Choice of Major: The Changing (Unchanging) Gender Gap
Author(s): Sarah E. Turner and William G. Bowen
Source: Industrial and Labor Relations Review, Vol. 52, No. 2 (Jan., 1999), pp. 289-313
Published by: Cornell University, School of Industrial & Labor Relations
Stable URL: http://www.jstor.org/stable/2525167 .
Accessed: 04/03/2011 22:32
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=cschool. .
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].
Cornell University, School of Industrial & Labor Relations is collaborating with JSTOR to digitize, preserve
and extend access to Industrial and Labor Relations Review.
http://www.jstor.org
CHOICE OF MAJOR:
THE CHANGING (UNCHANGING) GENDER GAP
SARAH E. TURNER
and WILLIAM
G. BOWEN*
Within the arts,sciences, and engineering fields, differencesbetween
men and women in choice of college major have not lessened in the past
two decades. In this paper, detailed data on choice of major and
individual scores on the Scholastic Aptitude Test (SAT) are used to
examine the extent to which observed differences between men and
women reflect the effectsof pre-collegiate preparation (as reflected in
SAT scores), as contrasted with a panoply of other forces. One conclusion is that there is a widening divide between the life sciences and
math/physical science fields in their relative attractiveness to men and
women. Differences in SAT scores account for only part of the observed
gap, and an array of residual forces-including differences in preferences, labor market expectations, and gender-specific effects of the
college experience-account for the main part of today's gender gaps in
choice of academic major.
D
ifferences
betweenmenandwomenin
field of study chosen at the undergraduate level mayrepresentdifferencesin
the skills that these groups bring to the
labor marketand maypartiallyexplain observed differencesin wages. Choice of
major, as well as decisions about where to
attend college, is an importantlink in the
chain of decisions and events that build
human capital forthose fortunateenough
*Sarah Turner is AssistantProfessorof Education
and Economics at the Universityof Virginia and
William Bowen is presidentof the AndrewW. Mellon
Foundation. The authorsthankJohnBound, Charles
Brown,Ronald Ehrenberg,Michael McPherson,Paula
Stephan, Yu Xie, and Harriet Zuckerman forreading
earlier draftsand providingthoughtfulcommentson
this research.
to go on to higher education. Choice of
school, choice of major, and academic performance coalesce to influence options
available to studentsforfurthereducation
and career development. Choice of major
is both an immediate outcome of the educational process and a determinantof later
outcomes of many kinds. Understanding
the factorsthatinfluence choice of major,
for men and women, is one part of the
larger process of understanding personal
as well as societal returnsto varied investments in human capital.
In addition to probing present-daypatternsin choice of major,we willinvestigate
whether,and in what ways,these patterns
have been changing. Have the choices of
majors made by men and women converged? Or have traditional differences
solidified or become even more pronounced in recent years?
In(ldlstrial
and LaborRelationsReview,Vol. 52, No. 2 (January1999). ? by Cornell University.
0019-7939/99/5202 $01.00
289
290
INDUSTRIAL AND LABOR RELATIONS REVIEW
Among the manyfactorsthat influence
men's and women's choice of major, and
consequentlygender differencesin careers
and wages, three factorsrelated to gender
maybe particularlyimportant: a student's
preparation and achievement at precollegiate levels of education, especiallyin
mathematics; an individual's preferences
forvarious courses of study,which maybe
encouraged byparentaland societal expectations; and the labor market prospects
associated witha given set of skills,which
mayprovide more encouragement forone
sex than the other to pursue certain fields
of study. Our focus in this studyis on the
firstof those factors-the extent to which
differencesbetween men and women in
precollegiate achievement (measured by
SAT verbal and math scores) account for
differencesin choice of major at the college level. Afterexamining the national
trend in the choice of undergraduate major by men and women, we turnour attentionto thechoices made bymen and women
at a small set of selectiveresearch universities and liberal artscolleges.
National Trends
and Descriptive Findings
The Dissimilarity Index:
Trends in National Data
One startingpoint for measuring dissimilaritiesbetween women and men in
choice of major is to calculate the absolute
value of the differencesbetween the percentages of women and men majoring in
each of the fields into which we classify
students,sum these differences,and divide
by two. This measure, referredto in other
literatures (such as those on residential
and occupational segregation) as an index
of dissimilarity,
captures the percentage of
studentswho would need to change majors
in order for parityto be achieved in the
distributions,with 100% indicating complete segregationand 0% indicatingidentical distributions.1
'Jacobs ( 1995) appears to be the firstresearcherto
have applied this measure to the differentdistribu-
Movementsover the past 30 years in an
all-inclusive dissimilarityindex and in a
dissimilarity
index limited to arts-sciencesengineering (A-S-E) are shownon Figure 1
forthe entirepopulation of B.A. recipients
in the United States.2 The all-inclusive,
"All-Fields"index is calculated using the
numberofB.A. recipientsat all U.S. institutions in 10 broad field-of-study
categories,
including an array of professional fields
that togetheraccounted forjust over half
(53%) of all B.A. recipients in 1995. Just
two fields-education and business-account fornearly2/3 of these B.A.s outside
the arts-sciences-engineering
fields.
The "arts-sciences-engineering"
(A-S-E)
index is based on field choices made by
studentswithin the arts-science-engineering disciplines (the remaining47% of B.A.
recipients). The A-S-Eaggregate captures
the range of concentrationsofferedat the
undergraduate level at the institutionsincluded in thisanalysis. In the main, these
selectiveliberal artsand research universities do not award degrees in professional
preparatoryprograms such as business or
education. Engineering programs,generally part of distinct engineering schools
within the institution,represent the primaryundergraduate professional degrees
offeredat theresearchuniversitiesincluded
in this study;moreover,in manyinstances
these programsof studycould as easily be
called "applied science" as "engineering."
tions of men and women byfield of study. The basic
descriptive results presented here are substantively
similarto his,withdifferencesattributableto the use
of differentfield taxonomies.
2The 10 fieldcomposites used in the calculation of
the "All Fields" dissimilarityindex are engineering,
physicalsciences-math,lifesciences, psychology,economics, politics and other social sciences, scienceengineering technologies, humanities, education,
business and communications,and a broadlydefined
"other"categoryincluding the social servicesprofessions and vocational studies. The 7 field groups
included in the "A-S-E"dissimilarityindex are engineering, physical sciences-math, life sciences, psychology, economics, politics and other social sciences, and the humanities.
GENDER
AND CHOICE
OF MAJOR
291
Figure 1. Index of Dissimilarityamong B.A. Recipientsin theUnitedStates,1965-1994.
0.400.35- TotalDissimilarity
Index
0.30
=
0.25
Education-Business-
a
0
0.20
~S 0.15
Arts-Sciences0.10
Engineering
0.05
I
0-00
65
IlI
i1-
70
I
I
75
80
I
I
I
85
90
95
Note:Basedon BA degreesconferred
byU.S. collegesand universities
reported
in theHEGIS/IPEDS
surveyscompiledin theCASPARdata archive."All BAs awarded"is used as thedenominator
in the
computation
offieldsharesused to calculatethetotalindexofdissimilarity,
as well as thecomponents
attributable
to thearts-sciences-engineering
subsetoffieldsand theeducation-business-other
subsetof
fields.
The All-Fieldsdissimilarity
index showsa
pronounced decline, froma high of nearly
40% in the 1965-66 academic yearto about
19% in 1994-95 (Figure 1). In aggregate,
the largest decline in the overall index of
dissimilaritytook place during the 1970s,
with a drop of slightlymore than 10 percentage pointsoccurringbetween1973 and
1983. Focusing on the 1980s,Jacobs(1995)
found verysimilar results for B.A. recipients,as well as advanced degree recipients.
While that analysisalso suggested that the
narrowingof the gender gap came to a halt
in the mid-1980s, it did not distinguishthe
role of the professionalfieldsfromthatof
the arts and sciences in thisdynamic.
The relationship between the piece of
the dissimilarity
index attributableto business-education-otherand thatattributable
to arts-sciences-engineering
changes over
time, as a result of both changes in the
degree of gender segregation within the
component fieldsand the relativebalance
between these two loosely defined subsectors of higher education. The dissimilarityindex declined until 1975 for both
components. In business-educationit continued itsdownwardcourse after1975 (until 1985), but in arts-sciences-engineering
it took an upward turn.
Extremelylarge movements of women
away from the field of education and the
associated migrationof women into business programs account for much of the
index. In the
reduction in the dissimilarity
mid-1960s, gender segregationoutside artssciences-engineeringaccounted fornearly
two-thirdsof the total value of the dissimilarityindex. In this period, the undergraduate business major was more segre-
292
INDUSTRIAL AND LABOR RELATIONS REVIEW
gated than some fieldswithinthe physical
sciences, withwomen receivingonlyabout
10% of business degrees. By the mid1980s,women had entered undergraduate
business programs in record numbers.
Because of greatlyexpanded representation in business and greatlyreduced participation in the field of education, the
portion of the dissimilarityindex attributable to fields outside the A-S-Ecomposite,
primarilythese two fields, fell to a little
more than 1/3.
The movementsof the more restricted
A-S-E index follow a verydifferenttrajectory,particularlyafter1975. This index of
dissimilarityincreases through the early
1980s, before leveling off and declining
moderatelyin the early 1990s.
