1. style and fashion
2. men 's fashion

GENDER EARNINGS DIFFERENTIALS IN TOTAL PAY,
BASE PAY,AND CONTINGENT PAY
KEITH W. CHAUVIN and RONALD A. ASH*
Using data froma 1988 surveyof business school graduates, the authors
analyze gender differentialsin earningsbyformof pay-total pay,base pay,
and contingentpay-with controlsforhuman capital, occupation,job level,
and individual characteristics. The results indicate that within narrowly
defined occupations and jobs, most of the unexplained differencein total
pay between the men and women in the sample was due to gender differences in the portion of pay thatwas contingenton job performance. The
greaterimportanceof contingentpayin the earningsof the men than of the
women may reflect differentialtreatmentof men and women by firms,
gender differencesin performance,gender differencesin riskpreferences,
or some other sortingmechanism.
M
ostattempts
toexplainthegap inpay
between men and women have focused on sets of explanatoryvariables included in wage equations. Typically,annual earningsor hourlywagesare regressed
on human capital and other productivityrelated measures of thejob and individual.
These variablesaccount forpartbut not all
of the gender-paygap. The unexplained
portion of the pay gap is ofteninterpreted
as theeffectofmarketdiscrimination.Since
itis possible thatthe failureof these regressions to explain all of the pay gap is due to
omittedproductivity-related
variables,how-
*KeithChauvin is Associate Professorof Business
and Ronald Ash is Professorof Business, both at the
Universityof Kansas. This research was supported, in
part,byfundsfromthe General Research Fund of the
Universityof Kansas. The authors thank James
Guthrie,JudyOlian, and Sara Rynesforhelpful commentson an earlier draftof the paper.
ever, researchers are constantlysearching
forbettermeasures of productivity.So far,
better measures of the explanatory variables have not significantlyreduced the
unexplained portion of the pay gap.
In thispaper we attemptto move toward
an understandingofthecauses oftheunexplained portion of the gender-paygap by
exploringa differentdimension of thegap.
Rather than simply tryingto account for
the gender differencein pay, we analyze
how the pay gap varies by typeof pay. The
data setwe use, drawnfroma 1988 surveyof
business school graduates, includes measures of individuals' base pay and pay contingent on job performance. We test
usedtogenerate
Copiesofthedataand programs
the resultswillbe availableafterMay1995 forpurfromtheauthorsat theSchoolof
posesofreplication
of Kansas, Lawrence,Kansas
Business,University
66045.
Industrialand LaborRelationsReview,Vol. 47, No. 4 (July1994). ? by Cornell University.
0019-7939/94/4704 $01.00
634
Sage Publications, Inc.
is collaborating with JSTOR to digitize, preserve, and extend access to
Industrial and Labor Relations Review
www.jstor.org
®
GENDER EARNINGS DIFFERENTIALS
635
gap decline significantly.Reviews of the
literatureshow convincinglythat most of
thegender gap in pay,holding constantthe
traditionalmeasures of human capital, occurs across occupations. Furthermore,
nearlyall of the relativelysmall difference
in pay that exists withinnarrowlydefined
jobs and occupations occurs across firms
(Blau 1977; Groshen 1991). Studies that
look at pay differenceswithinjobs, within
firms,findlittleor no gender differencein
pay. Gerhart (1990), for example, found
that when human capital and other relevant
factorswere controlled for, the feResearch on the Gender-PayGap
male/male pay ratio withinjobs in a single
firmwas between .95 and .98. We know
Studies of why men earn, on average,
fromthe existingresearch, therefore,that
more thanwomen continue to
significantly
find that a large percentage of the differ- nearlyall of the differencein pay between
ences in earnings and wages cannot be
men and women is due to women being
explained bydifferencesin human capital,
disproportionatelyrepresented in lowerdemographic characteristics,or firmchar- paying occupations and in lower-paying
acteristics.Even aftercontrollingforbroad
firmswithinoccupations. This conclusion
is supported byreviewsof recent studiesof
occupational categories, approximately
50% of the earningsdifferentialcannot be
thegender-earningsgap (forexample, Cain
accountedforbydifferencesin endowments 1986; Blau and Ferber 1987a; Gunderson
1989).
betweenmen and women. Since measures
that allow a comparison of productivity
The extent to which marketdiscriminaacrossworkers,jobs,occupations,and firms tion is or is not a cause of the differencesin
are not available, the approach used in
mobility across jobs and in the occupaalmost all attemptsto explain the "unextionaldistributionbetweenmen and women
plained" portionof the paygap has been to
cannot be inferredfromthese studies. Altryto obtain marginallybetterproxies for
though the findingsindicate that much of
the gender differencein pay occurs across
productivity.The qualityof education, actuallabormarketexperience,interruptions occupations and jobs, Blau and Ferber
in labor marketexperience, and firmten(1987a, 1987b) pointed out thatwhen controlssuch as occupation orjob are included
ure, for example, have been included in
in the estimate of the wage equation, the
recent studiesof the pay gap. Measures of
productionin thehousehold have also been
gender-paygap tendsto be underestimated.
included in earningsmodels. Responding
This underestimation occurs because the
to the argument by Becker (1985) that
choice ofjob or occupation has likelybeen
gender differencesin household responsi- influenced by market discrimination. If
bilitiesmayaccount forpartof the paygap,
the rate of return on educational investHersch (1991), for example, used a new
ments in engineering is lower for women
data set thatincluded measures of the inthan for men because of marketdiscrimitensityof household and child care duties.
nation against female engineers, for exShe found that these factors,and specific
ample, then women will invest less than
job characteristics,added verylittleto the
men in thistypeof education. In thiscase,
the "choice" of women not to go into engiexplanation of the pay gap.
Only when controlvariablesfordetailed
neering is partlythe result of marketdiscrimination. At best, a lower bound estioccupational category or the job are included in the estimatesof the gender-pay mate of the effectsof discriminationis obtained when variables that may be influgap does the unexplained portion of the
whether the unexplained portion of the
pay gap varies between these two typesof
pay. The processes and managerial practices that determine base pay are verydifferentfrom those that are used to determineperformance-contingent
pay. Because
of these differences,we argue that understanding how the pay gap and the unexplained portion of the gap are distributed
between these two formsof pay is potentiallyuseful for identifyingor narrowing
the possible causes of the pay gap.
636
INDUSTRIAL AND LABOR RELATIONS REVIEW
enced bydiscriminationthroughfeedback
effectsare included in the analysis.
To betterunderstand the causes of the
pay gap, and notjust the size of the gap, it
is necessaryto understandhow the choices
of occupations and firmsdifferbetween
men and women, how firms'recruitment,
selection, promotion, and pay determination practices vary across jobs, and how
these practicesvaryacross firms.If market
discriminationis a significantfactorin the
segregationofwomen acrossjobs or across
firms,then these practices are the mechanisms through which discriminationand
the related feedback effectsoccur.
In thisstudywe askwhetheror not,within
narrowlydefined occupations or jobs, the
gender-paygap varieswiththe formof pay.
