CHOICE OF MAJOR - Earlham Collegelegacy.earlham.edu/~pardhan/research/refs/choice_of_major.pdf · Among the many factors that influence men's and women's choice of major, and consequently

Choice of Major: The Changing (Unchanging) Gender GapAuthor(s): Sarah E. Turner and William G. BowenSource: Industrial and Labor Relations Review, Vol. 52, No. 2 (Jan., 1999), pp. 289-313Published by: Cornell University, School of Industrial & Labor RelationsStable 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 unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay 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 printedpage of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

Cornell University, School of Industrial & Labor Relations is collaborating with JSTOR to digitize, preserveand extend access to Industrial and Labor Relations Review.

http://www.jstor.org

http://www.jstor.org/action/showPublisher?publisherCode=cschool

http://www.jstor.org/stable/2525167?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp

http://www.jstor.org/action/showPublisher?publisherCode=cschool

CHOICE OF MAJOR: THE CHANGING (UNCHANGING) GENDER GAP

SARAH E. TURNER and WILLIAM G. BOWEN*

Within the arts, sciences, and engineering fields, differences between 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 effects of 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 between men and women in field of study chosen at the under-

graduate level may represent differences in the skills that these groups bring to the labor market and may partially explain observed differences in wages. Choice of major, as well as decisions about where to attend college, is an important link in the chain of decisions and events that build human capital for those fortunate enough

*Sarah Turner is Assistant Professor of Education and Economics at the University of Virginia and William Bowen is president of the Andrew W. Mellon Foundation. The authors thankJohn Bound, Charles Brown, Ronald Ehrenberg, Michael McPherson, Paula Stephan, Yu Xie, and Harriet Zuckerman for reading earlier drafts and providing thoughtful comments on this research.

to go on to higher education. Choice of school, choice of major, and academic performance coalesce to influence options available to students for further education and career development. Choice of major is both an immediate outcome of the educational process and a determinant of later outcomes of many kinds. Understanding the factors that influence choice of major, for men and women, is one part of the larger process of understanding personal as well as societal returns to varied investments in human capital.

In addition to probing present-day patterns in choice of major, we will investigate 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 Labor Relations Review, Vol. 52, No. 2 (January 1999). ? by Cornell University. 0019-7939/99/5202 $01.00

289

290 INDUSTRIAL AND LABOR RELATIONS REVIEW

Among the many factors that influence men's and women's choice of major, and consequently gender differences in careers and wages, three factors related to gender may be particularly important: a student's preparation and achievement at precollegiate levels of education, especially in mathematics; an individual's preferences for various courses of study, which may be encouraged by parental and societal expectations; and the labor market prospects associated with a given set of skills, which may provide more encouragement for one sex than the other to pursue certain fields of study. Our focus in this study is on the first of those factors-the extent to which differences between men and women in precollegiate achievement (measured by SAT verbal and math scores) account for differences in choice of major at the college level. After examining the national trend in the choice of undergraduate major by men and women, we turn our atten- tion to the choices made by men and women at a small set of selective research universities and liberal arts colleges.

National Trends and Descriptive Findings

The Dissimilarity Index: Trends in National Data

One starting point for measuring dis- similarities between women and men in choice of major is to calculate the absolute value of the differences between the per- centages of women and men majoring in each of the fields into which we classify students, sum these differences, and divide by two. This measure, referred to in other literatures (such as those on residential and occupational segregation) as an index of dissimilarity, captures the percentage of students who would need to change majors in order for parity to be achieved in the distributions, with 100% indicating com- plete segregation and 0% indicating identical distributions.1

'Jacobs ( 1995) appears to be the first researcher to have applied this measure to the different distribu-

Movements over the past 30 years in an all-inclusive dissimilarity index and in a dissimilarity index limited to arts-sciences- engineering (A-S-E) are shown on Figure 1 for the entire population of B.A. recipients in the United States.2 The all-inclusive, "All-Fields" index is calculated using the number of B.A. recipients at all U.S. institutions in 10 broad field-of-study categories, including an array of professional fields that together accounted for just over half (53%) of all B.A. recipients in 1995. Just two fields-education and business-account for nearly 2/3 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 students within the arts-science-engineering disciplines (the remaining 47% of B.A. recipients). The A-S-E aggregate captures the range of concentrations offered at the undergraduate level at the institutions included in this analysis. In the main, these selective liberal arts and research universities do not award degrees in professional preparatory programs such as business or education. Engineering programs, generally part of distinct engineering schools within the institution, represent the primary undergraduate professional degrees offered at the research universities included in this study; moreover, in many instances these programs of study could as easily be called "applied science" as "engineering."

tions of men and women by field of study. The basic descriptive results presented here are substantively similar to his, with differences attributable to the use of different field taxonomies.

2The 10 field composites used in the calculation of the "All Fields" dissimilarity index are engineering, physical sciences-math, life sciences, psychology, economics, politics and other social sciences, science- engineering technologies, humanities, education, business and communications, and a broadly defined "other" category including the social services profes- sions and vocational studies. The 7 field groups included in the "A-S-E" dissimilarity index 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 Dissimilarity among B.A. Recipients in the United States, 1965-1994.

0.40 -

0.35 - Total Dissimilarity Index

0.30

= 0.25 Education-Business-

a 0.20 0

~S 0.15

Arts-Sciences- 0.10 Engineering

0.05

0-00 I I i1- lI I I I I I

65 70 75 80 85 90 95

Note: Based on BA degrees conferred by U.S. colleges and universities reported in the HEGIS/IPEDS surveys compiled in the CASPAR data archive. "All BAs awarded" is used as the denominator in the computation of field shares used to calculate the total index of dissimilarity, as well as the components attributable to the arts-sciences-engineering subset of fields and the education-business-other subset of fields.

The All-Fields dissimilarity index shows a pronounced decline, from a high of nearly 40% in the 1965-66 academic year to about 19% in 1994-95 (Figure 1). In aggregate, the largest decline in the overall index of dissimilarity took place during the 1970s, with a drop of slightly more than 10 percentage points occurring between 1973 and 1983. Focusing on the 1980s,Jacobs (1995) found very similar results for B.A. recipients, as well as advanced degree recipients. While that analysis also suggested that the narrowing of the gender gap came to a halt in the mid-1980s, it did not distinguish the role of the professional fields from that of the arts and sciences in this dynamic.

The relationship between the piece of the dissimilarity index attributable to business-education-other and that attributable to arts-sciences-engineering changes over

time, as a result of both changes in the degree of gender segregation within the component fields and the relative balance between these two loosely defined sub- sectors of higher education. The dissimilarity index declined until 1975 for both components. In business-education it con- tinued its downward course after 1975 (until 1985), but in arts-sciences-engineering it took an upward turn.

Extremely large movements of women away from the field of education and the associated migration of women into business programs account for much of the reduction in the dissimilarity index. In the mid-1 960s, gender segregation outside arts- sciences-engineering accounted for nearly two-thirds of the total value of the dissimilarity index. In this period, the undergraduate business major was more segre-


gated than some fields within the physical sciences, with women receiving only about 10% of business degrees. By the mid- 1980s, women had entered undergraduate business programs in record numbers. Because of greatly expanded representation in business and greatly reduced participation in the field of education, the portion of the dissimilarity index attributable to fields outside the A-S-E composite, primarily these two fields, fell to a little more than 1/3.

The movements of the more restricted A-S-E index follow a very different trajec- tory, particularly after 1975. This index of dissimilarity increases through the early 1980s, before leveling off and declining moderately in the early 1990s.