The "storyline"of thispaper is what did
not happen to the dissimiliarity index
within the arts-science-engineeringfields
after 1975. During a period when other
measures of economic outcomes and opportunitiesformen and women were conindex failed
verging,theA-S-Edissimilarity
to move towardgreaterparity.The remainder of thispaper uses new data thatinclude
standardized testscores for individual students to explain more fullythan has been
possible heretoforethe choices of majors
bymen and women withinthe arts-scienceengineeringfields.
ated (mostlyin the classes of 1955, 1980,
and 1993). The advantage of these data for
a micro-levelinquiryon choice of major is
that we observe the full set of studentsat
each of the twelveinstitutions.3
There are obviouslymanywaysin which
areas of studycan be classifiedand aggregated. In keepingwithnormal practice,we
have chosen to group all of the majors
commonly considered to comprise the
Humanities (including classics, English,
foreignlanguages and literatures,history,
philosophy, and religion); but we break
withnormal practice in our decision not to
group in single categories all of the Social
Sciences or all of the Natural Sciences.
Within the Social Sciences, there are pronounced gender-related differences between economics and psychology,and between each of these fields and the other
social sciences. These patternsmayreflect,
at least in part, differencesamong these
fields in math-intensity.In the remainder
of this paper, we workwith eight fields of
study,some broader than others: the humanities; economics; psychology;politics
and othersocial sciences; biologyand other
lifesciences; mathematicsand physicalsciences; engineering;and a small number of
students in "other" fields (principally architecture, communications, education,
and speciallyconstructedmajors).4
The College and Beyond Database
The data forthisinquiryare taken from
theCollege and Beyonddatabase assembled
by the Andrew W. Mellon Foundation in
cooperation with34 colleges and universities. The part of the database used here
consistsof detailed recordsof entrance test
scores and majors subsequentlychosen by
undergraduates at twelveacademically selectivecolleges and universities:threeuniversities (Princeton, Stanford,and Yale);
six coeducational colleges (Hamilton,
Kenyon,Oberlin, Wesleyan,Williams,and
Swarthmore);and three women's colleges
(Bryn Mawr, Smith, and Wellesley). The
database includes all matriculantsin the
1951, 1976, and 1989 enteringcohorts,and
we restrictour analysisto those who gradu-
3While we present some descriptive data for the
1951 enteringcohort,we restrictthe analyticalpartof
this studyto the 1976 and 1989 cohorts. Although
analysis of the choices of the 1951 cohort mightbe
particularlyilluminating,the analysisis restrictedto
the later twocohorts fortworeasons. First,standardized testing was far from universal in this period,
leading to a large share of missingcases in thiscohort
thatare unlikelyto be randomlydistributed.Second,
the interpretationof shiftsin parametersbetween the
1951 cohort and later cohorts would be complicated
bythe factthatseveral of these colleges and universities shifted from single-sex to coed status between
1951 and 1976.
4Whilethe humanitiespresentthe broadest aggregate of the fields listed, disaggregatingthis category
does not add to the substance of the analysis. History
and English are the largestsubfields,and the gender
distributionsforthese fieldsare included in Table 1.
A large numberof small fields,including philosophy,
religion,specificforeignlanguages, and comparative
GENDER AND CHOICE OF MAJOR
293
professionalfields. However, withinthe AThe large size and census-likecharacter
of the database, itsrelatively"fine"level of
S-E fields, the national distributionsfor
aggregation,and the strongsimilaritiesin
both men and women are surprisingly
congruent with the distributions at our 12
admissionsstandardsand curriculaamong
the 12 colleges and universitiespermit a
schools (Appendix Table 1, top panel).5
This similaritysuggeststhat the forces incloser, more intensive, examination of
male/female differencesin choice of mafluencing choice of major and the differences between men and women in making
jors thanis possible in studiesusingsamples
of individuals from a larger and more dithis choice extend well beyond this small
verse arrayof institutions. To the extent
set of selectiveprivateinstitutions.
that choice of major is a decision made
Distinguishingamong the 12 schools in
thisanalysisbytyperevealspredictable diflargelyin a given institutionalcontext,it is
ferencesin distributionsofmajorsbetween
conceptually appealing to have data in
the universitiesand the liberal artscolleges
which we observe the full population of
studentsat each oftheseinstitutions.Inter- (Appendix Table 1, bottom panel). With
actions between choice of major and type significantnumbers of degrees conferred
of institutionmaybias estimatesof returns in engineeringat the threeuniversities,the
overall share of studentschoosing to major
to specificfieldsofstudybygender and may
in the humanities is notably lower at the
confound the interpretationof aggregate
researchuniversitiesthan at the liberal arts
trendsin the choice of major formen and
women.
colleges, and this differenceis greaterfor
The other side of the proverbialcoin is
men than forwomen since more men major in engineering. The general patterns
that these 12 institutionsare not at all
representativeof American higher educaare consistentwiththe notion that the cotion. They represent a highly selective, educational liberal arts colleges and the
entirelyprivate,subsetofinstitutionsoffer- women s colleges have similar"production
functions,"with an emphasis on subjects
ing what is generally regarded as a basic
liberal artscurriculum.The threeuniversi- thatare more labor-intensivethan capitalties (and some of the colleges) do offer intensive,while the research universities
undergraduate programs in engineering, offerundergraduates more opportunities
in fields that require complex infrastrucand these schools also offera smallnumber
of "other"majors. Still,the overwhelming tures and that attracthigh levels of sponsored research ("big science").
majority of all students in this database
In most fields,the choices of majors by
(roughly90%) majored in traditionalfields
the women at the women's colleges can be
withinthe humanities,the social sciences,
said to be intermediatebetweenthechoices
and the natural sciences. Unlike studies
based on High Schooland Beyond,thisstudy made by women at the coeducational colis limitedbyan incomplete representation leges and the choices made by men at the
of the population attending college from coed colleges. The share of women in the
humanities is markedly lower in the
anyhigh school cohort. From the perspective of national norms, the professional women's colleges than in the coed colleges
fields (especially business and education)
(41% versus 49%), while the shares of
are grosslyunder-represented.We cannot,
women in economics and in math-physical
then, on the basis of these data, say anything about decisions to major in these
literature,make up the remainderof the fieldsin the
humanities. The largestchanges in gender compositionwithinthehumanitieshave occurred withinthese
small and specialized fields.
5We compare our 1989 entering cohort with the
1993 B.A. recipients nationally because the largest.
number ofour graduates earned theirdegrees in four
years. There are some "s'ystemic" differences b)etwteen
the "national" and "local" distributions,though tliesc
differencesapply to both sexes.
INDUSTRIAL
294
AND LABOR RELATIONS
REVIEW
Table 1. Changes over Time in Choice of Major, Graduates Only.
(12 Institutions,Collegeand BeyondData)
1951 Cohort
1976 Cohort
1989 Cohort
Field ofStudy
W%
M%
Diff
W%
M%
Diff
W%
M%
Diff
Humanities
English
History
Other
Humanities
Economics
Psychology
Politics & 0th
Soc Sci
Tot Soc Sci
Life Sciences
Math/PhySci
Tot Natural Sci
54.8%
14.9%
11.3%
39.7%
11.4%
13.2%
15.1%
3.6%
-1.9%
43.1%
12.4%
6.8%
31.9%
8.1%
10.1%
11.2%
4.3%
-3.3%
42.4%
12.5%
8.9%
33.6%
8.8%
10.6%
8.8%
3.6%
-1.7%
28.6%
2.4%
5.2%
15.1%
10.1%
4.4%
13.4%
-7.7%
0.8%
23.9%
7.5%
7.9%
13.6%
12.4%
5.2%
10.2%
-4.9%
2.7%
21.0%
5.4%
9.0%
14.2%
9.9%
4.1%
6.8%
-4.5%
4.9%
18.0%
25.7%
3.3%
4.5%
7.9%
11.9%
6.1%
26.4%
-0.8%
6.4% -3.0%
10.9% -6.4%
17.3% -10.7%
15.8%
31.2%
11.7%
6.7%
18.4%
13.6%
31.1%
11.8%
9.7%
21.5%
2.2%
0.1%
-0.1%
-3.0%
-3.1%
19.6%
34.0%
11.9%
5.7%
17.6%
18.9%
32.9%
9.5%
10.0%
19.5%
0.6%
1.1%
2.4%
-4.3%
-1.9%
Engineering
0.3%
10.9% -10.7%
Other/NEC
11.4%
5.7%
5.7%
TOTAL
100.0% 100.0%
Index of Dissimilarity:a
Broad
21.5%
Disaggregated
27.7%
2.7%
4.6%
100.0%
11.4%
4.1%
100.0%
-8.7%
0.5%
3.6%
2.4%
100.0%
11.8%
2.2%
100.0%
-8.2%
0.3%
11.8%
16.7%
10.1%
17.0%
aThe broad index of dissimilaritytreatsthe social sciences and natural science subtotalsas single entries;the
disaggregated index treatsthe components of these disciplines as individual fields (for example, for the social
sciences, economics, psychologyand politics/othersocial sciences).
science in the women's colleges are higher
than the correspondingshares in the coed
colleges. However, because selection bias
maybe at work,judgmentsabout themeaning of thispatternshould be suspended, as
the women who choose to attend the
women's colleges maybe more inclined to
major in scientificfieldsthan theirpeers at
coeducational institutions.