Is the level of pay the only differencebetweenjobs in high-payingfirmsemploying
men andjobs in lower-payingfirmsemploying women? One characteristicthatnearly
all of the studies of the pay gap have in
common is theiruse of the level of hourly
wages or earnings as the dependent variable. Although there is a growingbody of
empiricalliteraturefocusingon compensation and the incentive effectsof various
forms of pay (for example, Lazear 1986;
Abowd 1990; Brown1990; Kahn and Sherer
1990), only one attempthas been made to
test whether the gender-pay differential
varies across differentformsof pay. This
factis partlydue to the nature of the data
setsthatare regularlyanalyzed forexplanations of the gender-paydifferential.Most,
if not all, large public data sets provide
measures of totalearningsor hourlywages,
but they do not provide the information
necessaryto distinguishamong the forms
of pay.
The one studythat does provide some
informationon differencesin the formof
payis Goldin (1986). Using data from1890
to 1940 formanufacturingand clericaljobs,
Goldin found thatwomen were more likely
than men to be paid piece rates. This
findingis consistentwitha model in which
workerswithhigh turnoverratesare sorted
intojobs or occupations withpiece rates as
opposed to timerate methods of pay determination. Goldin's model shows that this
sorting reduced monitoring costs associated withhigher turnoverrates.
The analysisin thispaper uses data from
a surveyin which respondents reported
their total annual pay, theirbase pay, and
the part of their total pay thatwas contingent on theirjob performance. Examples
of contingent pay are sales commissions
and performancebonuses. Using thisdata
set,we estimatethe gender gap in totalpay,
base pay,and contingentpay. We attempt
to answer the simple question: does the
unexplained portion of the gender gap in
totalpay,withinjobs,varybetweenbase pay
and contingentpay?
Base Pay and ContingentPay
The importantrespect in which thispaper departs from previous studies is our
disaggregationof the dependent variable,
total pay,byformof pay. Total pay equals
the sum of base pay and contingent pay.
Base pay is the partof the totalsalaryor pay
thatdoes not depend in a givenyearon the
individual'sjob performance. Base pay is
determined largely by the level of an
individual's job in the firm's hierarchy.
Changes in base pay are determined by
year-to-yearadjustments from merit pay
awards and frompromotions (demotions)
to higher (lower) job levels. Merit pay
adjustments are based on a supervisor's
performance evaluation of the individual
duringa givenperiod of time;theybecome
a permanentpartofthebase paybeginning
in the nextpayperiod. Base pay,therefore,
varieswithjob performanceover time,but
not during a given pay period. Although
meritpay adjustmentsaccount forthevariance in base pay withina specificjob level
in a givenfirm,mostof thevariance in base
payreflectsdifferencesamongjobs in their
levels in the hierarchy.
Contingentpay awards such as bonuses,
commissions,and profitand gain sharing
awards,on the other hand, depend primarilyon the individual'sjob performance,or
the performanceof a group in the case of
gain sharing and profitsharing, during a
given period of time. When the individual
(or the individual's group) achieves a pre-
GENDER EARNINGS DIFFERENTIALS
specified performancestandard,he or she
receives a pre-specifiedpay award. Unlike
meritpay adjustments,these payawardsdo
not become part of the base pay and are
paid in subsequent timeperiods onlyifthe
individual (or group) continues to achieve
the performancestandard. All else equal,
as contingentpayincreases as a percentage
of total pay, the individual's risk of shortrunvariationsin payas a resultofvariations
in hisor herjob performancealso increases.
This differencebetween base pay and
contingentpay allows for a unique test of
the extent to which the gender-pay gap
varieswiththe amount of riskin pay thatis
borne by the individual. If firmsdiscriminate against women because of a priori expectationsthatwomen are more likelythan
men to leave thefirmduringanygiventime
period forfamilyor other non-job-related
reasons, then firmsmay attempt to force
women to bear more of the riskfor shortrun variations in theirjob performance.
This relationshipwould be consistentwith
the findings of Goldin (1986). The expected relationship between the genderpaygap and theformofpay,however,is not
clear. Systematicdifferencesin riskpreferences between men and women, for example, could cause women to select into
jobs and firmswith less pay risk than the
jobs and firmsselected by men, or vice
versa. Subich et al. (1989), based on a study
of college juniors and seniors who were
about to enterthe labor market,found that
women were less likelyto take risksin their
occupation andjob choices thanwere men.
Based on theirreviewof the literature,they
also provided a substantialamount of evithan
dence thatwomenare morerisk-averse
men in theirbehavior in the labor market
and in other activities.
The purpose of this studyis not to estimate the effectsof discrimination. We
attempt,instead, to estimatethe extent to
which the unexplained portion of the pay
gap varies between base pay and contingent pay. Since the base pay/contingent
pay mix is likelyto varywiththe level of the
job in the firm,we hold constant the job
level byincluding controlsforjob mobility
and access thatmayresultfromchoices and
637
abilities of individuals,barriersto job mobilityconfrontingwomen,feedback effects
fromsuch discrimination,or some combination of these factors. Controlling for
these endogenous factorsis likelyto result
in underestimation of the effectsof discrimination. Once we control for thejob
level, however,we willbe able compare the
remainingunexplained portion of the pay
gap in base pay withthe unexplained portion in contingentpay. If the gender-pay
gap does not varyby formof pay, then we
would expect, controllingforthejob level,
thattheunexplained percentage of thegap
would be the same forbase pay and contingent pay.
The Data
The data analyzed in this paper come
froma surveyof business school graduates
of twolarge stateuniversities.This data set
was originallycollected by George Dreher
and Ronald Ash forthepurpose ofstudying
gender differencesin mentoring. The results of that studyare reported in Dreher
and Ash (1990). The data setwas collected
in 1988 through a mail survey of 1,000
graduates fromthe MBA and undergraduate classes of 1978 and 1983. The sample
was drawn to generate an approximately
equal numberof graduates byschool, year,
gender, and degree. 440 surveyformswere
returned.' For purposes of the current
studywe did not include the 48 respondentswho reportedworkingpart-time(less
than 35 hours per week) or the 28 individuals who were self-employed.Of the remaining 364 responses, there are 312 usable
observations. There are 168 men and 144
women in the sample.
Given the population surveyed,several
sources of variation in pay are eliminated
(or reduced) in the data set. Variances in
quantityand qualityofeducation, lengthof
workexperience,and occupation are greatly
constrained. Since onlythegraduatesfrom
122of the surveyswere returnedbecause of wrong
addresses. The response rate, therefore,was 45%,
whichis similarto response ratesin other surveysthat
use similar methodologies.
638
INDUSTRIAL AND LABOR RELATIONS REVIEW
1978 and 1983 were sampled, the data set is
restrictedto a fairlywell-definedage/experience cohort. The average age of those
sampled is 32, and 95% of the individuals
are between 26 and 40 years old.
The data set includes an extensiveset of
measures of human capital, firm,occupation, and individual characteristics. Also,
the respondentsto the surveywere asked to
report the total annual pay theyreceived
and to disaggregatetheirtotalpay between
base pay and contingentpay.2
Below is a description of the variables
used in our analysis. The discussion of the
variables is divided into measures of pay,
human capital, job and firmcharacteristics, and individual characteristics. Variable definitionsand sample means, bygender, are summarized in Table 1.
