Gender inequities throughout STEM: Women with higher grades drop STEM majors while men persist
GGender inequities throughout STEM: Women with higher grades drop STEM majorswhile men persist
Kyle M. Whitcomb and Chandralekha Singh
Department of Physics and Astronomy, University of Pittsburgh, Pittsburgh, PA, 15260 (Dated: April 2, 2020)Efforts to promote equity and inclusion using evidence-based approaches are vital to correct long-standing societal inequities that have disadvantaged women and discouraged them from pursuingstudies, e.g., in many STEM disciplines. We use 10 years of institutional data at a large publicuniversity to investigate trends in the majors that men and women declare, drop after declaring,and earn degrees in as well as the GPA of the students who drop or earn a degree. We find thatthe majors with the lowest number of students also have the highest rates of attrition. Moreover,we find alarming GPA trends, e.g., women who drop majors on average earn higher grades thanmen who drop those majors, and in some STEM majors, women who drop the majors were earningcomparable grades to men who persist in those majors. These quantitative findings call for a betterunderstanding of the reasons students drop a major and for making learning environments equitableand inclusive.
INTRODUCTION AND THEORETICALFRAMEWORK
Increasingly, Science, Technology, Engineering, andMathematics (STEM) departments across the US arefocusing on using evidence to improve the learning ofall students, regardless of their background and mak-ing learning environments equitable and inclusive [1–13].However, women are still severely underrepresented inmany STEM disciplines [14, 15]. In order to understandthe successes and shortcomings of the current state of ed-ucation, the use of institutional data to investigate pastand current trends is crucial. In the past few decades,institutions have been keeping increasingly large digitaldatabases of student records. We have now reached thepoint where there are sufficient data available at manyinstitutions for robust statistical analyses using data an-alytics that can provide invaluable information for trans-forming learning environments and making them moreequitable and inclusive for all students [16, 17]. Studiesutilizing many years of institutional data can lead to anal-yses that were previously limited by statistical power.This is particularly true for studies of performance andpersistence in STEM programs that rely on large samplesizes [4–7, 18–26].In this study, we use 10 year institutional data froma large state-related research university to investigatehow patterns of student major-declaration and subse-quent degree-earning may differ for men and women. Thetheoretical framework for this study has two main foun-dations: critical theory and expectancy value theory.Critical theories, e.g., of race and gender, focus on his-torical sources of inequities within society, that is, soci-etal norms that perpetuate obstacles to the success ofcertain groups of disadvantaged people [3, 27–34]. Crit-ical theory tells us that the dominant group in a societyperpetuates these norms, which are born out of their in-terests, and pushes back against support systems that seek to subvert these norms [28–30]. In our case, criti-cal gender theory provides a historical perspective on themuch-studied gender inequities in STEM.Much important work has been done that relates tocritical theory of gender in STEM education [1–3, 19, 33–41]. One mechanism by which historical societal stereo-types and biases about gender can influence studentchoice of major is proposed by Leslie et al. , who showedthat disciplines with a higher attribution of “brilliance”also have a lower representation of women [42] due to per-vasive stereotypes about men being “brilliant” in thosedisciplines. These brilliance-attributions affect all levelsof STEM education, starting with early childhood wheregirls have already acquired these notions that girls arenot as brilliant as boys [43, 44], which can later influencetheir interest in pursuing certain STEM disciplines [45],and even affect how likely they are to be referred for em-ployment in these disciplines in professional contexts [46].Expectancy value theory (EVT) is another frameworkthat is central to our investigation and states that a stu-dent’s persistence and engagement in a discipline are re-lated to student’s expectancy about their success as wellas how the student values the task [47–49]. In an aca-demic context, “expectancy,” which refers to the individ-ual’s beliefs about their success in the discipline, is closelyrelated to Bandura’s construct of self-efficacy, defined asone’s belief in one’s capability to succeed at a particulartask or subject [47–55].There are four main factors that influence students’expectancy or self-efficacy, namely vicarious experiences(e.g., instructors or peers as role models), social per-suasion (e.g., explicit mentoring, guidance, and sup-port), level of anxiety [50–55], and performance feedback(e.g., via grades on assessment tasks). Women