The Dissimilarity Index Again:
Trends within the 12 Institutions
The dissimilarity
index providesa convenientwayto assess both thedegree towhich
male-femaledifferencesin choice of major
have narrowedover the past threedecades
or so within these 12 institutionsand the
sizes of the gender gaps thatremain. (For
presentpurposes,we aggregatethedata for
all 12 institutions;looking at subsets does
not alter the main conclusions.) In brief,
considerable convergence occurred be-
tween 1951 and 1976: our disaggregated
which treatseach of
index of dissimilarity,
the eight fields listed above as a separate
field,fallsfrom27.7 for the 1951 entering
cohort to 16.7 for the 1976 entering cohort. Driving this transformationwas the
migration of women out of the humanities and into fields like economics and
the life sciences, withthe share of women
choosing economics risingfromless than
2.5% to 7.5% and the share choosing the
life sciences growingfrom3.3% to 11.7%
(Table 1).
More surprising,perhaps, is the lack of
furtherchange in the index between 1976
index is at 17.0
and 1989-the dissimilarity
for the 1989 cohort. There were, however, interesting movements below this
apparently placid surface. Within the
sciences, gender differences at the field
level widened in both the life sciences
(where women moved into a position of
over-representation) and math-physical
GENDER AND CHOICE OF MAJOR
sciences (where the over-representation
of men increased).
To the extent that men and women do
not enter college with the same math and
verbal skills,as measured bythe SAT, some
differencesin patternsoffieldchoice would
be expected. Replicatinga patternthathas
been well established based on national
data, theaveragemathSAT scoreforwomen
at these 12 schools lags behind the average
mathSAT score formen byabout 50 points
in both the 1976 and 1989 cohorts;average
verbal SAT scores forwomen and men were
essentiallyequal in both the cohorts.6
Yet,the "gender gap" in math SATs does
not translate into a proportionate difference in representationin each of the sciences, including the quantitativesocial sciences. To illustratethe need to consider
the component fieldsof the sciences separately,Figure 2 shows the distributionof
men and womenwithveryhighmathscores
by field. In this bar chart, it is plain that
women with high SAT scores are much
more likelythan men to choose to major in
the lifesciences and the humanitiesrather
than engineering, math, or the physical
sciences. If the paucityofwomen withvery
high math achievement were the sole explanation fordifferencesin representation
in the sciences and quantitativefields,then
one would expect these distributionsto be
identical for men and women. Relatively
high mathSAT scores provide men withan
advantage relativeto women in fields that
require substantialquantitativeskills. Multivariateanalysis,based on testscores and
choice of major byindividuals,providesan
6Atthe national level, the gap in SAT math scores
between men and women was about the same as the
gap withinthese institutions: the average SAT math
score was 500 for men versus 454 for women (ETS
data forcollege-bound seniors in 1989). The combination of highermathscores formen than forwomen
and roughlyequal verbalscores means thatcombined
SAT scores were higher for the men than for the
women at our 12 schools (and nationally). There is
some evidence that the "gender gap" in the mathematics SAT is narrowingin the 1990s, as the 1994
gap, nationwide,was 41 points,withaverage scores of
501 and 460 for men and women, respectively.
295
effectivemeans of describing howmuchof
the gender gap is associated with differences in SAT scores.
Other Studies and Approaches
Differencesbetweenmen and women in
choices of undergraduatemajor have been
the subject of research by numerous sociand economistsover
ologists,psychologists,
the past several decades. In their 1997
surveyarticle,Leslie and Oaxaca identified
over 120 theoreticaland empiricalanalyses
of the under-representationofwomen and
minoritiesin the science and engineering
disciplines. Science policy analysts have
often expressed the concern that the under-representationofwomen in thesedisciplines limitsthe pool of talententeringthe
science and engineering work force. Beyond chartingaggregatesand flows,a number of researchershave attemptedto identifythe behavioral factorsexplaining why
this choice process leads to such different
observed outcomes formen and women.
Theoretical explanations forthe differing outcomes tend to emphasize eitherdifferences in skills or differencesin preferences and environmental determinants.
Men and women (or boys and girls) with
the same measured skill sets may differ
dramaticallyin theirpreferencesfordifferent typesofoccupations or courses ofstudy.
Such differences may reflect biological
forces or cultural factorssuch as the sexrole socialization forcescited byEccles and
Hoffman(1984). Accordingto the sex-role
socialization hypothesis,sex-role patterns
and norms conveyed by schools and families in childhood affectlaterinvestmentsin
education and training. Women may be
encouraged to pursue fields and studies
emphasizing nurturingwhile men are encouraged in domains emphasizing quantitative reasoning. While identificationof
such sex-rolefactorsis inherentlydifficult,
Corcoran and Courant (1985) suggested
that variation in factorsaffectingsex-role
socialization within gender groups (for
example, variationin familystructure)provides one means ofdistinguishingthevariation in outcomes associated with sex-role
INDUSTRIAL AND LABOR RELATIONS REVIEW
296
Than750
ofMenand WomenwithMathSAT ScoresGreater
Figure 2. Distribution
byChoiceofMajor.
30
Li Men
25
a
A
Women
E
c8OSHM20
0
ut
.
Huaiis
c
Eooic
15~~~~~~~~~~~Scecs
i
s
h
l0
scooy
n
Ohr
oil
BoLf
Sine
Mt
hsc
niern
20
norms from differencesin outcomes reflecting the presence of gender-specific
barriersin employmentor wages.
Expected labor force commitmentmay
well affectthe economic rewards to any
field of study, as suggested by Polachek
(1978). Men and women maydifferin their
intended labor forcecommitment,leading
to markedlydifferentoptimal investments
in human capital as captured bythe choice
of major variable. For this reason, estimates ascribing gender differencesin occupational choice to employerdiscrimination may be overstated. As such, women
withintermittent
expected labor forceparticipation may favor fields with low "skill
atrophy"or the lowest cost to labor force
interruption, though available evidence
suggests that this factor is not a primary
empirical determinantof the observed differencesin choice of major Jacobs 1995).
Experiences at the undergraduate level,
including mentoringand peer influences,
may have differentialeffectsfor men and
women throughthe relative "costs" of any
course of study. Several researchers, including Solnick (1995) and Baileyand Rask
(1996), have focused on the role of the
undergraduate institutionand the collegiate environment in determining the
choice ofmajorbymen and women. Solnick
(1995) presented an innovative study of
how a student's desired field of study
changes during college, and how the pattern of change varies by type of college
attended. She found that women at allfemaleschools wereappreciablymore likely
to switchto fields traditionallydominated
bymen than were women at coeducational
institutions.Since these resultswere based
on transitionsduring college, Solnick argued that the cultural and academic environments associated with all-women's
schools facilitatewomen's entryinto the
sciences.
The identificationofspecificeducational
and instructionalfactorssuch as introductoryclass size and facultymentorshipdifferentially
affectingchoice of majorbymen
and women has received considerable at-
GENDER AND CHOICE OF MAJOR
297
sorting men and women into particular
fieldsof study. In particular,are the forces
leading to differentchoices of major similar enough across fieldsthata model positing a dichotomous choice between science
and non-science fields adequately characterizesobserved variation? Our answer to
thisquestion is a clear "no."
tentionbecause college administratorsare
assumedto exercisesome controloverthese
variables. Canes and Rosen (1995) investigated the extent to which increases in the
share of female facultymembersin a given
department-providing,in principle,more
role models for female students-corresponded withgrowthin the representation
offemalestudentsin thatdepartment.They
found no evidence of such a link.7
While there is ample evidence that factorsother than achievementcontributeto
the observed gender differencesin choice
ofmajor at the undergraduatelevel,persistentdifferencesbetween men and women
in high school courses and measured
precollegiate achievementprovide at least
a partial explanation for the gap. Skillbased explanations fordifferencesin field
choices at the undergraduate level often
focus on differences between men and
women in measured mathematical skills
and the number of high school courses
completed in science and math.8A number
of studies,including Ware and Lee (1988)
and Ware, Steckler,and Leserman (1985),
have included measures of achievement,
particularlyin quantitative fields, as explanatoryvariablesin discretechoice models of the decision to major in a scientific
field. However, such studies do not consider whether these measures varyin the
magnitude of theireffectsacross the component fields of the sciences or across a
more differentiatedset of fields.
The observationthatwomen have demonstratedsizable gains in representationin
some fields,notablythe lifesciences, while
theirrelativeshareshave stagnatedin some
quantitativesciences, leads to a reconsideration of the role of precollegiate skillsin
In thispaper, our focus is on the extent
to whichacademic achievement,measured
by entering SAT scores, explains gender
differencesin choice of major. While this
parsimonious specificationof explanatory
variables is plainlylimited,our data afford
considerable power in explaining the behaviorof men and women across a range of
fields of study. We quantifythe degree to
which the observed gender differencesin
choice of major can be attributed to
precollegiateacademic achievement. Each
individual'sSAT scoresare proxiesforskills
brought to college, with the verbal and
mathscoresrepresentingseparate (though
oftencorrelated) typesof skills.
Nevertheless, the most interestingbehavioral phenomenon is the piece thatremains unexplained. This residual element
reflectssome combination of differentopportunities in the labor market for men
and women and differencesin long-term
occupational aspirations associated with
gender.
The multinomiallogitformulationiswell
suited to the measurementoffactorsaffecting qualitative choices such as choice of
major.9 In theory,choice of major foreach
individual representsa deterministicout-
7Bailey and Rask (1996) tested the same hypothesis with micro data froma differentprivate liberal
arts college and found modest support for the role
model hypothesis.
8Since long-termexpectations maybe formedwell
in advance of college enrollment,ascribing a causal
interpretationto differencesassociated withachievement levels is unwarranted if these levels of
precollegiate skill are endogenously related to long-
term goals. For example, differencesin observed
mathematicsachievement between men and women
at the precollegiate level mayreflectdifferentiallevels of parental encouragement of sons and daughters
in these courses at the secondary level.
