance rating between 1980 and 1986.
In contrast, women received an additional
.33 promotions for each adclitional point.
This finding is consistent with Olson and
Becker's (1983) suggestion that women may
be held to higher promotion standards than
men.
The fact that women experienced a lower
salary disadvantage in the cross-sectional
analysis yet received better salary increases
and more promotions over time offers an
interesting contrast. One implication may
be that because men are not favored in the
salary increase and promotion process, the
explanation for women's salary level dis-
advantage must be sought elsewhere. If so,
one alternative avenue of investigation shoul
PAY EQUITY: EMPIRICAL INQUIRIES
perhaps be the recruitment and initial place-
ment of men and women.
Aside from possible differences in initial
placement, it may be that women have not
always been favored in terms of salary in-
creases and promotions. In fact, the greater
number of promotions and larger salary
increases of women may reflect an attempt
by the firm to reduce what was perceives!
to be an inequitable salary and job level
structure. Rosenbaum (1985), for example,
found a reduction in the negative impact of
percentage female on salary between 1965
and 1975. Further, he found that the impact
of percentage female on promotions changed
from negative to positive over the same
period. In the case of the Rosenbaum stu(ly,
the firm had implemented a "serious" af-
firmative action program during the period
of the study.
The firm that we studied ha(l an ongoing
affirmative action plan. It also covere(l EEO
issues in its management training. Further,
although EEO litigation pertaining to these
issues among exempt employees does not
appear to have been important, the firm
did settle a case pertaining to such practices
vis-a-vis hourly employees for a substantial
amount. Although no formal goals or new
practices resulted, one might speculate that
this event enhanced or at least reinforced
the vigilance with which progress toward
affirmative action and EEO goals was mon-
itore(l.
Any progress in the affirmative action area
may have been facilitated by the general
financial success of the company during the
periocl of our study. Research by Rosen-
baum (1979) suggests that promotion op-
portunities may be greater during periods
of organizational ((lefine(1 as employment)
growth. The general financial success an
growth of the firm we studied may have
facilitated affirmative action progress of fe-
male employees. It is interesting to note,
however, that the bulk of employment growth
in exempt jobs seems to have been among
OCR for page 39
SALARIES, SALARY GROWTH, AND PROMOTIONS
women rather than men. i7 As cliscussed
earlier, this appears to have been partly
due to a brief period of reductions in force
during the early 1980s.
The promotion and salary growth advan-
tage of women implies that the salary gap
would eventually remedy itself if past trends
were to continue. Note, however, that even
a small initial salary clisadvantage can take
many years to be eliminated. As an example,
a projection of salary growth rates over the
6-year period into the future indicates that
the 1980 female/male salary ratio would not
equal unity until the year 2003. Within job
levels 1, 2, and 3, 1980 salaries would
equalize in 1989, 1996, and 1992, respec-
tively. Of course, as this equalization pro-
cess works to its conclusion, women con-
tinue to receive lower salaries.
In addition to our focus on human capital,
job level, and performance ratings, we also
examined the possibility that percentage
female in a job group was a structural factor
contributing to attainment differences. Con-
sistent with Rosenbaum (1985), controlling
job level generally reduced the impact of
percentage female by a substantial amount.
Consistent with Hartmann (1987), percent-
age female had a small, positive effect on
women's salaries (in 1986~. Unlike Hart-
mann's results, however, percentage female
had a negative impact on salaries of men.
With respect to salary growth, the lack of
a negative impact of percentage female for
both men and women was consistent with
Hartmann's findings. Regarding promotion,
although Hartmann found a negative impact
of percentage female for women, Rosen-
baum found a positive impact by the end
of his study period. Our results suggest no
stable impact of percentage female on wom-
i'This is not to say that there were not also a
substantial number of newly entering men as well.
The net growth in the number of women seems to
have been higher because there were fewer women
in exempt jobs at retirement age, for example.
39
en's promotion rates. In contrast, a positive
effect was found for men.