gener-ally have lower self-efficacy than men in many STEMdisciplines because these four factors negatively influ-ence them [1, 2, 33–37]. For example, in many STEMfields women are underrepresented in their classrooms, a r X i v : . [ phy s i c s . e d - ph ] A p r and less likely to have a female role model among thefaculty [14, 15]. Further, the stereotypes surroundingwomen in many STEM disciplines can affect how theyare treated by mentors, even if such an effect is sub-conscious [1, 56–62]. Moreover, women are susceptibleto stress and anxiety from stereotype threat (i.e., thefear of confirming stereotypes about women in manySTEM disciplines) which is not experienced by their malepeers [1, 56–62]. This stress and anxiety can rob themof their cognitive resources, especially during high-stakesassessments such as exams.Expectancy can influence grades earned as well as thelikelihood to persist in a program [50–55]. Stereotypethreats that women in many STEM disciplines experi-ence can increase anxiety in learning and test-taking sit-uations and lead to deteriorated performance. Since anx-iety can increase when performance deteriorates, thesefactors working against women in STEM can force theminto a feedback loop and hinder their performance fur-ther, which can further lower their self-efficacy and cancontinue to affect future performance [50–55].In EVT, value is typically defined as having four facets:intrinsic value (i.e., interest in the task), attainment value(i.e., the importance of the task for the student’s iden-tity), utility value (i.e., the value of the task for futuregoals such as career), and cost (i.e., opportunity cost orpsychological effects such as stress and anxiety) [47–49].In the context of women’s enrollment and persistence inmany STEM disciplines, the societal stereotypes can in-fluence all facets of the students’ value of these STEMdisciplines. Intrinsic value can be informed by societalstereotypes and brilliance-attributions of the STEM dis-ciplines, and attainment and utility values can be fur-ther tempered by these stereotypes. Utility value is animportant facet of student education in STEM, since adegree in a STEM field provides many job opportunitiesfor graduating students. In addition, the psychologicalcost of majoring in these disciplines can be inflated bythe stereotype threat. All of these effects can conspire tosuppress the likelihood of women choosing and/or per-sisting in various STEM disciplines.In order to measure the long-term effects of these sys-temic disadvantages, we investigate the differences in at-trition rates and choices of major of men and womenover the course of their studies at one large public re-search university using 10 years of institutional data.Since these disadvantages to students can be context-dependent, we will consider the attrition rates in manydifferent STEM majors and non-STEM majors in orderto understand the trends in each discipline. Research Questions
Our research questions regarding the relationships be-tween gender, degree attainment, attrition, performance and persistence pertaining to a college degree over a 10year period are as follows.
RQ1.
How many students major in each discipline? Howmany men and women major in each discipline?
RQ2.
Do rates of attrition from the various majors differ?Do rates of attrition from the various majors differfor men and women?
RQ3.
Among those students who drop a given major,what degree, if any, do those students earn? Howdo these trends differ for men and women?
RQ4.
What fraction of declared majors ultimately earn adegree in that major in each STEM subject area?How do these trends differ for men and women?
RQ5.
What are the GPA trends over time among stu-dents who earn a degree in a given major and thosewho drop that major? How do these trends differfor men and women?
METHODOLOGYSample
Using the Carnegie classification system, the univer-sity at which this study was conducted is a public, high-research doctoral university, with balanced arts and sci-ences and professional schools, and a large, primarily res-idential undergraduate population that is full-time andreasonably selective with low transfer-in from other in-stitutions [63].The university provided for analysis the de-identifiedinstitutional data records of students with InternationalReview Board approval. In this study, we examined theserecords for N = 18 ,
319 undergraduate students enrolledin two schools within the university: the School of Engi-neering and the School of Arts and Sciences. This sampleof students includes all of those from ten cohorts who metseveral selection criteria, namely that the students hadfirst enrolled at the university in a Fall semester from Fall2005 to Fall 2014, inclusive, had provided the universitywith a self-reported gender, and the students had eithergraduated and earned a degree, or had not attended theuniversity for at least a year as of Spring 2019. Thissample of students is 49.9% female and had the followingrace/ethnicities: 77.7% White, 11.1% Asian, 6.8% Black,2.5% Hispanic, and 2.0% other or multiracial.