The "storyline" of this paper is what did not happen to the dissimiliarity index within the arts-science-engineering fields after 1975. During a period when other measures of economic outcomes and opportunities for men and women were con- verging, the A-S-E dissimilarity index failed to move toward greater parity. The remainder of this paper uses new data that include standardized test scores for individual students to explain more fully than has been possible heretofore the choices of majors by men and women within the arts-science- engineering fields.

The College and Beyond Database

The data for this inquiry are taken from the College and Beyond database assembled by the Andrew W. Mellon Foundation in cooperation with 34 colleges and universities. The part of the database used here consists of detailed records of entrance test scores and majors subsequently chosen by undergraduates at twelve academically selective colleges and universities: three universities (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 matriculants in the 1951, 1976, and 1989 entering cohorts, and we restrict our analysis to those who gradu-

ated (mostly in the classes of 1955, 1980, and 1993). The advantage of these data for a micro-level inquiry on choice of major is that we observe the full set of students at each of the twelve institutions.3

There are obviously many ways in which areas of study can be classified and aggre- gated. In keeping with normal practice, we have chosen to group all of the majors commonly considered to comprise the Humanities (including classics, English, foreign languages and literatures, history, philosophy, and religion); but we break with normal 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 patterns may reflect, at least in part, differences among these fields in math-intensity. In the remainder of this paper, we work with eight fields of study, some broader than others: the humanities; economics; psychology; politics and other social sciences; biology and other life sciences; mathematics and physical sciences; engineering; and a small number of students in "other" fields (principally ar- chitecture, communications, education, and specially constructed majors).4

3While we present some descriptive data for the 1951 entering cohort, we restrict the analytical part of this study to the 1976 and 1989 cohorts. Although analysis of the choices of the 1951 cohort might be particularly illuminating, the analysis is restricted to the later two cohorts for two reasons. First, standardized testing was far from universal in this period, leading to a large share of missing cases in this cohort that are unlikely to be randomly distributed. Second, the interpretation of shifts in parameters between the 1951 cohort and later cohorts would be complicated by the fact that several of these colleges and universities shifted from single-sex to coed status between 1951 and 1976.

4While the humanities present the broadest aggregate of the fields listed, disaggregating this category does not add to the substance of the analysis. History and English are the largest subfields, and the gender distributions for these fields are included in Table 1. A large number of small fields, including philosophy, religion, specific foreign languages, and comparative


The large size and census-like character of the database, its relatively "fine" level of aggregation, and the strong similarities in admissions standards and curricula among the 12 colleges and universities permit a closer, more intensive, examination of male/female differences in choice of majors than is possible in studies using samples of individuals from a larger and more di- verse array of institutions. To the extent that choice of major is a decision made largely in a given institutional context, it is conceptually appealing to have data in which we observe the full population of students at each of these institutions. Inter- actions between choice of major and type of institution may bias estimates of returns to specific fields of study by gender and may confound the interpretation of aggregate trends in the choice of major for men and women.

The other side of the proverbial coin is that these 12 institutions are not at all representative of American higher education. They represent a highly selective, entirely private, subset of institutions offer- ing what is generally regarded as a basic liberal arts curriculum. The three universities (and some of the colleges) do offer undergraduate programs in engineering, and these schools also offer a small number of "other" majors. Still, the overwhelming majority of all students in this database (roughly 90%) majored in traditional fields within the humanities, the social sciences, and the natural sciences. Unlike studies based on High School and Beyond, this study is limited by an incomplete representation of the population attending college from any high school cohort. From the perspec- tive of national norms, the professional fields (especially business and education) are grossly under-represented. We cannot, then, on the basis of these data, say any- thing about decisions to major in these

literature, make up the remainder of the fields in the humanities. The largest changes in gender composition within the humanities have occurred within these small and specialized fields.

professional fields. However, within the A- S-E fields, the national distributions for both men and women are surprisingly congruent with the distributions at our 12 schools (Appendix Table 1, top panel).5 This similarity suggests that the forces in- fluencing choice of major and the differences between men and women in making this choice extend well beyond this small set of selective private institutions.

Distinguishing among the 12 schools in this analysis by type reveals predictable differences in distributions of majors between the universities and the liberal arts colleges (Appendix Table 1, bottom panel). With significant numbers of degrees conferred in engineering at the three universities, the overall share of students choosing to major in the humanities is notably lower at the research universities than at the liberal arts colleges, and this difference is greater for men than for women since more men major in engineering. The general patterns are consistent with the notion that the coeducational liberal arts colleges and the women s colleges have similar "production functions," with an emphasis on subjects that are more labor-intensive than capital- intensive, while the research universities offer undergraduates more opportunities in fields that require complex infrastruc- tures and that attract high levels of spon- sored research ("big science").

In most fields, the choices of majors by the women at the women's colleges can be said to be intermediate between the choices made by women at the coeducational colleges and the choices made by men at the coed colleges. The share of women in the humanities is markedly lower in the women's colleges than in the coed colleges (41% versus 49%), while the shares of women in economics and in math-physical

5We compare our 1989 entering cohort with the 1993 B.A. recipients nationally because the largest. number of our graduates earned their degrees in four years. There are some "s'ystemic" differences b)etwteen the "national" and "local" distributions, though tliesc differences apply to both sexes.


Table 1. Changes over Time in Choice of Major, Graduates Only. (12 Institutions, College and Beyond Data)

1951 Cohort 1976 Cohort 1989 Cohort

Field of Study W% M% Diff W% M% Diff W% M% Diff

Humanities 54.8% 39.7% 15.1% 43.1% 31.9% 11.2% 42.4% 33.6% 8.8% English 14.9% 11.4% 3.6% 12.4% 8.1% 4.3% 12.5% 8.8% 3.6% History 11.3% 13.2% -1.9% 6.8% 10.1% -3.3% 8.9% 10.6% -1.7% Other Humanities 28.6% 15.1% 13.4% 23.9% 13.6% 10.2% 21.0% 14.2% 6.8%

Economics 2.4% 10.1% -7.7% 7.5% 12.4% -4.9% 5.4% 9.9% -4.5% Psychology 5.2% 4.4% 0.8% 7.9% 5.2% 2.7% 9.0% 4.1% 4.9% Politics & 0th Soc Sci 18.0% 11.9% 6.1% 15.8% 13.6% 2.2% 19.6% 18.9% 0.6%

Tot Soc Sci 25.7% 26.4% -0.8% 31.2% 31.1% 0.1% 34.0% 32.9% 1.1% Life Sciences 3.3% 6.4% -3.0% 11.7% 11.8% -0.1% 11.9% 9.5% 2.4% Math/Phy Sci 4.5% 10.9% -6.4% 6.7% 9.7% -3.0% 5.7% 10.0% -4.3%

Tot Natural Sci 7.9% 17.3% -10.7% 18.4% 21.5% -3.1% 17.6% 19.5% -1.9% Engineering 0.3% 10.9% -10.7% 2.7% 11.4% -8.7% 3.6% 11.8% -8.2% Other/NEC 11.4% 5.7% 5.7% 4.6% 4.1% 0.5% 2.4% 2.2% 0.3%

TOTAL 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% Index of Dissimilarity:a

Broad 21.5% 11.8% 10.1% Disaggregated 27.7% 16.7% 17.0%

aThe broad index of dissimilarity treats the social sciences and natural science subtotals as single entries; the disaggregated index treats the components of these disciplines as individual fields (for example, for the social sciences, economics, psychology and politics/other social sciences).

science in the women's colleges are higher than the corresponding shares in the coed colleges. However, because selection bias may be at work,judgments about the mean- ing of this pattern should be suspended, as the women who choose to attend the women's colleges may be more inclined to major in scientific fields than their peers at coeducational institutions.