9As discussed below, while the multinomial logit
formulationoffersthe most tractableestimationof a
qualitative choice problem, the functionalformalso
leads to implicitassumptions associated withthe pa-
AnalyticApproach
and Estimation Strategy
INDUSTRIAL AND LABOR RELATIONS REVIEW
298
come that maximizes utility. As researchers, we do not know enough about all the
factorsaffectingthe choice-or their actual values-to replicate this calculation
withouterror. Nonetheless, observations
of individual achievement measures and
choices of major provide sufficientinformation to make probabilistic statements.
In particular,theprobabilitythatindividual
i chooses field j over the alternative k is
represented by
(1)
PmF=
+
Prob(VMF(Z, f3 )
e..> vM'F(Z.,
=Prob (e..- e.k>
-
+ ek)
3M[)
VMF(Z.,
fMF)
MF)r),
VMF(Z.,
where individual characteristicsin the vector Z. represent the observable determinantsoftheutilityfunction(V), theweights
on these explanatoryfactorsare estimated
as Pi,and the unknowncharacteristicsare
captured in eY. In this framework,the
gender-specificsuperscripts (M,F) recognize that men and women may differin
theirunderlyingpreferencefunctionsand
in the benefits(or costs) associated witha
in a particularfield.
particularcharacteristic
Withan assumptionabout the distribution
(Type I Extreme Value) of the residuals
(ei ), we can estimatethe parametersof the
model. In particular,if the epsilon terms
are independentlyand identicallydistributed, the estimationprocedure followsthe
multinomial logit formulation and the
choice probabilitiestake the form
VMF
id
....................
.........=................
.......
....
jE Jnexp
M,F
~~~V
id
Answers to the interestingquestion of
how we would expect individuals to sort
themselvesamong fields, given their test
rameter estimates. In particular, it is assumed that
the residuals follow a well-behaved Type I extreme
value distribution. A further(testable) restrictionis
that the "independence of irrelevant alternatives"
restrictionis maintained such thatthe elimination of
one option (forexample, economics) does notchange
the relative probabilities of the choices of other options (for example, humanities or math).
scores, follow from the estimation of the
coefficientsin this model. First, if the
coefficientson math and verbal SATs are
nonzero, then the model providessupport
forthe hypothesisthatmore than one skill
set mattersin the determinationof choice
offieldand related investmentsin building
human capital. Second, therelativemagnitude of the coefficientsindicates the degree towhichselection ofparticularmajors
depends on mathand verbalproficiencyon
entryto college.
The objectiveofthisanalysisis to address
the counterfactualquestion of how the distributionof the undergraduate majors of
women would be expected to change if,
holding preferencesconstant,women had
the same distribution of SAT scores (X
values) as men (Bound, Schoenbaum, and
Waidman 1996).1o This typeof calculation
allows us to estimate the effectson choice
of major of differencesbetween men and
women in both (a) theirprecollegiate academic preparation, measured here by the
SAT scores that they bring with them to
school, which are the X values, and (b) the
waysin which women and men separately
"convert"theirrespectiveSAT scores into
choices of majors. These "residuals"are in
turn the product of other forces and variables listed previously, including labor
marketopportunitiesforwomen and men,
wealth, other anticipated familycircumstances (which affectboth the desire for
labor force participation and the need to
earn sizable amounts of income), social
and parental expectations, and attitudes
and interestsstimulatedbyfacultyand fellow studentsin college.
Empirically,thedifferencesbetweenmen
and women in observed field distributions
can be seen in the context of the familiar
"0Thealternative
is to considerhowthedistributionof majorsformen woulddifferif,giventheir
of SAT
preferences,
theyhad the same distribution
inthemagnitude
ofthe
scoresas women.Differences
twocomponentpieces,dependingon whethermen
the
orwomenareusedas thereference
group,reflect
indexnumberproblemwiththistypeof
underlying
analysis.
GENDER AND CHOICE OF MAJOR
Oaxaca decomposition
(2)
pPFXF _ pPMXM
+
(PPFXM
=
(pFXF
PpFXM)
_PMXM)
wherepMXM is the mean of the predicted
probabilityof choosing field j using the
coefficientsand values of the explanatory
variablesformen;pPFXF is the mean of the
predicted probabilityof choosing field j
using the coefficientsand values of the
explanatoryvariablesforwomen; and pPFXM
is the mean of the predicted probabilityof
choosing fieldj using the coefficientsfor
women and values of the explanatoryvariables for men.
The firstterm to the rightof the equal
sign can be interpretedas the share of the
observed differenceassociated withdifferences in SAT scores,while the second term
shows the combined effectsof the differences associated with the coefficientsnot
attributable to differences in measured
achievement. In the case of a logit model
or other nonlinear function, the magnitude of the effectof a change in any single
exogenous factoris a functionof the level
of the other variables used in this evaluation,and such nonlinear choice models do
not allow for the decomposition of the
effectsof particularexogenous variables.
Empirical Results
We begin by estimatingseparate multinomial logit models for men and women,
with verbal and math SAT scores as the
explanatoryvariables. Specifyingthe test
score variables as categorical variables limits the functionalformassumptionswithin
the logitframework(though italso permits
some unconventional "hillsand valleys"in
the implied distributionof majors by test
scores). For the 1989 entering cohorts,
estimated coefficientsand the associated
standarderrorsare shownin Tables 2a and
2b." The significance of the parameter
estimates(excepting thecase ofthe "other"
"By aggregating across schools, we assume that
individualsjointly select institutionand major. For
299
field group) supports the proposition that
SAT scores help explain choice of major.12
For both men and women, high verbal
scores in the presence of low quantitative
scores have a strongand positiveeffecton
the likelihood of majoring in the humanities (the reference group) relative to the
probabilitiesof choosing all other majors.
Stated the other way around, the nearly
alongunbrokenrunofnegativecoefficients
side the verbal SAT variables tells us that
relativeto thegroup ofstudentswithverbal
SAT scores below 550, ceteris
paribus,studentswiththeserelativelyhighverbalscores
are less likely to major in each of these
fields than to major in the humanities.
Relative magnitudes of the verbal coefficients also mediate choices among fields
outside the humanities; for example, an
increase in SAT verbal scores holding the
example, since the liberal arts schools do not offer
engineering programs, choosing to attend one of
these institutionsis effectively
choosing not to major
in engineering. Aggregationalso disregardsthe question of whether the two-wayselection process-students choosing schools and schools choosing among
applicants-affects the choice of major. Given these
selection issues lurking near the surface, the coefficients should be interpretedas representingthe net
effectof the institutionalselection process and the
"true" test score effecton the choice of major. An
alternative theoretical frameworkwould posit that
institutionalchoice and choice of major are sequential. That type of analysis would be institution-specific.
12Asone early reader noted, it may be surprising
that test scores seem to be such significantexplanatoryvariables determiningchoice of major and the
differencesbetween men and women in choice of
major. Two explanations meritconsideration. First,
testscores are likelyto be correlated witha range of
factors,including both other measures of academic
achievement (such as high school rigoror high school'
GPA) and socioeconomic variables. To this end, a
causal interpretationof the parameter estimates as
"explaining" differencesin choice of major is inappropriate,particularlyifthe magnitude of the correlation with omitted variables differsfor men and
women. A second explanation, warrantingfurther
investigation,is thatdifferencesin achievementmeasured by testscores are more powerfuldeterminants
of choice of major in a constructwhere the achievementthresholdsforeach major are reasonablyhomogeneous frominstitutionto institution.
300
INDUSTRIAL AND LABOR RELATIONS REVIEW
Table 2a. Multinomial Logit Estimatesof Choice of Major forWomen.
(1989 Cohort, 12 Institutions)
Math Scores
ChoiceofMajor
Humanities
Economics
Psychology
ExplanatoryVariable
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Politicsa OtherSocial Sciences Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Sciences
Biology-Life
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Math-PhysicalSciences
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Constant
Engineering
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
OtherFields(NEC)
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
No. Obs.
X2(70)
CoefficientStd.Error
Reference Group
-1.9159
(0.2107)
2.3380
(0.3770)
1.8124
(0.2746)
1.2625
(0.2552)
1.0963
(0.2544)
-0.0250
(0.3055)
-1.3916
(0.1588)
0.8581
(0.3442)
0.9206
(0.2074)
0.2793
(0.1980)
0.5001
(0.1891)
0.0976
(0.2005)
-0.4786
(0.1149)
0.5440
(0.2452)
0.2106
(0.1571)
0.0147
(0.1409)
0.1523
(0.1379)
-0.1180
(0.1436)
-1.8376
(0.1873)
2.2891
(0.2809)
1.7375
(0.2201)
1.3758
(0.2085)
0.9013
(0.2169)
0.6547
(0.2230)
-2.9104
(0.2996)
3.0099
(0.3805)
2.1529
(0.3327)
1.4245
(0.3281)
1.1945
(0.3335)
0.7917
(0.3495)
-3.6820
(0.4312)
4.2719
(0.4923)
3.0523
(0.4576)
1.8321
(0.4617)
0.9739
(0.5004)
0.2542
(0.5645)
-2.5102
(0.2728)
0.4750
(0.6720)
0.3794
(0.4095)
0.6545
(0.3413)
0.3272
(0.3560)
0.3922
(0.3458)
4290
533.14 Log Likelihood
math score constant would increase the
probabilityof majoring in biology relative
to theprobabilityofmajoringin economics
formen and women. Similarly,the higher
VerbalScores
Coefficient Std.Error
-3.0289
-2.5770
-1.7221
-1.7222
-0.6350
(0.6122)
(0.3191)
(0.2192)
(0.2400)
(0.1934)
-1.2115
-1.3127
-1.0657
-0.5789
-0.0057
(0.3433)
(0-2331)
(0.1924)
(0.1808)
(0.1655)
-0.9104
-0.9321
-0.6279
-0.2932
-0.0617
(0.2518)
(0.1689)
(0.1398)
(0.1352)
(0.1315)
-1.0460
-1.0620
-0.8131
-0.6795
-0.2224
(0.2766)
(0.1949)
(0.1689)
(0.1730)
(0.1658)
-0.9453
-1.0464
-0.7585
-0.5320
-0.1984
(0.3641)
(0.2707)
(0.2363)
(0.2400)
(0.2363)
-1.5511
-1.8584
-1.3056
-0.5633
-0.0514
(0.4618)
(0.3676)
(0.3096)
(0.3009)
(0.2949)
-1.7445
-0.9632
-1.0056
-0.9063
-0.8437
(0.7548)
(0.3588)
(0.3129)
(0.3152)
(0.3194)
-6,962.2844
the student's math SAT, the greater the
probabilitythat the student will major in
fields other than the humanities. But the
magnitude of the effectof math SATs dif-
GENDER
AND CHOICE
OF MAJOR
301
Table 2b. Multinomial Logit Estimatesof Choice of Major forMen.
