Bridging the divide: examining field of study gender earnings disparities in the early career

The majority of the gender earnings gap emerges in the first five years of a career. With the early career being a vital time in establishing an earnings trend for oneself, understanding the factors that influence the early career earnings gap is vital. This paper uses a newly released Federal Score Card data that reports median three years postgraduation earnings data disaggregated by gender and degree program. This data set is used to identify institution and programs that have large impacts of the gender earnings gap. Using an Ordinary Least Squares (OLS) regression I find that schools in less dense and more rural regions are associated with a higher earnings gap compared to those in more dense regions. The OLS regression also identified that public colleges have a lower wage gap then private colleges and that the wage gap increases with attainment of advanced degrees. To address the high dimensionality of the whole data set (nearly 2500 variables), Lasso Regression is used. This analysis reveals which higher education institutions and degree programs have the largest impacts on the gender earnings gap.

ABSTRACT The majority of the gender earnings gap emerges in the first five years of a career. With the early career being a vital time in establishing an earnings trend for oneself, understanding the factors that influence the early career earnings gap is vital. This paper uses a newly released Federal Score Card data that reports median three years postgraduation earnings data disaggregated by gender and degree program. This data set is used to identify institution and programs that have large impacts of the gender earnings gap. Using an Ordinary Least Squares (OLS) regression I find that schools in less dense and more rural regions are associated with a higher earnings gap compared to those in more dense regions. The OLS regression also identified that public colleges have a lower wage gap then private colleges and that the wage gap increases with attainment of advanced degrees. To address the high dimensionality of the whole data set (nearly 2500 variables), Lasso Regression is used. This analysis reveals which higher education institutions and degree programs have the largest impacts on the gender earnings gap. ii TABLE OF CONTENTS ABSTRACT II INTRODUCTION 1 LITERATURE REVIEW 2 DATA/METHODS 8 DISCUSSION OF RESULTS 11 CONCLUSION 21 APPENDIX 23 REFERENCES 26 iii 1 INTRODUCTION This paper examines the gender earnings gap three years after individuals attain their highest college degree. The three-year post-graduation timeline gives new graduates time to find a job in their degree field and begin working. Additionally, the three-year post-graduation time period fills a gap in the literature, the wage gap has been studied in the one-year post college graduation (Corbett, 2012) and many years after. Much of the literature around the gender wage gap has focused on aggregate gender wage gap (Ventre, 2022). While this research is vital it is often faced with structural difficulties like women being sorted into lower paying career fields from college major choice (Ventre, 2022). Examining data within individual skill groups has traditionally been difficult as much of the post college graduation salary data is collected by individual institutions through alumni surveys that do not have mandatory data reporting requirements. This leads to biases in the data and incomplete data sets. Recently, the US Department of Education released data as part of their Federal Scorecard program that features threeyear median earnings for men and women at the degree program level. This new dataset will provide insight into the gender earnings gap that forms in skill specialties. It will allow for examining which schools and majors at those schools have equal outcomes and which do not. Using this new dataset to examine the gender wage gap and its relationship to degree programs and school will provide incoming higher education students, school administrators, and state legislators with insights into how to help schools and programs that have strong or weak outcomes for women. 