The Dissimilarity Index Again: Trends within the 12 Institutions

The dissimilarity index provides a conve- nient way to assess both the degree to which male-female differences in choice of major have narrowed over the past three decades or so within these 12 institutions and the sizes of the gender gaps that remain. (For present purposes, we aggregate the data 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 index of dissimilarity, which treats each of the eight fields listed above as a separate field, falls from 27.7 for the 1951 entering cohort to 16.7 for the 1976 entering cohort. Driving this transformation was the migration of women out of the humanities and into fields like economics and the life sciences, with the share of women choosing economics rising from less than 2.5% to 7.5% and the share choosing the life sciences growing from 3.3% to 11.7% (Table 1).

More surprising, perhaps, is the lack of further change in the index between 1976 and 1989-the dissimilarity index is at 17.0 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


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 by the SAT, some differences in patterns of field choice would be expected. Replicating a pattern that has been well established based on national data, the average math SAT score for women at these 12 schools lags behind the average math SAT score for men by about 50 points in both the 1976 and 1989 cohorts; average verbal SAT scores for women and men were essentially equal in both the cohorts.6

Yet, the "gender gap" in math SATs does not translate into a proportionate difference in representation in each of the sciences, including the quantitative social sciences. To illustrate the need to consider the component fields of the sciences separately, Figure 2 shows the distribution of men and women with very high math scores by field. In this bar chart, it is plain that women with high SAT scores are much more likely than men to choose to major in the life sciences and the humanities rather than engineering, math, or the physical sciences. If the paucity of women with very high math achievement were the sole explanation for differences in representation in the sciences and quantitative fields, then one would expect these distributions to be identical for men and women. Relatively high math SAT scores provide men with an advantage relative to women in fields that require substantial quantitative skills. Mul- tivariate analysis, based on test scores and choice of major by individuals, provides an

6At the national level, the gap in SAT math scores between men and women was about the same as the gap within these institutions: the average SAT math score was 500 for men versus 454 for women (ETS data for college-bound seniors in 1989). The combination of higher math scores for men than for women and roughly equal verbal scores means that combined 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 narrowing in the 1990s, as the 1994 gap, nationwide, was 41 points, with average scores of 501 and 460 for men and women, respectively.

effective means of describing how much of the gender gap is associated with differences in SAT scores.

Other Studies and Approaches

Differences between men and women in choices of undergraduate major have been the subject of research by numerous soci- ologists, psychologists, and economists over the past several decades. In their 1997 survey article, Leslie and Oaxaca identified over 120 theoretical and empirical analyses of the under-representation of women and minorities in the science and engineering disciplines. Science policy analysts have often expressed the concern that the under-representation of women in these disciplines limits the pool of talent entering the science and engineering work force. Be- yond charting aggregates and flows, a number of researchers have attempted to iden- tify the behavioral factors explaining why this choice process leads to such different observed outcomes for men and women.

Theoretical explanations for the differ- ing outcomes tend to emphasize either differences in skills or differences in preferences and environmental determinants. Men and women (or boys and girls) with the same measured skill sets may differ dramatically in their preferences for different types of occupations or courses of study. Such differences may reflect biological forces or cultural factors such as the sex- role socialization forces cited by Eccles and Hoffman (1984). According to the sex-role socialization hypothesis, sex-role patterns and norms conveyed by schools and fami- lies in childhood affect later investments in education and training. Women may be encouraged to pursue fields and studies emphasizing nurturing while men are encouraged in domains emphasizing quantitative reasoning. While identification of such sex-role factors is inherently difficult, Corcoran and Courant (1985) suggested that variation in factors affecting sex-role socialization within gender groups (for example, variation in family structure) provides one means of distinguishing the variation in outcomes associated with sex-role


Figure 2. Distribution of Men and Women with Math SAT Scores Greater Than 750 by Choice of Major.

30

a 25 Li Men A

E Women c8 20 0 i l0 .

ut c s h OSHM- n

20

Huaiis Eooic scooy Ohr oil BoLf Mt niern 15~~~~~~~~~~~Scecs Sine hsc

norms from differences in outcomes reflecting the presence of gender-specific barriers in employment or wages.

Expected labor force commitment may well affect the economic rewards to any field of study, as suggested by Polachek (1978). Men and women may differ in their intended labor force commitment, leading to markedly different optimal investments in human capital as captured by the choice of major variable. For this reason, estimates ascribing gender differences in occupational choice to employer discrimina- tion may be overstated. As such, women with intermittent expected labor force participation 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 determinant of the observed differences in choice of major Jacobs 1995).

Experiences at the undergraduate level, including mentoring and peer influences, may have differential effects for men and women through the relative "costs" of any

course of study. Several researchers, including Solnick (1995) and Bailey and Rask (1996), have focused on the role of the undergraduate institution and the collegiate environment in determining the choice of major by men 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 all- female schools were appreciably more likely to switch to fields traditionally dominated by men than were women at coeducational institutions. Since these results were based on transitions during college, Solnick ar- gued that the cultural and academic envi- ronments associated with all-women's schools facilitate women's entry into the sciences.

The identification of specific educational and instructional factors such as introduc- tory class size and faculty mentorship dif- ferentially affecting choice of major by men and women has received considerable at-


tention because college administrators are assumed to exercise some control over these variables. Canes and Rosen (1995) investi- gated the extent to which increases in the share of female faculty members in a given department-providing, in principle, more role models for female students-corre- sponded with growth in the representation of female students in that department. They found no evidence of such a link.7

While there is ample evidence that factors other than achievement contribute to the observed gender differences in choice of major at the undergraduate level, persistent differences between men and women in high school courses and measured precollegiate achievement provide at least a partial explanation for the gap. Skill- based explanations for differences in 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.8 A number of studies, including Ware and Lee (1988) and Ware, Steckler, and Leserman (1985), have included measures of achievement, particularly in quantitative fields, as explanatory variables in discrete choice models of the decision to major in a scientific field. However, such studies do not consider whether these measures vary in the magnitude of their effects across the component fields of the sciences or across a more differentiated set of fields.

The observation that women have dem- onstrated sizable gains in representation in some fields, notably the life sciences, while their relative shares have stagnated in some quantitative sciences, leads to a reconsid- eration of the role of precollegiate skills in

7Bailey and Rask (1996) tested the same hypothesis with micro data from a different private liberal arts college and found modest support for the role model hypothesis.

8Since long-term expectations may be formed well in advance of college enrollment, ascribing a causal interpretation to differences associated with achievement levels is unwarranted if these levels of precollegiate skill are endogenously related to long-

sorting men and women into particular fields of study. In particular, are the forces leading to different choices of major similar enough across fields that a model posit- ing a dichotomous choice between science and non-science fields adequately charac- terizes observed variation? Our answer to this question is a clear "no."

Analytic Approach and Estimation Strategy

In this paper, our focus is on the extent to which academic achievement, measured by entering SAT scores, explains gender differences in choice of major. While this parsimonious specification of explanatory variables is plainly limited, our data afford considerable power in explaining the be- havior of men and women across a range of fields of study. We quantify the degree to which the observed gender differences in choice of major can be attributed to precollegiate academic achievement. Each individual's SAT scores are proxies for skills brought to college, with the verbal and math scores representing separate (though often correlated) types of skills.

Nevertheless, the most interesting behavioral phenomenon is the piece that re- mains unexplained. This residual element reflects some combination of different opportunities in the labor market for men and women and differences in long-term occupational aspirations associated with gender.

