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 153
7
Occupational Segregation, Compensating
Differentials, and Comparable Worth
RANDALL K. FILER
In surveying the relative positions of men
and women in the labor market, two facts
stand out. First, wage differences between
the two sexes are substantial. Second, dif-
ferences in occupational structure are sig-
nificant; men and women are concentrated
in different occupations and heavily female
occupations tend to be Tower paying. Recent
years have seen substantial shifts in both of
these factors. Median weekly earnings of
fulI-time female workers over age 16 have
risen from 61 percent of those for men in
1978 to 71 percent of male earnings in the
second quarter of 1987. Thus, in the past
10 years approximately 25 percent of the
difference in mate and female earnings has
been eliminated. At the same time, although
there are methodological difficulties in mea-
suring changes in the degree of occupational
sex segregation over time, numerous recent
studies have documented that the degree
of sex segregation in occupations has de-
cTined since at least 1970 (and probably
since 1960. As one example, Bianchi and
iThe methodological difficulties arise from changes
over time in the categories into which occupations are
classified (England, 1981~.
Rytina (1986) found that the index of dis-
similarity fell from 67.9 in 1970 to 59.4 in
980.2
Despite these improvements in the rel-
ative economic position of women over the
past few years, significant differences in
occupational (listributions an(l earnings be-
tween the sexes persist. There are important
policy implications to be derived from an
understan(ling of why these differences ex-
ist. To the extent that they arise from un-
equal opportunities caused by unfair hiring
or promotional practices, the economy has
failed to make appropriate use of human
resources and has created inefficiencies. In
this case there is justification for interven-
tion to facilitate greater sex equality in the
labor market. On the other hancl, to the
extent that sex (differences in occupational
structure and earnings arise from differences
in individual productivity or choices, (le-
spite equal labor market opportunity, in-
terventions to change either employment
2The dissimilarity index is roughly equal to the
percentage of workers who would have to change jobs
in order to create an equal distribution of the sexes
in all jobs.
153
OCR for page 154
154
or earnings patterns would leac] to distor-
tions in resource allocation and the creation
of inefficiencies.
POSSIBLE CAUSES OF
SEX DIFFERENCES
The observed pattern of the genders being
concentrated in different occupations, cou-
pled with lower average wages in the oc-
cupations that are heavily female, is con-
sistent with a number of possible explanations
that have been proposed by economists.
Those explanations provide the framework
for the analysis in this paper.
Differences in Productivity
It may be that one gender has lower
average levels of productivity and has con-
centrated in occupations in which it has a
comparative advantage. Primary among the
factors that may contribute to differences
in productivity between typical men and
women is past work experience. Previous
research has established that between one-
quarter and one-half of the gender gap in
wages may be due to differences in the
extent of previous employment (Corcoran,
1979; Mincer and Ofek, 1982; Mincer and
Polachek, 1978; Sandell and Shapiro, 1978~.
Physical differences may also contribute
to slivering occupational comparative ad-
vantages and overall productivity. In one
setting, Hoffmann and Hoffmann (1987) found
that upper body strength and lifting re-
quirements limited women's bidding on and
accepting "mate" jobs even though they
were actively encouraged to clo so by their
employer. Similarly, several authors (e. g.,
Daymont and Andrisani, 1984; Filer, 1983;
Greenfield et al., 1980) have observed that
women and men in the labor market have
substantially different personality patterns
with respect to such characteristics as em-
pathy and aggression, which may lead to
PAY EQUITY: EMPIRICAL INQUIRIES
different job choices an(l, consequently, dif-
ferent reward structures.3
Differences in Utility Functions
Men and women may make rational choices
in the job market based on differences in
utility functions that create differing pref-
erences for certain types of work and other
(luties. For example, some evi:lence in(li-
cates that women attach greater importance
to various forms of attractive working con-
clitions and that men place relatively greater
emphasis on incomes (Forgionne and Pe-
ters, 1982; Harvey, 1986; Murray an(l At-
kinson, 1981~. Such a difference in pref-
erences, coupled with the fact that the
market forces employers to pay compen-
sating differentials to those workers who fill
jobs with relatively unattractive working
conditions, will, even given equal produc-
tivity, result in women being concentrates!
in lower paying but otherwise more attrac-
tive jobs. Evidence presented in Filer (1985)
suggests that such compensating differen-
tials may be responsible for up to one-
quarter of earnings differences between men
and women.
Much of the literature regarding (liffer-
ences between men ant] women in labor
market preferences starts with the fact that
there are differences in home duties. Filer
(1985) reports that jobs typically held by
women are those from which it is easier to
take time off for personal reasons and are
typically located closer to their homes.4
3~t should be noted that nothing has been said about
how these personality differences may have arisen.
Some maintain that they are innately linked to bio-
logical differences between the sexes; others believe
they are the result of childhood conditioning. The
reality is probably some combination of these and other
sources.
4Such locational differences contribute to wage dif-
ferentials because commuting time is a negative char-
acteristic that must be compensated for and because,
by restricting the opportunity set from which a job
may be chosen, individuals who desire to work close
OCR for page 155
OCCUPATIONAL SEGREGATION
Others (O'Neill, 1983, 1985; Waite ant] Ber-
ryman, 1985) have pointed out that female-
dominated jobs require less overtime, are
less likely to have rotating shifts, ant] are
more likely to be part time. All of these
findings are consistent with women assum-
ing responsibility for child rearing.
Perhaps the most frequently advanced
reason why Mitering home responsibilities
might lead to occupational segregation comes
from the fact that women tend to have more
discontinuous work histories. This should
lead them to choose jobs that require little
firm-specific human capital and in which
there is relatively little atrophy of skills
when not in use. Such a rational investment
reason for occupational segregation has been
most forcefi~ly advocated by Polachek (1979,
1981, 1985, 1987~. It can also be found in
AneshenseI and Rosen (1980), Blakemore
and Low (1984), Matthaei (1982), Mincer
and Ofek (1982), and Waite and Berryman
(19851. Evidence casting doubt on this hy-
pothesis has been presenter] by Corcoran
et al. (1984), England (1982, 1985), and
England et al. (19861. A related but more
complex explanation has been advanced by
Goldin (1986), who argues that a higher
turnover rate for women makes structuring
of contracts preventing shirking more dif-
ficult ant] therefore results in the concen-
tration of women in occupations in which
such contracts are less critical.
Employers may respond to a greater pro-
pensity of women to leave the labor force
to home limit their ability to seek out their highest
productivity and highest paying match. Occupational
segregation will result from the fact that workers who
desire employment close to home will only be available
to those industries that are well suited to decentralized
production in residential areas. Firms requiring large
work forces (therefore having to draw workers from a
wide geographic area) or not able to locate in residential
neighborhoods (due to the need to be in a central
place or because of production externalities) will tend
to have a disproportionate share of male workers. To
the extent that such firms are (as would appear likely)
better paying establishments, this segregation will also
contribute to wage differentials.
155
by investing less in training women and
being less likely to promote women (see
Lazear and Rosen, 1989~. Such theories
of"statistical discrimination" rest on the
inability of employers to distinguish be-
tween those women who will remain on
the job and those who will leave. They do
not, however, explain why women, who
presumably know whether they intend to
leave their employer to assume respon-
sibilities at home, clo not negotiate con-
tingent claims contracts insuring employ-
ers against lost investments. Finally, Becker
(1985) provides a theoretical rationale for
why differing home duties will result in
men providing greater levels of effort on
the job.
Discrimination
If lower wages for one group are not the
result of lower productivity and are not fully
compensated by nonwage aspects of the job,
the labor market is not in equilibrium and
members ofthe group receiving lower wages
should move into higher wage occupations.