The preceding discussion does not really
provide support for the idea that percentage
female is an important structural property
that negatively affects women's (and perhaps
men's) attainments. A better research strat-
egy would be to examine its impact con-
trolling for other characteristics of jobs or
occupations that may be related to per-
centage female. This strategy was demon-
stratec] by Treiman and Hartmann (1981)
using national survey data at the occupa-
tional level. A stronger test, however, would
make use of firm-level data wherein simi-
larity of occupational titles is more likely to
correspond to actual similarity of work con-
tent. The effect of percentage female in
different firms is another possible avenue
of investigation (Pfeffer and Davis-Blake,
1987) in cases in which job content is stan-
lar(lized.
Given our use of data from a single firm,
the study has a disadvantage relative to the
coverage en cl external valiclity possible with
marketwide or national surveys. Along these
lines, replication studies would be necessary
before attempting to answer the question
of how typical our results are of other large
firms' relative treatment of men and women.
Nevertheless, the better coverage obtained
using national survey data comes at the
expense of not being able to measure pro-
ductivity in the way that many firms actually
measure it. Finally, recall that the policies
anti practices of the firm we studie(l ten(le
to be consistent with those reported in
surveys of other large firms.
Besides the job-specific productivity mea-
sures, the results of our study were strength-
ened by the following. First, the use of
longitudinal (lata should have recluced the
impact of any unobserved, constant incli-
viclual differences in procluctivity. Second,
firm-specific differences in (leterminants of
salaries (e.g., pay policies and practices) are
obviously not an issue. Finally, the use of
OCR for page 40
40
data from personnel records rather than self-
reports typical of national surveys may have
reduced the potential for reporting errors
(Duncan and Hill, 19851.
In their agenda for basic research on
comparable worth, Hartmann et al. (1985:7)
emphasizer] that "we need to understand
better how wages are set within enterprises
and how they are affected by other employer
practices, such as job assignment...." They
stressed the importance, especially in large
firms, of internal labor markets and pro-
motion from within as aspects of job as-
signment. We hope that our research con-
tributes to a better understanding of these
processes.
ACKNOWLEDGMENTS
We would like to thank Donald Schwab,
the other members of the Panel, the other
authors, and our colleagues at Cornell Uni-
versity for their helpful comments on earlier
drafts of this chapter.
REFERENCES
BielLy, W. T., and J. N. Baron
1986 Men and women at work: Sex segregation
and statistical discrimination. American Jour-
nal of Sociology 91:759-799.
Blinder, A. S.
1973 Wage discrimination: Reduced form and
structural estimates. Journal of Human Re-
sources 8:436-455.
Brown, C.
1982 Estimating the determinants of employee
performance. Journal of Human Resources
17:178-194.
Bureau of National Affairs, Inc.
1983 Performance appraisal programs. Personnel
Policies Forum, no. 135.
1985 Training and development programs. Per- Jones, F. L.
sonnet Policies Forum, no. 140. 1983
Cain, G. G.
1986 The economic analysis of labor market dis-
crimination A survey. Pp. 693-785 in O.
Ashenfelter and R. Layard, eds., Handbook
of Labor Economics. Amsterdam: North-
Holland.
PAY EQUITY: EMPIRICAL INQUIRIES
Conference Board
1984 Pay and performance: The interaction of com-
pensation and performance appraisal. Re-
search Bulletin, no. 115. New York.
Dipboye, R. L.
1985 Some neglected variables in research on dis-
crimination in appraisals. Academy of Man-
agement Review 10:116-127.
Duncan, G. J., and D. H. Hill
1985 An investigation of the extent and conse-
quences of measurement error in labor-eco-
nomic survey data. Journal of Labor Eco-
nomics 3:508-532.
Flanders, D. P., and P. E. Anderson
1973 Sex discrimination in employment: Theory
and practice. Industrial and Labor Relations
Review 26:938-955.
Fuchs, V.
1971 Differences in hourly earnings between men
and women. Monthly Labor Review 94:9-15.
Guilford, Joy Paul
1954 Psychometric Methods. New York: McGraw-
Hill.
Halaby, C. N.
1979 Job-specific sex differences in organizational
reward attainment: Wage discrimination vs.
rank segregation. Social Forces 58: 108-
127.