The multinomial logit formulation is well suited to the measurement of factors affecting qualitative choices such as choice of major.9 In theory, choice of major for each individual represents a deterministic out-

term goals. For example, differences in observed mathematics achievement between men and women at the precollegiate level may reflect differential levels of parental encouragement of sons and daughters in these courses at the secondary level.

9As discussed below, while the multinomial logit formulation offers the most tractable estimation of a qualitative choice problem, the functional form also leads to implicit assumptions associated with the pa-


come that maximizes utility. As researchers, we do not know enough about all the factors affecting the choice-or their actual values-to replicate this calculation without error. Nonetheless, observations of individual achievement measures and choices of major provide sufficient information to make probabilistic statements. In particular, the probability that individual i chooses field j over the alternative k is represented by

(1) PmF= Prob(VMF(Z, f3 ) + e..> vM'F(Z., 3M[) + ek)

=Prob (e..- e.k > VMF(Z., fMF)

- VMF(Z., MF)r),

where individual characteristics in the vec- tor Z. represent the observable determinants of the utility function (V), the weights on these explanatory factors are estimated as Pi, and the unknown characteristics are captured in e Y. In this framework, the gender-specific superscripts (M,F) recog- nize that men and women may differ in their underlying preference functions and in the benefits (or costs) associated with a particular characteristic in a particular field. With an assumption about the distribution (Type I Extreme Value) of the residuals (ei ), we can estimate the parameters of the model. In particular, if the epsilon terms are independently and identically distributed, the estimation procedure follows the multinomial logit formulation and the choice probabilities take the form

VMF .........=................ ....... .. .. . .. .. .... ....... .. . .

id ~~~V M,F jE Jnexp id

Answers to the interesting question of how we would expect individuals to sort themselves among 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) restriction is that the "independence of irrelevant alternatives" restriction is maintained such that the elimination of one option (for example, economics) does not change the relative probabilities of the choices of other options (for example, humanities or math).

scores, follow from the estimation of the coefficients in this model. First, if the coefficients on math and verbal SATs are nonzero, then the model provides support for the hypothesis that more than one skill set matters in the determination of choice of field and related investments in building human capital. Second, the relative magnitude of the coefficients indicates the degree to which selection of particular majors depends on math and verbal proficiency on entry to college.

The objective of this analysis is to address the counterfactual question of how the distribution of the undergraduate majors of women would be expected to change if, holding preferences constant, women had the same distribution of SAT scores (X values) as men (Bound, Schoenbaum, and Waidman 1996).1o This type of calculation allows us to estimate the effects on choice of major of differences between men and women in both (a) their precollegiate academic preparation, measured here by the SAT scores that they bring with them to school, which are the X values, and (b) the ways in which women and men separately "convert" their respective SAT scores into choices of majors. These "residuals" are in turn the product of other forces and variables listed previously, including labor market opportunities for women and men, wealth, other anticipated family circum- stances (which affect both the desire for labor force participation and the need to earn sizable amounts of income), social and parental expectations, and attitudes and interests stimulated by faculty and fel- low students in college.

Empirically, the differences between men and women in observed field distributions can be seen in the context of the familiar

"0The alternative is to consider how the distribution of majors for men would differ if, given their preferences, they had the same distribution of SAT scores as women. Differences in the magnitude of the two component pieces, depending on whether men or women are used as the reference group, reflect the underlying index number problem with this type of analysis.


Oaxaca decomposition

(2) pPFXF _ pPMXM = (pFXF PpFXM)

+ (PPFXM _PMXM)

where pMXM is the mean of the predicted probability of choosing field j using the coefficients and values of the explanatory variables for men; pPFXF is the mean of the predicted probability of choosing field j using the coefficients and values of the explanatory variables for women; and pPFXM is the mean of the predicted probability of choosing field j using the coefficients for women and values of the explanatory variables for men.

The first term to the right of the equal sign can be interpreted as the share of the observed difference associated with differences in SAT scores, while the second term shows the combined effects of the differences associated with the coefficients not attributable to differences in measured achievement. In the case of a logit model or other nonlinear function, the magnitude of the effect of a change in any single exogenous factor is a function of the level of the other variables used in this evalua- tion, and such nonlinear choice models do not allow for the decomposition of the effects of particular exogenous variables.

Empirical Results

We begin by estimating separate multinomial logit models for men and women, with verbal and math SAT scores as the explanatory variables. Specifying the test score variables as categorical variables limits the functional form assumptions within the logit framework (though it also permits some unconventional "hills and valleys" in the implied distribution of majors by test scores). For the 1989 entering cohorts, estimated coefficients and the associated standard errors are shown in Tables 2a and 2b." The significance of the parameter estimates (excepting the case of the "other"

"By aggregating across schools, we assume that individuals jointly select institution and major. For

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 strong and positive effect on the likelihood of majoring in the humanities (the reference group) relative to the probabilities of choosing all other majors. Stated the other way around, the nearly unbroken run of negative coefficients along- side the verbal SAT variables tells us that relative to the group of students with verbal SAT scores below 550, ceteris paribus, students with these relatively high verbal scores 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 institutions is effectively choosing not to major in engineering. Aggregation also disregards the question of whether the two-way selection 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 interpreted as representing the net effect of the institutional selection process and the "true" test score effect on the choice of major. An alternative theoretical framework would posit that institutional choice and choice of major are sequen- tial. That type of analysis would be institution-specific.

12As one early reader noted, it may be surprising that test scores seem to be such significant explanatory variables determining choice of major and the differences between men and women in choice of major. Two explanations merit consideration. First, test scores are likely to be correlated with a range of factors, including both other measures of academic achievement (such as high school rigor or high school' GPA) and socioeconomic variables. To this end, a causal interpretation of the parameter estimates as "explaining" differences in choice of major is inap- propriate, particularly if the magnitude of the corre- lation with omitted variables differs for men and women. A second explanation, warranting further investigation, is that differences in achievement measured by test scores are more powerful determinants of choice of major in a construct where the achievement thresholds for each major are reasonably homo- geneous from institution to institution.


Table 2a. Multinomial Logit Estimates of Choice of Major for Women. (1989 Cohort, 12 Institutions)

Math Scores Verbal Scores

Choice of Major Explanatory Variable Coefficient Std. Error Coefficient Std. Error

Humanities Reference Group Economics Constant -1.9159 (0.2107)

SAT> 750 2.3380 (0.3770) -3.0289 (0.6122) SAT 700-750 1.8124 (0.2746) -2.5770 (0.3191) SAT 650-700 1.2625 (0.2552) -1.7221 (0.2192) SAT 600-650 1.0963 (0.2544) -1.7222 (0.2400) SAT 550-600 -0.0250 (0.3055) -0.6350 (0.1934)

Psychology Constant -1.3916 (0.1588) SAT> 750 0.8581 (0.3442) -1.2115 (0.3433) SAT 700-750 0.9206 (0.2074) -1.3127 (0-2331) SAT 650-700 0.2793 (0.1980) -1.0657 (0.1924) SAT 600-650 0.5001 (0.1891) -0.5789 (0.1808) SAT 550-600 0.0976 (0.2005) -0.0057 (0.1655)

Politics a Other Social Sciences Constant -0.4786 (0.1149) SAT> 750 0.5440 (0.2452) -0.9104 (0.2518) SAT 700-750 0.2106 (0.1571) -0.9321 (0.1689) SAT 650-700 0.0147 (0.1409) -0.6279 (0.1398) SAT 600-650 0.1523 (0.1379) -0.2932 (0.1352) SAT 550-600 -0.1180 (0.1436) -0.0617 (0.1315)