2 LITERATURE REVIEW The gender wage gap has been a frequently explored topic. There are many facets to be explored in the gender wage gap as it spans entire careers and is impacted by a variety of societal factors. In this Literature Review I will focus on research done on early career and post college gender wage gap. This will specifically focus on how the work done in this paper fits into the broader landscape of gender wage gap research. Research done by economists at the London School of Economics examined the early career gender wage gap from college graduates in Finland in 1990. The researchers discovered that in Finland, “gender differences in background characteristics account about 27 percent for the average early career gender wage gap among university graduates” (Napari, 2008, p. 28). These findings compel a further investigation into why men have stronger background characteristics after graduating from college than women. Early career differences in background characteristics could be attributed to universities providing better experiences to male students when compared to female students. It is likely that some universities still have structures for programs that give men more opportunities than women or make it easier for men to take advantage of opportunities. Universities’ programs providing better experience for men over women is something that policy makers and school administrators would be keenly interested in knowing about. Having statistics on which programs offer equal or unequal opportunities for men and women would give decision makers the tools they need to correct bad outcomes for women. The researchers also found that, “the gender wage gap increases rapidly during the first 5 years after labor market entry. In fact, pretty much all the lifetime increases in 3 the gender wage gap takes place during the first 15 years of the working career. This holds for every education group” (Napari, 2008, p.4-5). These findings compel a further analysis of the early career gender wage gap in the US to determine if this trend holds. These findings are particularly compelling as the gender wage gap in the United States is much larger than Nordic countries like Finland. With the wage gap appearing mostly in the first 5 years in countries with lower wage gaps it is likely that the same trend occurs in the US. It also has the potential to be even more prominent in the United States since there is a higher wage gap. Looking at the first three years of earnings post college graduation in the US will provide insight into if early career gender wage gap is consistent across countries. Another study done by the Institute of Labor Economics in 2002 looked at early career wage gaps in Germany for apprenticeship workers. The researchers found that, “in the German data that the male-female differential in entry wages is approximately 25 percent. Throughout the early career, it stays virtually constant at this level. This is in contrast to human capital theory which predicts zero gap between equally qualified workers” (Kunze, 2002, p. 6). These findings illuminate the stark problem of the early career gender wage gap and how their existence of the early career wage gap contradicts commonly used economic models. The human capital theory has risen in popularity as we have entered a knowledgebased economy. The theory suggests that employers base hiring and pay off of employees’ skills and knowledge (Robinson, 2023). In a person's early career where they do not have specialized skills and have not had time to differentiate their knowledge from their peers, the gender wage gap should not exist. Yet, clearly despite what the theory 4 says, the early career wage gap still exists. With many top organizations like the World Bank (Corbett, 2012) using the human capital theory to guide them in the knowledge economy there is a rising risk of the gender pay gap going unsolved in the knowledge economy. Having transparency into which knowledge gaining programs lead to equal outcomes in pay for men and women is vital as policies are made. Institute of Labor Economics researchers also determined that, “large permanent wage disadvantages during the early career are formed by the occupational qualification while other detailed background characteristics and differences in individual work histories are only of minor importance” (Kunze, 2002, p. 24). Career choice is commonly used to justify the gender wage gap but as time has opened more career choices for women there needs to be new analysis of wage gaps. Since the 1990’s the number of women in the science, technology, engineering, and math (STEM) career fields has skyrocketed. Now over 50% of people working STEM jobs are women (Ventre, 2022). This is a big change from the early and mid-19th century when there were very few career options for women. As women’s career choices have expanded it has brought with it a need for new data and analysis. There is a need to examine the wage gap not at the aggregate level of all working people but at the granularity of individual career fields. Having information on the gender wage gap in each career field would cement the existence of a wage gap while controlling for career