Women’s Colleges and Economics Major Choice: Evidence from Wellesley College Applicants

Abstract

Many observers argue that diversity in Economics and STEM fields is critical, not simply because of egalitarian goals, but because who is in a field may shape what is studied by it. If increasing the rate of majoring in mathematically intensive fields among women is a worthy goal, then understanding whether women’s colleges causally affect that choice is important. Among all admitted applicants to Wellesley College, enrollees are 7.2 percentage points (94 percent) more likely to receive an Economics degree than non-enrollees (a plausible lower bound given negative selection into enrollment on math skills and major preferences). Overall, 3.2 percentage points – or 44 percent of the difference between enrollees and non-enrollees – is explained by college exposure to women instructors and students, consistent with a wider role for women’s colleges in increasing women’s participation in Economics.

HIGHLIGHTS

• In US colleges, men are more than twice as likely as women to major in Economics.

• Enrollees are twice as likely as non-enrollees to major in Economics.

• The difference is a plausible lower bound given patterns of selection on observed variables.

• Over 40 percent of the difference is explained by the gender of students and faculty.

KEYWORDS:

• Women economists
• women’s colleges
• major choice

JEL Codes:

• I2

ACKNOWLEDGMENTS

We are grateful to administrators and staff of Wellesley College for supporting the research, especially Joy St. John, Pamela Taylor, and Hui Xiong. We are also grateful to many colleagues and former students for comments and ideas, including Yo-Jud Cheng, Emily Cuddy, Rebecca Fraenkel, Phil Levine, Elaine Liu, Lucie Schmidt, Karen Scott, Olga Shurchkov, Micah Villarreal, Amy Wickett and Heidi Williams. Views expressed here are those of the authors and do not reflect those of the Federal Reserve Bank of Chicago, the Federal Reserve System, or any other entity.

Notes

1 See Karen E. Dynan and Cecilia E. Rouse (1997); Amanda Bayer and Cecilia E. Rouse (2016); Tatyana Avilova and Claudia Goldin (2018); Kasey Buckles (2019); and Shelly Lundberg and Jenna Stern (2019).

2 Chang-Tai Hsieh et al. (2019) find that convergence in the occupational distribution across men and women explains up to 40 percent of growth in GDP per capita since the 1960s. Mathematically-intensive fields of study increase male and female wages, even after controlling for individuals’ occupations (Zafar 2013; Altonji, Arcidiacono, and Maurel 2016), and gender gaps in fields of study account for over half of the male-female wage gap in the United States (Brown and Corcoran 1997; Altonji, Arcidiacono, and Maurel 2016; Sloane, Hurst, and Black 2021).

3 Catherine Porter and Danila Serra (2020) and Thomas Breda et al. (2021) show that transitory exposure to women role models in Economics or STEM courses – entwined with the provision of information about careers – increases the probability that women pursue similar undergraduate specializations. At the US Air Force Academy, high-achieving women exposed to women instructors in the first year were far more likely to choose STEM majors and careers (Carrell, Page, and West 2010; Mansour et al. 2020). Similarly, South Korean students exposed to women math teachers in middle schools increased their subsequent engagement with STEM fields (Lim and Meer 2020).

4 Random assignment to single-sex classrooms increased women’s classroom performance in economics (Booth, Cardona-Sosa, and Nolen 2018) and mathematics (Eisenkopf, Hessami, and Fischbacher 2015), while women peer mentors improved retention of women in an engineering program (Dennehy and Dasgupta 2017). However, more women peers in gender-mixed settings of business schools had either zero effects on outcomes (Oosterbeek and van Ewijk 2014) or negative effects on women’s propensity to choose male-dominated majors (Zölitz and Feld 2021).

5 The female proportion of Economics faculty between 2009 and 2018 – including tenure-track and non-tenure-track faculty – is from the CSWEP survey. The female proportion of college-wide faculty between 2009 and 2018 – including all full-time instructional staff – is from IPEDs surveys. See Appendix for details.

6 See, for example, C. Kirabo Jackson (2012), Soohyung Lee et al. (2014), and Hyunjoon Park, Jere R. Behrman, and Jaesung Choi (2013).

7 Avery Calkins et al. (2021) use an event study design using differences in the timing of co-educational conversion across comparable schools to identify the effect of single-sex education on women’s major choices. Billger (2002) uses a difference-in-differences strategy to examine what happens to women’s major choice when one formerly women’s college transitioned to being co-educational and finds that women become less likely to major in male-dominated fields when men are admitted.