Biology-Life Sciences Constant -1.8376 (0.1873) SAT> 750 2.2891 (0.2809) -1.0460 (0.2766) SAT 700-750 1.7375 (0.2201) -1.0620 (0.1949) SAT 650-700 1.3758 (0.2085) -0.8131 (0.1689) SAT 600-650 0.9013 (0.2169) -0.6795 (0.1730) SAT 550-600 0.6547 (0.2230) -0.2224 (0.1658)

Math-Physical Sciences Constant -2.9104 (0.2996) SAT> 750 3.0099 (0.3805) -0.9453 (0.3641) SAT 700-750 2.1529 (0.3327) -1.0464 (0.2707) SAT 650-700 1.4245 (0.3281) -0.7585 (0.2363) SAT 600-650 1.1945 (0.3335) -0.5320 (0.2400) SAT 550-600 0.7917 (0.3495) -0.1984 (0.2363)

Engineering Constant -3.6820 (0.4312) SAT> 750 4.2719 (0.4923) -1.5511 (0.4618) SAT 700-750 3.0523 (0.4576) -1.8584 (0.3676) SAT 650-700 1.8321 (0.4617) -1.3056 (0.3096) SAT 600-650 0.9739 (0.5004) -0.5633 (0.3009) SAT 550-600 0.2542 (0.5645) -0.0514 (0.2949)

Other Fields (NEC) Constant -2.5102 (0.2728) SAT> 750 0.4750 (0.6720) -1.7445 (0.7548) SAT 700-750 0.3794 (0.4095) -0.9632 (0.3588) SAT 650-700 0.6545 (0.3413) -1.0056 (0.3129) SAT 600-650 0.3272 (0.3560) -0.9063 (0.3152) SAT 550-600 0.3922 (0.3458) -0.8437 (0.3194) No. Obs. 4290 X2(70) 533.14 Log Likelihood -6,962.2844

math score constant would increase the probability of majoring in biology relative to the probability of majoring in economics for men and women. Similarly, the higher

the student's math SAT, the greater the probability that the student will major in fields other than the humanities. But the magnitude of the effect of math SATs dif-


Table 2b. Multinomial Logit Estimates of Choice of Major for Men. (1989 Cohort, 12 Institutions)




SAT> 750 1.9498 (0.3564) -1.5184 (0.3553) SAT 700-750 1.2388 (0.3329) -1.6997 (0.2456) SAT 650-700 0.9797 (0.3279) -1.4903 (0.2186) SAT 600-650 0.5725 (0.3439) -0.7848 (0.2035) SAT 550-600 0.2407 (0.3821) -0.6173 (0.2070)

Psychology Constant -1.6224 (0.3473) SAT> 750 0.5688 (0.4708) -1.3777 (0.5648) SAT 700-750 0.1497 (0.4165) -1.3382 (0.3534) SAT 650-700 0.2710 (0.3994) -0.8578 (0.2832) SAT 600-650 0.3733 (0.4048) -0.7690 (0.2827) SAT 550-600 0.4199 (0.4285) -1.0075 (0.3111)

Politics & Other Social Sciences Constant -0.1919 (0.1945) SAT> 750 0.2581 (0.2543) -1.0043 (0.3070) SAT 700-750 -0.0627 (0.2214) -0.7779 (0.1969) SAT 650-700 -0.1009 (0.2171) -0.5172 (0.1735) SAT 600-650 0.0528 (0.2222) -0.2087 (0.1702) SAT 550-600 -0.2014 (0.2482) -0.2095 (0.1735)




Other Fields (NEC) Constant -2.5706 (0.5259) SAT> 750 1.3844 (0.6395) -1.1917 (0.6796) SAT 700-750 0.6665 (0.6010) -1.5164 (0.4897) SAT 650-700 0.6173 (0.5850) -0.9247 (0.3973) SAT 600-650 -0.2292 (0.6698) -0.6124 (0.3868) SAT 550-600 0.5061 (0.6366) -0.8416 (0.4283) No. Obs. 3,373 x2(70) 579.36 Log Likelihood -5,881.4223

fers appreciably across the fields."3 For both men and women, increases in the

13More generally, tests of the hypothesis that all of the coefficients are equal to zero for each major and

math SAT increase the probability of majoring in engineering or math-physical sci-

tests of the equality of coefficients across majors are rejected nearly across the board (there are 28 pairwise


ences relative to any other field, holding the verbal score constant. Note that in these fields, relatively large changes in the choice probabilities occur near the top of the scale. This result reinforces the intu- itively plausible proposition that moving from a 700 to a 750 math score is likely to have a much larger impact on the probability that a student will major in a field like theoretical physics than will a shift from 600 to 650.

Increases in math SAT scores have a uni- formly positive effect on the probability of majoring in economics and the life sciences relative to the probability of majoring in the humanities, as indicated by the positive parameter estimates. However, changes in math scores do not have as large an effect on the expected probabilities of choosing these fields as they do in the physical sciences and engineering. Little can be said about the variation in the probability of choosing to major in psychology associated with test scores, as the parameter estimates are small in magnitude for women (though statistically significant) and generally indistinguishable from zero for men.

Calculating the marginal probabilities- the predicted change in the probability associated with an incremental change in an explanatory variable-indicates the relative sensitivity of each field choice to changes in math and verbal scores. A key facet of the multinomial estimation is 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 relatively high SAT scores (particularly in the quantitative dimension) increase the probabilities of majoring in both the life sciences and the physical sciences for women, while for men there is greater separation between the life sciences and the physical sciences in the upper dimensions of the SAT range.

fact that the magnitude of the marginal effects-illustrated as the slope of a function relating the explanatory variable to the expected probability of majoring in any given field-depends on the level of the individual's scores. Since the models use categorical explanatory variables, a discrete parallel is the measurement of the expected change in the probability associated with a movement between achievement score categories. Figures 3 and 4 trace, for women and men, respectively, the expected probabilities associated with each field choice and combination of math and verbal achievement categories. A flat horizontal plane would be indicative of a field in which the choice probabilities did not vary with test scores, while a plane with a constant slope would indicate a field in which the choice probability increased or decreased steadily with achievement scores.

The "shapes" of the predicted functions are, with some 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 maxi- mum in the high verbal, low math corner. As expected, the probability of majoring in engineering and math-physical science rises at an increasing rate with high math SAT scores. For both men and women, the highest probability of choosing math-physical sciences is associated with top category math and verbal scores, while the probability of choosing engineering bears less rela- tion to the verbal score. Comparing the pictures for men and women, note that the rise in probabilities associated with moving from the penultimate to the highest category in math SATs is noticeably larger for men than for women. The predicted probability of majoring in engineering for women with modest math scores (between 600 and 650) is close to zero (the picture resembles a throw-rug) , while the predicted probability for men with scores in this range is decidedly higher.

The graphs for economics and the life sciences suggest a topography with more hills and valleys and somewhat greater differences by gender than are found for the


other fields. For men, two groups are most likely to major in economics: those with moderately high SAT scores (650 to 750) and low verbal scores and those with both high verbal and high quantitative scores. For women, those with the highest quantitative scores (greater than 750) and lowest verbal scores (less than 550) are most likely to choose to major in economics, while those with high scores on both dimensions of the SAT are less likely to 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 test score categories until they reach a "hilltop" with verbal scores in the 650 to 700 range. For women, the plane continues to slope modestly upward as both math and verbal scores rise, reaching its peak in the category reflecting the highest combined scores. These figures provide fascinating insights into the sometimes different ways in which women and men "convert" math and verbal aptitudes into choices of majors.