choice. Additionally, it would illuminate the existing gender biases in many career fields. Research published in the Journal of Economic Perspectives examined how much of a role college major and occupation plays in the gender wage gap. The researchers found that, “controlling for major and occupation sorting explains roughly 60 percent of 5 this gap. These patterns are a highly salient topic for future research” (Sloane, 2021, p. 24). Demonstrating that 60% of the gender wage gap can be explained by major and career choice is a promising start, but it overlooks wage differences that still exist between men and women in the same career field. One line summary statistics do not explore the intricacies in the wage gap that exists between the plethora of different major and school characteristics that students have to choose from. These days prospective students have a myriad of options to choose from when it comes to school type and major. There are large and small state schools, large and small private non-for-profit universities and for-profit universities. In addition, there are roughly 2,000 different majors offered between all these universities (Learning, 2024). With this massive number of majors and school types there is a need to examine the wage gap at a much more granular level to build a stronger understanding of how students' choice of major and school impacts the gender wage gap they face. The researchers additionally discovered that, “women are systematically sorted into majors with lower potential wages relative to men” (Sloane, 2021, p. 2). The focus on the sorting of women into low paying gender stereotyped jobs obscures the deeper issue of women in majors with high potential wages earning less than their male counterparts. There are likely many societal reasons that women tend to end up in these low paying majors but without controlling for major choice gender wage gap statistics can become hard to interpret and easy to poke holes in. I will not attempt to explain why women tend to choose majors that do not pay as well when compared to their male counterparts. Instead, this research will focus on isolating the wage gap for each individual major to control for general major choice. This will allow for insights into 6 where the wage gap is still prevalent despite men and women having the same specialized skills from their degrees. This focus on only comparing similarly skilled men and women will provide transparency into which careers have deeply structural gender discrimination. Recent research published in the Academy of Management Studies examined the commonly believed idea that women get paid less because they are less likely to negotiate their salaries than men. The researchers discovered, “the percentage of women who reported that they negotiated their offer (54%) was greater than that of men (44%)” (Kray, 2024, pg. 19). Having data demonstrating that women do negotiate at a higher rate than men validate policies and programs looking to decrease the wage gap. The idea that women get paid less because they don’t negotiate their pay as much as men has been a commonly believed and cited idea. Work published in many major sources including the Harvard Business Review (Bowles, 2020) cite this idea as a major reason for the Gender wage gap. There is even a book published in 2003 tilted Women Don’t Ask, Negotiating and the Gender Divide. With this theory being popularly used it can make garnering support for policies and programs designed to reduce the wage gap difficult. Now that this new research has debunked this idea it has taken the blame for the wage gap off women and helped shift the narrative. The researcher also found that, “the gender gap in compensation in this MBA population remained, with women earning less than men” (Kray, 2024, p. 24). These findings demonstrate that the wage gap persists even after advanced degree programs like an MBA. 7 Seeing that women still make less than their male counterparts even after graduating from advanced degree programs shows how structurally deep the wage gap goes. With many MBA programs being highly competitive MBA graduates are generally highly motivated and highly skilled employees. Human capital theory would predict equally high earnings for those with the same subset of advanced skills post MBA graduation. Clearly since this is not the case on the aggregate level for MBA programs gender is playing a large role. This warrants further exploration of which advanced degree programs have equal outcomes for men and women. Research done by the American Association of University Women looked at the gender wage gap one year after graduation. The researchers found that, “a hypothetical pair of