Measuresofpay. The means in Table 1
show that there is a large and statistically
significantdifferencein pay between men
and women in thissample. The mean pay
formen is $59,597,and themean forwomen
is $45,044. Althoughthisratio of female to
male pay, .76, may seem small given the
narrowpopulation sampled, itis consistent
with recent surveydata of the gender-pay
ratio in professionaloccupations. Bureau
of Labor Statisticsfiguresfor1988 indicate
thatthe median weeklyearnings of female
professionalsand managerswas 70% of the
median weeklyearningsof men. The ratio
in sales and technicaloccupations was 65%.
The base pay ratio in this sample is .87.
Of the $14,553 gap in total pay, $6,100 is
due to the differencein base pay. The
contingentpay ratio is only.31. Of the gap
in totalpay,$8,467 is due to the difference
in contingent pay. Despite the fact that
contingentpay is a relativelysmall portion
oftotalpay,12% formen and 6% forwomen,
the differencein contingent pay between
2The surveyasked respondents to report total directearningsfromthe employer,excluding non-cash
benefits,and also to report the components of total
earnings, including annual salary, annual income
fromcommissions,and supplementaryincome such
as bonuses and profitsharing. Our measure of contingentpay is the sum of the reported commissions
and supplementaryincome.
men and women accounts for 58% of the
differencein total pay. These raw data
suggestthatthere maybe importantdifferences between the waysin which base pay
and contingent pay contribute to the differentialin total pay.
Human capital.The data setcontainsone
measure of formal education and several
measures of work experience. Sampling
only business school graduates simplifies
the measure of education: DEGREE is a
dummyvariable indicatingwhetheror not
the respondent has a graduate degree in
business. The measures of work experience include EXP, the totalnumber ofyears
of actual workexperience, YRS, the number
of years of work experience since graduation frombusiness school, and TENURE, the
numberofyearswiththecurrentemployer.
Four additional measures of workexperience help qualifythe more standardmeasureslistedabove. INTERRUPTIONS iS the total
number of months since graduation duringwhichtherespondentwasnotemployed.
JOB CHANGES
is
the numberof timesthe
respondent changed firmssince gradua-
tionfrombusinessschool. JOBS
REJECTED is
the numberofjob offers(including withinfirmtransfersor promotions) since graduation that the individual rejected because
the offerwas not compatible withfamilyor
spouse needs. This variable is included to
control for the possibilitythat women accept lower rates of returnon theirhuman
capital in order to "followtheirhusbands."
The means of thisvariables indicate, however, that women in this sample rejected
fewerjobs for family-relatedreasons than
did the men in the sample. This resultmay
indicate thatmen had morejobs to reject,
thata higher percentage of the men in the
sample than of the women are married,or
thatmarriedwomen in thisage cohort are
no longer followingtheirhusbands.3
3This differencebetween men and women in the
number of jobs rejected for family-relatedreasons
also holds bymaritalstatus. The mean numberofjobs
rejected forfamilyreasons by married women in the
sample is .46, whereas the mean for married men is
.72. For single women and single men, the means are
.17 and .5, respectively.
GENDER
EARNINGS
639
DIFFERENTIALS
Table 1. Variable Definitionsand Means of a Sample
of 312 Business School Graduates fromTwo Large State Universities.
(Standard Deviations in Parentheses)
Variable
Definition
TOTAL PAY
=
total annualjob earnings ($$)
BASE PAY
=
annual base pay ($$)
CONTINGENT PAY
=
annual pay thatwas contingenton job performance ($$)
DEGREE
=
1 formaster's degree and 0 for bachelor's degree
EXP
=
total yearsof workexperience
YRS
=
yearsof workexperience since graduation
TENURE
=
yearswithcurrentemployer
INTERRUPTIONS
=
number of months out of the workforce since graduation
MENTOR
=
amount of mentoringreceived between 0 and 5
JOB CHANGES
=
number ofjob (firm) changes since graduation
JOBS REJECTED
=
number ofjobs rejected for family-relatedreasons
FIRM SIZE
=
0 for less than 50 employees ... 7 formore than 50,000
INDUSTRY
= 1
for manufacturing;0 otherwise
PROFESSIONAL
= 1
for professional occupation; 0 otherwise
TECHNICAL
=
1 for technical occupation; 0 otherwise
SALES
=
1 for sales occupation; 0 otherwise
HOURS
=
average weeklyhours
MARITAL
=
1 if married; 0 otherwise
DEPENDENTS
=
number of dependents
FAMILY TIME
=
% of time spent withfamily
SUPERVISION
=
number of supervisorsreportingto the respondent
PROMOTIONS
=
number of promotions received since graduating
%CONTINGENT
=
contingentpay/totalpay
n
Men
Women
59596
(44445)
47309
(22587)
12286
(37586)
.649
(.479)
10.185
(3.999)
7.929
(2.650)
5.292
(3.720)
1.774
(3.989)
2.922
(.770)
1.375
(1.503)
.655
(1.309)
4.179
(2.309)
.268
(.444)
.393
(.490)
.071
(.258)
.196
(.398)
50.804
(8.377)
.750
(.434)
.857
(1.011)
25.827
(13.795)
.976
(1.209)
3.494
(2.235)
.125
(.179)
45044**
(21848)
41210**
(17261)
3819**
(6922)
.514*
(.502)
9.465
(4.090)
7.340
(2.600)
4.819
(3.870)
2.771
(5.673)
3.006
(.855)
1.472
(1.496)
.354
(.642)
4.229
(2.317)
.181
(.386)
.486
(.502)
.035
(.184)
.215
(.412)
47.938*
(6.444)
.604*
(.491)
.444*
(.834)
26.611
(16.492)
.340*
(.681)
3.278
(1.815)
.062*
(.086)
168
*Significantdifferencein means between men and women at the .05 level; **at the .01 level.
144
640
INDUSTRIAL AND LABOR RELATIONS REVIEW
Finally,the data set includes a measure
of the intensityand quality of mentoring
received by the respondent (MENTOR).
Respondents indicated the extent to which
they had experienced various forms of
mentoringfromformallydesignated mentors,sponsors,or influentialmanagerssince
their graduation from business school.
The data set includes two other measures that help to control for the level of
the individual'sjob in the firm'shierarchy.
SUPERVISION is a measure of supervisoryresponsibilities, indicating the number of
supervisorswho report to the respondent.
It is coded 0 ifno supervisorsreportto the
individual; 1 if 1-2 supervisorsreport; 2 if
3-5 report; 3 if 6-8 report; and 4 if 9 or
MENTOR rangesbetween1 and 5, and is the
average score on 18 items on a mentoring more report. PROMOTIONS is the total numscale used in the survey.4 Since mentors ber of promotions the individual has received since graduation. Although these
usuallyaid subordinateswithcareer develvariables may be closely related to the huopment, involve them in networking,and
man capital measures and may proxy for
provide them with informationabout the
firm,including mentoringin the analysis job productivity,
differencesin supervisory
should augment the measure of firmten- responsibilitiesand promotions may also
ure as a proxy for the amount of firm- be the result of discriminationthat limits
the upward mobilityof women. We have
specifictrainingtheindividualhas received.