(1989 Cohort, 12 Institutions)
Math Scores
ChoiceofMajor
ExplanatoryVariable
Humanities
Economics
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Psychology
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Politics& OtherSocial Sciences Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Sciences
Constant
Biology-Life
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Constant
Math-PhysicalSciences
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Engineering
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
OtherFields(NEC)
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
CoefficientStd.Error
Reference Group
-1.3075
(0.3003)
1.9498
(0.3564)
1.2388
(0.3329)
0.9797
(0.3279)
0.5725
(0.3439)
0.2407
(0.3821)
-1.6224
(0.3473)
0.5688
(0.4708)
0.1497
(0.4165)
0.2710
(0.3994)
0.3733
(0.4048)
0.4199
(0.4285)
-0.1919
(0.1945)
0.2581
(0.2543)
-0.0627
(0.2214)
-0.1009
(0.2171)
0.0528
(0.2222)
-0.2014
(0.2482)
-1.5807
(0.3326)
2.1025
(0.3803)
1.2334
(0.3637)
0.7161
(0.3645)
0.2614
(0.3889)
0.3398
(0.4127)
-2.0508
(0.4045)
3.1494
(0.4407)
1.7709
(0.4314)
0.9211
(0.4371)
0.0453
(0.4891)
0.3394
(0.5031)
-2.5625
(0.5190)
4.0940
(0.5457)
2.6755
(0.5382)
1.7179
(0.5424)
1.1144
(0.5657)
1.0124
(0.5950)
-2.5706
(0.5259)
1.3844
(0.6395)
0.6665
(0.6010)
0.6173
(0.5850)
-0.2292
(0.6698)
0.5061
(0.6366)
No. Obs.
x2(70)
3,373
579.36 Log Likelihood
VerbalScores
Coefficient Std.Error
-1.5184
-1.6997
-1.4903
-0.7848
-0.6173
(0.3553)
(0.2456)
(0.2186)
(0.2035)
(0.2070)
-1.3777
-1.3382
-0.8578
-0.7690
-1.0075
(0.5648)
(0.3534)
(0.2832)
(0.2827)
(0.3111)
-1.0043
-0.7779
-0.5172
-0.2087
-0.2095
(0.3070)
(0.1969)
(0.1735)
(0.1702)
(0.1735)
-1.6352
-1.1582
-0.7252
-0.5337
-0.8396
(0.4119)
(0.2496)
(0.2223)
(0.2242)
(0.2493)
-1.2464
-1.0619
-0.9664
-0.8358
-0.4635
(0.3534)
(0.2527)
(0.2401)
(0.2489)
(0.2502)
-1.8427
-1.5820
-1.0985
-0.8272
-0.6815
(0.3586)
(0.2400)
(0.2178)
(0.2218)
(0.2317)
-1.1917
-1.5164
-0.9247
-0.6124
-0.8416
(0.6796)
(0.4897)
(0.3973)
(0.3868)
(0.4283)
-5,881.4223
fers appreciably across the fields."3 For
both men and women, increases in the
math SAT increase the probabilityof majoring in engineeringor math-physicalsci-
13More generally,testsof the hypothesisthatall of
the coefficientsare equal to zero for each major and
testsof the equality of coefficientsacross majors are
rejected nearlyacross the board (there are 28 pairwise
302
INDUSTRIAL AND LABOR RELATIONS REVIEW
ences relative to any other field, holding
the verbal score constant. Note that in
these fields,relativelylarge changes in the
choice probabilities occur near the top of
the scale. This result reinforcesthe intuitivelyplausible proposition that moving
froma 700 to a 750 math score is likelyto
have a much largerimpacton the probabilitythat a studentwill major in a field like
theoretical physics than will a shiftfrom
600 to 650.
Increases in mathSAT scores have a uniformlypositiveeffecton the probabilityof
majoring in economics and the life sciences relative to the probabilityof majoring in the humanities,as indicated by the
positive parameter estimates. However,
changes in mathscores do not have as large
an effecton the expected probabilities of
choosing these fields as they do in the
physical sciences and engineering. Little
can be said about the variationin the probabilityof choosing to major in psychology
associated with test scores, as the parameter estimatesare small in magnitude for
women (thoughstatistically
significant)and
generally indistinguishablefrom zero for
men.
Calculating themarginalprobabilitiesthe predicted change in the probability
associated with an incremental change in
an explanatoryvariable-indicates therelative sensitivityof each field choice to
changes in math and verbal scores. A key
facet of the multinomialestimationis the
tests for each gender group). At the 1% level of
significance,the exceptions are psychology-political
sciences (men and women), other-humanities (men
and women), other-psychology(men and women),
other-politics(men and women), other-biology(men
and women), math-physical sciences-engineering
(men only), economics-other (men only), psychology-politics (men only), and biology-math-physical
sciences (women only). These tests suggest that
relativelyhigh SAT scores (particularlyin the quantitativedimension) increase the probabilitiesof majoring in both the lifesciences and the physicalsciences
forwomen, while formen there is greaterseparation
between the life sciences and the physicalsciences in
the upper dimensions of the SAT range.
fact that the magnitude of the marginal
effects-illustratedas the slope of a function relating the explanatory variable to
the expected probabilityofmajoringin any
given field-depends on the level of the
individual's scores. Since the models use
categoricalexplanatoryvariables,a discrete
parallel is themeasurementoftheexpected
change in the probabilityassociated witha
movementbetween achievementscore categories. Figures 3 and 4 trace,forwomen
and men, respectively,the expected probabilities associated with each field choice
and combination of math and verbal
achievement categories. A flathorizontal
plane would be indicativeofa fieldin which
the choice probabilities did not varywith
test scores, while a plane with a constant
slope would indicate a field in which the
choice probabilityincreased or decreased
steadilywithachievementscores.
The "shapes" of the predicted functions
are, withsome exceptions,quite congruent
for men and women, though the levels of
the planes differ. In the case of the humanities, the surface moves up from the
low verbal, high math corner to its maximum in the high verbal, low math corner.
As expected, the probabilityof majoringin
engineering and math-physical science
rises at an increasing rate withhigh math
SAT scores. For both men and women, the
highestprobabilityofchoosingmath-physical sciences is associated withtop category
mathand verbal scores,while the probabilityof choosing engineeringbears less relation to the verbal score. Comparing the
picturesformen and women, note thatthe
rise in probabilitiesassociated withmoving
from the penultimate to the highest categoryin math SATs is noticeablylargerfor
men than forwomen. The predicted probability of majoring in engineering for
women withmodest math scores (between
600 and 650) is close to zero (the picture
resemblesa throw-rug)
,whilethepredicted
probabilityformenwithscoresin thisrange
is decidedly higher.
The graphs for economics and the life
sciences suggest a topographywith more
hills and valleysand somewhatgreaterdifferencesby gender than are found for the
GENDER AND CHOICE OF MAJOR
otherfields. For men, twogroups are most
likelyto major in economics: those with
moderatelyhigh SAT scores (650 to 750)
and low verbal scores and those withboth
high verbal and high quantitative scores.
For women, those withthe highestquantitativescores (greater than 750) and lowest
verbal scores (less than 550) are mostlikely
to choose to major in economics, while
those withhigh scores on both dimensions
of the SAT are less likelyto choose economics than those with low scores on both dimensions. In the life sciences, the choice
probabilities for men rise modestly with
math and verbal testscore categories until
theyreach a "hilltop"withverbal scores in
the 650 to 700 range. For women,theplane
continues to slope modestlyupwardas both
math and verbal scores rise, reaching its
peak in the categoryreflectingthe highest
combined scores. These figures provide
fascinatinginsightsinto the sometimesdifferentwaysin whichwomen and men "convert"mathand verbalaptitudesintochoices
of majors.
We now employthe decomposition strategy outlined in the previous section in order to evaluatetheimportanceofSAT scores
in explainingdifferencesin the majorschosen bymen and women. The resultsforthe
1989 cohort,usingwomen as the reference
group, are presentedin Table 3. The table
shows three sets of differences: the total
observed differencesin the actual distributions;14 the differencesattributableto differencesbetween women and men in SAT
scores (or "attributes"or X values); and the
remaining differences,attributableto differences in "preferences" and other unspecified variables. These differencesare
computed fromthe observed distributions
of men and women and the distribution
predicted using the SAT scores observed
for men and the estimatedparametersfor
women. (We made analogous calculations
'4From the firstorder conditions of the log likelihood function,it can be shown that the mean of the
predicted values is equal to the population shares.