graduates—one man and one woman—from the same university who majored in the same field. One year later, both were working full time, the same number of hours each week, in the same occupation and sector … the woman would earn about 7 percent less than the man would earn” (Corbett, 2012, p. 2). These findings build a strong foundation of gender bias in the post college graduation gender pay gap that further research can build upon. This research focused on the very near-term gender wage gap. The data set I will use in this research will allow the analysis done by the American Association of University Women to be expanded to three years post-graduation. Considering this longer time scale will demonstrate if these trends in wage gaps are consistent once graduates have had more time to settle into their careers. Measuring wages three years after graduation instead of one will allow graduates more time to find jobs in their respective degree fields. This will provide more robust results. 8 The authors also found that, “when we compare men and women with similar grades, men earned more than women did, on average, at every level” (Sloane, 2021, p. 11). Despite outperforming their male counterparts in college classes women receive worse wage outcomes, exposing how entry level wages are not based on the competence of the applicant. With women having a 2.85 GPA on average and men having a 2.67, women have been outperforming their male counterparts in the classroom (Conger, 2010). Despite their superior academic performance, they are still falling behind on wages. Additionally, with record high student loan debt for college graduates, the wage gap stretches women even thinner after graduation. With their lower pay women must use a larger portion of their paychecks to stay current on their student loans. This further compounds the impact of the wage gap as women begin to accumulate less wealth. DATA/METHODS Two data sources were used to create the data set used in this analysis. First, the 2017-2019 Cohort Federal Scorecard Data. This is the only published Federal Scorecard Cohort that contains median earnings three years post-graduation broken down by male and female Cohort members and degree programs. Federal Scorecard data contains information collected from those who receive federal aid to attend college. These data fields are only populated if a program has more than 30 graduates who reported data to the Federal Scorecard. This is the data set used to calculate the gender wage gap. The second data source used is the Integrated Postsecondary Education Data System (IPEDS). The following date fields from IPEDS were merged with the Scorecard Data: HBCU Stats (Boolean), Tribal College Status (Boolean), School Size (categorical), and Degree 9 of Urbanization (categorical, ex. Midsize City, Large City, etc.). Rows that were missing one of these data points were dropped. The final data set has 10,036 entries spread across all degree types. Each row represents one degree program at one school. Earnings gaps were calculated as a percentage based on the median male and female earnings three years after graduation (positive number indicates men make more and a negative means women make more). The final data set has 9 columns which are examined in Table I. Table I: Description of Dataset Variables Variable Name Description EARNINGS GAP (median male earning 3 years post grad/median female earnings 3 years post grad)/ median female earnings 3 years post grad *100 CIPDESC (Classification of instructional A degree program, for example: programs) Communications and Media Studies or Economics INSTITUTION TYPE Classification of the university: Public, Private Nonprofit, or Private for Profit INSTITUTION NAME Name of the University, for example: University of Utah DEGREE TYPE Type of degree awarded: Bachelor’s, Master’s, Doctoral, Associate’s, Undergrad Certificate, First Professional 10 Degree, or Graduate/Professional Certificate HISTORICALLY BLACK COLLEGE Is the institution an HBCU OR UNIVERSITY (HBCU) TRIBAL COLLEGE Is the institution a Tribal College URBAN STATUS Urbanization of the area the college is in, for example: City: Large, Rural: Fringe, Town: Remote, etc. SCHOOL SIZE The number of students at an institution: Under 1,000, 1,000-4,999, 5,000-9,999, 10,000-19,999, 20,000 and above, not reported or not applicable Before modeling was done all categorical variables were transformed using one hot encoding, a technique that represents each category in a categorical variable as its own binary variable. To avoid the dummy variable trap, the dummy variable that occurs the most in each category was dropped. The EARNINGS GAP (dependent variable) was normalized to be a percentage. This