Job and firm characteristics.There are
included these twomeasures in our discussion of job characteristics because they
severalvariablesrelatingto theindividual's
clearly proxy for mobility and the
occupation, job, and firm. FIRM SIZE is a
measure of the numberofemployeesin the
individual's level in the job hierarchy,rerespondent'sfirm.5INDUSTRY is a dummy gardlessofwhetherjob level is the resultof
discrimination or productivity. As exvariable indicatingwhetherthe firmis in a
manufacturingindustry.HOURS is the aver- plained in the previous section, we are interestedin controllingforjob level even if
age weekly hours of work. PROFESSIONAL,
job level may be determined partlyby disTECHNICAL, and SALES are occupation dummies. The excluded occupation is managecrimination.
rial. These occupation variablesconstitute
Individual characteristics.This data set
contains informationabout maritalstatus
relativelynarrow occupational categories,
and number of dependents. These meareflectingthe fact that the population includes only business school graduates.
sures are traditionallyincluded in analyses
of the gender-paygap. The data set also
includes a measure of the time each individual spends with his or her family. The
4Aglobal measure of mentoringpractices was desurveyasked respondents to indicate what
veloped in Dreher and Ash (1990) by selecting items
of the total amount of time
percentage
used in previous research (Noe 1988; Whitelyet al.
spent on work,recreation, and familyac1988). The items were selected to cover the various
tivitieswas devoted to their family(FAMILY
careerand psychologicalfunctionsdescribed byKram
(1985). When answering these questions, responOne reason for the gender-paygap
TIME).
dents were asked to "consider your career history
be that women have more familyremay
since graduating from [your] program and the desponsibilities than men and, therefore,
gree to which influential managers have served as
lowerlevelsof effortavailable forjob activiyoursponsor or mentor (this need not be limited to
one person)." The response formatranged from a
ties (Becker 1985). The average of FAMILY
low of "1 = not at all" to a high of "5 = to a verylarge
TIME is one percentage point higherforthe
extent." the internalconsistency(coefficientalpha)
women in the sample than forthe men, but
forthe scale was .95. See Dreher and Ash (1990) for
the differenceis not statistically
significant.
a more complete discussion of the mentoring variAs mentioned above, Hersch (1991) inable and fora list of the 18 mentoringitems used to
calculate the measure.
cluded controls for household responsi5FIRM SIZE is coded as zero for1-49 employees; 1 for
bilitiesand found thatthe reduction in the
50-99; 2 for 100-499; 3 for 500-999; 4 for 1,000unexplained portion of the pay gap attrib4,999; 5 for 5,000-9,999; 6 for 10,000-49,999; and 7
utable to these factorswas verysmall.
for 50,000+ employees.
GENDER EARNINGS DIFFERENTIALS
Empirical Models
For total pay and base pay, we estimate
separatelyfor men and women the equation
(1I)
ln(PAYx)= B X. + e,,
where ln(PAY)is the natural log of the pay
variable, B is a vectorof coefficients,and X
is a vector of the human capital,job, and
individual characteristicsdescribed above.
We use the Blinder (1973) method for
decomposing the paygap forthe measures
of total pay and base pay in this sample.
This is the standard method used for decomposing the pay gap between the portion of the gap due to differencesbetween
men and women in their endowments
(means) of the explanatoryvariables and
the portion due to differencesin the rates
of return (coefficients) to those endowments.
The measure of contingent pay, however,does not allow us to use a comparable
method of analysisof the gap forthisform
of pay. Contingent pay is censored at a
lowerbound ofzero. Althoughindividuals
can earn positive pay awards by achieving
or exceeding specific performance standards, compensation policies generallydo
not specifythatthe individualmustreturn
pay for failing to meet the performance
standard. In this sample of 312 observations, the contingent pay variable has a
value of zero in 132 cases: 59 out of 168
cases formen and in 73 out of 144 cases for
women.
Because of thiscensoring of the contingent pay variable, we estimate the gender
differencein contingent pay with a standard tobit model, using a dummyvariable
forgender. Since contingentpay is zero in
a large number of cases, we do not take the
normal log transformationof the pay variable. The focus of our analysis,therefore,
is on total pay and base pay, and on how
the unexplained portion of the pay gap
differs between total and base pay. A
tobit model of contingent pay is estimated simply to test whether or not the
difference in contingent pay, holding
other things equal, between men and
641
women in this sample is statisticallysignificant.
The tobit model of contingent pay is
specified as
(2)
PAY>
B X.
+
+ V.
b2(GENDER)
PAY. = PAY*
if PAY* >
0
=0 if PAY*< 0,
where PAY* is the value of contingent pay
earned, PAY.is the observed level of contingent pay, and GENDER is a dummyvariable
equal to 0 forwomen and 1 formen. Under
theassumption
thatv.is
distributed
N (0,cy2),
we estimatethe coefficientsbymaximizing
a tobitlikelihood function.
OLS Results and Decomposition of the
Differentialin Total Pay and Base Pay
OLS estimatesofmodel (1) are presented
in Tables 2 and 3. Table 2 reports the
results for the estimates of the total pay
model, and the results for base pay are
reportedin Table 3. We firstestimatedthe
equations withoutthe SUPERVISION and PROMOTIONS variables. In both tables, these
resultsare reportedin columns (1) and (2)
for men and women, respectively.Results
for estimates including
MOTIONS
SUPERVISION
and
PRO-
are reported in columns (3) and
(4).
Of the experience and tenure variables,
only years since graduation (YRs) is signifi-
cant. Including quadratic terms for the
experience and tenure variables did not
change theresults,and thequadratic terms
were not significant.These resultsmaybe
due to the limited range of experience in
the sample. They may also be due to the
otherexperience-relatedvariablesincluded
in the equation. Interruptions in work
experience have a significantnegative effecton both totalpay and base pay. Except
fortheeffectofthe numberofjobs rejected
for familyreasons, the coefficientson the
human capital variablesare in the expected
direction. The coefficientsonJOBS REJECTED
are positiveforboth men and women, but
insignificant. This finding may indicate
INDUSTRIAL AND LABOR RELATIONS REVIEW
642
Table 2. Factors Affecting Total Pay:
OLS Results for ln(TOTAL
PAY).