303
Table 3. Decomposition Analysis
of Model of Undergraduate Choice
of Major: College Graduates Only,
1989 Cohort, Nonlinear Model.
Field
(A)
(B)
8.6%
-4.5%
5.0%
5.1%
-0.7%
0.4%
3.6%
-3.8%
4.6%
0.8%
1.8%
-1.0%
(C)
Unexplained
Difference bySAT
Observed Due to
Scores
SAT Scores [Exp %M
Difference,
% W- [Actual %W -Actual
%M
-Exp %M]
%M]
Humanities
Economics
Psychology
Politics & Other
Social Sciences
Biology-Life
Sciences
Math-Physical
Sciences
Engineering
Other
2.4%
-2.2%
4.6%
-4.3%
-8.4%
0.3%
-2.0%
-2.7%
0.3%
-2.3%
-5.7%
0.0%
Index of
Dissimilarity
17.2%
7.5%
12.8%
Notes: Calculations are based on a multinomial
specificationthatincludes SAT Verbal and SAT Math
scores as categorical variables. The 12 institutions
include Stanford,Yale, Princeton, Kenyon, Oberlin,
Swarthmore, Hamilton, Williams, Wesleyan, Bryn
Mawr, Smith, and Wellesley. Column (A) presents
the actual differencebetween the share of women
majoring in a field and the share of men majoring in
the field; column (B) is the differencebetween the
actual share of women in a field and the predicted
share of men, using the coefficientsfromthe multinomial estimation for women (Table 2a) and the
actual SAT scores of the men at the 12 universities;
and column (C) is,the differencebetween the predicted distributionof men and the observed distribution for men.
using the SAT scores observed forwomen
and the estimatedparametersformen; the
resultsare not qualitativelydifferent.)
In the case of the humanities,we find
thatwhile the observeddifferencebetween
the actual distributionsforthe women and
men was about 8.6 percentage points (42%
of the women majoring in the humanities,
versus 34% of the men), about 5 percentage points of thisgap can be attributedto
SAT scores; the remaining3.6 points indi-
304
INDUSTRIAL
AND LABOR RELATIONS
REVIEW
Figure3. PredictedChoice of Major: ProfilesforWomen,1989 Cohort.
0275
/'
....0.25
Humanities
77
Biology- LifeSciences
VERBAL
725
0.10
0.15
Econoicis
010-
0.05~
-
_2
575
Sciences
Biolog Phyical
0,25
~~~~~~~~~~~~~~~~0.05
0.05
0-w
725AL~
syhoog
0.10
cate the degree to which women with the
same math and verbal SATs as the men are
more likelythan the men to major in the
humanities. Similarly,the much greater
tendencyformen thanforwomen to major
in math-physicalsciences and in engineering also appears to be thejoint product of
variationsin SAT scores (highermathscores
for the men) and the relativelystronger
preferences among men for these fields,
though the relative magnitudes of these
Egieein
Eninern
effectsare not at all equal. Test score
differencesaccount for about 45% of the
total gender gap in the math-physicalsciences fieldsand forabout 32% in engineering. In economics and psychology,differences between women and men in SAT
scores explain onlya verysmall part of the
gender gap-about 16% in economics and
less than 8% in psychology. Instead, it is
pronounced differencesin "preferences"
and otherresidual forcesthatgenerate the
GENDER AND CHOICE OF MAJOR
305
Figure 4. Predicted
forMen,1989Cohort.
ChoiceofMajor:Profiles
*00S
X.5
:0:.25
Humanities
Humanities
Economics
Pyh725
o
yAL
large gender differencesin representation
in these fields. In biology-lifesciences, the
twoforcespull in opposite directions. Differencesin SAT scores alone would lead us
to expect more men than women to major
in these fields; however,women's preferences for the life sciences relativeto other
fieldsare so much strongerthan men's that
the net effectis that women are modestly
over-represented.
In summary,we find that differencesin
0.20
525
Biology- LifeSciences
cience
Math
- Physical
Sciences
Biolgyifeer
e725 E
VERBAL
SAT scores account forless than halfof the
total gender gap. If men had the same
preferencesas women but differedonly in
the distributionof SAT scores,the index of
dissimilaritywould drop to 7.5. As noted
above, the qualitativeconclusions reached
are much the same whetherwe use women
or men as thereferencegroup. It should be
emphasized that the share of the differential associated with test scores is appreciablylowerusing thistaxonomythanitwould
INDUSTRIAL AND LABOR RELATIONS REVIEW
306
Table 4. Decomposition Analysis
of Model of Undergraduate Choice
of Major: College Graduates Only,
1976 Cohort, Nonlinear Model.
(A)
Field
(C)
Unexplained
Difference bySAT
Scores
Observed Due to
SAT Scores [Exp %M
Difference,
%W- [Actual %W -Actual
%M
%M]
-Exp %M]
Humanities
11.8%
-4.9%
Economics
2.7%
Psychology
Government& Other
Social Sciences
2.0%
Biology-Life
Sciences
-0.2%
Math-Physical
Sciences
-2.8%
Engineering
-8.9%
Other
0.3%
Index of
Dissimilarity
16.8%
(B)
5.2%
-1.0%
1.0%
6.5%
-4.0%
1.8%
2.3%
-0.3%
-2.2%
2.0%
-2.9%
-2.6%
0.2%
0.1%
-6.2%
0.1%
8.7%
10.5%
Notes: See explanation in notes to Table 3 and
estimationresultsin Appendix Tables 2a and 2b.
be ifwe used a simple dichotomybetween
science and non-science fields.'5
The same cross-sectionalanalysisapplied
to data forthe 1976 enteringcohort yields
generally-but not exactly-similar results
(Table 4). (Logit estimatesappear in Appendix Tables 2a and 2b.) For these individuals,differencesin SAT scores explain a
slightlyhigher fractionof the gender gap,
particularlyin the physicalsciences, where
the observed differenceis more than accounted forbydifferencesin testscores. In
Table 5, we summarize the changes over
thisintervalby comparing the decompositions for the 1976 and 1989 entering co-
horts; changes in the overall size of the
gender gap foreach field are apportioned
between a component attributable to
changes in the relativedistributionsof SAT
scores for men and forwomen and a component attributable to shifts in preferences.'6 In general, changes in thedistribution of testscores formen and women had
verylittleeffecton gender gaps. Changes
in the preferences or forces other than
measured achievement were the driving
force in wideninggaps in psychology,the
life sciences, and math-physicalsciences.
For example, while the gap between the
shareofwomenand theshareofmen choosing to major in biology-life sciences increased from a small over-representation
of
ofmen in 1976 to an over-representation
women by nearly 2.5 percentage points,
almost none of this change is attributable
to changes in the SAT scores for men and
women.
In understandingthese transformations,
a key piece of the puzzle is to distinguish
the changing "weights"used by men and
womenin matchingachievementlevelswith
fieldsof studyfromchanges in overall levels of demand. For example, one might
consider changes in the discipline of economics requiring greater levels of mathematics achievementas leading to changes
over time thatare independent of gender,
'6Combining the period-specific decomposition
expressions, we obtain a measure of the extent to
which the change in the gap between cohorts is
related to changes in SAT scores or to changes in
preferencesand other residual factors. In each field
j at time t,
G
(pRFtXFt
I
=p=F
XF _
pPFtXMt)
+
i
pPMtXMt=
(p5RXMt
I
_.p5MtXMt)
I
The change in the gap can be writtenas
C -G.
+
'5The dichotomous distinction between non-science and science majors yields a total differenceof
10.26 percentage points, with 6.7 percentage points
attributableto differencesin testscores and about 3.5
percentage points attributableto differencesin the
residual.
-
[(p
= [ (pI5FXFI
F1XM_
pIFIXMt)
pPMtXMt)
-
(p
(pI3F -XFtJ
pF3Ft
_Ft-)XMt-_
p
,XM1 ,)]
Mt-IXMt-))]
j
I
I
The firsttermto the rightof the equal sign represents
the portion of the change due to changes in the
respectivedistributionsof SAT scores, while the second termin square bracketscaptures the portion of
the change attributableto changes in the difference
in the relative magnitude of the beta coefficients.
GENDER
AND CHOICE
OF MAJOR
307
Table 5. Decomposition of the Change in the Gender Gap in Field Preferences:
1976 to 1989 Entering Cohorts, 12 Institutions,College and Beyond Data.
Field ofStudy
Humanities
Economics
Psychology
Politics & Other Social Sciences
Biology-Life Sciences
Math-Physical Sciences
Engineering
Other
Level of
Gap in
1976
0.1176
-0.0493
0.0273
0.0197
-0.0017
-0.0278
-0.0888
0.0031
Sign of
Gapfor
1976
Direction
ofChange,
1976-89
Total
Change
in Gap
W>M
W<M
W>M
W>M
W<M
W<M
W<M
W>M
Narrow
Narrow
Widen
Narrow
Ch Sign
Widen
Narrow
Narrow
-0.0313
0.0044
0.0228
-0.0114
0.0262
-0.0154
0.0050
-0.0002
Changein
Gap Due to
Changein
SA7'Scores
-0.0017
0.0022
-0.0056
-0.0046
0.0004
0.0095
-0.0005
0.0003
Changein
Gap Unexplained
byChange
in SAT
Scores
-0.0296
0.0022
0.0284
-0.0068
0.0258
-0.0248
0.0055
-0.0006
Note: Components of change are calculated fromTables 3 and 4. See textfor details of calculation.
while factors such as the opening of the
medical profession to women would be
likelyto have particularlystrongeffectson
women's choices. A confounding force is
the change in the overall demand for particular fields of study,which can serve to
expand or contractthemeasured gaps without affectingthe relativerepresentationof
men and women by field.