The absence of such equilibrating move-
ment (and thus a stability over time in the
extent of occupational sex segregation) would
suggest that women have been involuntarily
lenie(l access to certain occupations (see
Bergmann, 1974; Blau, 1984; Madclen, 1975;
Stevenson, 1984~. Obviously, such con-
scious denial of access to occupations, whether
through the actions of employers, other
workers, customers, or legislative action,
would create occupational segregation.5 This
fit is an unanswered question to what extent ob-
served patterns of occupational segregation result from
impositions from outside the labor market. An obvious
example is the law preventing women from serving in
combat specialties in the armed forces. Until very
recently, many states had protective legislation that
limited the exposure of women to hazardous working
conditions and restricted the schedules and number
of hours women could work.
It has been very difficult to develop models of how
denial of access can be stable over time and not create
OCR for page 156
156
could explain Tower wages for women through
one of two mechanisms. "Crowding" of
women into a limited number of jobs could
artificially increase supply and depress wages
(see Bergmann, 1974; Johnson and Solon,
1984~. Alternatively, employers may con-
sciously take the sex composition of jobs
into account when setting pay levels (see
England et al., 1982; Treiman and Hart-
mann, 1981~.
If discriminatory differences in occupa-
tional structures are not being eliminatecl
by labor market mobility, some structural
barrier must be preventing such movement.
This would suggest two possible courses of
action. Either the barrierks) to mobility may
be removed so that rational mobility cle-
cisions on the part of workers will create
equality of compensation, or the occupa-
tional distribution may be taken as fixed
and an attempt made to raise wages in jobs
heavily filled by women.6
It is the latter policy that has come to be
known as "comparable worth." Advocates
of comparable worth call for pay to be
administratively set so that differences in
wages (or full compensation, including the
value of fringe benefits) not based on dif-
ferences in productive skills, effort, re-
sponsibility, and working conditions are
eliminated. An excellent review of the de-
velopment and implementation of the con-
cept of comparable worth is presented in
Weiler (1986).
incentives for nondiscriminatory employers (including
women themselves) to employ women in the previously
denied occupations. Indeed, most models of discrim-
inatory actions on the part of employers lead to firm
segregation rather than occupational segregation, a
result consistent with the finding of Bielby and Baron
(1984) that segregation among firms is generally greater
than that across occupations.
in. .. .
. . ..
bThese policies are. to a certain extent, mutually
exclusive. If policies are enacted to raise wages in
women's occupations to a level commensurate with
their productivity and working conditions, this reduces
the incentive for women to move into jobs previously
held by men.
PAY EQUITY: EMPIRICAL INQUIRIES
ANALYTIC FRAMEWORK
This paper investigates the extent to which
there exist (differences in the wages paid in
various occupations that are not related to
levels of effort and responsibility, working
conditions, or the productive characteristics
of incumbents in them, but which are re-
lated to the sex composition of the occu-
pation. Much recent work has applied a
similar procedure to micro-level data, re-
gressing inctiviclual wages on personal char-
acteristics and the percentage of women in
an individual's occupation.7 Studies such as
England (1982, 1985), England et al. (1986),
Ferber and Lowry (1976), Jusenius (1977),
and Stevenson (1975) have found a negative
relationship between the proportion of fe-
male workers in an occupation and its av-
erage wage.
Other studies (Aldrich and Buchele, 1986;
England et al., 1982; Fuchs, 1971; Treiman
et al., 1984) have used occupations as the
unit of analysis, regressing average wages
in an occupation (either separately for men
and women or combined) on a set of ex-
planatory variables as well as the occupa-
tion's sex composition. These studies have
been han(licapped by their ability to in-
clucle, at most, a small subset of the factors
encompassed in an occupation's effort, re-
sponsibility, working conditions and pro-
uctive requirements."
Estimation Issues
Comparable worth studies traditionally
estimate an equation whereby wages (w)
are a function of productive characteristics
(P), job characteristics (effort, responsibility,
anti working conditions) (C), and the sexual
composition of the job (F) such that
7See Polachek (1987) for an explanation of why this
procedure is unable to distinguish adequately between
human capital and occupational sex segregation ex-
planations for sexual differences in earnings.
OCR for page 157
OCCUPATIONAL SEGREGATION
Wi = aPi + bci + cFi + Ui. (1)
A number of issues must be addressed
in the estimation of this relationship.
First, some authors (see Treiman and
Hartmann, 1981; Treiman et al., 1984) have
regressed average wages for men and wom-
en combined in an occupation on its char-
acteristics as specified in Equation (11. Such
a strategy, however, is inherently incapable
of addressing comparable worth issues be-
cause it confounds male-female wage dif-
ferentials that result from differences in pay
according to sex composition with differ-
ences due solely to pay differentials within
each occupation. Suppose women were paid
75 percent of men's earnings in each oc-
cupation no matter what its sex composition.
It would still be the case that an occupation
that was almost 100 percent female would
have an average wage three-quarters of that
for an occupation that was almost 100 per-
cent mate. Thus, separate equations for men
and women must be estimated to determine
whether the gender composition of an oc-
cupation affects earnings.
Second, in a comparable worth study it
is especially incumbent on the researcher
to be sure that the hedonic wage equations
are properly specified. Consider the issue
of omitted variables. Suppose that instead
of Equation (1) a researcher estimated an
equation of the form
Wi = CtFi + U i
A classic case of omitted variables bias then
exists such that
c' = c + bEcov~ci' Fi~lvar(Fi)]
+ atcov(Pi, Fi~lvar(Fi)~.
Thus, if P and C are measured so that the
expected signs of a and b are positive, and
if women have lower levels of the productive
attributes or choose to enter occupations
with fewer undesirable job characteristics
that elicit compensating differentials so
that cov(Pi, Fi) and/or cov(Ci, Fi) are neg-
157
ative, c' will be less than c and estimates
will overstate the negative impact of the
proportion female on the average level of
earnings in an occupation. Given that, as
was discussed above, gender is likely to be
causally related to a number of important
characteristics of occupations through pro-
ductivity and occupational choice consid-
erations, little confidence can be placed on
estimates of gender effects derived from
equations that do not fully reflect the jobs
under analysis. Given the complexity of
occupations, this is a daunting task and
ought to endow claims for perceived effects
of an occupation's gender composition on
its wages with a strong dose of agnosticism.
Third, comparable worth equations of the
form of Equation (1) are in fact hedonic
wage or price equations resulting from the
interaction of workers' demand for and firms'
supply of various job characteristics. As is
typical in the compensating differentials lit-
erature, this relationship is estimated here
using ordinary least squares (OLS). It must
be recognized, however, that it is only in
special cases that such estimates can be
consistent (Epple, 19871. Although some
recent empirical work has suggested that
the available instruments that could be used
to "solve" this problem in the context of
compensating differentials studies are so
poor as to render it impossible to reject
statistically the accuracy of the OLS esti-
mate, such a conclusion would raise serious
questions were such studies to become the
basis for legal actions rather than academic
research.
Fourth, it is important to recognize that
unless the hedonic wage equation is strictly
linear in all its regions, there is no reason
that returns to various factors should be the
same for different groups even if wages for
both were determined by a common func-
tion. Thus, an interpretation of differences
in coefficients between male and female
wage equations as indicating structural dif-
ferences in how the sexes are treated in the
OCR for page 158
158
labor market, although frequently macle, is
fundamentally inappropriate.
Finally, it must be recognizes! that the
estimated impact of gentler composition on
wages from Equation (1) overstates the po-
tential effect of the adoption of comparable
worth policies. The proposals for the United
States, unlike the situation in Australia, do
not require job comparability to be estab-
lished across firms but rather within par-
ticular firms. Thus, to the extent that the
gentler impact represents different pay
policies among firms, it is immune to com-
parable worth remedies.