(Standard Errors in Parentheses)
(1)
Variable
DEGREE
EXP
YRS
TENURE
INTERRUPTIONS
MENTOR
JOB CHANGES
JOBS REJECTED
FIRM SIZE
INDUSTRY
PROFESSIONAL
TECHNICAL
SALES
HOURS
MARITAL
DEPENDENTS
FAMILY TIME
Men
(2)
Women
.209** .200**
(.074)
(.058)
(4)
Women
.210** .207**
(.073)
(.057)
.0001
.012
(.011)
(.009)
.059** .053**
(.016)
(.013)
-.005
.005
(.014)
(.009)
-.019*
-.010*
(.009)
(.005)
.066
.053*
(.045)
(.031)
-.045
-.012
(.033)
(.023)
.040
.028
(.027)
(.041)
.009
.037**
(.016)
(.013)
-.040
-.028
(.077)
(.067)
-.026
.025
(.079)
(.067)
-.093
.249
(.138)
(.161)
.118
.123
(.094)
(.079)
.015** .015**
(.004)
(.005)
.104
.027
(.099)
(.062)
.011
-.005
(.040)
(.037)
-.003
.002
(.003)
(.002)
.002
(.011)
.046**
(.017)
-.006
(.013)
-.017
(.009)
.047
(.045)
-.044
(.033)
.035
(.026)
.012
(.016)
-.070
(.077)
.029
(.080)
-.037
(.139)
.187
(.097)
.014
(.009)
.039**
(.014)
.009
(.009)
-.008
(.005)
.034
(.032)
-.018
(.023)
.031
(.040)
.035**
(.012)
-.026
(.065)
.059
(.067)
.219
(.159)
.145
(.079)
.012**
(.004)
.134
(.099)
.005
(.040)
.001
(.003)
.060
(.032)
.027
(.017)
.012*
(.005)
.053
(.061)
-.010
(.038)
-.003
(.002)
.059
(.041)
.040*
(.017)
9.181** 8.966**
(.309)
(.261)
9.239** 9.027**
(.306)
(.260)
SUPERVISION
PROMOTIONS
CONSTANT
(3)
Men
Adj. R2
SEE
.349
.275
.413
.580
.523
.294
.379
.299
.406
.604
.543
.288
n
168
144
168
144
R2
*Statisticallysignificantat the .05 level; **at the
.01 level (2-tailed tests).
thatJoBsREJECTED is measuringjob mobility
(more offersto move) ratherthanjob immobilitydue to familyties.
The resultson the individual characteristicsare consistentwithfindingsof other
studies. The coefficientson maritalstatus
are positive in each specification,and in
each case the coefficientfor men is larger
than thatforwomen. In most of the specificationsthe coefficienton the family-time
formen.
variableis positivebutinsignificant
The resultsare negativeand significantfor
women in the total pay equation that includes SUPERVISION and PROMOTIONS, and
negativeand significantforwomen in both
base pay equations.
The coefficientson the SUPERVISION and
PROMOTION variables are positive for men
and women in both the total pay and base
pay equations. The coefficientson the
PROMOTION
variable are significantin each
case. The coefficientson the SUPERVISION
variable are significantat the 10% level for
men and not quite significantforwomen.
Despite the factthatthe data set is small
and was drawn from a relativelynarrow
segment of the workforce, the resultsare
consistent,more or less, withexpectations
based on previous studies. Furthermore,
the resultsforthe total pay model are very
similarto those for the base pay model.
FollowingBlinder (1973), we decompose
the estimated pay gap into the effectsof
differencesin means and differencesin
coefficients. We calculate and report the
unadjusted pay ratio and the adjusted pay
ratio. The adjusted pay ratio is calculated
using the expected female pay evaluated at
the male means of the independent variables. We also calculate the percentage of
thegap attributableto differencesin means,
and the percentage of the gap attributable
to differencesin coefficients.6 The per6See Cain (1986) and Blau and Ferber (1987) for
a careful discussion of this decomposition method
and a reviewof the discriminationliteraturein which
this analysishas been applied.
The expected pay gap between men and women
based on the regression resultsis calculated as
Pay gap = IBiM(X- -XX) + 12Xi(Bim
-Bi),
where B. is the estimated coefficientfor the ithvariable and X. is the mean of the ithvariable from the
OLS estimatesreported in Tables 2 and 3. The fand
GENDER EARNINGS DIFFERENTIALS
centage of the gap due to coefficientsis the
"unexplained" portion of the pay gap. If
the estimatedmodels of pay have fullycontrolled for productivityfactors,then this
unexplained percentage of the gap is a
measure of discrimination. As mentioned
above, the unexplained portion of the pay
gap is a lowerbound estimateof the effects
of discriminationwhen endogenous variables thathave likelybeen affectedbywage
discriminationare used in the analysis.
The results of the decomposition for
both the total pay and base pay models are
reported in Table 5. The values fromthe
decomposition have been calculated using
both the male coefficientsand the female
coefficients.Resultsusing the female coefficientsare reported in parentheses below
the values based on the male coefficients.
Accordingto theresultsin Table 5, there
is a large differencein the unexplained gap
in pay between the total pay and base pay
models. When human capital, firm,occupation, and individual characteristicsare
included in the estimate of total pay, the
payratio,adjusted fordifferencesin means,
is .88. When controls for supervisoryresponsibilitiesand number of promotions
are included, the adjusted ratio increases
to .91. The percentage of the gap in total
pay that is attributable to differencesin
coefficientsis 44% and 34% in those respective specifications. Given the narrow
sample used and the number of controlsin
the equation, including controlsfor occupation, supervisoryresponsibilities, and
promotions, a 9-12% gap in pay that is
attributableto differencesin coefficientsis
large and representsa significanteconomic
m subscriptsdenote the female and male equations,
respectively. The firstterm in the equation is the
differencein paydue to differencesbetween the male
and female means, and the second termis the difference in pay due to differencesbetween the male and
female coefficients. Cain (1986) showed that the
decomposed values depend on the choice ofwhether
the male coefficientsor the female coefficientsare
used as the basis of the decomposition. Based on the
assumption that the male equation represents the
prevailing rate at which endowmentswould be compensated ifmarketdiscriminationwere eliminated,it
is the convention to decompose the paygap using the
male coefficients.
643
Table 3. Factors Affecting Base Pay:
OLS Results for ln(BASE
PAY).
(Standard Errors in Parentheses)
(1)
Variable
DEGREE
EXP
YRS
TENURE
INTERRUPTIONS
Men
.195**
(.065)
.008
(.010)
.046**
(.014)
-.002
(.012)
(.029)
.024
(.023)
.003
(.014)
.069
(.068)
.027
(.069)
.035
(.121)
.211**
(.052)
.012
(.008)
.050**
(.012)
.0004
(.0084)
-.009*
(.004)
.031
(.028)
-.018
(.021)
.033
(.037)
.033**
(.011)
-.043
(.060)
.009
(.060)
.253
(.144)
-.036
(.083)
.014**
(.004)
.101
(.087)
-.033
(.036)
.0002
(.0026)
.045
(.071)
.013**
(.004)
.020
(.056)
-.003
(.033)
-.003
(.002)
-.017*
MENTOR
(.008)
.024
(.039)
JOB CHANGES
-.066*
JOBS REJECTED
FIRM SIZE
INDUSTRY
PROFESSIONAL
TECHNICAL
SALES
HOURS
MARITAL
DEPENDENTS
FAMILY TIME
SUPERVISION
-.052
PROMOTIONS
-.046**
CONSTANT
R2
Adj. R2
SEE
n
(2)
Women
9.352** 9.177**
(.234)
(.271)
.579
.347
.273
.523
.362
.264
168
144
(3)
Men
(4)
Women
.201 **
(.062)
.011
(.009)
.026
(.014)
-.003
(.011)
-.013
(.008)
-.003
(.038)
-.066*
(.028)
.017
(.022)
.007
(.013)
.025
(.066)
.089
(.069)
.088
(.118)
.215**
(.051)
.014
(.008)
.035**
(.013)
.005
(.008)
-.007
(.004)
.011
(.028)
-.023
(.020)
.036
(.036)
.031**
(.011)
-.041
(.058)
.042
(.060)
.216
(.142)
.067
.028
(.083)
(.070)
.011** .010*
(.004)
(.004)
.045
.148
(.084)
(.055)
-.004
-.045
(.034)
(.034)
-.0005 -.004*
(.0025) (.002)
.048
(.027)
(.036)
.043**
(.014)
(.015)
9.390** 9.251**
(.261)
(.231)
.609
.412
.550
.336
.346
.257
168
144
*Statisticallysignificantat the .05 level; **at the
.01 level (2-tailed tests).