8 Non-enrollees graduated from a range of highly selective colleges and universities, including 50 percent from Amherst, Barnard, Brown, Chicago, Cornell, Dartmouth, Duke, Georgetown, Harvard, MIT, Northwestern, Pennsylvania, Princeton, Smith, Stanford, Williams, and Yale.

9 The NSC matches applicant data to degree attainment data using applicants’ names and birthdates, and typically covers over 90 percent of enrollment at selective colleges and universities (Dynarski, Hemelt, and Hyman 2015). NSC coverage depends on whether students graduate from college, on whether colleges submit data, on matching errors, and on the suppression of student-level records under the Family Educational Rights and Privacy Act (FERPA).

10 Our typology departs in three ways from DHS (2016), in order to ensure that Wellesley College is not unduly favored. First, we use a four-digit CIP code (4506) to indicate economics majors, rather than a STEM-eligible six-digit code for Econometrics and Quantitative Economics (450603). Wellesley College began reporting Economics majors with the STEM code in Spring 2017, the final year of our degree data, although the required courses did not change. However, not all colleges and universities necessarily use the STEM-eligible code, even when they require similar courses. Second, we identify an expanded list of mathematically-intensive business majors that includes finance and business economics, in addition to the STEM-eligible codes for management science. All are plausible substitutes for Economics at colleges and universities with business programs. Third, we conservatively omit the six-digit CIP code for a popular STEM-eligible major at Wellesley (Media Arts and Sciences) that nonetheless requires only three computer science courses.

11 There are 64 dummies indicating preferred majors; these are not mutually-exclusive since applicants can identify up to two preferred majors. There are 9 mutually-exclusive race-and-ethnicity dummy variables, and 13 mutually-exclusive home language dummy variables.

12 One must further make an assumption about the proportion of variance in the dependent variable that is explained by the full set of observed and unobserved controls. The maximum is 1, but this tends to dramatically inflate bounds, and it is high in most empirical settings due to measurement error or other idiosyncratic variation in the dependent variable (Oster 2019). We adopt a benchmark of $\min \left(1,1.5R^{,22}\right)$, where $R^2$ is obtained from a regression with a full set of controls. Using estimates from randomized experiments, Emily Oster (2019) shows that 86 percent of bound estimates are within 2.8 standard errors of the experimental estimate with an assumption of $\min \left(1,1.5R^2\right)$. An assumption of $\min \left(1,1.25R^2\right)$ increases this to 91 percent, while an assumption of 1 decreases it to 37 percent.

13 Recall that 0.01 and 16.5 percent of the initial Wellesley enrollees and non-enrollees, respectively, were not matched to NSC degree and major data. Among non-enrollees, the means of observed variables are similar between missing and non-missing observations (see Table A2). We also re-calculated the comparisons in Table 2 but including missing observations in the calculation of mean differences. As before, test scores are still 32 percent to 35 percent of a standard deviation lower among enrollees, and other differences are similar to those in Table 2. These similarities between missing and non-missing observations in the comparison group suggests that missingness is not a straightforward proxy for dropping out of college. A more likely explanation is that many NSC records are privacy-blocked, even at selective institutions in the comparison group. The NSC provides an aggregate matching report that cannot be matched to individual data. More than half of unmatched records among admitted Wellesley applicants are privacy-blocked. These include, for example, all degree records from Columbia University (which enrolls the fifth-highest number of admitted Wellesley applicants).

14 This assumes equal selection and a maximum R² of 0.21 ( = 0.14×1.5).

15 A referee pointed out that students might be substituting away from another business-related field that is not included in our definition of mathematically-intensive business majors (see Table A1). We calculated a dummy dependent variable indicating degrees in either Economics or any business field (that is, a two-digit CIP code of 52). The coefficient and standard error in a specification like that of column (2) in Table 3 is 0.031 (0.007). While smaller – given the absence of any business degrees at Wellesley College – the coefficient is still large and suggests the Economics effect is not wholly the result of substitution away from business-related fields.

16 Our definition of under-represented minority includes students who self-identify as African-American, Latinx, and/or Native American.