The Data
This analysis combines data from a num-
ber of sources to obtain a more exhaustive
portrait of the occupations and their incum-
bents than is available from any single data
set. Earnings and personal characteristics
for each three-digit occupation were derived
from the 5 percent sample of the 1980
census.8 Measures of working conditions,
effort, and responsibility were obtained fiom
the Dictionary of Occupational Titles (U. S.
Department of Labor, 1977), the 1977 Qual-
ity of Employment Survey (Quin and Staines,
1979), the 1976 Survey of Time Use in
Economic and Social Accounts (data tape
provided by the Institute for Social Re-
search, University of Michigan), and the
May 1983 Current Population Survey (data
tape provided by Bureau of the Census).
Measures of personality were taken from a
proprietary data set (leveloped by the author
in the mid-1970s (see Filer, 1981~. Union-
ization is a 3-year moving average centered
on 1980 taken from Kokkelenberg and Sock-
ell (1985~. The Duncan measure of occu-
pational prestige (see Duncan and Stafford,
1980) and a measure of occupational dead-
encledness (Brown, 1982) were added. Fi-
8Thus, the results presented here miss any of the
substantial changes in the labor market status of women
that have occurred since 1980.
PAY EQUITY: EMPIRICAL INQUIRIES
nally, since the typical imputation of ex-
perience as age minus years of schooling
minus six is not particularly useful for wom-
en, average levels of experience for women
were~ computed for each occupation using
the National Longitu(linal Surveys (NLS) of
Mature Women ant] GirIs.9 (See notes to
appendix table for full data sources.)
Since job characteristics were observer!
for occupations and not workers, there would
be no added variability from attributing
these common measures to incTividual-level
data. No micro-level data set contains suf-
ficient measures of the concepts important
for comparable worth to enable indivi(lual-
level cross-sectional estimates. The process
of integrating the data was complicate`] by
the fact that the sources used several distinct
sets of occupational codes that had to be
mapped into the 1980 census scheme. it
When small sample sizes meant that there
9The procedure involved combining all of the waves
of both panels into a single data set so that sample
sizes would be large enough and then estimating for
each census occupation an equation of the form: Ex-
perience = f (age, ages, marital status, race, and the
year of observation). Since all of the right-hand side
variables in these equations are available in the census,
the estimated coefficients could be used to obtain a
predicted average level of experience for those women
actually in any occupation in 1980. In theory, a similar
procedure could have been followed for men; however,
the improvement that this would offer over the more
traditional approximation was judged not to be worth
the additional computational effort.
i°The Bureau of the Census double-coded a sample
of occupations each time the codes were revised,
thereby enabling each earlier-year code to be divided
among one or more later-year codes according to
population proportions. Similarly, the May 1981 Cur-
rent Population Survey was double-coded using both
1970 census occupation codes and DOT codes. Thus,
the DOT codes and the Filer data (which used DOT
codes) could be merged into the 1970 census data.
The Institute for Survey Research provided a matching
of the codes used in the Time Use Survey with 1970
census codes. Thus, by appropriate chaining, each data
set could be bridged into 1980 census codes. The
method used was to compute average values for each
variable in its "native" code and then to calculate
values for the 1980 coding scheme as weighted averages
of the native code values.
OCR for page 159
OCCUPATIONAL SEGREGATION
were too few observations in any three-digit
code to enable reliable computation of av-
erage values for the characteristics, the av-
erage values were computed at the two-
digit levelly and then imputed back to the
three-digit occupation.
The final data set contained over 225
measures of various characteristics of oc-
cupations and the workers who held them. i2
There are obvious difficulties with the in-
clusion of all 225 available job character-
istics. Given that there are only 430 three-
ctigit occupations in the final sample, the
reduction in degrees of freedom if all pos-
sible explanatory variables were included
wouIc! be substantial. Therefore, estimated
results are reported in the appendix for a
specification that maximizes the adjusted R2
of the wage equation.
Estimated Wage Equations
The 5 percent sample from the 1980
census contains data on 6,490,318 men and
4,986,538 women who were employed for
all or part of 1979. The average wage for
men in this sample was $9.23 an hour and
that for women was $5.87.~3 Thus, women
Census provides a set of two-digit codes and their
constituent three-digit codes for the 1980 scheme.
After a conversation with John Priebe at the Bureau
of the Census failed to uncover any similar amalgam-
ations for the 1960 and 1970 censuses, they were
created by the author following as closely as possible
the groupings in the 1980 scheme. Generally, this
procedure was invoked if there were fewer than 5
observations in any cell in which averages were com-
puted or fewer than 50 observations in the NLS cells
in which the equations for imputing female work
experience were estimated.
t2A complete list of the variables in the data set is
available from the author.
i3Wages were calculated as labor earnings (the sum
of wage or salary income in 1979, nonfarm self-
employment income in 1979," and "farm self-employ-
ment income in 1979" divided by the product of"weeks
worked in 1979" and "usual hours worked per week
in 1979"~. Since, for confidentiality reasons, the Census
Bureau codes all incomes in excess of $75,000 as
$75,000, "top-coded" incomes were replaced by an
159
in the sample earne(l approximately 64 per-
cent of what men earned. The first acJjust-
ment that needs to be macle recognizes that
men and women typically work different
numbers of hours.~4 Results are reporter]
below only for full-time, full-year workers
(those who averaged at least 32 hours of
work per week for at least 45 weeks in
19791.
RESULTS
Table 7-1 reports the results of a number
of hedonic wage equations that forcefully
illustrate the points macle above in the
discussion of estimation considerations. Re-
sults are reported for regressions using lev-
els of wages as the dependent variable. Both
this specification and a semilogarithmic one
in which the dependent variable is the
natural logarithm of wages are common in
the literature. The results for levels of wages
are reported because their interpretation is
patently obvious. Results using the natural
logarithm of wages do not (lifter in any
significant way and are available from the
author.
The first column in Table 7-1 reports the
estimated impact on the wages in a job if
it were to move from O percent female to
100 percent female as estimated from a
combiner] sample of men and women. Col-
umns two and three show results for men
and women separately. Differences be-
tween them and column one represent the
extent to which women's lower wages within
estimate of the mean income among those earning
over $75,000 (calculated by fitting a Pareto distribution
to reported incomes for those earning less than $75,000
within each income category and occupation) before
total labor incomes and average wages were calculated.
i4Census data cannot be adjusted for the fact that
the typical longer work week of men means that more
of them are likely to have hit statutory limits requiring
the payment of overtime. Weller (1986) reports that
comparisons of base pay of men and women rather
than average hourly wages may eliminate as much as
one-third of the apparent gender gap in earnings.
OCR for page 160
160
PAY EQUITY: EMPIRICAL INQUIRIES
TABLE 7-1 Estimated Coefficients of Gender Composition on Wages, All Full-Time,
FulI-Year Workers
Estimated Coef~cienta (Standard Error)
Men and Women Women Men
Equation Specification Combined Only Only
Gender composition only - 4.41 - 1.73 - 2.32
(.38) (.25) (.42)
Gender composition and - 3.30 - 1.92 - 1.38
demographic variables (.30) (.19) (.32)
Gender composition, demographic - 3.13 - 1.59 - 1.29
variables, and union coverage (.30) (.19) (.32)
Gender composition, demographic -1.70 0.30 0.31
variables, union coverage, (. 31) (.24) (.32)
effort, responsibility, and
working conditions Maximum
R2)
Gender composition, demographic - 1. 35 - 0.55 0.18
variables, union coverage, ef- (.50) (.37) (.53)
fort, responsibility, and work-
ing conditions (all variables)
aThe estimated impact of an occupation shifting from entirely male to entirely female composition. Multiply by
the difference in the femaleness of the job held by the typical woman and that held by the typical man to obtain
the estimated gender contribution to the gross wage differential.
occupations bias the gender effect when it
is estimated using a combined sample. To
calculate the impact of gender composition
on wage differentials, one must calculate
the change in wages that would occur if
each occupation exactly mirrored the pro-
portion female in the work force. Forty-two
percent of the workers in the 5 percent
census sample were women. The average
man was in a job that was 23 percent female
and the average woman was in a job that
was 68 percent female.