644
INDUSTRIAL
AND LABOR RELATIONS
REVIEW
Table 4. OLS Resultsfor in (TOTAL
PAY), Controllingfor the Percentage of Contingent Pay.
(Standard Errorsin Parentheses)
Variable
DEGREE
EXP
Men
Women
Variable
.189**
(.057)
.217**
(.050)
SALES
(.008)
(.008)
(.013)
(.013)
(.011)
(.008)
.007
.031*
YRS
TENURE
INTERRUPTIONS
MENTOR
JOB CHANGES
JOBS REJECTED
FIRM SIZE
INDUSTRY
PROFESSIONAL
TECHNICAL
.001
-.013
(.007)
.013
HOURS
.032*
MARITAL
.003
DEPENDENTS
-.006
FAMILY TIME
(.035)
(.004)
-.003
(.029)
(.026)
(.020)
.008
-.046
.013
(.021)
.013
(.012)
-.013
(.060)
.055
-.027
PROMOTIONS
Women
-.009
(.078)
.011**
(.003)
.029
(.072)
(.077)
(.054)
.166*
-.037
(.032)
-.0003
(.0023)
.008*
(.004)
.049
-.002
(.025)
(.033)
-.004*
(.002)
.041
(.036)
(.013)
(.015)
.043
.039**
.044**
.037
%CONTINGENT
1.511**
1.772**
.028*
CONSTANT
9.346**
9.380**
R2
Adj. R2
.624
.699
.317
.252
n
168
(.035)
(.011)
-.049
(.057)
.039
(.063)
(.059)
.051
.216
(.109)
SUPERVISION
Men
(.139)
SEE
(.154)
(.239)
.573
(.284)
(.234)
.650
144
at the.05 level;**atthe.01 level(2-tailedtests).
*Statistically
significant
differencein pay. This finding,however,is
very consistent with other studies of the
gender-paygap that carefullycontrol for
occupation or estimate the pay gap using
occupation-specificdata sets. Cain (1986),
forexample, listedfourstudiescontrolling
fordetailed occupations, and the adjusted
pay ratios in those studies ranged from.86
to .93.
The decomposition of the resultsfrom
thebase paymodels is verydifferent.When
SUPERVISION and PROMOTIONS are not included
in the base pay regressions,the adjusted
pay ratio is .98, almost unity. Only 13% of
the unadjusted pay gap is due to differences in coefficients. When SUPERVISION
and PROMOTIONS are included in the estimates, the adjusted pay ratio is 1.0. As we
pointedout above, thisresultdoes not mean
thatthe estimateof discriminationin base
pay is zero. At best, this is a lower bound
estimateofdiscriminationin base pay,since
we have controlled for variables that may
be the resultof discrimination.The values
of SUPERVISION, PROMOTIONS, the occupation
variables,MENTOR, TENURE, and possiblyother
variables may result from discrimination.
In fact,given that differencesin base pay
are determinedmostlybylevel ofjob in the
hierarchy,we would expect a model ofbase
pay that includes variables reflectingjob
level to explain the gender differencesin
base pay.
The most interestingresultin Table 5 is
the differencebetweenthedecompositions
of the total pay and base pay estimates.
Although the unexplained portion of the
gap in base payvariesbetweenzero and 2%
in the two differentspecificationsof the
model, controllingfor the same variables
leaves a 9-12% unexplained gap in total
pay. Since the remaining 9% gap in total
pay cannot be accounted for by the variables that account for all of the gender
GENDER
EARNINGS
DIFFERENTIALS
645
Table 5. Decomposition of Gender-PayDifferentialsby Model and Type of Pay.
Pay Ratio
TotalPay
Modela
Base Pay
UnMeansc
adjusted Adjusted' (%)
Coeffd
(%)
UnMeansc
adjustedAdjustedb (%)
Coeffd
(%)
(1)
HUMAN CAPITAL,
FIRM/OCCUPATION, and
INDIVIDUAL CHARACTERISTICS
.77
(.84)
.88
(51)
56
(49)
44
.86
(.98)
.98
(84)
87
(16)
13
(2)
HUMAN CAPITAL,
FIRM/OCCUPATION, INDIVIDUAL
CHARACTERISTICS, SUPERVISION,
and PROMOTION
.78
(.91)
.91
(65)
66
(35)
34
.87
(1.01)
1.00
(108)
103
(-8)
-3
HUMAN CAPITAL,
FIRM/OCCUPATION, INDIVIDUAL
CHARACTERISTICS, SUPERVISION,
PROMOTION, and
% CONTINGENT PAY
(1.01)
97
3
(3)
.78
.99
(106)
(-6)
Note: Male coefficientsare used as the basis of the decomposition. Results using female coefficientsare
reported in parentheses.
aModel(1) is basedon theresultsfromequations(1) and (2) in Tables 2 and 3 forTotalPayand Base Pay,
respectively.Model (2) is based on the resultsfromequations (3) and (4), and model (3) is based on the results
reportedin Table 4.
bPayratio afteradjusting for differencesin means of explanatoryvariables.
cPercentageof pay differencedue to differencein means of explanatoryvariables.
dPercentageof pay differencedue to differencein coefficientsof explanatoryvariables.
differencein base pay,this9% gap in total
pay must be due to gender differencesin
contingentpay.
To testwithinthe same analyticframeworkwhetheror not the remaining9% gap
in total pay is due to differencesin contingentpay,we re-estimated
thetotalpaymodel
holding constant the proportion of total
pay thatwas earned in the formof continThe regression
gent pay (%CONTINGENT).
resultsfromthisestimationare reportedin
Table 4. The resultsof the decomposition
of these estimatesare reportedin the third
row of resultsin Table 5.
When %CONTINGENT is included in the
estimateof totalpay,the adjusted pay ratio
increases from .91 to .99 (Table 5). The
resultsof thisdecomposition are verysimilar to the previousresultsfromthe base pay
model. Controlling for contingent pay,
there is basically no evidence of a gender
gap in total pay. This findingis consistent
with the hypothesisthat the unexplained
9% gap in totalpay,whichwe observewhen
controlling for human capital, job level,
and other factors,is due to gender differences in contingentpay.
Tobit Results
Results from the estimates of the tobit
model of contingentpay are presented in
Table 6. Estimates of the level of contingent pay as the dependent variable are
reported in columns (1) and (2), and estimates of contingentpay as a proportionof
total pay are reported in columns (3) and
(4). The gender pay difference is estimated here by including a GENDER dummy
variable equal to 0 for women and 1 for
men.
The coefficient
on GENDERiS positiveand
significanteven when we control forjob
level variables. This result holds for both
the level of contingentpay and contingent
pay as a proportion of total pay. Although
we cannot compare these resultsdirectlyto
the OLS estimatesor decompose the tobit
estimates, these results support the findings above thatindicate a significantdiffer-
INDUSTRIAL AND LABOR RELATIONS REVIEW
646
ence in contingent pay between the men
and women in this sample.