Measures
Gender
In this study, we focus on gender differences in stu-dent trajectories as they progress towards degrees. Weacknowledge that gender is not a binary construct, how-ever in self-reporting their gender to the university stu-dents were given the options of “male” or “female” andso those are the two self-reported genders that we areable to analyze. The student responses to this questionwere included in the institutional data provided by theuniversity. Very few students opted not to provide a gen-der, and so were not considered in this study. We usedthe answers of those students who chose either “male”(“M”) or “female” (“F”) to group students in order tocalculate summary statistics on the measures describedin this section.
Academic Performance
Measures of student academic performance were alsoincluded in the provided data. High school GPA wasprovided by the university on a weighted scale from 0-5 that includes adjustments to the standard 0-4 scalefor Advanced Placement and International Baccalaureatecourses. The data also include the grade points earnedby students in each course taken at the university. Gradepoints are on a 0-4 scale with A = 4, B = 3, C = 2, D = 1,F = 0, where the suffixes “+” and “ − ” add or subtract,respectively, 0 .
25 grade points (e.g., B − = 2 . Declared Major and Degree Earned
For each student, the data include their declared ma-jor(s) in each semester as well as the major(s) in whichthey earned a degree, if any. The data were transformedinto a set of binary flags for each semester, one flag foreach possible STEM major as well as psychology and ageneral non-STEM category for all other majors. A sim-ilar set of flags was created for the degrees earned bystudents. From these flags, we tabulated a number ofmajor-specific measures in each semester, including • current number of declared majors, • number of newly declared majors from the previoussemester, • number of dropped majors from the previoussemester, Major Short LabelChemistry ChemComputer Science CSEngineering EngrMathematics MathPhysics and Astronomy PhysBiological Sciences BioEconomics EconGeology and Other STEMEnvironmental ScienceNeuroscience Other STEMStatistics Other STEMNon-STEM Non-STEMPsychology PsychTABLE I. A list of the majors considered in this study andthe shortened labels used to refer to those majors in tablesand figures. • number of retained majors from the previoussemester.The total number of unique students that ever declaredor dropped a major were also computed. The subset ofstudents that dropped each major were further investi-gated and the major in which they ultimately earned adegree, if any, was determined.Throughout this paper we group the STEM majorsinto two clusters: chemistry, computer science, engineer-ing, mathematics, and physics and astronomy; and bi-ological sciences, economics, geology and environmen-tal science, neuroscience, and statistics. When order-ing majors (i.e., in figures and tables), the majors willbe presented in the order they are listed in the previoussentence (first by group, then alphabetically within eachgroup), followed by non-STEM and psychology. Further,we group the final three STEM majors (geology and en-vironmental science, neuroscience, and statistics) into acategory labeled “Other STEM” for figures and tables.Similarly, “engineering” groups together all engineeringmajors for departments in the School of Engineering atthe studied university. These majors include chemical,computer, civil, electrical, environmental, industrial, andmechanical engineering as well as bioengineering and ma-terials science.Finally, we will make use of shortened labels for themajors in figures and tables. These shortened labels aredefined in Table I. Year of Study
Finally, the year in which the students took each coursewas calculated from the students’ starting term and theterm in which the course was taken. Since the sampleonly includes students who started in fall semesters, each“year” contains courses taken in the fall and subsequentspring semesters, with courses taken over the summeromitted from this analysis. For example, if a studentfirst enrolled in Fall 2007, then their “first year” occurredduring Fall 2007 and Spring 2008, their “second year”during Fall 2008 and Spring 2009, and so on in thatfashion. If a student is missing both a fall and springsemester during a given year but subsequently returns tothe university, the numbering of those post-hiatus yearsis reduced accordingly. If instead a student is only miss-ing one semester during a given year, no corrections aremade to the year numbering.