Adding Demographic and Personal
Characteristics
The first row of Table 7-1 shows the
estimated effect of a job's being 100 percent
female on the wages of full-time workers in
that job if no other characteristics of either
the worker or the job are taken into account.
The most standard adjustment is to rec-
ognize that men and women do not, on
average, bring the same levels of productive
attributes to the labor market. Census data
provide a limited set of personal and de-
mographic characteristics that may capture
these productivity differences and can be
included in the regression. Among them
are racial group, marital and citizenship
status, education, and crude measures of
the type of employer. In addition, estimated
actual work experience (as discussed above)
was included in this specification. Results
from this estimation are reported in the
second row of Table 7-1. Increased female-
ness of an occupation still implies signifi-
cantly lower wages after controlling for these
characteristics. Moving from being 100 per-
cent male to being entirely female would,
according to these estimates, bring a re-
duction of $1.92 an hour in average wages
for women and $1.38 an hour for men. The
relationship is significant for both sexes,
although, unlike results found by other re-
searchers, it appears to be stronger for
women than for men.
It is often asserted that one reason women
earn less than men is their lower partici-
pation in unions that obtain higher than
OCR for page 161
OCCUPATIONAL SEGREGATION
competitive wages. The third row of Table
7-1 reports estimated gentler coefficients for
an equation including personal character-
istics and the proportion of the occupation
covered by a collective bargaining agree-
ment. There is some support for the hy-
pothesis that lack of female participation in
unions contributes to the estimated gender
impact on wages. When unionization is
added, the gentler impact in the female
equation falls by almost 20 percent.
Due to space considerations, coefficients
on other variables are not reported in full. is
In general, however, they are as expected.
Education effects are somewhat stronger
than found in micro-level studies. An ad-
clitional year of schooling is estimated to
result in between 51 and 76 cents an hour
in additional earnings. Estimated union ef-
fects are consistent with those from micro-
leve} studies. The estimated benefits to join-
ing unions are substantially greater for wom-
en than for men. This raises the question
of why women's unionization rates have
traclitionally been lower than men's. The
answer may lie in discrimination within
unions, higher costs to women in joining
unions, or the fact that women must am-
ortize organizational costs over shorter ex-
pectec] periods of job tenure.
No relationship was found between the
imputed average level of experience of wom-
en in an occupation and its wages. For men,
a positive relationship was found. These
findings provide at least partial support for
the Polachek analysis noted above. Results
regarding the impact of the ethnic com-
position of an occupation on its wages will
be discussed below.
Adding the Full Set of
"Comparable Worth" Factors
Even after the addition of census de-
mographic characteristics, the effect of the
proportion female in an occupation on its
They are available from the author.
161
average wages remains substantial. We can
turn now to the extent to which this is a
statistical bias resulting from the omission
of significant characteristics that are cor-
related with the proportion female in an
occupation but which advocates of compa-
rable worth recognize as compensable in
their own right, such as a job's working
conditions ant] levels of effort and respon-
sibility.
The results of two versions of the "com-
plete" comparable worth specification are
presented. The fourth row of Table 7-1
contains estimates ofthe gentler impact from
an equation constrained to include the vari-
ables discussed in the previous section plus
the job characteristics that maximized the
adjusted R2 of the linear hedonic wage equa-
tion for men and women combined (since
there were slight differences in the set that
was entered for the sexes separately). t6 The
actual estimates for this equation are pre-
sented in the appendix. The reader is cau-
tioned, however, that where there are sev-
eral measures relating to any comparable
worth factor, patterns of multicollinearity
make interpretation of any one coefficient
impossible. It is only the effect of the full
set taken jointly that has meaning. What is
of interest here is not the coefficients in
and of themselves (for a discussion of their
meaning see Filer, 1987), but rather the impact
that their inclusion has on estimates of the
i6As an alternative method of data reduction, factor
analysis was tried on the full set of raw variables. Even
when rotated in several alternative ways, however, a
large number of factors were required to capture even
a moderate portion of the complexity in the data.
Given the pattern of loadings on the factors, it proved
difficult to assign any meaningful interpretation to
them. An attempt was then made to group the variables
on an ad hoc basis into 25 distinct sets based on what
they apparently measured and then extract factors only
within each set. Results from this experiment were
highly ungratifying, and the resulting factors were of
little use in explaining average wages in occupations.
Thus, the decision was reached to retain variables in
their raw form and reduce the number of variables
used through a stepwise procedure.
OCR for page 162
162
elect of gender composition on wages in an
occupation. Finally, the last row of Table 7-
1 reports the result when all 225 job char-
acteristics were entered into the wage equa-
tion.
The impact of adjusting for compensable
job characteristics is striking. Once com-
pensating differentials for a iob's effort re-
sponsibility, fringe benefits, and working
conditions are taken into account there is
no significant relationship between an oc-
cupation's gender composition and its wages
for either men or women. What appears to
be an effect in the combined equation results
from lower wages for women within each
occupation, which, to the extent that they
represent other than legitimate compen-
sation practices, can be addressed by equal r ., .
employment laws but which are immune
to comparable worth adjustments.
Some of the results for other variables in
this equation are worthy of note. For a more
thorough analysis, the reader is referred to
Filer (1987~. The pattern of the census vari-
ables remains the same. There is an ap-
proximately 25 percent reduction in the
estimated impact of education on wages,
although this coefficient is still highly sig-
nificant. Examining comparable worth con-
cepts such as effort and responsibility is
complicated by the fact that there are several
related ant] highly intercorrelated measures
of each. When taken as a group (say by
assuming a change of one standard! deviation
in each), the results suggest that occupations
requiring more effort or responsibility, ex-
posure to worse working conditions, or long-
er commutes pay higher wages for both men
and women; while those with higher levels
of fringe benefits, more interaction with
other people, or employment in smaller
establishments can pay lower average - ~ -
wages. i7
PAY EQUITY: EMPIRICAL INQUIRIES
Results for Other Ethnic Groups
It is important to remember that the
principle behind comparable worth laws
shouIc! apply to all groups who might have
been the victim of discrimination, not just
to women. Thus, the question to be asked
is whether there is a significant relationship
between the average wages in an occupation
ant] the proportion of that occupation filled
by ethnic groups, such as blacks, Hispanics,
and Asians.
Table 7-2 reports the results of such es-
timates for a sample that includes all work-
ers. i8 Once job characteristics are controlled
for, the estimated ethnic impacts are not
significant. Before controlling for the nature
ot the JOUS and any compensating differ-
entials but after standardizing for average
levels of productive attributes, it would
appear that occupations held by blacks and
Asians paid higher wages than those held
by whites. The reality, however, is that
these groups have entered better paying
but otherwise less attractive jobs. Thus, with
respect to other protected groups, as with
women, there is no evidence that there are
compositional effects on wages that could
be remedied by the adoption of comparable
worth policies.
i7Fringe benefits are one of the clearest cases for
which compensating differentials theory suggests that
wages will adjust to job characteristics. For example,
a job that provides health coverage can attract workers tion.
at a lower wage than one that does not provide health
benefits since workers will not have to pay for them
out of pocket. Although one might alternatively add
fringe benefits to wages to obtain a measure of full
compensation, there are difficulties with this approach.