Table 6. Tobit Coefficientsfor
Contingent Pay Models.
(Standard Errorsin Parentheses)
Contingent
Pay
Variable
(1)
Contingent
Pay!
TotalPay
(2)
(3)
(4)
6723
.021
.020
6759
(.027)
(5422) (5422)
(.027)
-365
-.002
-.002
EXP
-385
(871)
(.004)
(.004)
(867)
.010
YRS
.012
2689* 2649*
(.006)
(.007)
(1220) (1321)
TENURE
-436
-482
-.002
-.003
(976)
(.005)
(.005)
(970)
-585
-.004
-.005
INTERRUPTIONS
-590
(575)
(.003)
(.003)
(571)
7141* 7157*
.039*
.043**
MENTOR
(.016)
(.016)
(3149) (3234)
-180
.003
.002
JOB CHANGES
-121
(.012)
(.012)
(2380) (2381)
.019
3244
.020
JOBS REJECTED
3294
(.011)
(.011)
(2228) (2232)
.003
.003
FIRM SIZE
1063
1120
(.006)
(.006)
(1178) (1181)
-.004
INDUSTRY
-.001
-1242 -1287
(.029)
(.029)
(5736) (5782)
-.003
.001
PROFESSIONAL
-3818 -2991
(.030)
(.031)
(5995) (6149)
-.087
-.078
TECHNICAL
-15390 -14284
(.064)
(.064)
(12602) (12717)
SALES
.157** .166**
12173 13459
(.034)
(.035)
(6677) (6959)
580
660
.003
.003
HOURS
(356)
(.002)
(.002)
(335)
MARITAL
-.022
-.025
-2588 -2386
(.032)
(.033)
(6393) (6458)
DEPENDENTS
.025
.024
1899 17587
(.016)
(.016)
(3164) (3173)
184
FAMILY TIME
.001
.001
194
(207)
(.001)
(.001)
(206)
SUPERVISION
1758
.014
(2627)
(.013)
-87
-.007
PROMOTIONS
(1349)
(.007)
.083** .077**
GENDER
12826* 12019*
(.026)
(.027)
(5200) (5335)
37095 37086
.191
.190
Scalea
(.011)
(.011)
(2005) (2005)
CONSTANT
-96530**-95145** -.422** -.414**
(.119)
(.119)
(23524) (23614)
Log-likelihood -2219 -2219 -40.98 -39.95
DEGREE
n
312
312
312
312
*Statistically
significantat the .05 level; **at the .01
level (2-tailedtests).
aIn the tobitequation the errortermis scaled byan
unknownparameter.
The coefficientreportedhere is the estimateof the
scale parameter.
Discussion and Conclusions
We have analyzed the gender-paygap by
formof pay using new data froma surveyof
business school graduates. We take a new
approach to the analysisof the gender-pay
gap. Rather than focusing on the righthand-side variables in the earnings equation,we examine the paygap bytypeofpay.
The principal findingof the analysis is
thata significantportionof the gap in total
payis due to differencesin thatpartof total
pay thatis contingenton job performance.
The gender differencein contingent pay
accounts for 58% of the $14,500 gender
differencein averagetotalpayin thesample;
gender differencesin base pay account for
only42%. When human capital, individual
and occupation andjob level
characteristics,
are controlled for,it appears thatapproximately34% of the unadjusted gap in total
pay that is not explained by gender differences in endowmentsis due to gender differences in contingent pay. We believe
these results are interestingand informative about a dimension of the gender-pay
gap that has previouslyreceived verylittle
attention.
This study has some clear limitations.
Because the data set is small and was drawn
froma rathernarrowsegmentof the population, we cannot extrapolate the results
from this studyto the entire population.
The resultsfromthe estimatesof the totalpay model, however, are consistent with
results from studies that rely on broader
and larger samples and include controls
fornarrowlydefined occupations.
Although we stated no predictions concerning the relationshipbetween the gender-paygap and the form of pay, we will
here explore severalpossible explanations
of the findings.The largewithin-jobdifference in contingent pay estimated in this
data set, controllingfor those factorsthat
explain all of the differencesin base pay,
might occur withinfirmsor across firms.
Since we do not have a controlforthe firm
in thisstudy,we cannot testwhetherthisis
GENDER EARNINGS DIFFERENTIALS
a within-firm
or across-firm
effect.Possible
explanations for the within-job'difference
in contingent pay vary,however,depending on where the differenceoccurs.
If gender differencesin contingentpay,
withinjobs, occur withinfirms,the causes
mustbe one or more ofthefollowing.First,
firmsmaynot offermen and women equal
opportunitiesto gain contingentpay. Second, men and women may be treated differently
injob performanceevaluationsthat
are used to determinecontingentpay. And
third, there may be differencesbetween
men and women in theirmarginallevels of
job performance. One of these causes or
some combinationofthemwould explain a
within-firm,
within-job,differencein contingentpay.
Given the findingsof previous studies,
however,we believe that the observed difference in contingentpay more likelyoccurs across firmsthan within firms. As
mentionedabove, previousstudiesin which
controls for the firm are included have
found that withinnarrowlydefined occupations, gender differencesin pay occur
across firmsrather than withinfirms(for
example, Blau 1975; Groshen 1991). These
studieshave found thatwithinjobs,women
are disproportionately
representedin lowerpayingfirms.Ifthe differencefoundin this
studyoccurs across firms,then otherwise
equallyqualifiedwomenmustbe morelikely
to be employed in firmswithlowerlevelsof
contingentpay. To explain whywomen are
over-representedin firmswithlower levels
of contingent pay, we must focus on the
sortingmechanismsthatcause thisparticular distributionof employment. It is possible, for example, that firmsthat relyon
contingentpay incentivemechanismsalso
use recruitment,selection, and career development mechanisms that sort women
out of thejobs providingopportunitiesfor
high levels of contingentpay.
Finally,one possible explanation forthe
findingsin this studyis thatwomen prefer
less payriskthando men and theyare more
likely than men to choose those occupationsand firms,everything
else being equal,
thatoffercompensationplanswithlessvariation in pay between pay periods. At least
647
part of the observed differencein totalpay
between men and women, therefore,may
reflecta premiumto men forbearing more
of the risk of short-runvariationsin their
job performancethan, on average,women
bear. This interpretationof the resultsis
consistentwiththe findingsof Subich et al.
(1989) thatwomen are less willingto take
risksin thejob marketthan are men. Further support for this conclusion can be
found in studiescomparing the risk-taking
behaviorofmen and womenwithinspecific
jobs and occupations. Muldrowand Bayton
(1979) compared severaldimensionsofthe
decision making process of male and female executives and found that the decisions of the women in the studywere significantlyless riskythan were those of the
men. Sexton and Bowman-Upton (1991)
tested the psychologicaltraitsof male and
female entrepreneurswhose firmswere of
similar size. Among the very few differences that were found between the men
and the women tested was a significantly
higherscore on riskaversionforthewomen.