17 We use the NSC data to identify students with graduate degrees in our undergraduate estimation sample. If students are not observed to receive graduate degrees, we assume that they did not. However, the latter group may include students whose data is privacy-blocked or otherwise unmatched.

18 Michael Inzlicht and Talia Ben-Zeev (2000) and Denise Sekaquaptewa and Mischa Thompson (2003) show that women’s performance in mathematics declines when women complete tasks in the presence of men perhaps because it causes women to view themselves through the lens of a negative stereotype. Porter and Serra (2020) show that the presence of female role models might encourage or inspire women to pursue similar paths.

19 We thank an anonymous referee for this observation.

20 Valentina A. Paredes M. Daniele Paserman, and Francisco Pinto (2020) measured gender attitudes among Chilean Economics students. They were more gender-biased than students in other fields upon entry. The gap increased over time, especially for men students, but the gap was attenuated when students were exposed to more women instructors and peers. Alice Wu (2018) shows that women are described with derogatory and sexualized language in an anonymous online forum for nominally professional economists.

21 A referee pointed out that, ideally, we would also want measure variation in exposure to women faculty in introductory Economics courses, but we do not have access to this data on institutions in the comparison group.

22 Amanda Bayer, Syon O. Bhanot, and Fernando Lozano (2019) found that randomly sent emails with an encouragement to take economics courses and additional information about the types of research conducted by economists had modest but imprecisely-estimated effects on the likelihood of taking additional courses. Hsueh-Hsiang Li (2018) evaluated a treatment that provided information about careers and earnings in economics along with targeted encouragement of women with above-median scores in economics. The combined intervention had large effects on the probability that women with higher grades chose a major in economics.

23 An observational literature shows that women at coeducational schools who receive lower grades in introductory economics courses are less likely to major in economics than men with similar grades (Rask and Tiefenthaler 2008; Goldin 2015). F.M. Antman, E. Skoy, and N. E. Flores (2020) randomly provided students with information about their location in the grade distribution, and found that it reduced grade sensitivities among women at a large coeducational university. At Wellesley College, a regression-discontinuity design showed that women just above letter-grade cutoffs in introductory economics courses are much more likely to major in Economics (McEwan, Rogers, and Weerapana 2021), suggesting that grades are important signals used to update beliefs about major-specific abilities. Related theory also suggests that closing the difference in mean grades across departments might also close gaps in major choice (Ahn et al. 2019).

24 The explained portion of a college covariate is obtained by multiplying two estimates: (1) the coefficient on the college covariate from the fully-specified regression in column (2) of Table 6, and (2) the mean difference in the college covariate between enrollees and non-enrollees, conditional on the covariates already included in the base regression in column (1) of Table 6. The latter is obtained by regressing the covariate on an enrollment indicator, while controlling for covariates in the base specification (which are, in this case, all the applicant and high school variables included in the base specification). The explained portion for a group of college covariates (that is, gender-related) is obtained by summing the individual contribution of each variable. Gelbach (2016) further describes formulas for obtaining the standard errors of the components of the decomposition, which are reported in column (3) of Table 6.

25 As an additional exercise, we included an indicator of whether the college or university has a business program (inferred in the IPEDs data by any reported undergraduate degrees with the two-digit CIP code of 52). The coefficient on this variable was very small and not statistically different from zero. It did not substantively affect the decomposition results in Table 6.

26 One possibility is that Economics is a relatively popular major, with at least 15 percent of students, which may influence major choices via information or interactions with peers. On the other hand, the popularity of the major may simply be an outcome that results from the effects of unmeasured variables.

27 https://www.research.gov/grfp/AwardeeList.do?method=loadAwardeeList.

28 https://ideas.repec.org/c/boc/bocode/s457992.html.

Additional information

Notes on contributors

Kristin F. Butcher

Kristin F. Butcher is Vice President and Director of Microeconomic Research at the Federal Reserve Bank of Chicago. She is also the Marshall I. Goldman Professor of Economics at Wellesley College. She is interested in the economics of immigration and the economics of childhood obesity.

Patrick J. McEwan

Patrick J. McEwan is the Luella LaMer Slaner Professor in Latin American Studies and Professor of Economics at Wellesley College. He is interested in the economics of education, Latin American education and social policy, and applied econometrics.

Akila Weerapana

Akila Weerapana is Professor of Economics at Wellesley College. He is interested in macroeconomics and the economics of higher education.