Hartmann, H. I.
1987 Internal labor markets and gender: A case
study of promotion. Pp. 59-97 in C. Brown
and J. A. Pechman, eds., Gender in the
Workplace. Washington, D.C.: Brookings In-
stitution.
Hartmann, H. I., P. A. Boos, and D. J. Treiman
1985 An agenda for basic research on comparable
worth. Pp. 3-33 in H. I. Hartmann, ea.,
Comparable Worth: New Directions for Re-
search. National Research Council, Com-
mittee on Women's Employment and Related
Social Issues. Washington, D. C.: National
Academy Press.
The Hay Group
1986 Trends and issues in total compensation.
Compensation Quarterly October 6:3.
Hildebrand, G. H.
1980 The market system. Pp. 79-106 in E. R.
Livernash, ea., Comparable Worth: Issues
and Alternatives. Washington, D. C.: Equal
Employment Advisory Council.
On decomposing the wage gap: A critical
comment on Blinder's method. Journal of
Human Resources 18:126-130.
King, L. M., J. E. Hunter, and F. L. Schmidt
1980 Halo in a multi-dimensional forced choice
performance evaluation scale. Journal of Ap-
plied Psychology 65:507-516.
OCR for page 41
SALARIES, SALARY GROWTH, AND PROMOTIONS
Lewis, G. B.
1986 Gender and promotions. Journal of Human
Resources 21 406-419.
Malkiel, B. G., and J. A. Malkiel
1973 Male-female pay differentials in professional
employment. American Economic Review
63 693-705.
Medoff, J. L., and K. G. Abraham
1981 Are those paid more really more productive?
The case of experience. Journal of Human 1985
Resources 16 186-216.
Megdal, S. B., and M. R. Ransom
1985 Longitudinal changes in salary at a large
public university: What response to equal
pay legislation? American Economic Review
75 271-274.
Milkovich, G. T.
1980 Comparable worth: The emerging debate.
Pp. 23-47 in E. R. Livernash ea., Compa-
rable Worth: Issues and Alternatives. Wash-
ington, D.C.: Equal Employment Advisory
Council.
Milkovich, G. T., and J. M. Newman
1987 Compensation. Plano, Tex.: Business Publi-
cations.
Mundlak, Y.
1978 On the pooling of time series and cross-
sectional data. Econometrica 46 69-85.
Oaxaca, R. L.
1973 Male-female wage differentials in urban labor
markets. International Economic Review
14 693-709.
Olson, C. A., and B. E. Becker
1983 Sex discrimination in the promotion process.
Industrial and Labor Relations Review 36 624-
641.
Pfeffer, J., and A. Davis-Blake
1987 The effect of the proportion of women on
salaries: The case of college administrators.
Administrative Science Quarterly 32 1-24.
Reskin, B. F., and H. I. Hartmann, eds.
1985 Women's Work, Men's Work: Sex Segregation
41
on theJob. National Research Council, Com-
mittee on Women s Employment and Related
Social Issues. Washington, D. C.: National
Academy Press.
Rosenbaum, J. E.
1979 Organizational career mobility: Promotion
chances in a corporation during periods of
growth and contraction. American Journal of
Sociology 85 21-48.
Jobs, job status, and women's gains from
affirmative action: Implications for compa-
rable worth. Pp. 116-136 in H. I. Hartmann,
ea., Comparable Worth: New Directions for
Research. National Research Council, Com-
mittee on Women's Employment and Related
Social Issues. Washington, D. C.: National
Academy Press.
Rynes, S. L., and G. T. Milkovich
1986 Wage surveys: Dispelling some myths about
the "market wage." Personnel Psychology
39 71-90.
Sanborn, H.
1964 Pay differences between men and women.
Industrial and Labor Relations Review 17 534-
550.
Stewart, L. P., and W. B. Gudykunst
1982 Differential factors influencing the hierar-
chical level and number of promotions of
males and females within an organization.
Academy of Management Journal 25 586-
597.
Treiman, D. J., and H. I. Hartmann, eds.
