









































Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Community
Ask the community for help and clear up your study doubts
Discover the best universities in your country according to Docsity users
Free resources
Download our free guides on studying techniques, anxiety management strategies, and thesis advice from Docsity tutors
The relationship between gender and salary distribution among college graduates. It discusses how gender differences in college major choice, industry choice, and internship experience contribute to the gender pay gap. The document also introduces the concept of a 'glass escalator' for female graduates and the persistence of the gender gap at the top of the salary distribution.
What you will learn
Typology: Study notes
1 / 49
This page cannot be seen from the preview
Don't miss anything!
Submitted to the Distinguished Majors Program Department of Economics University of Virginia April 17, 2020 Advisor: Amalia R. Miller
Many thanks are in order : First and foremost, to my advisor, Professor Amalia Miller. For her support of my research goals and ambitious ideas, for her expertise and thoughtful suggestions, and for her belief in my ability, I will always be grateful. This project would not have been possible without her. I also thank my DMP cohort - though our time together was cut short, I am thankful for their friendship and for their invaluable advice on every step of this project. To Professor Sarah Turner, for her thoughtful and honest feedback on many stages of this project. Her guidance this year has surely made me a better economist. And to Professor Sandip Sukhtankar, my first research mentor, for inspiring my love of economics and teaching me to be a curious, compassionate, and rigorous researcher. To Julia Lapan, Director of Engineering Career Development at the UVA Career Center. Her work to improve gender equity in career outcomes inspired this study, and her persistent advocacy allowed me to access the data I used. I also thank Everette Fortner, Kim Sauerwein, and Nathan Hunsaker from the UVA Career Center and Sarah Schultz Robinson from the UVA Office of Institutional Research and Analytics for their assistance with data access. To Professor Ken Elzinga, Sarah Lapp, and the Marshall Jevons fund for providing resources for this project that I ultimately didn’t use. And, to Debby Stanford and Ashley Watkins, for assistance and support when my research plan changed. I also thank the University of Virginia Economics Department as a whole for providing me the excellent training that brought me to this point. And finally, to my loved ones. To my parents, for providing me with the world but encouraging me to be humble and work hard. To my grandparents, for pushing me to be the best I can be. To my twin brother Keshav, for being my partner in crime, and to my little sister Leela, for the title of this thesis. To Jacob Cunningham, for supporting me in all of my endeavors. Lastly, to Miss Julia, for inspiring my feminism and reminding me to be an advocate for women in all I do.
Long-term trends in the United States indicate substantial reductions in the gender pay gap since the 1950s (Blau & Kahn, 2017). This convergence of male and female earnings is driven in part by changing trends in college attendance: The gender gap in college attendance has decreased continuously since the 1950s, with female college graduates now outnumbering males (Goldin et al., 2006). In addition to changes in college attendance trends, gender differences in collegiate schooling content have also narrowed. Goldin (2005) finds that the gender gap in college majors closed significantly between 1970 and 1985, a period she termed the “quiet revolution.” Gender gaps in the skills that students develop during college, however, have failed to converge since the 1980s (Shauman, 2016; Turner & Bowen, 1999). Furthermore, convergence in the mapping of major to occupation has been modest (Sloane et al., 2019). Progress in closing the gender pay gap has also slowed since the 1980s, suggesting an important link between gender differences in schooling content, industry choice, and earnings. Previous literature demonstrates that a significant portion of the existing gender pay gap for college educated workers can be explained by gender differences in college major choice and industry selection (Blau & Kahn, 2017). It is unclear, however, whether this explanation holds for all college-educated workers. In particular, the extent to which gender differences in schooling content and industry selection can explain gender differences in earnings may vary across the earnings distribution. This thesis examines the impact of gender differences in educational investments on gender differences in career outcomes for graduates of a public flagship university. I use new self-reported data from the University of Virginia Career Center’s “First Destinations” Survey to analyze the gender gap in starting salary for UVA students who are employed full- time immediately after graduation. Using linear Oaxaca-Blinder decomposition models as well as unconditional quantile decomposition methods, I analyze the gender pay gap across the starting salary distribution.
The term “glass ceiling” is used to describe explicit or implicit barriers that exclude women from the highest-paying jobs. Glass ceilings are often found even in developed West- ern nations and could be due to a scarcity of senior women in high-skilled, high-paying oc- cupations; women’s higher propensity to exit the labor force temporarily and interrupt their acquisition of skills; women’s hesitation to negotiate salary offers (Babcock & Laschever, 2003); and the exclusionary aspects of “boys’ club networks that are prominent in certain industries (Xiu & Gunderson, 2014). Glass ceilings are likely to be a particular concern for female graduates of elite universities, many of whom will graduate with extremely high earnings potential. As such, my work provides important insight into the pre-market human capital specialization and subsequent labor market treatment of highly qualified female college graduates. The remainder of the paper proceeds as follows: Section 2 presents an overview of relevant theoretical and empirical literature; Section 3 describes the empirical specifications used; Section 4 summarizes the data; Section 5 details results; and Section 6 concludes.