This interpretationis also consistentwith
the-findings of Dreher and Ash (1990).
Using the same data set we have used, and
controllingformanyofthesame individual,
firm,and job characteristicsheld constant
in our analysis,Dreher and Ash found that
the men and women in this sample were
equally satisfiedwiththeirlevel of pay despitethe significantdifferencein totalpay.7
One problem with this explanation is
thatifthe differencerepresentsa riskpremium,thenwe would expect base pay to be
lower for men than forwomen. Our findings indicate that,all else equal, base pay is
equal. The resultsfor the decomposition
of the differencein base pay reported in
Table 5, based on the model controllingfor
promotions and supervisoryresponsibilities, do indicate, however,that the coefficientshave a positiveeffecton women's pay
relative to men's pay. The coefficienteffectis -3%, meaning that3% of the differential in base pay cannot be explained by
7Dreher and Ash looked only at total pay in their
analysisof pay satisfaction.
648
INDUSTRIAL AND LABOR RELATIONS REVIEW
differencesin means, and that this unexplained 3% favorswomen's pay. Although
thedirectionofthiscoefficienteffectwould
be consistent with men accepting lower
base pay in exchange for higher (and
riskier) expected returns on contingent
pay, 3% of the unadjusted difference is
too small to be meaningful in an economic sense. Given such a small difference in base pay, we would expect even
highlyrisk-averseworkersto be willing to
choose firmsusing contingentpay mechanisms.
Obviously,the reason whysuch a large
part of the unexplained gender difference
in total pay can be accounted for by the
gender differencein contingentpay is not
clear. Regardlessofwhetherthedifference
is due to risk preferences,it is clear that
withinthissample mostofthe unexplained
differencein pay,withinjobs,is due to that
partofpaythatis at risk. This findingposes
important questions for future research.
First,does the relationship observed here
between the gender-paygap and contingent pay hold within other narrowlydefined occupations? Second, does thisrelationship hold for differencesin pay that
occur across occupations? Finally,do differences in riskpreferencesbetween men
and women explain any of the pay difference? -And if so, to what extent is this
differencein taste caused by pre-market
socialization and education processes, or
bywomen's response to theirtreatmentin
the labor marketor how theyexpect to be
treated there? A better understandingof
whydifferencesin contingentpay can account forso much of the differencein total
pay is necessaryin order to determinehow
labor market policy can or cannot affect
the gender-paygap.
REFERENCES
Abowd, John M. 1990. "Does Performance-Based
Managerial Compensation AffectCorporate Performance?" Industrialand LaborRelationsReview,Vol.
43, No. 3 (February), pp. S52-S73.
Becker, GaryS. 1971. TheEconomicsofDiscrimination,
2nd ed. Chicago: Universityof Chicago Press.
. 1985. "Human Capital, Effort,and the Sexual
Division ofLabor." JournalofLaborEconomics,
Vol. 3,
No. 1, Part 2 (January),pp. S33-S58.
Blau, Francine D. 1977. Equal Pay in the Office.
Lexington, Mass.: D.C. Heath.
Blau, Francine D., and Marianne A. Ferber. 1987a.
"Discrimination: Empirical Evidence from the
United States." AmericanEconomicReview,Vol. 77,
No. 2 (May), pp. 316-20.
_. 1987b. TheEconomicsofMen, Womenand Work.
Englewood Cliffs,NJ.: Prentice-Hall.
Blinder, Alan S. 1973. "Wage Discrimination: Reduced Form and StructuralEquations." Journalof
Human Resources,
Vol. 8, No. 4 (Fall), pp. 436-55.
Brown,Charles. 1990. "Firms' Choice of Method of
Pay." Industrialand LaborRelationsReview,Vol. 43,
No. 3 (February), pp. S165-S182.
Cain, Glen G. 1986. "The Economic Analysis of
Labor Market Discrimination: A Survey." In 0.
Ashenfelterand R. Layard, eds., HandbookofLabor
Economics.Amsterdam: North-Holland, pp. 693785.
Dreher, George F., and Ronald A. Ash. 1990. "A
Comparative Studyof MentoringAmong Men and
Women in Managerial, Professional,and Technical
Positions." JournalofAppliedPsychology,
Vol. 75, No.
5 (October), pp. 539-46.
Gerhart,Barry. 1990. "Gender Differencesin Current and Starting Salaries: The Role of Performance, College Major, andJob Title." Industrialand
Labor RelationsReview,Vol. 43, No. 4 (April), pp.
418-33.
Goldin, Claudia, 1986. "MonitoringCosts and Occupational Segregation bySex: A HistoricalAnalysis."
JournalofLaborEconomics,Vol. 4, No. 1 (January),
pp. 1-27.
Groshen, Erica L. 1991. "The Structure of the
Female/Male Wage Differential."JournalofHuman
Vol. 26, No. 3 (Summer), pp. 457-72.
Resources,
Gunderson, Morley. 1989. "Male-Female Wage Differentialsand PolicyResponses." JournalofEconomic
Literature,
Vol. 27, No. 1 (March), pp. 46-72.
Hersch, Joni. 1991. "Male-Female Differences in
Hourly Wages: The Role of Human Capital, Working Conditions, and Housework." Industrialand
LaborRelationsReview,Vol. 44, No. 4 (July),pp. 74659.
Kahn, Lawrence M., and Peter D. Sherer. 1990.
"Contingent Pay and Managerial Performance."
Industrialand LaborRelationsReview,Vol. 43, No. 3
(February), pp. 107-20.
Kram,KathyE. 1985. Mentoring
at Work:DevelopmentalRelationships
in OrganizationalLife. Glenview,Ill.:
Scott, Foresman.
Lazear, Edward P. 1986. "Salaries and Piece-Rates."
JournalofBusiness,Vol. 59, No. 3 (July),pp. 405-31.
GENDER EARNINGS DIFFERENTIALS
649
Muldrow, Tressie W., and James A. Bayton. 1979.
Subich, Linda Mezydlo, Gerald V. Barrett, Dennis
"Men and Women Executivesand Processes Related
Doverspike, and Ralph A. Alexander. 1989. "The
EffectsofSex-Role-RelatedFactorson Occupational
to Decision Accuracy." JournalofAppliedPsychology,
Vol. 64, No. 2 (April), pp. 99-106.
Choice and Salary." In Robert T. Michael, Heidi I.
Noe, Raymond A. 1986. "An Investigationof the
Hartmann, and Brigid O'Farrell, eds., Pay Equity:
Determinants of Successful Assigned Mentoring
EmpiricalInquiries. Washington, D.C.: National
Relationships." PersonnelPsychology,
Vol. 41, No. 3
Academy Press.
(Autumn), pp. 457-79.
Whitely, William., Thomas W. Dougherty, and
Sexton, Donald L., and NancyBowman-Upton. 1991.
George F. Dreher. 1988. "The Relationship of
Mentoring and Socioeconomic Origin to Manag"Female and Male Entrepreneurs: Psychological
ers' and Professionals' Early Career Progress."
Characteristicsand Their Role in Gender-Related
Discrimination." JournalofBusinessVenturing,
Vol.
AcademyofManagementBestPaper Proceedings(August), pp. 58-62.
5, No. 1 (January),pp. 29-36.