We now employ the decomposition strategy outlined in the previous section in order to evaluate the importance of SAT scores in explaining differences in the majors chosen by men and women. The results for the 1989 cohort, using women as the reference group, are presented in Table 3. The table shows three sets of differences: the total observed differences in the actual distributions;14 the differences attributable to differences between women and men in SAT scores (or "attributes" or X values); and the remaining differences, attributable to differences in "preferences" and other un- specified variables. These differences are computed from the observed distributions of men and women and the distribution predicted using the SAT scores observed for men and the estimated parameters for women. (We made analogous calculations

'4From the first order conditions of the log likelihood function, it can be shown that the mean of the predicted values is equal to the population shares.

Table 3. Decomposition Analysis of Model of Undergraduate Choice of Major: College Graduates Only,

1989 Cohort, Nonlinear Model.

(A) (B) (C) Unexplained

Difference by SA T Observed Due to Scores

Difference, SAT Scores [Exp %M % W- [Actual %W -Actual

Field %M -Exp %M] %M]

Humanities 8.6% 5.1% 3.6% Economics -4.5% -0.7% -3.8% Psychology 5.0% 0.4% 4.6% Politics & Other Social Sciences 0.8% 1.8% -1.0%

Biology-Life Sciences 2.4% -2.2% 4.6%

Math-Physical Sciences -4.3% -2.0% -2.3%

Engineering -8.4% -2.7% -5.7% Other 0.3% 0.3% 0.0% Index of Dissimilarity 17.2% 7.5% 12.8%

Notes: Calculations are based on a multinomial specification that includes 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 difference between the share of women majoring in a field and the share of men majoring in the field; column (B) is the difference between the actual share of women in a field and the predicted share of men, using the coefficients from the multinomial estimation for women (Table 2a) and the actual SAT scores of the men at the 12 universities; and column (C) is, the difference between the predicted distribution of men and the observed distribution for men.

using the SAT scores observed for women and the estimated parameters for men; the results are not qualitatively different.)

In the case of the humanities, we find that while the observed difference between the actual distributions for the 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 this gap can be attributed to SAT scores; the remaining 3.6 points indi-


Figure 3. Predicted Choice of Major: Profiles for Women, 1989 Cohort.

Humanities Biology - Life Sciences

0275 ....0.25 /'

0.10 010- 0.15 0.05~

77 725 VERBAL 575 725AL~

Econoicis Biolog - Phyical Sciences

_2 - 0,25

0.10 Eninern

0.05 ~~~~~~~~~~~~~~~~0.05

0-w syhoog Egieein

cate the degree to which women with the same math and verbal SATs as the men are more likely than the men to major in the humanities. Similarly, the much greater tendency for men than for women to major in math-physical sciences and in engineering also appears to be the joint product of variations in SAT scores (higher math scores for the men) and the relatively stronger preferences among men for these fields, though the relative magnitudes of these

effects are not at all equal. Test score differences account for about 45% of the total gender gap in the math-physical sciences fields and for about 32% in engineering. In economics and psychology, differences between women and men in SAT scores explain only a very small part of the gender gap-about 16% in economics and less than 8% in psychology. Instead, it is pronounced differences in "preferences" and other residual forces that generate the


Figure 4. Predicted Choice of Major: Profiles for Men, 1989 Cohort.

X.5 :0:.25 525

Humanities Biology - Life Sciences *0 0S 0.20

Economics Math - Physical Sciences

Pyh725 o yAL E e725 VERBAL

Humanities Biolgyifeer cience

large gender differences in representation in these fields. In biology-life sciences, the two forces pull in opposite directions. Dif- ferences in 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 relative to other fields are so much stronger than men's that the net effect is that women are modestly over-represented.

In summary, we find that differences in

SAT scores account for less than half of the total gender gap. If men had the same preferences as women but differed only in the distribution of SAT scores, the index of dissimilarity would drop to 7.5. As noted above, the qualitative conclusions reached are much the same whether we use women or men as the reference group. It should be emphasized that the share of the differential associated with test scores is appreciably lower using this taxonomy than it would


Table 4. Decomposition Analysis of Model of Undergraduate Choice of Major: College Graduates Only,

1976 Cohort, Nonlinear Model.

(A) (B) (C) Unexplained

Difference by SAT Observed Due to Scores

Difference, SAT Scores [Exp %M %W- [Actual %W -Actual

Field %M -Exp %M] %M]

Humanities 11.8% 5.2% 6.5% Economics -4.9% -1.0% -4.0% Psychology 2.7% 1.0% 1.8% Government & Other Social Sciences 2.0% 2.3% -0.3%

Biology-Life Sciences -0.2% -2.2% 2.0%

Math-Physical Sciences -2.8% -2.9% 0.1%

Engineering -8.9% -2.6% -6.2% Other 0.3% 0.2% 0.1%

Index of Dissimilarity 16.8% 8.7% 10.5%

Notes: See explanation in notes to Table 3 and estimation results in Appendix Tables 2a and 2b.

be if we used a simple dichotomy between science and non-science fields.'5

The same cross-sectional analysis applied to data for the 1976 entering cohort yields generally-but not exactly-similar results (Table 4). (Logit estimates appear in Ap- pendix Tables 2a and 2b.) For these individuals, differences in SAT scores explain a slightly higher fraction of the gender gap, particularly in the physical sciences, where the observed difference is more than accounted for by differences in test scores. In Table 5, we summarize the changes over this interval by comparing the decompositions for the 1976 and 1989 entering co-

'5The dichotomous distinction between non-science and science majors yields a total difference of 10.26 percentage points, with 6.7 percentage points attributable to differences in test scores and about 3.5 percentage points attributable to differences in the residual.

horts; changes in the overall size of the gender gap for each field are apportioned between a component attributable to changes in the relative distributions of SAT scores for men and for women and a component attributable to shifts in preferences.'6 In general, changes in the distribution of test scores for men and women had very little effect on gender gaps. Changes in the preferences or forces other than measured achievement were the driving force in widening gaps in psychology, the life sciences, and math-physical sciences. For example, while the gap between the share of women and the share of men choosing to major in biology-life sciences increased from a small over-representation of men in 1976 to an over-representation of 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 understanding these transformations, a key piece of the puzzle is to distinguish the changing "weights" used by men and women in matching achievement levels with fields of study from changes in overall levels of demand. For example, one might consider changes in the discipline of economics requiring greater levels of mathematics achievement as leading to changes over time that are 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 preferences and other residual factors. In each field j at time t,

G = p=F XF _ pPMtXMt=

(pRFtXFt - pPFtXMt) + (p5RXMt _.p5MtXMt) I i I I

The change in the gap can be written as

C - G. = [ (pI5FXFI pIFIXMt) (pI3F -XFtJ pF3Ft ,XM1 ,)]

+ [(p F1XM_ pPMtXMt)

- (p _Ft-)XMt-_

p Mt-IXMt-))] j I I

The first term to the right of the equal sign represents the portion of the change due to changes in the respective distributions of SAT scores, while the second term in square brackets captures the portion of the change attributable to changes in the difference in the relative magnitude of the beta coefficients.


Table 5. Decomposition of the Change in the Gender Gap in Field Preferences: 1976 to 1989 Entering Cohorts, 12 Institutions, College and Beyond Data.