Human capital theory provides insight into why gender differences in educational invest- ments may persist. In a simple human capital model, an individual will invest in education up to the point at which the benefits from schooling are equal to the costs (Becker, 1964). The benefits and costs of schooling, and consequently the ideal schooling decision, may dif- fer between men and women, causing gender gaps in decisions such as college major choice. Gender differences in the costs and benefits of college majors may arise for several reasons. First, men and women may possess innate differences in academic ability; thus, the return to investing in academically rigorous majors with potentially high salary returns may be lower for women than for men. This notion of comparative advantage influencing occupational
choice and subsequent earnings has been explored extensively in the theoretical literature, beginning with Roy (1951). The theory of comparative advantage amongst college educated workers has also been tested empirically. Willis and Rosen (1979) use data on male World War II veterans’ IQ and test scores to show that comparative advantage partly determines the decision to invest in higher education. Using SAT and GRE test scores, Paglin and Rufolo (1990) find that mathematical ability is an important determinant of college major choice, and that differences in earnings across fields are largely explained as returns to the use of quantitative abilities that not all students possess. The theory of comparative advantage may be particularly useful in the context of quan- titative majors such as STEM (science, technology, engineering, and mathematics) fields, which continue to be male dominated despite changing gender attitudes towards women’s work. If women have lower mathematical aptitude than men on average, then women will likely experience greater costs to majoring in STEM fields, such as time spent studying and stress imposed by examinations. Furthermore, women may also experience smaller salary returns to majoring in STEM fields if they have lower mathematical aptitude. Previous literature, however, finds that gender differences in mathematical ability, as measured by quantitative standardized test scores, are likely too small to explain existing gender gaps in college major choice (Ceci et al., 2014; Hyde et al., 2008; Riegle-Crumb et al., 2012). An alternative explanation for self-selection of men and women into different college majors may be that men and women simply have different preferences regarding college majors and the associated career outcomes (Zafar, 2012). For example, women may prefer majors with less quantitative coursework or majors associated with career fields that have less demanding hours. This could be the result of innate differences in preferences or of “sex role socialization” (Corcoran & Courant, 1985; Eccles & Hoffman, 1984), the conditioning of women to prefer certain occupations over others due to social factors. Women may also anticipate differential treatment in the labor market, due to discrimination in hiring or discrimination in promotions (Lazear & Rosen, 1990). Economic theory, however, predicts
Graduates (NSCG) and find that between 44 and 73% of the gender wage gap is accounted for by highest degree, major, and age (Black et al., 2008). A study by Glassdoor economists finds somewhat contradictory results using linear decomposition methods on anonymous, self-reported salary data from the Glassdoor platform. While 68% of the gender gap in base salary was accounted for by gender differences in observable characteristics, only 14% was accounted for by gender differences in education and experience. The remaining 54% was accounted for by sorting of men and women into different industries and occupations (Chamberlain, 2016). Relevant to my analysis, additional literature explores college major choice and the gen- der wage gap specifically for starting salary offers. This eliminates the potential influence of gender differences in promotions, in parental leave time, and in years of work experience (although expected years of labor force participation may still differ). McDonald and Thorn- ton use data from the annual surveys of the National Association of Colleges and Employers (NACE) and find that as much as 95% of the overall gender gap in starting salary offers can be attributed to differences in college major choice (McDonald & Thornton, 2007).