1981 Women, Work, and Wages: Equal Pay for
Jobs of Equal Value. National Research Coun-
cil, Committee on Occupational Classification
and Analysis. Washington, D. C.: National
Academy Press.
Tsui, A. S., and B. A. Gutek
1984 A role set analysis of gender differences in
performance, affective relationships, and ca-
reer success of industrial middle managers.
Academy of Management Journal 27 619-
635.
Appendixes A and B follow.
OCR for page 42
42
,'
.
o
C~
U:
;~
V2
o
x
C: O CD ~ X c~ c~ ~ LO CC X ~ OC
cc X CM ~ O O ~ ~ O ~ CO ~ CO C~ G) ~ ~
0 C~ o o 0 0 )~ ~ o CD O ~ O , CS) ~ 1 CC)
. .
_ CM ~ C~ ~ ~ C~ CM Z)
C~ ~ _ ~ CO
U: C~
U) o C~ C~ ~ C~ CO o o X ~ ~ C~ ~ ~ ~ ~ C~
_ ~ C~ ~ ~ _ o C~ ~ ~ C~ ~ ~ C~ C] ~ o _
O ~ CM O O O O C~ C~ 00 ~ X CD O CC C~
........ ....... ..
X ~ o o C~ o CM C>) C~
_ ~ C~
~ ~ O C] co ~ U) CD ~ CO CD C~ oo ~ O ~ CC ~ O
CM ~ o U: U) o o CM ~ o ~ ~ ~ X U: ~ CO ~ ~
~4 0 ~ C~ O O O O ~ ~ c~ ~ 0 1 =4 ~ )~ U:
[f: ~ oo ~ _ 1 ~ c~ c~ u:^
C~ X X o ~ C~ o U: CC C~ ·c ~ o ~ o
a; ~ c~ co ~ ~ c~ 0 _ 0 ~ ~ cc c~ oo ~ _ c~ X ~
CD ~ ~ ?~ °. ° °. ? ~ ~ ° ~ ~ ~ ° 1 ~ ~ ~ X
oo 0 ~ co LO _ ~ ~ · 1 c~ c~
;> C~ C~ _
~ ~ ~ C~
_ ~ ~ O CM C~ ~ U: CC CD O CD ~ C~ C~ O ~ C~ O
C~ C~ O ~ ~ O O C~
.= ~ ~ o ~ c~ o 0 0 0 cs' ~ c~ ~ ~ c~ 0 1 ~4 U) 1
~ a; x 1~ ~ c~ c~ X 1 c~ c~ 1
~ ·~7 ~ _~_= ~
C~ CO C~ X X O ~ ~ CO ~ O ~ oo U) ~ C~ C~ O
.- ~ ~ ~ ~ ~ ~ ~ O ~ O ~ ~ ~ O ~ ~ O
bt O ~ ~ o _ ~ 0 0 0 0 ~ x 0 ~ 0 ~ 0 1 ~ ~ 1
o ~ =1 ~ ~ ~ ~ ~o ~ 1 ~ ~ 1 1
C) .'
~ .-
C~
= V 5
~ ~ ~ ~ C . V ~ ~ ~ ~ F _
OCR for page 43
43
.o~
~ 2
* * * * * * * *
* * * * * * * * * *
U) ~ CS ~ ~ o ~ ~ o ~ ~ ~ o CO ~ ;o ~
CM ~ C~ ~ ~ ;o X ~ o ~ ~ Lr: ~ oo CD CO ~ X
G: O c~ ~ O c~ O r: c5) ~ ~ O ~ 0 0 0 c~ c~
~ o o o o o ~ o ~ o o o o o o o o
1 1 1 1 1 1 1
* ** ** ****
* * * * * * * * * *
X ~ ~ ~ ~ c~ ~ ~ C~ C~ 5) G~ X ~ r~
C~ ~ X ~ O ~ ~ G. C~ CD
o CM ~ o C~ o ~ ~ o C~ o o C] o ~ ~ C~
~, o o o o o o o o o o o o ~ o o o
1 1 1 1 1 1
* ,
* * * *
C~ ~ ~ X ~ G) O ~ CM ~ CD O C~ C~
(= C~ ~ o ~ ~ CM C~ oo ~ o X U:
o ~ o C~ ~ ~ ~ C~ ~ ~ o ~ ~ o
C~ o o o o o o o U) o o o o o o
1 1 1 1 1 1 1
* -x- ~
CD
CM
* * * * * * * *
* * * * * * * * *
oo ~ ~ C~ o X ~ ~ o ~ C~ _ ~ ~ X
cc x ~ 0 ~ c, ~ co c~ X ~ ~ ~ O r~
o o ~ ~ o CM ~ ~ o ~ o o o ~ C~
C~ o o o o o o o o o o o o X o
1 1 1 1 1 1 1 1 1
U:
o
·~,
U:
;~
bt
o
·~,
o
?