Since fringe benefits are typically offered as a take-it-
or-leave-it package, there can be no assumption that
any given worker values them at their full cost to the
employer (e.g., consider the value of maternity leave
to a single man). Thus, the value of a fringe benefit
package to employees is best established not by ac-
counting costs but rather by the wage reductions
workers are willing to accept in order to obtain the
package.
i8Thus, as was discussed above, these results include
both the impact of the ethnic composition of an
occupation and any within-occupation discrimina-
OCR for page 163
OCCUPATIONAL SEGREGATION
163
TABLE 7-2 Estimated Coefficients of Ethnic Group on Wages, All Full-Time, Full-Year
Workers
Estimated Coefficienta (Standard Error)
Equation Specification Blacks Hispanics Asians
Ethnic composition only - 19. 39 - 39.07 33.64
(1.75) (3.69) (9.79)
Ethnic composition and 3.47 0.69 24.05
demographic variables (1.30) (3.84) (7.21)
Ethnic composition, demographic 2.72 1.12 26.08
variables, and union coverage (1.32) (3.82) (7.22)
Ethnic composition, demographic .88 - 6.05 4.41
variables, union coverage, (1.15) (3.03) (5.36)
effort, responsibility, and
working conditions (maximum
R2)
Ethnic composition, demographic .27 2.50 2.16
variables, union coverage, (1.64) (4 90) (8.26)
effort, responsibility, and
working conditions (all
variables)
aResults are from an estimating equation that combines men and women.
Changes in Sex Composition
To what extent are mobility patterns of
women consistent with a labor market mov-
ing toward an equilibrium resulting in
equality of wages? Women can be expected
to enter those jobs for which pay is greatest
for women, no matter what the extent of
any (liscriminatory pay gap in that occu-
pation.
Women can also be expected, all else
being equal, to enter those jobs for which
there is the least degree of penalty for being
female. The size of such a potential gap can
be estimated by comparing the actual earn-
ings of women in an occupation with the
earnings they would be predicted] to have
if they were rewarded in the same manner
as men (the sum of the average levels of
independent variables for women times the
coefficients for men). If these predicted
earnings are equal to their actual earnings,
women in the occupation are being exactly
compensated for their productive attributes,
effort, responsibility, and working condi-
tions faced. If predicted earnings exceed
actual earnings, women are being un(ler-
compensated in that occupation (i.e., there
is a discriminatory gap). The greater the
gap, the greater shoul(1 be the incentive for
women to move out of that occupation and
into ones in which they do not face such a
disadvantage. Finally, if pre(lictecl earnings
are lower than actual earnings, the incentive
goes in the opposite direction, the discrim-
inatory gap favors women, and women shouIc!
esire to enter the occupation.
Thus, if occupational mobility is serving
to equalize wages between men and women,
a positive relationship between actual wages
pai(1 to women in an occupation and the
movement of women into that occupation
shouIcl be fount] as well as a negative re-
lationship between the gap between pre-
dicted and actual wages ant! the rate of
growth offemale employment. Both ofthese
results are seen. The correlation between
the percentage increase in the proportion
of workers in an occupation who were wom-
en between 1970 and 1980 and its average
wage in 1980 was .24, a result that is sta-
tistically significant at a better than .0001
conficlence level. Similar results are foun
when correlating growth in the percentage
OCR for page 166
166
Matthaei, Julie A.
1982 An Economic History of Women in the Amer-
icas: Women's Work, the Sexual Division of
Labor and the Development of Capitalism.
New York: Shocken Books.
Mincer, Jacob, and Haim Ofek
1982 Interrupted work careers: Depreciation and
restoration of human capital. Journal of Hu-
man Resources 17 3-24.
Mincer, Jacob, and Solomon Polachek
1978 An exchange: Theory of human capital and
the earnings of women: women's earnings
reexamined. Journal of Human Resources
13 118-134.
Murray, Michael A., and Jon Atkinson
1981 Gender differences in the correlates of job
satisfaction. Canadian Journal of Behavioral
Science 13 44-52.
O'Neill, June
1983 The Determinants and Wage Effects of Oc-
cupational Segregation. Washington, D.C.:
The Urban Institute.
1985 Role differentiation and the gender gap in
wages. Pp. 50-75 in Laurie Larwood, Ann
H. Stromberg, and Barbara A. Gutek, eds.,
Women and Work. Vol. 1. Beverly Hills,
Calif.: Sage.
Polachek, Solomon
1979 Occupational segregation among women:
Theory, evidence and a prognosis. Pp. 137-
157 in Cynthia B. Lloyd, Emily S. Andrews,
and Curtis L. Gilroy, eds., Women in the
Labor Market. New York: Columbia Uni-
versity Press.
1981 Occupational self-selection: A human capital
approach to sex differences in occupational
structure. Review of Economics and Statistics
63 60-69.
1985 Occupational segregation: A defense of hu-
man capital predictions. Journal of Human
Resources 20 437-440.
1987 Occupational segregation and the gender wage
gap. Population Research and Policy Review
6 47-67.
PAY EQUllrY: EMPIRICAL INQUIRIES
Quin, R. P., and G. L. Staines
1979 The 1977 Quality of Employment Survey:
Descriptive Statistics with Comparison Data
from the 1969-70 Survey of Working Con-
ditions and the 1972-73 Quality of Employ-
ment Survey. Ann Arbor: Institute for Social
Research, University of Michigan.
Sandell, Steven H., and David Shapiro
1978 An exchange: Theory of human capital and
the earnings of women: A reexamination of
the evidence. Journal of Human Resources
13 103-117.
Stevenson, Mary Huff
1975 Relative wages and sex segregation by oc-
cupation. Pp. 175-200 in Cynthia B. Lloyd,
ea., Sex, Discrimination and the Division of
Labor. New York: Columbia University Press.
1984 Determinants of Low Wages for Women
Workers. New York: Praeger.
Treiman, Donald J., and Heidi 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.
Treiman, Donald J., Heidi I. Hartmann, and Patricia
A. Roos
1984 Assessing pay discrimination using national
data. Pp. 137-154 in Helen Remick, ea.,
Comparable Worth and Wage Discrimina-
tion: Technical Possibilities and Political Real-
ities. Philadelphia: Temple University Press.
U. S. Department of Labor
1977 Dictionary of Occupational Titles. 4th ed.
Washington, D.C.: U.S. Government Print-
ing Office.
Waite, Linda J., and Sue E. Berryman
1985 Women in Nontraditional Occupations. Santa
Monica, Calif: Rand.
Weiler, Paul
1986 The uses and limits of comparable worth.
Harvard Law Review 99 1728-1807.
The appendix follows.