The previously described analyses of nationally representative data may obscure differ- ences between higher education institutions. These differences could include disparities in teaching quality, career services, alumni networks, and other factors potentially impacting labor market outcomes. Some existing literature controls for differences between institutions by evaluating college majors and career outcomes for graduates of individual colleges and universities. In particular, Todd Stinebrickner and Ralph Stinebrickner designed and administered the Berea Panel Study (BPS), a multipurpose longitudinal survey conducted on students at Berea College in central Kentucky. Respondents in two cohorts were surveyed 12 times per year while in school and annually thereafter about expectations towards uncertain outcomes,
the factors that might influence these outcomes, and eventually the outcomes themselves (T. Stinebrickner & Stinebrickner, 2011). This study has been the basis for research papers exploring college major choice (T. Stinebrickner & Stinebrickner, 2011); persistence within college major (R. Stinebrickner & Stinebrickner, 2014); and, relevant to my analysis, labor market outcomes such as the beauty wage premium (R. Stinebrickner et al., 2019) and the wage consequences to over-education (Agopsowicz et al., 2017). My work will contribute to this literature by examining career outcomes within a different institutional setting. Berea College is a small, private liberal arts college with a focus on providing an education to students from low income backgrounds. As such, Berea offers a full tuition subsidy to all students. UVA is a large, public flagship university with students coming from more diverse socioeconomic backgrounds and receiving a spectrum of financial aid packages. UVA is also slightly more selective, with an acceptance rate of 26% compared to Berea’s 38% in 2018.^1 Other literature specifically analyzes college major choice at individual institutions. (Za- far, 2012) and (Wiswall & Zafar, 2014) study gender differences in college major choice in the context of Northwestern University and New York University, respectively. Similarly to R. Stinebrickner and Stinebrickner (2014), Wiswall and Zafar collect data on students’ major preferences and their subjective expectations regarding the academic, personal, and profes- sional outcomes associated with various majors. The authors find that most of the gender gap in major choice is due to gender differences in preferences for major-specific outcomes, such as salary and work hours.
(^1) Acceptance figures were taken from the Integrated Postsecondary Education Data system (IPEDS) at https://nces.ed.gov/ipeds/about-ipeds.
term on the right hand side of the equation, [(XM − XF ) × βˆ], represents the portion of the gender gap in average salary that can be explained by observable outcomes, while the second term, [( βˆM − βˆF ) × X], represents the portion of the gender gap that cannot be explained using the variables in the data. This portion of the gap is due to gender differences in returns, rather than stock, of human capital. The unexplained portion of the gender gap is often attributed to discrimination; however, this portion of the gap may also be the result of variables omitted from the regression. Notably, I do not control for any measure of academic ability due to data constraints discussed in Section 4. I am primarily interested in the relative magnitude of [(XM − XF ) × βˆ]; that is, the portion of the gender wage gap that can be explained by gender differences in observable characteristics. One advantage of using a pooled decomposition model is that standard errors are much lower for the coefficients of sex-atypical majors using this approach. This is because, for majors with very few women or very few men, the pooled regression estimates draw from a much larger sample size and therefore produce more precise estimates (Brown & Corcoran, 1997).
There exists a significant amount of recent literature debating the merits of various quan- tile regression models and quantile decomposition methods. Albrecht et al. (2003) employ conditional quantile regression methods on 1998 data from Sweden containing a represen- tative sample of workers aged 15-75. They find that the log gender wage gap increases throughout the wage distribution and sharply accelerates at the upper tail even after con- trolling for gender differences in age, education, industry, and occupation. This result is interpreted by the authors as evidence of a glass ceiling for female workers. Machado and Mata (2005) propose applying conditional quantile regression techniques to generalize lin- ear decomposition models, a technique used in several subsequent papers to analyze wage gaps (Arulampalam et al., 2007; Lucifora & Meurs, 2006). Melly (2005) proposes a similar
method of applying conditional quantile regression to decomposition models. More recent literature has focused on the use of unconditional quantile regression models, such as the reduced influence function (RIF) regression model popularized by Firpo et al. (Firpo et al., 2009; Fortin et al., 2011). This procedure allows for the generalization of linear Oaxaca-Blinder models to distributional statistics other than the mean (Firpo et al., 2018), a technique used in recent literature on the gender wage gap in China (Chi & Li, 2008; Xiu & Gunderson, 2014), the United States (Kassenboehmer & Sinning, 2014), and various countries in Latin America (Carrillo et al., 2014). Unlike conditional quantile regression decomposition techniques, this method allows quantiles to be decomposed non-sequentially in the same way means can be decomposed using the conventional Oaxaca-Blinder methodology (Firpo et al., 2018). Due to its popularity in recent literature as well as its analogies to the standard Oaxaca- Blinder model, I use recentered influence function (RIF) regression analysis to evaluate the gender pay gap at various points along the salary distribution.^3 My RIF regression model replaces the dependent variable in a standard linear regression model with the recentered influence function of the quantile of interest.^4 In this regression model, the coefficients correspond to the marginal effect on the unconditional quantile of shifts in the distribution of covariates, holding everything else constant. For each quantile of interest τ , I estimate three RIF unconditional quantile regressions: one for the male earnings distribution, one for the female earnings distribution, and one coun- terfactual distribution in which females have the same characteristics as males. Analogous to OLS regressions, the RIF regression functions assume a linear specification as follows: (^3) To estimate my RIF decomposition models in Stata I use Fernando Rios-Avila’s oaxaca rif command (Rios-Avila, 2019). 4 An influence function of a distributional statistic represents the influence of a single observation on the value of that distributional statistic. Adding back the distributional statistic to the influence function yields the recentered influence function (RIF). Conveniently, the expectation of the RIF is equal to the distributional statistic. In this case, the distributional statistics of interest are quantiles of the salary distribution (Firpo et al., 2009).