was done to standardize coefficient values across degree programs and schools. Two types of regressions are used in our analysis. First, I will use an Ordinary Least Squares (OLS) Regression on subsets of the dataset. This will be done as OLS Regression offers the benefit of being able to calculate the p-values of coefficients. OLS Regression cannot be used on the whole data set due to the high dimensionality of the 11 dataset (2,500 variables after one hot encoding). Some of these nearly 2,500 variables are highly correlated. To perform dataset wide analysis a Least Absolute Shrinkage and Selection Operator (Lasso) Regression was used. Lasso Regression was selected as it effectively features selects by shrinking some coefficients to zero. This makes Lasso regression a good choice for datasets with lots of variables and data that is multicollinear. It does this by adding a L1 regularization term to the OLS loss function. This L1 regularization term is the sum of the absolute values of the model coefficients. This regularization term acts as a penalty on the model for having a large number of non-zero coefficients and large coefficient values. This causes coefficients with little impact to be set to zero during optimization. The impact of this regularization is controlled by a model parameter, alpha. The L1 regularization terms value is multiplied by alpha in the loss function. In this paper the optimal value of alpha is found through 5-fold cross validation. Model evaluation was done using R-squared metrics. All data cleaning and analysis was done in Python 3.13.1. The sklearn and statsmodels packages were used to create the regression models. All code and data used in this paper are available on this GitHub. DISCUSSION OF RESULTS When looking at the EARNINGS GAP differences across degree programs and schools we see vast differences in outcomes. The school with the largest EARNINGS GAP three years post-graduation is the Psychology degree program at Brigham Young University Idaho at 92.23% (Male graduates make 92.23% more than their female counterparts). The school program with the smallest EARNINGS GAP is Texas Barber 12 College’s Cosmetology Program which has a -370.8% wage gap (Women graduates earn 370.8% more than their male counterparts). Figure I provides violin plots of the distribution of EARNINGS GAPS in each degree type. Every degree type on average has a positive EARNINGS GAP. Figure I Violin plot of Earnings Gaps By Degree Type The average EARNINGS GAP tends to decrease as the level of education increases. Figure II illustrates the top five highest and lowest EARNINGS GAPS for the four main DEGREE TYPES. There is significant variation across schools and degree programs. The goal of the regression analysis will be to identify trends in the variation between schools and degree programs. Due to the high cardinality nature of the data a combination of OLS and Lasso regressions were used to generate accurate and interpretable results. The OLS regression analysis is done on a subset of the data to eliminate potential multi collinearity problems if the whole data set is used in an OLS 13 Figure II: Top Five Highest and Lowest Wage Gaps regression. Table II displays which categories were dropped during the one hot encoding process. The categories that were dropped were the ones that appear the most frequently in the data. These values are the base line that coefficients of other one hot encoded categories will be compared against. Table II: Description of Dataset Variables Category Dropped in One Hot Encoding Variable Name CIPDESC (Base Line Category) Business Administration, Management and Operations INSTITUTION TYPE Public 14 INSTITUTION NAME Arizona State University Campus Immersion DEGREE TYPE Bachelor’s Degree SCHOOL SIZE 20,000 and above URBAN STATUS City: Large The first regression to examine is an OLS regression on all data field values except INSTITUTION NAME. The dependent variable is the gender EARNINGS GAP as a percentage. The independent variables are CIPDESC, INSTITUTION TYPE, INSTITUTION NAME, DEGREE TYPE, HBCU, TRIBAL COLLEGE, URBAN STATUS, and SCHOOL SIZE. INSTITUTION NAME was excluded as it is correlated with the institutional characteristics like SCHOOL SIZE, and URBAN STATUS. Including it would lead to overfitting due to an extremely large number of categories. The data for this regression has 10,0036 samples with 238 model variables, giving it a model degree of freedom of 238 and a residual degree of freedom of 9797. The model explains 27.00% of the variance in the gender EARNINGS GAP (Rsquared=.270). The adjusted R-squared value is 0.252. The model is statistically significant with an F-statistic of 15.23 and its