OCR for page 167
OCCUPATIONAL SEGREGATION
APPENDIX Estimated Results Inclucling Personal anc] tote Characteristics
Parameter Estimates Parameter Estimates
(Standard Errors) for (Standard Errors) for Meanb (Standard
Variablea Women Men Deviation)
PERSONAE CHARACTERISTICS
Proportion female —.30 .31 .31
(.24) (.32) (.28)
Proportion black 2.34 .45 .09
(.74) (1.29) (.06)
Proportion Hispanic 3.01 - 5.03 .04
(1.82) (3.21) (.03)
Proportion Asian 6.11 - 3.13 .01
(3.33) (5.15) (.01)
Proportion other nonwhite ethnic - 2.40 10.67 .01
group (3.21) (8.77) (.01)
Proportion noncitizens —3.02 - 6.30 .03
(2.85) (4.74) (.03)
Proportion having English - 7.06 .42 .02
difficulty (4.41) (5.06) (.02)
Proportion married .95 - 1.62 .63
(.53) (1.40) (.13)
Proportion having work- 1. 83 - 12.61 .06
influencing disability (1.79) (4.05) (.02)
Proportion employed by federal .80 -.38 .05
government (.37) (.47) (.11)
Proportion employed by state and - .76 - 1.52 .12
local governments (.28) (.37) (.20)
Proportion self-employed - .90 .42 .08
(.51) (.55) (.12)
Proportion working part-time - . 06 - .002 .11
(.41) (.006) (.12)
Average years of education .51 .76 12.39
(.07) (.09) (1.84)
Average years of work experiences .008 .16 16.76
(.02) (.03) (4.86)
Proportion in a union (Kokkelen- 1.40 .55 .27
berg and Sockell, 1985) (.29) (.41) (.20)
Average number of children in - .41 1.66 1.12
workers' homes (.21) (.57) (.19)
FRINGE BENEFITS
Average number of vacation days - . 03 - .13 13.93
(Time Use Survey) (.01) (.02) (4.28)
Proportion having a K401 plan —. 57 - 2.60 .08
(CPS) (.67) (.85) (.07)
Proportion having employer- - .22 - 1.15 .71
provided medical insurance (.45) (.58) (.20)
(QES)
Proportion eligible for paid mater- - 1. 19 - 2.65 .07
nity leave (female workers (. 65) (.84) ( 09)
only) (QES)
Proportion having employer- .76 1.81 .59
provided life insurance (QES) (.49) (.62) (.22)
Proportion having employer- - .62 - 2.59 .63
provided retirement program (.44) (.56) (.21)
(QES)
167
Continued
OCR for page 168
168
APPENDIX Continued
PAY EQUITY: EMPIRICAL INQUIRIES
Parameter Estimates Parameter Estimates
(Standard Errors) for (Standard Errors) for Meanb (Standard
Variablea Women Men Deviation)
Proportion eligible for profit- .13 1.73 .44
sharing stock option or thrift (.37) (.47) (.18)
plans (QES)
EFFORT MEASURES
Proportion of workday not work- - 8.20 - 31.58 .16
ing (Time Use Survey) (2.88) (3.80) (.03)
Proportion of workday "goofing 21.72 63.83 .05
off"(Time Use Survey) (6.63) (8.63) (.01)
Anchored work effort 2.31 5.18 .55
(Duncan and Stafford, 1980) (.90) (1.17) (.07)
Worker determines pace - .53 - 1.54 3.78
(1-5 scale, QES) (.21) (.26) (.29)
Enough time to do work -.37 -.48 3.04
(1-4 Scale, QES) (.27) ( 34) (.25)
Worker not asked to do "exces- .21 .72 2.84
sive" amounts of work (1-4 (.21) (.26) (.26)
scale, QES)
Self-reported effort .29 - .84 3.66
(1-4 scale, QES) (.35) (.45) (.17)
Extra effort required —.03 .79 3.42
(1-4 scale, QES) (.26) ~ 33) (.23)
REsPoNs~s~TY MEASURES
Many people affected -.31 -.73 4.16
(1-5 scale, QES) (.19) ( 25) (.33)
Have a lot of say about work .30 .31 3.31
(1-5 scale, QES) (.18) (.22) (.58)
Worker feels personally respon- .35 .46 4.33
sible for work (1-5 scale, (.31) ( 40) (.20)
QES)
PERSONAL RELATIONS FACTORS
Average "friendliness' of workers - .11 - .14 16.48
in occupation (1-30 scale, (.04) (.05) (1.27)
Filer, 1981)
Co-workers take a personal inter- .10 -.34 2.88
est in respondent (1-4 scale, (.22) (.28) (.32)
QES)
Job requires a preference for .17 .22 .51
dealing with things and oh- (.15) (.19) (.46)
jects (DOT)
Job requires a preference for .22 -.40 .27
contact with people (DOT) (.16) (.20) (.42)
Job requires a temperament for .31 .98 .38
dealing with others (DOT) (.17) (.22) (.46,
PHYSICAL WORKING CONDITIONS
Proportion exposed to risk of 1.22 1.82 .22
attack by people or animals (.34) (.43) (.18)
(QES)
Proportion exposed to dangerous .62 .88 .18
work methods (QES) (.44) (.58) (.15)
OCR for page 169
OCCUPATIONAL SEGREGATION
APPENDIX COntinUeC/
169
Parameter Estimates Parameter Estimates
(Standard Errors) for (Standard Errors) for Meanb (Standard
Variablea Women Men Deviation)
Proportion working both indoors -.17 -.27 .20
and outdoors (DOT) (.12) ~ 16) (.34)
Proportion exposed to wet or .36 .65 .07
humid environment (DOT) (.20, (.25) (.21)
Proportion exposed to noise or .26 -.15 .24
vibration (DOT) (.13) (.17) (.36)
Proportion exposed to hazardous - .23 - .20 .21
conditions (DOT) (.14) ( 18) (.34)
INTELLECTUAL CHALLENGE
Job requires a preference for .05 .30 .34
routine activities (DOT) (.13) (.18) (.42,
Job requires a preference for -.38 -.52 .05
abstract activities (DOT) (.23) (.28) (.20)
Average level of "thoughtfulness" - .02 - .09 19.72
of workers in job (1-30 scale, (.04) (.05) (1.12)
Filer, 1981)
WORKERS SKILLS
Job requires worker to learn .59 1.20 4.04
new things (1-5 scale, QES) (.16) ( 20) (.54)
Level of spatial aptitude required .0004 .004 42.72
(population percentile, DOT) (.002) (.004) (20.89)
Level of finger dexterity required .0024 .004 36.65
(population percentile, DOT) (.003) (.004) (16.50)
Requirement for math ability .17 .14 2.76
(1-6 scale, DOT) (.06) (.08) (1.21)
WORKERS Goads ARE MAINLY
MONETARY
Workers' lack of interest in .14 .21 8.73
power (1-11 scale, Filer, (.06) ( 08) (.73)
1981)
Workers' lack of interest in - .19 - .26 2.90
family life (1-11 scale, Filer, (.07) (.09) (.60,
1981)
Workers' lack of interest in com- .18 .26 8.01
munity activities (1-11 scale, (.07) (.09) (.63)
Filer, 1981)
Workers' lack of interest in job's - .04 - .18 6.35
contribution to society (1-11 (.05) (.07) (.82)
scale, Filer, 1981)
Workers mainly working for .43 .57 3.04
money (1-5 scale, QES) (.17) (.22) ~ 57)
Workers have a lot invested in .05 .84 2.98
their job (1-5 scale, QES) (.14) (.18) (.39)
LABOR MARKET CONDITIONS
Worker is afraid of what might -.33 -.56 3.18
happen if he quit job (1-5 (.14) ( 18) (.40)
scale, QES)
Worker reports shortage of similar -.62 -.70 .36
jobs in area (QES) ( 31) (.40) (.16)
Continued
OCR for page 170
170
APPENDIX Continued
PAY EQUITY: EMPIRICAL INQUIRIES
Parameter Estimates Parameter Estimates
(Standard Errors) for (Standard Errors) for Meanb (Standard
Variablea Women Men Deviation)
OTHER CHARACTERISTICS OF
INTEREST
Average travel time to work .06 .09 23.37
(in minutes, census) (.01) (.01) (4.75)
Proportion working in establish- -1.03 . - 1.15 .57
ment of less than 100 (CPS) (.23) ~ 30) (.22)
Deadendedness of job .68 1.68 .61
(Brown, 1982) (.33) (.43) (.16)
Worker has freedom to decide - .26 - .38 3.24
what to do on the job (1-5 (.14) (.18) (.59)
scale, QES)
Job has aggravated illness or - 1.09 - 2.98 .05
physical condition (QES) (.90) (1.14) (.06)
Intercept —3.69 - 2.57
(2.68) (3.37)
Adjusted R2 . 81 .89
NOTE: The dependent variable is the average wage for all full-time, full-year workers in each occupation.
aVariables are from a 1980 Bureau of the Census data tape (5 percent sample) unless otherwise indicated. Other
data sources are either keyed to the Reference list or identified as follows:
CPS = Bureau of the Census, Current Populaton Survey, May 1983 (data tape).