~ o
~? =
C) ~
~ o
U: o^ o X
~ o X ~ X ~
~ ~ ~ ~ X ~ ~ v ° O -
* * * *
* * * * *
0 ~ ~ ~ ~ ~ [Q ~ cc ~ ~ cs
r~ ~ 0 0 c~ _ c~ _ ~ _ ~ c
_ 0 0 0 0 0 0 0 ~ 0 0 c
. . . . . . . . . . . .
O 0 0 0 0 0 0 0 0 0 0 c
1 1 1 1 1 1 1 1
* * * *
* * * * * * *
0 _
_ 0 0 0 ~ ~ c~ ~
c~ 0 0 0 0 0 0 0 c~:
0 0 0 0 0 0 0 0 c
1 1 1 1 1 1
* *
G) CD O O _ _
o ~ O O oC CD
0 _ 0 0 0 0
0 0 0 c 0 0
1 1
* * * *
* * * * *
0 x _ ~ c~ c: _ ,_
_ 0 0 0 L~ 0 0 x
0 0 0 0 c~ 0 0 0 0
. . . . . . . . .
O 0 0 0 0 0 0 0 0
1 1 1
* * * *
* * * * *
u: ~ ~ ~ c~ _ _ ~ ~r ~ c~ c~ X ~ ~ ~ ~
x ~r 0 0 ~ 0 ~ cc ~ c~ c~ 0 _ ~ 0 0 0
c~ 0 0 0 0 0 0 0 ~ 0 0 0 0 _ 0 0 _ c~
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 1 1 1 1 1 1 1 1
* * * * *
* * . * * * * *
0 ~ ~ ~ c~ cc ~ _ _ ~ ~ 0 c>;
_ c~ 0 0 c~ c~ ~ 5, 0 0 0 c
~ 0 0 0 0 0 0 0 c~ 0 0 c
0 0 0 0 0 0 0 0 0 0 0 0 c
1 1 1 1 1 1 1
* ~
* * *
0 ~ co _ ~ c~ 0 u:
0 _ Lf: c~ cc cc
0 0 0 0 0 0 c
. . . . . . . .
O 0 0 0 0 0 0 c
1 1 1 1 1
c~
*
.. * * *
co ~ ~ c~ c~
c~ c~ c~ 0 c: ~ 0 ~
5) 0 0 0 0 _ 0 _
0 0 0 0 ~o 0 0
1 1 1 1 1
* * * * * *
* * * * * * * * * *
'3) ~ ~ ~ ~ ~ _ O CO _ ~ O ~ 53 G)
c~ _ 0 _ c~ ~ ~ ~ c~ _ 0 0 0 c~ 0
cc 0 0 0 0 0 0 0 c~ 0 0 0 0 x 0
. . . . . . . . . . . . . . .
O 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 1 1 1 1 1 1 1 1
.~
c~
. _
0= 0
V V
=, -x~
* *
OCR for page 44
Commentary
CHRISTOPHER WINSHIP
Gerhart and Milkovich provide an anal-
ysis of sex differentials in salaries, raises,
and promotions in a large, diversified firm.