Change in Gap Unex-

Change in plained Level of Sign of Direction Total Gap Due to by Change Gap in Gap for of Change, Change Change in in SAT

Field of Study 1976 1976 1976-89 in Gap SA 7'Scores Scores

Humanities 0.1176 W>M Narrow -0.0313 -0.0017 -0.0296 Economics -0.0493 W<M Narrow 0.0044 0.0022 0.0022 Psychology 0.0273 W>M Widen 0.0228 -0.0056 0.0284 Politics & Other Social Sciences 0.0197 W>M Narrow -0.0114 -0.0046 -0.0068 Biology-Life Sciences -0.0017 W<M Ch Sign 0.0262 0.0004 0.0258 Math-Physical Sciences -0.0278 W<M Widen -0.0154 0.0095 -0.0248 Engineering -0.0888 W<M Narrow 0.0050 -0.0005 0.0055 Other 0.0031 W>M Narrow -0.0002 0.0003 -0.0006

Note: Components of change are calculated from Tables 3 and 4. See text for details of calculation.

while factors such as the opening of the medical profession to women would be likely to have particularly strong effects on women's choices. A confounding force is the change in the overall demand for particular fields of study, which can serve to expand or contract the measured gaps without affecting the relative representation of men and women by field.

Because data on SAT scores for the year 1951 are widely available for only one of the schools examined in our study, we will not burden the text of the paper with regres- sions, tables, or figures based on those data. The admittedly incomplete evidence suggests, however, that large-scale shifts in preferences (especially by women) 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 entering cohort, women with the same SATs as men made radically different choices of majors; however, by the time of the 1976 cohort, preferences had converged to a considerable extent. We interpret this result as a reflection of the opening up over this period of many morejob opportunities for women in traditionally male fields, the greater commitment of women to extended labor force participation, and a general "loosening" of stereotypes as to what women and men should do with their lives. There

were dramatic contractions in the sizes of the preference components of the gender gaps in all of the predominantly male fields-and also in the humanities.

It is against this background that we should interpret what did not happen between the times of the 1976 and 1989 cohorts. As we saw in Table 5, the sizes of the preference components of the gender gaps not only failed to continue to shrink, they rose in psychology and the life sciences (with women expressing ever stronger positive preferences), and they rose in math/ physical sciences too (with women expressing much less interest than men, as com- pared with the situation in 1976). In short, the social and economic forces that pushed for a convergence in choices of major after the mid-1950s either spent themselves by the late 1970s or were subsequently over- taken by new forces. More recently, differences between men and women in choice of major appear to have become en- trenched, in spite of modest continuing convergence in SAT scores; gender-specific preferences now appear to be widening a number of gaps rather than narrowing them.

Conclusion and Open Questions

The decompositions of gender gaps re-


ported in this paper are entirely consistent 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 striking conclusion is that differences in SAT scores are nothing like the full story. In fact, they capture much less of the dynamics of change over the past 35 or 40 years than do the panoply of residual forces, including differences in preferences, labor market expectations, gender-specific effects of the college experience, and unmeasured aspects of academic preparation. To answer directly one of the questions posed at the start of the paper, we do not see continuing movement toward gender neutrality in the skills (or credentials) that students take from college.

One possible interpretation is that women and men typically attach different "weights" (or values) to the after-college opportunities associated with each area of study. Recognizing potential family respon- sibilities, women may prefer fields in which skills are unlikely to atrophy or become obsolete-for example, for members of the 1976 cohort, a keen understanding of Shakespeare may provide more opportunities in the 1990s than knowledge of the (nearly obsolete) COBOL computer lan- guage. Yet, this "depreciation effect" may be more important in theory than in practice. As women are observationally less and less likely to interrupt their labor force participation for prolonged periods of time to raise children, one would expect the gender gap in field choice to narrow ac- cordingly-and it has not.

A second type of explanation is that there continue to be impediments in the labor market to women receiving the 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 vary by gender. Following wage regression estimates from Brown and Corcoran (1997) that use data from the

National Longitudinal Survey of the Class of 1972, we see that women with a college degree receive a higher premium than men in the humanities, a lower return in biology, a slightly lower return in math and the physical sciences, and a much greater return in engineering. If women in the 1989 entering cohort observed these returns as "prices" before choosing majors, then the over-representation of 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 differences at the undergraduate level. In this regard, the considerable widening of opportunities for women 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 relationships would require detailed study of the links joining field choice, occupa- tion choice, labor force participation, and remuneration.

Other possible explanations for the hardy persistence of gender gaps include the presence 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 that SAT scores do a poor job of measuring higher order math skills. Given the same SAT scores, men may exceed women in the full specification of math skills on college entrance, perhaps because they tend to take more advanced math courses in high school than their female counterparts. If this hypothesis is 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 of quantitative skill. However, the data do not provide enough information to address the counterfactual question


of how an individual would have performed academically if he or she had chosen a different field of study. Since observed undergraduate performance measured by GPA is conditional on field choice and course selection, the observed relationship between undergraduate GPA and SAT scores will not provide an answer to the counterfactual question of how a student who did not major in a particular field would be expected to perform in that subject.

The choices men and women have made in elective fields of study at the undergraduate level have not steadily converged. Differences in academic preparation of women and men measured crudely by test scores explain some of the persistent differences in characteristic choices of major, whether we are attempting to account for changes over the past three decades or for the cross-sectional gap for the most recent cohort. But such differences in SAT scores are only part of the story, and a modest part at that. An array of residual forces-including differences in preferences, labor mar-

ket expectations, gender-specific effects of the college experience, and unmeasured aspects of academic preparation-account for the main part of today's gender gaps in choice of academic major.

It seems abundantly clear that the simple science-non-science dichotomy, however helpful it once was, is no longer useful as a taxonomy if the objective is to understand field choices by women and men. Quite plainly, there is a widening divide between the life sciences, on the one hand, and math-physical science-engineering, on the other hand, in their attractiveness to women. The magnitude and persistence of the disproportionately heavy representation of men in economics, as well as in engineering, mathematics, and the physical sciences, pose the puzzling question of why these fields are so different from other fields also classified broadly as "science." Perhaps the findings reported in this paper will stimulate new, even more determined, efforts to unravel the forces at work-a hard task in part because of the need to work across disciplinary boundaries.


co N - - o t oo ro v:

F t .; ~~~~~~~~~~~~~~~~00 00 sz0 AOC

a14 co-

W~~~~~~ 00 Sc C O K n (= O:. r -

- - Co -A4

06 C; C; - -

8 g 0 r

~~~~~z b: b: bz b~ bs b0? b W U~~~~~~~~~~~~~~~n n 00 Y 0

C~~~~~~~~~~~~~~~~~~~~~~~~~~I o V -0 C; or o 0o t r 6 o

(I, - C;ttsto - O NOv

m =S .^ ~ - .. e egg gg e

@ 0 -0 0 t 00 GA ~ ~ ~~~~n 00 in r -

o . , o6 * bs z O 0 6 S i\? o Co - Co - OY

X~~~~~~~~~~~0 00 ,c 0 C1 00 X1

J @ S i-e, @ @ @ g @ g g eg xs~~~S !, b w ,? I .,o ce2,r2, eo ̂

F~~~~ o O i ,o - , - ? G? 0,,toG

v i r g i} 0 00 00 t~k 00 <;s 00 Ci 006 t o

CIA ~ ~ A c

n~~~~~~~~~~~~~~~~~~ cli c ti4,

;, - - ., g

~~~~ 0 - Q~~~~~064 cl ~ 1 6 - C~~~~~~~~~~~~~~~~~~~~~C

Cu -~~~~~~~~~~~~~~~~~~~~C

-0 - , o in C)