DOT = U. S. Department of Labor (1977~.
QES = Quality of Employment Survey (Quin and Staines, 1979~.
Time Use Survey = Survey of Time Use in Economic and Social Accounts, 1975-1976. Raw data tapes released
by the Institute for Social Research, University of Michigan, Ann Arbor.
bFor characteristics of occupations, means are the same for men and women in that occupation. For average
characteristics of workers, the means presented are weighted averages of means for men and women in the
occupation.
CCalculated for men as (age - schooling - six) from the 1980 census and for women as the mean predicted value
using characteristics from the 1980 census and estimates performed using actual work experience from the National
Longitudinal Survey.
OCR for page 171
Commentary
JAMES P. SMITH
My comments a(ldress, first, the paper
by Parcel and then the paper by Filer.
Parcel's analysis is a replication of an earlier
one on a similar theme using the 1970
census. Her current paper uses the 1980
census and deals with the basic empirical
question posed in comparable worth re-
search: Is the wage rate related to the
proportion of women in an occupation? The
dependent variable throughout is mean oc-
cupation-leve} earnings, converted to an
annual full-time earnings equivalent. In Par-
cel's framework, occupational incomes re-
flect four types of factors: an occupation's
intrinsic "worth," labor market conditions
specific to an occupation, the human capital
of the occupation's incumbents, and aspects
of social organization that are not reflected
in productivity differences.
Using a match of 1980 census data with
data from the Dictionary of Occupational
Titles (DOT) and factor analysis, Parcel de-
rived five variables to measure worth. Oc-
cupational labor market conditions are mea-
sured by several variables, such as the
occupation-specific unemployment rate, the
reserve labor market pool (defined as the
number of workers in that occupation who
171
are out of the labor force divicled by the
experienced civilian labor force), and the
number of government employees in an
occupation. Parcel argues that, at least to
some extent, there are distinct occupation-
specific labor markets. In essence, skill clif-
ferentiation produces labor market seg-
mentation along occupational lines. The
variables relater] to the quality of labor
supplied are standard human capital ones-
eclucation an(1 years of labor market ex-
perience.
The most important thematic concept,
especially for comparable worth, is the vari-
ables put under the hea(ling "dimensions
of social organization." Parcel argues that
these dimensions influence wage levels in-
lepenclent of productivity. Among these
social organization variables, the paper in-
clucles "the comparable worth variable"-
the percentage of workers who are women.
The paper also extends this concept to other
ethnic an(1 racial groups, adding the per-
cents black, Hispanic, and Asian as vari-
ables. The analysis also explores marital
status effects. The maintained hypothesis is
the higher the proportion of men in an
occupation, the higher the wages. The op-
OCR for page 172
172
posite prediction is advanced for the per-
centage of married women.
The main emphasis of the paper, of course,
concerns the variables measuring the de-
mographic composition of the workers in
an occupation. The comparable worth vari-
able, percent female, has the predicted neg-
ative coefficient. Moreover, the magnitude
of the effect is not small. An increase of 10
percentage points in the proportion female
would reduce average earnings by $710.
The other demographic variables also
matter. For example, a large proportion of
married men increased incomes, but at least
in this specification the proportion of mar-
ried women was not statistically significant.
The proportion who were black had no effect
on wages, but the fraction who were Spanish
speaking reduced incomes and the propor-
tion of Asians in the occupation increased
salaries.
In combination, Parcel interprets her re-
sults as indicating not only that dimensions
of social organizations are important, but
that there exists a richer array of these
dimensions than simply the fraction of the
workers in a job who happen to be women.
My criticisms of Parcel's paper relate to
the interpretation of the factors, the worth
variables, and the social organization vari-
ables. Some of the differences in interpre-
tation may simply reflect disciplinary dif-
ferences between Parcel and me. For
example, instead of arguing that the oc-
cupation-specific unemployment rate (at least
its Tong-term level) is a signal of excess
supply, it seems more plausible to me that
the causation runs in reverse. A number of
studies show that Tow-skill and low-wage
workers exhibit less attachment to their jobs
and to the labor market in general. In this
view, low wages are associated with high
levels of job turnover and, consequently,
high unemployment rates.
With respect to the worth variables, this
study, like many others in this area, does
not offer much condolence that occupational
worth is being adequately measured' by these
proxies. For example, the additional ex-
PAY EQUITY: EMPIRICAL INQUIRIES
planatory power addecT by the "worth" vari-
ables is modest incleect—an increase in acI-
justed R2 of only 1 percentage point. This
finding is consistent with micro-level studies
that show that only 1 to 5 percent of the
variance in wages is explained by DOT-
(lerive(l worth measures. Similarly, the
magnitude of the effects estimated for these
variables is quite small. A difference of one
standard deviation in working conditions or
substantive complexity, for example, raises
wages by only $145. These results cause
me to question seriously what seems to be
an implicitly maintained hypothesis. Even
putting aside whether the concept of in-
trinsic worth of an occupation has any mean-
ing, can we seriously claim that we have
meaningfully controlled for differences in
the worth of an occupation?
My question is not a criticism of Parcel's
work. Her construction of these worth vari-
ables was (lone with skill and goo(1 sense.
My question instead is a more generic crit-
icism of the research area as a whole.
My concerns with the social organization
(or comparable worth variables) are also
generic in that I have no particular (lifficulty
with the way Parcel con(lucte(1 her analysis.
It's the interpretation placed on the results
that troubles me. Does a variable such as
the proportion of workers in an occupation
who are women really a(l(1 anything to our
knowle(lge about how wages are set in labor
markets or whether there is discrimination
against any minority group? Although the
inclusion of such variables has become stan-
lar(1 (especially in papers centering around
comparable worth), I am puzzle(1 about its
meaning.
Let me illustrate with an example. Con-
sider two wage (distributions, one for men
and one for women. Although the distri-
butions overlap, men are paid significantly
more than women. All we know are the
(distributions, an(l for the moment we don't
know what causes the wage (distributions to
differ. Now let's consider drawing lines sep-
arating segments of the distribution. Ob-
viously, if I run a regression of the mean
OCR for page 173
COMMENTARY
wage in each segment on the proportion of
women in the segment, I would get a whop-
ping negative coefficient. If I call the seg-
ments occupations, I have the comparable
worth regression.
But have I learned anything from this
regression about why men anc] women are
paid clifferently? Any difference in the dis-
tributions guarantees this result, and I have
learned nothing more than what I started
with mate wage distributions lie above
female wage distributions.
Now, of course, we can control for other
factors that contribute to sex disparities in
wages, such as the worth and skill variables
Parcel includes. Our wage distributions would
obviously converge. But we also know that,
in the end, the conditioned wage distri-
butions will still indicate that male wage
distributions exceed those of women. Given
that, the same generic question can be
raised: What do we learn about the causes
of the remaining sex (lisparity by inclucling
a variable measuring the proportion female?
As long as male wage distributions lie above
women's (a fact that nobody disputes), the
average wage (for women and men, sepa-
rately or together) will be correlated with
proportion female.
In one way, Parcel's paper may put an
exclamation mark on my point. Her results
indicate that an increasing proportion of
Asians in an occupation raises wages. No
one will seriously argue that occupations
are more highly valued because they have
more Asians in them. If not, are we going
to be asymmetric in our interpretation of
these demographic variables? If a protected
minority group has a positive coefficient on
a comparable worth variable, is it unob-
served components of skill? And if it is a
negative one, is it discrimination?