Analysis
For each student, we calculate their grade point av-erage (GPA) across courses taken in each year of studyfrom their first to sixth years. In addition, we calcu-late the student’s STEM GPA in each year, that is, theGPA in STEM courses alone. The mean GPA as well asthe standard error of the mean is computed for variousgroupings of students [64].Further, proportions of students in various groups (i.e.,grouped by major and/or gender) are calculated alongwith the standard error of a proportion [64]. In particu-lar, the proportions we report are • the proportion of students in each major that aremen or women, • the proportion of men and women, respectively,that declare each subject as a major, • the proportion of declared majors that drop themajor, • the proportion of those who drop each major thatearn a degree in another major, and • the proportion of all declared majors that ulti-mately earn a degree in that major.All analyses were conducted using R [65], making useof the package tidyverse [66] for data manipulation andplotting. RESULTSMajor Declaration Patterns
There are many angles with which we can approach
RQ1 and investigate patterns of student major declara-tion. First, Fig. 1 shows the number of students thatever declared each major. This is done both overall(Fig. 1a) and for female students (Fig. 1b) and malestudents (Fig. 1c) separately. These results provide animportant context for the upcoming analyses that may be partially explained by the number of students in eachmajor.Figures 1b and 1c begin to hint at gender differencesin enrollment patterns, such as a higher proportion ofwomen majoring in non-STEM than men, or a higherproportion of men majoring in engineering than women.These gender patterns are explored further in Fig. 2 bystandardizing the scales in two ways. In Fig. 2a, we con-sider the populations of each major separately and cal-culate the percentage of that population that are men orwomen. This provides insight into what these studentsmight be seeing in the classes for their major. For in-stance, in the biological sciences there is a roughly evensplit, so students in biology classes for biology majorsmight see a classroom that is equally representative ofmen and women. On the other end of the spectrum,around 80% of both computer science and physics andastronomy majors are men.Another way to represent the population of these ma-jors is to consider what percentage of all men or womenchoose each major, as seen in Fig. 2b. While the mainsignal of this plot mimics that of Figs. 1b and 1c, we cannow see the differences noted earlier more clearly. In par-ticular, the clearest differences in this view (Fig. 2b) arein engineering (32% of men and 11% of women declare anengineering major), non-STEM (27% of men and 36% ofwomen declare a non-STEM major), and psychology (6%of men and 16% of women declare a psychology major).Finally, another piece of information about enrollmentpatterns that is missing from Figs. 1 and 2 is when thesestudents declare each major. Figure 3 shows, for eachmajor, the average term in which students added themajor as well as the peak term (that is, the term withthe highest number of new students adding the major).As with Fig. 1, this is done separately for all students(Fig. 3a), female students (Fig. 3b), and male students(Fig. 3c).For the majority of majors in Fig. 3a, the peak of stu-dents adding the major is in the third term (that is, thestart of their second year), with an average of three tofour terms. Two majors, economics and psychology, de-part slightly from this general trend, each with a peakin the fourth term and an average between four and fiveterms. Two other majors, computer science and engi-neering, depart more significantly from the general trendin ways that can be explained by their particular imple-mentation at the studied university.Engineering has a peak in the first term in Fig. 3a,with an average only slightly later. Since all students whoenroll in the School of Engineering are considered “un-declared engineering” majors (the specific sub-disciplinewithin engineering is not assigned in the first year), themajority of engineering students can be identified in theirfirst term. Computer science instead has the latest peakterm in Fig. 3a, namely in the fifth term with a slightlyhigher average. This is due to the structure of the com- ll l ll l l ll l