p-value less then 0.001. The first variable to examine is the URBAN STATUS. These coefficients are relative to the City: Large value so a positive coefficient indicates that an urbanization category has a higher gender wage gap than a large city. The data suggests that schools in 15 less dense and more remote regions have a larger earnings gap compared to those in a large city. Table III displays the statistically significant coefficient values. Table III: URBAN STATUS’s Impact on the Gender EARNINGS GAP URBAN STATUS Coefficient p-value 95% CI Town: Distant 3.09 0.000 [1.66, 4.52] Town: Fringe 3.57 0.002 [1.34, 5.80] Town: Remote 2.60 0.007 [0.71, 4.50] City: Small 1.04 0.030 [0.10, 1.97] Other categories Not significant p > 0.05 NA Next, the impacts of SCHOOL SIZE on the gender wage gap will be examined. The coefficients for SCHOOL SIZE are relative to the 20,000 students and above classification. The data shows that schools under 1,000 students have a gender EARNINGS GAP that’s 1.29% higher than 20,000 students or larger schools and that those schools that don’t report size or size statistics do not apply to have a 6.61% and 4.93% higher gender EARNINGS GAP then the largest schools. The rest of the coefficients are not statistically significant at a p-value > 0.05 level. Table IV displays all coefficients and p-values for the SCHOOL SIZE category. Table IV: SCHOOL SIZE’s Impact on the Gender EARNINGS GAP SCHOOL SIZE Coefficient p-value 95% CI 1,000 - 4,999 -0.85 0.088 [-1.83, 0.13] 5,000 - 9,999 -0.58 0.283 [-1.63, 0.48] 10,000 - 19,999 -0.24 0.606 [-1.14, 0.66] Not applicable 4.93 0.000 [3.24, 6.62] Not reported 6.61 0.002 [2.51, 10.71] Under 1,000 1.29 0.025 [0.16, 2.41] The impact of INSTITUTION TYPE on the gender EARNINGS GAP is clear from the data. Public institutions have a 1.00% lower EARNINGS GAP than private nonfor-profit and a 4.75% lower EARNINGS GAP than private-for-profit institutions. Both 16 results are statistically significant with a p-value < 0.05. Table V displays the regression results for the INSTITUTION TYPE category. Table V: INSTITUTION TYPE’s Impact on the Gender EARNINGS GAP INSTITUTION TYPE Coefficient p-value 95% CI Private, for-profit 4.75 0.000 [3.69, 5.81] Private, nonprofit 1.00 0.026 [0.12, 1.88] Next, the impact of DEGREE TYPE on the gender EARNINGS GAP will be explored. The coefficients for DEGREE TYPES in the regression are relative to the EARNINGS GAP of a bachelor’s degree. Table VI shows that the gender EARNINGS GAP is higher for all advanced degrees except for Graduate/Professional Certificates but that result is not statistically significant. With doctoral degrees having a 4.23% higher EARNINGS GAP and a master’s degree having 1.74% higher EARNINGS GAP the EARNINGS GAP does not close as women attain advanced degrees beyond a bachelors. Interestingly, the EARNINGS GAP is also 1.34% higher for associate’s degrees but -6.57% lower for an undergraduate certificate. Table VI: DEGREE TYPE’s Impact on the Gender EARNINGS GAP DEGREE TYPE Coefficient p-value 95% CI Associate’s degree 1.34 0.021 [0.21, 2.48] Doctoral Degree 4.23 0.024 [0.56, 7.90] First Professional Degree 5.75 0.005 [1.77, 9.73] Graduate/Professional Certificate -4.29 0.112 [-9.57, 1.00] Master's Degree 1.74 0.000 [0.80, 2.68] Undergraduate Certificate or Diploma -6.57 0.000 [-8.50, -4.64] The last institutional characteristics to explore are TRIBAL COLLEGE and HBCU status. The results for TRIBAL COLLEGES were not statistically significant at a p-vale < 0.05 level. The HBCU status coefficient indicated that colleges that are HBCU’s 17 have a 7.61% lower earnings gap when compared to those that are not HBCU’s. This result is statistically significant at a p-value < .001. Table VII contains the coefficient information. Table VII: HBCU and TRIBAL COLLEGE Status’s Impact on the Gender EARNINGS GAP Institution Type Coefficient p-value 95% CI HBCU -7.61 0.000 [-11.22, -4.00] Tribal College 0.88 0.951 [-26.99, 28.74] Now that institutional characteristics impact has been laid out, analysis will shift to that of degree program (CIPDESC). The base line degree program that all coefficients are relative to is the Business Administration, Management and Operations CIPDESC. There are 104 CIPDESC categories coefficients that are statistically significant at a pvale < 0.05 level. Out of the 104 coefficients that are significant at a p-value < 0.05, seventy five are significant at a p-value < 0.01. Figure III presents a scatter plot of all Figure III: Scatter Plot of the Statistically Significant CIPDESC Programs 18 statistically significant coefficients. There are some outlying data points, but most coefficients are clustered in the -25% to 25% range. The outlying