Because data on SAT scores for the year
1951 are widelyavailable foronlyone ofthe
schools examined in our study,we will not
burden the text of the paper with regressions,tables,or figuresbased on thosedata.
The admittedlyincomplete evidence suggests,however,thatlarge-scaleshiftsin preferences (especiallybywomen) were mainly
responsible for the pronounced shrinking
of gender gaps in choice of major between
the mid-1950s and the late 1970s. In the
1951 enteringcohort,womenwiththesame
SATs as men made radically different
choices of majors; however,by the time of
the 1976 cohort,preferenceshad converged
to a considerable extent. We interpretthis
resultas a reflectionof theopening up over
thisperiod ofmanymorejob opportunities
forwomen in traditionallymale fields,the
greatercommitmentofwomen to extended
labor force participation, and a general
"loosening"ofstereotypesas towhatwomen
and men should do withtheirlives. There
were dramatic contractionsin the sizes of
the preferencecomponents of the gender
gaps in all of the predominantly male
fields-and also in the humanities.
It is against this background that we
should interpretwhat did nothappen between the times of the 1976 and 1989 cohorts. As we saw in Table 5, the sizes of the
preferencecomponentsof thegender gaps
not only failed to continue to shrink,they
rose in psychologyand the life sciences
(withwomen expressingeverstrongerpositive preferences), and theyrose in math/
physicalsciences too (withwomen expressing much less interestthan men, as compared withthe situationin 1976). In short,
the social and economic forcesthatpushed
fora convergence in choices of major after
the mid-1950s either spent themselvesby
the late 1970s or were subsequently overtaken bynew forces. More recently,differences between men and women in choice
of major appear to have become entrenched, in spite of modest continuing
convergencein SAT scores;gender-specific
preferencesnow appear to be widening a
number of gaps rather than narrowing
them.
Conclusion and Open Questions
The decompositions of gender gaps re-
308
INDUSTRIAL AND LABOR RELATIONS REVIEW
ported in thispaper are entirelyconsistent
with the commonly held view that differences in the academic preparation of
women and men help explain observed
differences in characteristic choices of
major. The more strikingconclusion is
thatdifferencesin SAT scores are nothing
like the full story. In fact, they capture
much less of the dynamicsof change over
the past 35 or 40 yearsthan do the panoply
of residual forces,including differencesin
preferences, labor market expectations,
gender-specificeffectsof the college experience, and unmeasured aspects of academic preparation. To answerdirectlyone
of the questions posed at the startof the
paper, we do notsee continuingmovement
toward gender neutralityin the skills (or
credentials) that students take from college.
One possible interpretation is that
women and men typicallyattach different
"weights"(or values) to the after-college
opportunitiesassociated witheach area of
study.Recognizingpotentialfamilyresponsibilities,women maypreferfieldsin which
skills are unlikely to atrophy or become
obsolete-for example, formembersof the
1976 cohort, a keen understanding of
Shakespeare mayprovide more opportunities in the 1990s than knowledge of the
(nearly obsolete) COBOL computer language. Yet, this "depreciation effect"may
be more importantin theorythan in practice. As women are observationallyless and
less likely to interrupt their labor force
participationforprolonged periods oftime
to raise children, one would expect the
gender gap in field choice to narrow accordingly-and it has not.
A second typeofexplanation is thatthere
continue to be impediments in the labor
marketto women receivingthe same wages
and professional opportunities as men in
occupations related to particular majors.
To some extent,measures of wage premia
associated with undergraduate major are
indicative of observed market incentives
(or disincentives) to pursue certain fields
that may varyby gender. Following wage
regression estimates from Brown and
Corcoran (1997) that use data from the
National LongitudinalSurveyof theClass of
1972, we see that women with a college
degree receivea higherpremiumthanmen
in the humanities,a lower returnin biology,a slightlylowerreturnin mathand the
physical sciences, and a much greater returnin engineering. If women in the 1989
enteringcohort observed these returnsas
"prices" before choosing majors, then the
over-representationof women in the life
sciences and their under-representation
in both the physical sciences and engineering would have to be seen as very
surprising.
Nevertheless, if undergraduate field
choice is an intermediate step in transitions to professional fields such as law or
medicine and to advanced academic study,
then examination of the "final"labor force
opportunities in these fields may help to
explain gender differencesat the undergraduate level. In thisregard,theconsiderable widening of opportunitiesforwomen
in medicine over the past two decades is
consistent with the high participation of
women in the life sciences at the undergraduate level. Grasping the underlying
relationshipswould require detailed study
of the linksjoining field choice, occupation choice, labor force participation,and
remuneration.
Otherpossibleexplanationsforthehardy
persistenceofgendergaps include thepresence of unmeasured differences in
precollegiate preparation in math, and
marked variations in the experiences of
men and women during the college years.
Some have suggested thatSAT scores do a
poor job of measuring higher order math
skills. Given the same SAT scores,men may
exceed women in the full specificationof
math skills on college entrance, perhaps
because theytend to take more advanced
math courses in high school than their
female counterparts. If this hypothesisis
true,we would expect to see a significant
relationship between academic performance and gender in the upper ranges of
math SAT scores in courses requiring a
high degree ofquantitativeskill. However,
the data do not provide enough information to address the counterfactualquestion
GENDER AND CHOICE OF MAJOR
309
ofhowan individualwould have performed ket expectations,gender-specificeffectsof
the college experience, and unmeasured
academically if he or she had chosen a
differentfield of study. Since observed
aspects of academic preparation-account
forthe main part of today'sgender gaps in
undergraduate performancemeasured by
choice of academic major.
GPA is conditional on field choice and
It seems abundantlyclear thatthe simple
course selection,the observedrelationship
science-non-science dichotomy, however
between undergraduate GPA and SAT
helpfulit once was, is no longer usefulas a
scores will not provide an answer to the
counterfactualquestion of how a student
taxonomyif the objective is to understand
field choices by women and men. Quite
who did not major in a particular field
plainly,there is a wideningdivide between
would be expected to performin thatsubthe life sciences, on the one hand, and
ject.
The choices men and women have made
math-physicalscience-engineering,on the
in elective fields of study at the underother hand, in their attractiveness to
women. The magnitudeand persistenceof
graduate level have not steadilyconverged.
Differences in academic preparation of
the disproportionatelyheavy representation of men in economics, as well as in
women and men measured crudelyby test
scores explain some ofthe persistentdiffer- engineering, mathematics,and the physiences in characteristicchoices of major,
cal sciences, pose the puzzling question of
whetherwe are attemptingto account for whythese fieldsare so differentfromother
changes over the past three decades or for fields also classified broadly as "science."
the cross-sectionalgap forthe most recent
Perhaps thefindingsreportedin thispaper
cohort. But such differencesin SAT scores
will stimulatenew, even more determined,
are onlypartof the story,and a modest part
effortsto unravel the forces at work-a
at that.An arrayofresidual forces-includhard task in part because of the need to
ing differencesin preferences,labor mar- workacross disciplinaryboundaries.
INDUSTRIAL AND LABOR RELATIONS REVIEW
310
co
t
F
W~~~~~~
-
Sc
00
-A4
~~~~~z b: b: bz
U~~~~~~~~~~~~~~~n
b~
-
06 C;
o
C;
n
.^
~
i\?
So
o
@ !,
F~~~~
Co
*
b w,?I0
i-e, -0
@
@
~~~~
@
o
00
-c
0
0
i}
00
00
t~k
~
C1
~
,
NOv
g
-
,
-
00
-
~~
z
t
!0
L
cCO
o
t
6 -
CQ
o
o
C;
A
,2,O
i~~.
, ?,0xOm
?-OOe4
Cci?Q0ZQ0
?~~~~~
t
in C)
-~~~~~~~~~~~~~~~~~~~~C
-
CO
t-C; eee
?
'D
s
'r-
n o
= v
We:;
gg
,!,9
o
t
006
-.Oc
Y
)
,~~ .1ge
I
c
O2is
s
:,
,>
o
Cos
-i
Ci
cl 0e
CO
xs~~~S
G?0,,toG
?
-
uS
GAsN
eg
A
g
.
@ ~~~~~~~~~~~n
6
X1
.,
1
Cu
i,
<2
g
-0
o
C'~~
r-
cli
i
.w
-t o~~~~~~~~~~(
b
-
o6
0
00
ti4,
n~~~~~~~~~~~~~~~~~~
cl ~
Q~~~~~064
C~~~~~~~~~~~~~~~~~~~~~C
sb
0
t
in
@
,o
<;s
00
OY
-
g
i
O
CIA
r g
b0?bW
e
O
~~~~n
~ ~00
ce2,r2,
eo ^
GA
t.,o 00
0
Co
-
zo6.
bs
;,
i
O:.
Y
gg
egg
e
,c 0
X~~~~~~~~~~~0
v
O
-
(=
=S
-
..