@ ~~~~~~~~~~~n !0 uS cl 0e t- C; eee

sb b i, <2 t o , > :, s o o o s n o = v 'r-

-t .w i . - c o~~~~~~~~~~( CO CQ CO

C'~~ ~~ CO z - GAsN O2is .Oc

Cos L ) Y C;

,~~ .1ge gg We:;

-i I t A Cci?Q0ZQ0 ? s t U i~~. , ? , 0xOm ? -OOe4 ,!,9 ,2,O

?~~~~~ 0 - t 'D


APPENDIX TABLE 2A

Multinomial Logit Estimates of Choice of Major for Women, 1976 Cohort, 12 Institutions




SAT> 750 1.2917 (0.4340) -2.8186 (0.7388) SAT 700-750 1.4866 (0-2391) -1.3174 (0.2647) SAT 650-700 0.8805 (0.2262) -0.9264 (0.2158) SAT 600-650 0.9573 (0.2035) -0.5192 (0.1983) SAT 550-600 -0.0424 (0.2385) -0.1946 (0.1991)

Psychology Constant -1.3174 (0.1500) SAT> 750 -1.4079 (1.0269) -1.1709 (0.4048) SAT 700-750 0.5437 (0.2429) -1.5920 (0.2916) SAT 650-700 0.3728 (0.2081) -0.8491 (0.2035) SAT 600-650 0.2681 (0.1903) -0.4385 (0.1806) SAT 550-600 0.0879 (0.1906) -0.6276 (0.2011)

Politics & Other Social Sciences Constant -0.6176 (0.1159) SAT> 750 -0.1716 (0.3747) -1.4110 (0.3753) SAT 700-750 -0.1394 (0.1953) -0.7291 (0.1893) SAT 650-700 -0.0410 (0.1571) -0.4015 (0.1544) SAT 600-650 -0.0916 (0.1429) -0.3424 (0.1468) SAT 550-600 -0.1123 (0.1396) -0.2054 (0.1517)


Math-Physical Sciences Constant -2.4343 (0.2293) SAT> 750 2.4570 (0.3705) -1.5753 (0.4247) SAT 700-750 2.2397 (0.2680) -1.0976 (0.2646) SAT 650-700 1.7028 (0.2567) -0.5663 (0.2237) SAT 600-650 1.0233 (0.2608) -0.5675 (0.2270) SAT 550-600 -0.0593 (0.3153) -0.4093 (0.2380)


Other Fields (NEC) Constant -2.0654 (0.2031) SAT> 750 0.3228 (0.5213) -0.2329 (0.4144) SAT 700-750 0.3536 (0.3012) -0.4738 (0.3013) SAT 650-700 0.0399 (0.2753) -0.4006 (0.2621) SAT 600-650 0.2437 (0.2406) -0.2791 (0.2477) SAT 550-600 -0.0608 (0.2512) -0.4484 (0.2750) No. Obs. 3,742 X2(7O) 478.37 Log Likelihood -6,199.9045


APPENDIX TABLE 2B

Multinomial Logit Estimates of Choice of Major for Men, 1976 Cohort, 12 Institutions




SAT> 750 3.1139 (0.3088) -2.6888 (0.4380) SAT 700-750 2.3100 (0.2847) -1.9104 (0.2218) SAT 650-700 1.8069 (0.2850) -1.4658 (0.1891) SAT 600-650 1.5430 (0.2814) -1.3178 (0.1961) SAT 550-600 0.9604 (0.3077) -0.8984 (0.1938)

Psychology Constant -1.0490 (0.1958) SAT > 750 0.5327 (0.3929) -3.2838 (1.0319) SAT 700-750 0.7194 (0.2820) -1.7121 (0.3205) SAT 650-700 0.0205 (0.2979) -1.5078 (0.2738) SAT 600-650 0.0044 (0.2793) -0.9599 (0.2543) SAT 550-600 -0.3277 (0.3171) -1.0449 (0.2739)

Politics & Other Social Sciences Constant -0.6701 (0.1596) SAT> 750 0.6775 (0.2511) -1.3547 (0.3959) SAT 700-750 0.2605 (0.2092) -0.5890 (0.2051) SAT 650-700 0.3058 (0.2020) -0.7094 (0.1909) SAT 600-650 0.0588 (0.2013) -0.3019 (0.1856) SAT 550-600 -0.0002 (0-2160) -0.0559 (0.1839)




Other Fields (NEC) Constant -1.8945 (0.2718) SAT> 750 1.3688 (0.4122) -1.3552 (0.5230) SAT 700-750 0.9792 (0.3562) -1.1728 (0.3167) SAT 650-700 0.7507 (0.3532) -1.1115 (0.2838) SAT 600-650 0.6043 (0.3447) -1.0985 (0.3004) SAT 550-600 0.4877 (0.3641) -0.9142 (0.3035) No. Obs. 3328 X2(70) 597.74 Log Likelihood -6,052.4772


REFERENCES

Bailey, Elizabeth, and Kevin Rask. 1996. "Are We Role Models? Major Choice in an Undergraduate Institution." Colgate Economics Working Paper.

Bound, John, Michael Schoenbaum, and Timothy Waidman. 1996. "Race Differences in Labor Force Attachment and Disability Status." The Gerontologist, Vol. 36, No. 3, pp. 311-21.

Brown, Charles, and Mary Corcoran. 1997. "Sex- Based Differences in School Content and the Male- Female Wage Gap." Journal of Labor Economics, Part 1, Vol. 15, No. 3, pp. 431-65.

Canes, Brandice, and Harvey Rosen. 1995. "Follow- ing in Her Footsteps? Faculty Gender Composition and Women's Choices of College Majors." Indus- trial and LaborRelations Review, Vol. 48, No. 3 (April), pp. 486-504.

Corcoran, Mary, and Paul Courant. 1985. "Sex-Role Socialization and Labor Market Outcomes." Ameri- can Economic Review, Vol. 75, No. 2 (May), pp. 275- 78.

Eccles,Jacquelynne, and Lois Hoffman. 1984. "Sex Differences in Preparation for Occupational Roles." In H. Stevenson and A. Siegal, eds., Child Develop- ment and Social Policy. Chicago: University of Chi- cago Press, pp. 367-420.

Jacobs, Jerry. 1995. "Gender and Academic Special- ties: 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 Minorities in Science and Engineering." In J. C. Smart, ed., Higher Education: Handbook of Theory and Research, Vol. 12. New York: Agathon.

Paglin, Morton, and Anthony Rufolo. 1990. "Het- erogeneous Human Capital, Occupational Choice, and Male-Female Earnings Differences." Journal of LaborEconomics, Vol. 8, No. 1, Part 1, pp. 123-44.

Polachek, Solomon. 1978. "Sex Differences in Col- lege Major." Industrial and Labor Relations Review, Vol. 31, No. 4 (July), pp. 498-508.

Solnick, Sara. 1995. "Changes in Women's Majors from Entrance to Graduation at Women's and Co- educational Colleges." Industrial and 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 De- grees Conferred." Science, Vol. 250 (October 26), pp. 517-21.

Ware, Norma, and Valerie Lee. 1988. "Sex Differ- ences in Choice of College Science Majors." Ameri- can Educational ResearchJournal, Vol. 25, No. 4 (Win- ter), pp. 593-614.

Ware, Norma, Nicole Steckler, andJane Leserman. 1986. "Undergraduate Women: Who Chooses a Science Major?" Journal of HigherEducation, Vol. 56, No. 1 (January-February), pp. 73-84.

Documents

CHOICE OF MAJOR - Earlham Collegelegacy.earlham.edu/~pardhan/research/refs/choice_of_major.pdf · Among the many factors that influence men's and women's choice of major, and consequently