CIPDESC Programs are plotted in Figure IV as a bar chart. The largest coefficient is 38.62 and the smallest is -84.60. There is a large spread in the CIPDESC programs impact on the gender EARNINGS GAP. Program selection in college can have a major impact on earnings equality after graduation. Figure IV shows that religion focused degrees tend to the highest EARNINGS GAPS. The data set has rich information about which EARNINGS GAP coincides to which specific degree programs at specific schools. This aspect of the data set is high dimensional making OLS a bad candidate for analyzing it. Instead of OLS Regression, Lasso Regression will be used to help quantify which schools are leading the way in Figure IV: Seven Highest and Lowest Earnings Gap CIPDESC 19 eliminating the EARNINGS GAP. A Lasso Regression is well suited for this data as its variable selection will help reduce noise and reduce overfitting. For the Lasso Regression the dependent variable is again the gender EARNINGS GAP as a percentage and the independent variables are CIPDESC, INSTITUTION NAME, and DEGREE TYPE. All the independent variables are one hot encoded with the same drop out method as in the OLS Regression. The Lasso regression model has an R-squared value of 0.5714 indicating that it explains 57.14% of the variance in the gender EARNINGS GAP. The Lasso penalty term, alpha, has a value of 0.084. This level of penalization effectively shrinks many coefficients to exactly zero, enhancing model interpretability. Out of 2,471 total predictors, the Lasso model retains 1,334 non-zero coefficients, meaning that 1,137 Figure V: Top 20 Intuitions with the Largest Impact on the Gender Earnings Gap (All Degree Types) 20 variables are either redundant due to multicollinearity or provide limited explanatory power for the gender earnings gap. The first Lasso Regression variable to examine is INSTITUTION NAME. These coefficient values are all relative to the dropped one hot encoding column which was the Arizona State University Campus Immersion Institution. For all degree types I find that Cosmetic Schools decrease the earnings gap while BYU Idaho, Utah Valley and Utah State have the largest increasing impact on the post-graduation EARNINGS GAP. The Top 20 Institutions with the largest magnitude of impact are shown in Figure V. When filtering for schools that grant bachelor’s degrees BYU Idaho, Utah Valley, and Utah State University still have the largest increasing impact on the gender EARNINGS GAP. Interestingly, design schools had the largest impact in reducing the EARNINGS GAP for bachelor’s degree granting universities. These results are plotted in Figure VI. For the Figure VI: Top 15 Intuitions with the Largest Impact on the Gender Earnings Gap (Bachelor’s Degree Granting) 21 remaining DEGREE TYPES the plots are available in the Appendix (Figures VII to XII). CONCLUSION The statistical analysis done in this paper reveals statistically significant patterns in the gender earnings gap across institutions and degree programs three years postgraduation. It was identified that larger schools tend to have a lower earnings gap then smaller ones and ones that do not report their size. Additionally, public universities have lower earnings gaps then their private non for profit and private for-profit counterparts. This statistical analysis has further confirmed the trend of gender earnings gap not closing even for women with advanced degrees. These findings are consistent with existing literature on early career earnings gaps. Previous studies have demonstrated that the earnings gap emerges on entering the labor force, regardless of field of study or occupation in Germany and Finland (Napari, 2008, pg. 4-5; Kunze, 2002, pg. 6). This paper demonstrates that these results also hold in the United States. The persistence of the earnings gap for all degree types including master’s and doctoral programs add additional evidence against the human capital model as even highly specialized programs still have earnings gaps. Previous research had noted that major and occupation choice explained about 60% of the gender wage gap (Sloane, 2021, pg. 24). The results here illustrate that large disparities still exist even within graduates with the same degrees. The degree program level findings shed light onto what the impacts of degree programs are on the earnings gap. This underscores that gender pay differences cannot only be attributed to different educational choices. The Federal Score Card data set enabled a granular analysis of how specific degree programs and institutions impact the early career gender earnings gap. Due to data 22 gaps in the Federal Score Card data, I am not able to do time series analysis to