,
@ S
J
00
0o
C~~~~~~~~~~~~~~~~~~~~~~~~~~I
(I,
C;ttsto
m
Kn
-
bs
or
C;
r
-0
00 sz0 AOC
O
C
Co
V
oo ro v:
t
o
~~~~~~~~~~~~~~~~00
.;
g 0 r
o
co-
r
-
8
N
a14
s
0
t U
-
GENDER AND CHOICE OF MAJOR
311
APPENDIX TABLE 2A
MultinomialLogit Estimates of Choice of Major for Women, 1976 Cohort, 12 Institutions
Math Scores
ChoiceofMajor
ExplanatoryVariable
Humanities
Economics
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Psychology
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Politics& OtherSocialSciences Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Biology-Life
Sciences
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Math-PhysicalSciences
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Constant
Engineering
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
OtherFields(NEC)
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
No. Obs.
X2(7O)
CoefficientStd.Error
VerbalScores
Coefficient Std.Error
Reference Group
-1.8367
1.2917
1.4866
0.8805
0.9573
-0.0424
-1.3174
-1.4079
0.5437
0.3728
0.2681
0.0879
-0.6176
-0.1716
-0.1394
-0.0410
-0.0916
-0.1123
-1.7144
1.9805
1.6226
1.4808
1.0326
0.6048
-2.4343
2.4570
2.2397
1.7028
1.0233
-0.0593
-4.0769
4.5923
3.7628
2.7454
2.0844
0.5316
-2.0654
0.3228
0.3536
0.0399
0.2437
-0.0608
3,742
478.37
(0.1794)
(0.4340)
(0-2391)
(0.2262)
(0.2035)
(0.2385)
(0.1500)
(1.0269)
(0.2429)
(0.2081)
(0.1903)
(0.1906)
(0.1159)
(0.3747)
(0.1953)
(0.1571)
(0.1429)
(0.1396)
(0.1665)
(0.3267)
(0.2178)
(0.1943)
(0.1899)
(0.1968)
(0.2293)
(0.3705)
(0.2680)
(0.2567)
(0.2608)
(0.3153)
(0.5153)
(0.6147)
(0.5552)
(0.5566)
(0.5595)
(0.6772)
(0.2031)
(0.5213)
(0.3012)
(0.2753)
(0.2406)
(0.2512)
Log Likelihood
-2.8186
-1.3174
-0.9264
-0.5192
-0.1946
(0.7388)
(0.2647)
(0.2158)
(0.1983)
(0.1991)
-1.1709
-1.5920
-0.8491
-0.4385
-0.6276
(0.4048)
(0.2916)
(0.2035)
(0.1806)
(0.2011)
-1.4110
-0.7291
-0.4015
-0.3424
-0.2054
(0.3753)
(0.1893)
(0.1544)
(0.1468)
(0.1517)
-1.8184
-1.1520
-0.9220
-0.5109
-0.2273
(0.3854)
(0.2114)
(0.1822)
(0.1692)
(0.1717)
-1.5753
-1.0976
-0.5663
-0.5675
-0.4093
(0.4247)
(0.2646)
(0.2237)
(0.2270)
(0.2380)
-2.4713
-1.7830
-1.5429
-1.0884
-0.5073
(0.6571)
(0.3821)
(0.3401)
(0.3312)
(0.3236)
-0.2329
-0.4738
-0.4006
-0.2791
-0.4484
(0.4144)
(0.3013)
(0.2621)
(0.2477)
(0.2750)
-6,199.9045
INDUSTRIAL AND LABOR RELATIONS REVIEW
312
APPENDIX TABLE 2B
MultinomialLogit Estimates of Choice of Major for Men, 1976 Cohort, 12 Institutions
Math Scores
ChoiceofMajor
ExplanatoryVariable
Humanities
Economics
OtherFields (NEC)
VerbalScores
Coefficient Std.Error
Reference Group
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Psychology
Constant
SAT > 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Politics& OtherSocialSciences Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Sciences
Biology-Life
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Math-PhysicalSciences
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Engineering
CoefficientStd.Error
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
Constant
SAT> 750
SAT 700-750
SAT 650-700
SAT 600-650
SAT 550-600
No. Obs.
X2(70)
-1.6252
3.1139
2.3100
1.8069
1.5430
0.9604
-1.0490
0.5327
0.7194
0.0205
0.0044
-0.3277
-0.6701
0.6775
0.2605
0.3058
0.0588
-0.0002
-1.3874
2.2703
1.4690
1.2270
0.8169
0.5879
-1.6939
2.7773
1.5763
1.1829
0.4728
0.2192
-3.2957
4.9888
4.0782
3.3489
2.3605
1.5908
-1.8945
1.3688
0.9792
0.7507
0.6043
0.4877
3328
597.74
(0.2438)
(0.3088)
(0.2847)
(0.2850)
(0.2814)
(0.3077)
(0.1958)
(0.3929)
(0.2820)
(0.2979)
(0.2793)
(0.3171)
(0.1596)
(0.2511)
(0.2092)
(0.2020)
(0.2013)
(0-2160)
(0.2115)
(0.2818)
(0.2562)
(0.2535)
(0.2553)
(0.2744)
(0.2421)
(0.3055)
(0.2896)
(0.2906)
(0.3062)
(0.3373)
(0.5100)
(0.5425)
(0.5300)
(0.5311)
(0.5413)
(0.5815)
(0.2718)
(0.4122)
(0.3562)
(0.3532)
(0.3447)
(0.3641)
Log Likelihood
-2.6888
-1.9104
-1.4658
-1.3178
-0.8984
(0.4380)
(0.2218)
(0.1891)
(0.1961)
(0.1938)
-3.2838
-1.7121
-1.5078
-0.9599
-1.0449
(1.0319)
(0.3205)
(0.2738)
(0.2543)
(0.2739)
-1.3547
-0.5890
-0.7094
-0.3019
-0.0559
(0.3959)
(0.2051)
(0.1909)
(0.1856)
(0.1839)
-1.4368
-1.4364
-0.9901
-0.6880
-0.4564
(0.3495)
(0.2297)
(0.1962)
(0.1966)
(0.1987)
-1.1365
-1.2053
-0.9338
-0.6573
-0.6774
(0.3438)
(0.2437)
(0.2203)
(0.2228)
(0.2389)
-2.5161
-2.0634
-1.7096
-0.7798
-0.6250
(0.4099)
(0.2408)
(0.2136)
(0.2002)
(0.2099)
-1.3552
-1.1728
-1.1115
-1.0985
-0.9142
(0.5230)
(0.3167)
(0.2838)
(0.3004)
(0.3035)
-6,052.4772
GENDER AND CHOICE OF MAJOR
313
REFERENCES
Bailey, Elizabeth, and Kevin Rask. 1996. "Are We
Role Models? Major Choice in an Undergraduate
Institution." Colgate Economics WorkingPaper.
Bound, John, Michael Schoenbaum, and Timothy
Waidman. 1996. "Race Differencesin Labor Force
Attachmentand DisabilityStatus." TheGerontologist,
Vol. 36, No. 3, pp. 311-21.
Brown, Charles, and Mary Corcoran. 1997. "SexBased Differencesin School Content and the MaleFemale Wage Gap." JournalofLaborEconomics,Part
1, Vol. 15, No. 3, pp. 431-65.
Canes, Brandice, and HarveyRosen. 1995. "Following in Her Footsteps? FacultyGender Composition
and Women's Choices of College Majors." Industrialand LaborRelationsReview,Vol. 48, No. 3 (April),
pp. 486-504.
Corcoran, Mary,and Paul Courant. 1985. "Sex-Role
Socialization and Labor MarketOutcomes." American EconomicReview,Vol. 75, No. 2 (May), pp. 27578.
Eccles,Jacquelynne, and Lois Hoffman. 1984. "Sex
Differencesin PreparationforOccupational Roles."
In H. Stevenson and A. Siegal, eds., ChildDevelopmentand Social Policy. Chicago: Universityof Chicago Press, pp. 367-420.
Jacobs,Jerry. 1995. "Gender and Academic Specialties: Trends Among Recipients of College Degrees
in the 1980s." Sociology
ofEducation,Vol. 68 (April),
pp. 81-98.
Leslie, Larry,and Ronald Oaxaca. 1997. "Variables
Impacting the Supply of Women and Minoritiesin
Science and Engineering." In J. C. Smart, ed.,
HigherEducation: Handbookof Theoryand Research,
Vol. 12. New York: Agathon.
Paglin, Morton, and Anthony Rufolo. 1990. "Heterogeneous
Human Capital, Occupational
Choice, and Male-Female Earnings Differences."
JournalofLaborEconomics,Vol. 8, No. 1, Part 1, pp.
123-44.
Polachek, Solomon. 1978. "Sex Differencesin College Major." Industrialand Labor RelationsReview,
Vol. 31, No. 4 (July),pp. 498-508.
Solnick, Sara. 1995. "Changes in Women's Majors
fromEntrance to Graduation at Women's and Coeducational Colleges." Industrialand LaborRelations
Review,Vol. 48, No. 3 (April), pp. 505-14.
Turner, Sarah, and William Bowen. 1990. "The
Flight from the Arts and Sciences: Trends in Degrees Conferred." Science,Vol. 250 (October 26),
pp. 517-21.
Ware, Norma, and Valerie Lee. 1988. "Sex Differences in Choice of College Science Majors." AmericanEducationalResearchJournal,
Vol. 25, No. 4 (Winter), pp. 593-614.
Ware, Norma, Nicole Steckler, andJane Leserman.
1986. "Undergraduate Women: Who Chooses a
Science Major?" JournalofHigherEducation,
Vol. 56,
No. 1 (January-February),pp. 73-84.