Their sample is confined to employees in
the 7 lowest of 15 job levels for exempt
personnel. The sample is further restricted
to individuals working in the firm in 1980
who were still working with the firm in
1986. Their basic strategy is to estimate a
large variety of regression models, almost
exclusively of the single-equation type, with
the aim of assessing what factors account
for sex differences in salaries, raises, and
promotions. The analyses here are quite
straightforward. They carry out the usual
decomposition procedures for determining
how much of the sex difference in the cle-
pendent variable, that is, salaries, raises,
or promotions, can be explained by maTe-
female differences in the in(lepen(lent vari-
ables and how much can be explained by
differences across sex in the effects of these
variables on the depen(lent variable.
A number of Gerhart and Milkovich's
findings are of interest. The firm they study
The author benefited from a conversation with Dale
Mortensen in preparing these comments.
44
is unusual in that sex differences in salaries
are relatively small. In 1980 the ratio of
female to male salaries was .84. In 1986 the
ratio was .88. As they point out, these ratios
are among the highest in literature. They
also find that over the 6-year period, wom-
en, on average, received higher raises, at
least on a percentage basis, than men. Over-
all, female salaries increase(1 61 percent over
the 6-year period whereas mates' increased
54 percent. Promotions show a similar sex
(difference. From 1980 to 1986, women re-
ceived an average of 1.29 promotions and
men an average of .94. Put in other words,
women received, on average, nearly 40 per-
cent more promotions than men.
Gerhart an(l Milkovich also carry out
breakdowns by job level. As one might
expect, within job level the female/male
salary (differences are much smaller. For
instance, in 1980 they range from .93 to
.95. The overall (differences between women
and men in size of raises and number of
promotions hold within levels as well as
across the population. An important impli-
cation of this is that differences in the dis-
tribution of men and women across levels
does not appear to explain the larger raises
OCR for page 45
COMMENTARY
ancl higher rates of promotion for women.
Gerhart and Milkovich also examine the
effect of sex composition of jobs on salaries.
In many of their models they find that the
effect of percentage female on the salaries
of women is statistically insignificant. The
sign of the effect also changes across models.
In sharp contrast, however, men are pen-
alized in both 1980 and 1986 for holding
jobs that are predominantly hell] by women.
For example, in 1980, they estimate that
for a man, a change from a completely male
to a completely female job would be as-
sociatecI with an 8 percent lower salary.
It seems to me that one of the advantages
of studying a single firm is that it provides
the opportunity to think in detail about
what kinds of processes affect salaries, rais-
es, and promotions. It gives us a chance to
bring our general knowledge about how the
world works in a specific situation to our
statistical analyses.
The authors, however, tell us almost noth-
ing about the firm they are examining. In
evaluating the analyses, I would have found
it quite useful to know about the kinds of
jobs in the sample and, in particular, the
kinds of jobs that were associated with the
seven job levels they analyze. It would also
be quite useful to know something about
the organization of the personnel system
within the firm. lob level is a variable
frequently used in the analysis, but at the
end of the paper I was not quite sure what
it meant. I was particularly puzzled when
I read a footnote that an individual can be
promoted without changing job level and
also can change job level without being
promoted. It would also be valuable to know
something about the structure of the firm.
Do the jobs in the sample fall into different
divisions or departments? Are men and
women clifferentially distributed across these
units and does this have any effect on salary?
These considerations might also refocus
the analysis a bit. One major factor in de-
termining sex differences in salaries is dif-
ferences in the type of jobs men and women
45
hold, but this paper offers only a very partial
analysis ofthis factor. Gerhart and Milkovich
do examine the effects ofthe sex composition
of jobs on salaries, but they go no further
than that.
Perhaps the most serious variable that is
not even discussed is differences in hours
worked, or how demanding a job is. The
analysis here is in terms of salaries, not
wages. The sample inclucles only exempt
employees. As such, one might guess they
are all expected to work a full week and
that their weekly salary does not change
even if they work more or less in a particular
week. One wouIc] expect that, on average,
some jobs have fuller work weeks than
others or that some have more variability
in hours worked in a week and that this
would be reflected in salary differences. If
this is in fact the case, it would not be
surprising to fins! that women were dispro-
portionately in jobs with fewer hours ant]
with less variable hours and that this ex-
plained part of the sex difference in salaries.