look for any changes in earnings gaps for degree programs or institutions. Right now, the only cohort with three-year post-graduation earnings data at the male and female granularity is the 2019 cohort. If this changes in the future, time series analysis will become possible. As the Federal Score Card continues to collect data on each cohort the earnings gap will be able to be tracked further into careers. In a few years when the five-year postgraduation earnings data may be available updating this analysis could provide interesting insights into how the gender pay gap has changed as individuals progress in their careers. Due to the large number of institutions and degree programs in the data set it is impossible to include all coefficient data in this paper. Future work could aim to deploying an interactive web-based dashboard to allow people to filter and select data of interest to them. This could help potential students, administrators, and policy makers draw value from this data set. Future work could also incorporate gender debt gaps postgraduation as the Federal Score Card data contains this information broken down at the male and female granularity. This could be used to identify trends in which universities and programs are causing differential outcomes in debt in addition to earnings disparities. 23 APPENDIX Figure VII: Top 15 Intuitions with the Largest Impact on the Gender Earnings Gap (Graduate Certificate Granting) Figure VIII: Top 15 Intuitions with the Largest Impact on the Gender Earnings Gap (Undergraduate Certificate Granting) 24 Figure IX: Top 15 Intuitions with the Largest Impact on the Gender Earnings Gap (Associate’s Degree Granting) Figure X: Top 15 Intuitions with the Largest Impact on the Gender Earnings Gap (Doctoral Degree Granting) 25 Figure XI: Top 15 Intuitions with the Largest Impact on the Gender Earnings Gap (First Professional Degree Granting) Figure XII: Top 15 Intuitions with the Largest Impact on the Gender Earnings Gap (Master’s Degree Granting) 26 REFERENCES Aragão, C. (2023, March 1). Gender pay gap in U.S. hasn’t changed much in two decades. Pew Research Center. https://www.pewresearch.org/shortreads/2023/03/01/gender-pay-gapfacts/#:~:text=The%20gender%20gap%20in%20pay,%2D%20and%20part%2Dti me%20workers. Besen-Cassino, Y. (2018). The cost of being a girl working teens and the origins of the gender wage gap. Temple University Press. Bowles, H. (2020, February 12). Why women don’t negotiate their job offers. Harvard Business Review. https://hbr.org/2014/06/why-women-dont-negotiate-their-joboffers Conger, D., & Long, M. C. (2010). Why are men falling behind? gender gaps in college performance and persistence. The ANNALS of the American Academy of Political and Social Science, 627(1), 184–214. https://doi.org/10.1177/0002716209348751 Cook, J. (2024, March 6). The C-suite gender pay gap persists. it’s not about pay discrimination. Morningstar, Inc. https://www.morningstar.com/sustainableinvesting/c-suite-gender-pay-gap-persists-its-not-about-pay-discrimination Corbett, C., & Hill, C. (2012). Graduating to a Pay Gap The Earnings of Women and Men One Year after College Graduation. AAUW. https://files.eric.ed.gov/fulltext/ED536572.pdf 27 Kray, L. J., Kennedy, J. A., & Lee, M. (2024). Now, women do ask: A call to update beliefs about the gender pay gap. Academy of Management Discoveries, 10(1), 7– 33. https://doi.org/10.5465/amd.2022.0021 Kunze, A. (2002). The evolution of the early career gender wage gap. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.300654 Learning, L. (n.d.). College Majors. College Success. https://courses.lumenlearning.com/suny-collegesuccess-lumen1/chapter/collegemajors2/#:~:text=In%20United%20States%20colleges%20and%20universities%2C%20 roughly%202%2C000%20majors%20are%20offered Napari, S. (2008). The early‐career gender wage gap among university graduates in the Finnish Private Sector. LABOUR, 22(4), 697–733. https://doi.org/10.1111/j.14679914.2008.00429.x Robinson, C., & Pope, R. (2023). Human capital theory. Human Capital Theory - an overview | ScienceDirect Topics. https://www.sciencedirect.com/topics/socialsciences/human-capital-theory Sloane, C. M., Hurst, E. G., & Black, D. A. (2021). College majors, occupations, and the gender wage gap. Journal of Economic Perspectives, 35(4), 223–248. https://doi.org/10.1257/jep.35.4.223 Ventre, J., & Min, A. (2022, January). The past, present and future of women in STEM: Pcs edventures!. PCS Edventures. https://edventures.com/blogs/stempower/the- 28 past-present-and-future-of-women-instem#:~:text=The%20Present%3A%20Inspiring%20the%20Next,women%20gra duated%20with%20STEM%20degree Name of Candidate: Owen Koppe Date of Submission: April 23, 2025