My data come from the University of Virginia Career Center’s “First Destinations” Sur- vey (FDS)^5 for the years 2016-2018. Due to data privacy restrictions, the years in my data set have been anonymized: I refer to them as Year A, Year B, and Year C. The FDS data include self-reported information from recent UVA graduates. The First Destinations Sur- vey is made available to students beginning in December of their last year at the University of Virginia, and remains open for approximately one year. This means that students who graduate in May have seven months after graduation to report their post-graduation plans. The data includes the following information: primary school of enrollment, degree attained, major(s), minor(s), enrollment in higher education, undergraduate internship experience, post-graduation salary,^6 post-graduation job location, and post-graduation career industry, along with demographic information such as students’ race and gender. I restrict my analysis to UVA bachelor’s degree recipients, excluding graduates of UVA’s professional schools and graduate programs. Despite being self-reported, this data arguably provides the most accurate estimates of salary outcomes for recent University of Virginia graduates. The State Council of Higher Education for Virginia (SCHEV) reports wage outcomes using unemployment tax data from the Virginia Employment Commission (VEC)^7. This administrative data includes informa- tion only on graduates who are employed in the state of Virginia and who meet the following (^5) Information about the survey can be found on the UVA Career Center’s website at https://career. virginia.edu/uva-career-outcomes. The data is collected primarily for the purpose of constructing the Career Center’s annual reports on student outcomes. Data for Year A and Year B was self-reported in the UVA Student Outcome Activity Report (SOAR) and data for Year C was self-reported through Handshake, an online recruiting platform used by the University of Virginia. Students from the McIntire School of Commerce reported their outcomes in the McIntire Portfolio Destination Survey (PDS). The Career Center website states that, in a limited number of cases, information was captured “through other sources, including faculty, employers, and social media (LinkedIn).” 6 7 I measure earnings using log - base salary in all of my analysis. SCHEV salary information on University of Virginia graduates is available at https://research.schev. edu/iprofile.asp?UID=234076.
criteria:
My sample is fairly representative of the racial composition of the University of Virginia undergraduate student body, although white students are slightly overrepresented and black students are slightly underrepresented.^13 As of Fall 2016, 6.45% of all undergraduates at the University of Virginia were Black; 12.82% were Asian; 6.28% were Hispanic; 4.40% were Multi-Race; 4.75% were Non-Resident Aliens; and 59.47% were White. The remainder reported their race as “Unknown” or “Other,” with the “Other” category including Native American and Alaskan Native students. As demonstrated in Table 2 below, one variable that does differ across years is my set of industry dummies. In particular, a higher percentage of students identify their industry as “Other” in Year C than in Year A or Year B. This is likely due to changes in industry categorization options in the First Destination Survey. (^11) Year C graduates did not have the option to identify their ethnicity as “Non-Resident Alien” in the First Destinations Survey. 12 Since this data is collected less than a year after graduation, the “Other” category includes many students who are taking time off after graduation to apply to graduate and professional schools; participate in volunteer work or unpaid fellowship programs; travel; or seek employment. 13 Data on the racial composition of the UVA student body comes from Institutional Research and Analytics (IRA) at the University of Virginia. Data was retrieved online at https://ira.virginia.edu/ university-stats-facts/enrollment.
Table 2: Proportion of Students Working in Each Industry, by Year Industry Year A Year B Year C Accounting 0.0029 0.0023 0. Arts, Media, and Entertainment 0.0385 0.0153 0. Communications 0.0584 0.0397 0. Construction 0.0292 0.0351 0. Consulting 0.1803 0.2313 0. Consumer Products/Retail 0.0207 0.0260 0. Education 0.0520 0.0412 0. Financial Services 0.1326 0.1733 0. Government 0.0371 0.0443 0. Healthcare 0.0919 0.0710 0. IT and Engineering 0.2060 0.1702 0. Natural Resources^14 0.0100 0.0084 0. Nonprofit/NGO 0.0378 0.0313 0. Real Estate 0.0007 0.0084 0. Services 0.0656 0.0626 0. Other 0.0364 0.0397 0.
4,419 (55.81%) of the observations in my sample are female and 3.499 (44.19%) are male. This is representative of overall undergraduate enrollment at the University of Virginia: In Fall 2016, 54.71% of undergraduates were female and 45.29% were male.^15 Summary statistics for male and female graduates are presented in Table 3. (^14) Full industry title: Natural Resources, Agriculture, and Environmental Science (^15) This information comes from Institutional Research and Analytics (IRA) at the University of Virginia and was retrieved online at https://ira.virginia.edu/university-stats-facts/enrollment.