Filer's paper) deals with this same ques-
iMy comments are based on Filer's presentation of
an earlier version of his paper at a Pay Equity workshop
in September 1987 and have been somewhat revised
to take into account some of the revisions Filer made
. . .
in nls paper.
173
tion, which has become the central empir-
ical question posed in research on compa-
rable worth: Is the wage related to the sexual
composition of an occupation after you con-
trol for other relevant factors, including the
productivity of incumbents, the amount of
effort and responsibility involved, working
conditions, and tastes?
Filer argues that a major shortcoming of
previous work on this topic is that each
existing micro-level data set contains only
a subset of these relevant other factors that
may matter. He tries to overcome this short-
coming (and he does so with a vengeance)
by obtaining occupation-level data from a
variety of sources. These data are subse-
quently merged, and the analysis is per-
formed on the merged data set. For ex-
ample, mean earnings and the characteristics
of the incumbents in an occupation are
derived from the 1980 (lecennial census;
measures of working conditions, effort, and
responsibility are derived from a variety of
sources, inclu(ling the DOT, the Quality of
Employment Survey, the Current Popu-
lation Survey, ant] the Michigan Time Use
study; a measure of personality is derived
from his own research; and so on. The end
results were 400-o(ld occupational groups
and 225 potential covariates.
The thematic regression in Filer's paper
involves regressing the average occupation-
specific wage on an ever-expanding list of
variables. One variable that remains
throughout is the proportion of workers in
the occupation who are women. Indeed,
the main exercise being performed is to
track the estimated coefficient on the pro-
portion female to see if it (1) goes away or
(2) withstands the onslaught.
The correct answer is (2), but it does get
a little batterer] in the battle. Filer sum-
marizes his results very nicely in his Table
7-1. Starting with a coefficient of $4.41 for
full-time, fulI-year workers (moving from an
occupation with no women to one that says
"la(lies only"), the effect becomes progres-
sively smaller. It falls to $3.30 when census-
clerived productivity variables (eclucation,
OCR for page 174
174
PAY EQUITY: EMPIRICAL INQUIRIES
experience, etc.) are added; falls to $1. 70 vices a firmer foundation (in terms of prior
when all variables are added subject to a
statistical sicle condition of maximizing ad-
justec] R2; and to $1. 35 when all 225 variables
are inclucled. All coefficients are statistically
significant.
The last number, $1.35, implies that the
actual difference between men and women
in the sex composition of their occupations
would explain roughly 20 percent of the
wage gap. We may all react clifferently to
this. My own reaction is that this still con-
stitutes an impact of significant magnitude.
In the initial version of his paper, Filer
argued that if this negative coefficient results
from a crowding effect or because employers
devalue occupations with lots of women in
them, male and female members of an oc-
cupation should be similarly affected. With
this in mincI, the identical regression mode!
was reestimated for men and women sep-
arately. In these regressions, percent female
becomes small in magnitude and statistically
. r.
ns~gn~cant.
Filer argues that this result shows that
comparable worth is a solution for a problem
that floes not exist. Most of the female-maTe
wage gap that remains after productivity
and job characteristics are taken into account
can be attributed to within-occupation sex
differences.
Before getting into more substantive mat-
ters, let me ask why so much of the analysis
on comparable worth is performed on the
aggregate occupation level instead of the
individual level. In both the Parcel and
Filer papers, the dependent variable was
mean wages in an occupation. Instead, the
comparable worth variable (the fraction fe-
male) could simply be added to a standard,
gar :len-variety wage function (much as So-
rensen, in this volume, does). The inter-
pretation of the comparable worth variable
would be pretty much the same in either
case. The advantages run in favor of the
microapproach: It allows for richer possi-
bilities of statistical testing, avoids problems
of what is lost in the aggregation, and pro-
literature) for what to expect from other
covariates. The absence of many variables
in the micro-level data sets is no particular
barrier either. The average characteristics
(personality, working conditions) of occu-
pations can simply be merged into a micro-
leve! data set in much the same way Filer
does. The use of richer variation in micro-
leve} data would help us avoid some of the
problems that Filer notices in his paper—
the lack of any effect of work experience
on wages for women, a small effect of work
experience for men, and the large effect of
schooling on earnings. When results go
against well-establishe(1 research results, I
also become concerned about what else may
be happening to other variables that we are
taking quite seriously because they are sta-
tistically significant.
My first point about the method Filer
used (as opposed to one that he dill not) is
the large number of independent variables.
Although that may pose no problem if the
objective is to test how much firepower the
coefficient on percent female can withstand,
it (loes create preclictable problems with
the believability of estimates for other things.
Let me illustrate with some not-so-random
selections. If everyone has clifficulty speak-
ing English, wages in an occupation rise by
$11. 52. Similarly, if we enter an occupation
in which we all goof off, wages go up by
$69 an hour, a whoppingly significant effect!
(Perhaps the last two results go a long way
toward explaining why economists are pair]
so well.) These are not isolatecl examples.
My intent is to raise an issue concerning
the central purpose of the paper. If it is
simply to put the coefficient of percent
female to the test, then most of what hap-
pens to other coefficients is not important.
But our concern should increase if the coef-
ficients of other variables are taken seri-
ously. Then, the believability of the entire
exercise is questionable. The former objec-
tive is more important; in this revised ver-
sion of the paper, Filer has appropriately
OCR for page 175
COMMENTARY
Reemphasized the structural wage function
interpretations .
Let me now argue the other side of the
issue. Clearly, the approach taken by Parcel
in her paper is a much cleaner one. Instead
of presenting every subcomponent, she first
aggregates the worth variables into mega-
groups using factor analysis. Instead of all
the variables in Filer's paper, Parcel uses
five. But Filer's approach actually opens up
important issues underlying the comparable
worth debate, issues that also tend to sep-
arate economists and sociologists. The ques-
tion is how do we want to aggregate the
subcomponents measuring worth into me-
gabundIes. Parcel's approach looks first only
at statistical information within the sets of
variables, but never looks at what impact
the subcomponents have on market earn-
ings. The problem with that approach is
that variables that may win an internal sta-
tistical information battle may have little
prominence in the marketplace. To an econ-
omist's eye (this one included), the weights
attached in aggregate should bear some
relation to market earnings. In Filer's ap-
proach, all the coefficients are related to
market earnings because all variables enter
the wage equation.
175
paper. Let me make a few additional points
on some extensions Filer puts forth.
Filer states, and correctly, that his finding
that percent female in the mate salary anal-
ysis is insignificant is provocative. It cer-
tainly is. Perhaps it is most provocative in
that it runs counter to all the other estimates
in the literature, including many ofthe other
papers in this volume. If true, given my
reservations about interpreting the com-
parable worth regression in any case, I
would find this as stronger evidence of
discrimination than the comparable worth
result itself. In essence, this result suggests
that men and women in the same occupation
(i.e., the same percent female) are paid
differently. The problem, of course, with
this interpretation is that there is so much
sex segregation that men and women may
not really be in the same jobs, even when
occupations with the same percent female
are considered. Thus, Filer's interpretation
that much ofthe problem is wage differences
within occupations may be inaccurate, though
the mechanism he describes in the final
section of his paper (the movement of wom-
en into occupations in which the level of
pay is higher and the female-male pay gap
is lower) would work well for discrimination
My second major point about Filer's pa- for any cause. Filer's public policy rec-
per relates to the "comparable worth vari- ommendations would thus be to encourage
able," already discussed above for Parcel's the job mobility of women.
OCR for page 176
Representative terms from entire chapter:
labor market