Other types of job differences might also
be important for salary differences anc] cor-
relatec] with sex.
~ 1 ~ ~ ~ _
Let me turn to a different issuethe
sample. Gerhart and Milkovich have se-
lectecl their sample so that only inclividuals
who were working with the firm in 1980
ant] were still working with the firm in 1986
are inclucled in their analysis. As such, they
excluded individuals who were working with
the firm in 1980 but left before 1986. It
may well be that the exclusion of inclivicluals
who have left the firm has only a negligible
effect on the analysis. The question is not
whether women or men are more or less
likely to leave the firm, but rather what the
interaction is among sex, exiting, and the
depenclent variable, that is, salaries, raises,
or promotions. It is easy to tell stories about
how inclusion of former employees could
change the sex (lifferential in raises and
promotions. If men who left the firm were
in(lividuals who would or (lic1 receive mul-
tiple promotions and large raises and women
OCR for page 46
46
who left would or did receive no promotions
and small raises, then the sex differential
in raises anti promotions could disappear
by controlling for this differential selection.
My other criticism of the sample is the
authors' decision not to inclucle more than
averages of variables from the years between
1980 and 1986. One's salary at a particular
time and the degree to which it has changed
over the years is a result of year-to-year
changes. A better model, for instance, would
be one in which performance rating, and
perhaps changes in performance rating, af-
fected one's raise and one's chance of pro-
motion in each particular year. It is the
history of one's performance that is probably
most important. The inclusion of average
performance rating in predicting raises and
promotions over the 6-year period strikes
me as a misspecification of the model.
This paper could perhaps make its great-
est contribution in terms of the analysis of
the performance variable. Most research
(loos not inclu(le a measure of productivity,
but they would need to consider a richer
data structure in order to include perfor-
mance in their models appropriately. With
a richer set of data, lagged values of per-
formance might also be appropriate. As they
acknowledge in a footnote, their perfor-
mance variable probably contains a consid-
erable amount of measurement error. If
measurement error in the performance vari-
able was corrected, performance might have
a much larger effect. This would be im-
portant in analyzing the sex differences giv-
en that women have somewhat higher per-
formance measures than men.
Of consi(lerable interest in the paper is
the observation that women received higher
raises en c] more promotions than men in
this firm. As the authors note, other re-
searchers have also reporte(l this fincling.
Gerhart and Milkovich's findings indicate
that nearly half of these differences can be
explained by the lower levels of experience
and job tenure of women. This, however,
PAY EQUITY: EMPIRICAL INQUIRIES
leaves a sizable unexplained portion. Ger-
hart and Milkovich offer one explanation-
that the firm has seen the light an(l is trying
to make up for past inequities. The eco-
nomics literature suggests five other pos-
sible interpretations. All of them potentially
have to do with childbearing, and several
also have to do with the employer's inability
to (liscern which women will stay with the
firm Tong term and which will not.
First, there is a very simple supply-side
story. As women have children, their re-
servation wage may rise over time. As a
result, women in the lower paying jobs,
women who expect to receive small raises
and few promotions, will leave the firm.
Those with better prospects remain in the
firm and make other arrangements for child
care. As a result, women's salaries rise faster
than they would have if the lower pai
women had not left the firm.
A second selection story works with con-
stant reservation wages, an(1 has to do with
learning about the quality of the match of
the job and the employee. (For a formal
model of this nature, see the matching
models developed by lovanovic, 19791. All
that is needecl is for women to have higher
reservation wages than men. This might be
true because of their greater productivity
in the home. Here, neither workers nor
firms know how well they are matched. As
time goes on, they learn how good the match
is an(1 the poor matches (lissolve. As in the
first example, this selection process increas-
es the raises and promotions of women by
selecting women who have higher values
on their reservation wages.
With respect to both selection stories, it
is unclear whether enough women leave
the firm to explain the residual sex differ-
ence in raises and promotions. Recent anal-
yses by Tope! (1986) an(1 by Altonji and
Shakotko (1987) have found that selection
(lue to matching in the above sense explains
most of the growth in wages with job tenure
for men.
OCR for page 47
COMMENTARY
There are also a set of demand-side sto-
ries. A straight human capital story would
be that firms anticipate a higher quit rate
for women than for men. As such, they
might be leery of investing in tamale em-
ployees for fear they would not recover
their investment. As a result, they might
adopt a salary schedule whereby women
are paid less early on and more later.
A related story is that firms cannot tell
the difference between women who will
stay with the firm and those who will leave
after several years. They may adopt a salary
schedule that pays women below their mar-
ginal product initially and above their mar-
ginal product later, so that only women who
intent! to stay with the firm will take jobs
there. This is the adverse selection mode!
of Salop and Salop (1976~. Here, what hap-
pens is that firms specialize. Firms with low
turnover costs end up hiring women who
are likely to move in and out of the labor
force, and firms with high turnover costs
hire only those women who are likely to
remain with them. The hypothesis here is
that Gerhart and Milkovich's firm is a firm
with high turnover costs. This mode! is
consistent with Gerhart and Milkovich's ob-
servation that women are no more likely to
leave the firm than men.
There are also stories based on nonTin-
earities due either to risk aversion or non-
linearities in the production process. If there
is more ex ante unobserved variability for
women than for men in their productivity,
firms, if they are risk averse, will initially
pay women less than men until those pro-
ductivity differences are revealed. This idea
is presented in an early paper by Aigner
and Cain (1977~. A relatecl idea of Roths-
chiTcI and Stiglitz's (1970) is that if firms are
worse at matching women to jobs than men,
both in that they are likely to err by putting
women in jobs that they are overquatifie(l
for as well as in jobs they are unclerquali-
fied for, then as a result, women will be
less productive than men and will be paid
less.
, ~ . ., _ ~ . .
47
Finally, there is the possibility that wom-
en may be more likely to shirk on the job
than men. Perhaps this might be the case
because child-care responsibilities are more
of a distraction for women. As a result, firms
may want to offer women contracts in which
they are paid below their marginal product
early on and above it later.
Consi(lerably more about the institutional
structure of Gerhart and Milkovich's firm
would have to be known to evaluate the
appropriateness of these different stories.
Each of these stories implies some amount
of job segregation, since it would probably
be difficult for an employer to pay men an
women different wages for the same jobs.
These hypotheses could be studied using
firm-levl! data. More than two periods of
(lata would be nee(le(l, however, preferably
a panel of yearly (lata. In addition, single-
equation regression models would probably
not be adequate. Selection out of the firm
ant] unobserved heterogeneity wouicl have
to be accounted for.
There are now a considerable number of
sophisticated models of wage growth in the
economics literature. I would encourage the
authors to examine them and to consider
adopting a more structural model for their
subsequent analysis. Doing so could create
a greater un(lerstancling of why they see
the empirical relations that they (lo and
woul(1 enable them to interpret their results
in terms of a conceptual mo(lel. Their work
would help us all move forward in under-
standing why men and women are paid
differently.
REFERENCES
Aigner, D. J., and G. G. Cain
1971 Statistical theories of discrimination in the
labor market. Industrial and Labor Relations
Review 30:175-187.
Altonji, J. G., and R. A. Shakotko
1987 Do wages rise with job seniority? Review of
Economic Studies Uuly):437-45l.
OCR for page 48
48
Jovanovic, B.
1979 Job matching and the theory of turnover.
Journal of Political Economy 87:972-990.
Rothschild, M., and J. Stiglitz
1970 Increase risk I: A definition. Journal of Eco-
nomic Theory 2:225-243.
Salop, S. C., and J. Salop
1976 Self-selection and turnover in the labor mar-
PAY EQUITY: EMPIRICAL INQUIRIES
ket. Quarterly Journal of Economics 90:619-
628.
Topel, R.
1986 Job mobility, search, and earnings growth:
A reinterpretation of human capital earnings
functions. Pp. 199-233 in R. G. Ehrenberg,
ea., Research in Labor Economics. Vol 8.,
A. Greenwich, Conn.: JAI Press.
Representative terms from entire chapter:
salary growth