Major N u m be r o f U n i que S t uden t s (a) All Students ll l ll l l ll l Major N u m be r o f U n i que S t uden t s (b) Female Students ll l ll l l ll l Major N u m be r o f U n i que S t uden t s (c) Male Students FIG. 1. For each major on the horizontal axis, the number of unique students in the sample that ever declared that majoris listed. Since students may change majors or declare multiple majors, some students may contribute to the counts of morethan one major. These counts are calculated separately for (a) all students, (b) female students, and (c) male students. Notethat the scale of the vertical axes differs in (a) compared to (b) and (c). puter science program at the studied university, whichdoes not allow students to declare the major until theyhave completed five of the required courses for the ma-jor. These trends in engineering and computer scienceare important to keep in mind while considering the re-sults presented later in this paper, since in computer sci-ence we are not able to capture attrition that occurs (ofstudents intending to major) during the terms before astudent officially declares a major. Conversely in engi- neering, we are able to capture almost all attrition inthe first year due to the unique enrollment conditionsof engineering students, which is not possible for majorswithin the School of Arts and Sciences (where studentscan declare their major at any time after the first year).Turning then to Figs. 3b and 3c, we see almost identicaltrends as in Fig. 3a. The two exceptions are that thepeak declaration of physics majors for women occurs onesemester earlier in the second term (Fig. 3b). This is ll l ll l l ll l
60% 82% 67%74% 64% 79% 50%51% 73%58%49% 27%42%40% 18% 33%26% 36% 21% 50%
Major P e r c en t age o f S t uden t s i n E a c h M a j o r T ha t a r e M en and W o m en Gender l MenWomen (a) Percentage of Students in Each Major That are Men and Women ll l ll l l ll l
Major P e r c en t age o f M en and W o m en T ha t M a j o r i n E a c h S ub j e c t Gender l MenWomen (b) Percentage of Men and Women That Major in Each Subject
FIG. 2. In (a), the percentages of students in each major that are men or women are calculated (i.e., the percentages in eachcolumn will sum to roughly 100%). In (b), the percentages of men and women that major in each subject are calculated (i.e.,the percentages for each gender group will sum to roughly 100% in this case). Discrepancies in the sum of percentages mayoccur due to rounding the listed percentages to the nearest integer as well as, in (b), students declaring multiple majors. because the overall enrollment in physics is primarily intwo semesters, the second and third, which happens toresult in different peaks but similar averages for men andwomen. Similarly, the peak declaration of mathematicsmajors for men occurs one semester later in the fourthterm (Fig. 3c), again since mathematics majors overallare most likely to declare in the third or fourth term.Apart from these minor differences, these trends in majordeclaration term between men and women are virtuallyidentical.A more detailed accounting of the number of studentsthat enroll in each term for each major are reported in Ta-bles III and IV in Appendix A. Also, summaries of totalnumber of unique students as well as the peak term andnumber of concurrent students in each major, studentsadding each major, and students dropping each majorare available in Tables V, VII, and VI in Appendix B.
Attrition Rates
In order to answer
RQ2 , we further considered pat-terns of attrition rates by gender. In Fig. 4, we considerthe drop rates of different subsets of students (all stu-dents, male students, and female students) in each majoror group of majors. In Fig. 4a, we see that computer sci-ence, non-STEM, and psychology students are the leastlikely to drop their major, while physics and mathemat-ics students are the most likely to drop. We note that the relatively low drop rate of computer science majors couldbe due to the late declaration of the computer sciencemajor seen in Fig. 3. That is, attrition from computerscience prior to when students are allowed to declare themajor is not accounted for in Fig. 4.Though the patterns in each subset of students largelymimic the pattern overall in Fig. 4a, there are a few po-tential exceptions. Figure 4d shows the drop rates bygender without the error bars, which eases comparisonbetween men and women, however the size of the stan-dard error shown in Figs. 4b and 4c indicates that thesedifferences may not be statistically significant. In partic-ular, in physics women are less likely to drop than men,with 31% of female physics majors dropping the major(Fig. 4b) and . Similarly, only 18% of female economicsmajors drop that major, while 27% of male economicsmajors drop that major.
Trajectories of Students After Dropping a Major
After discussing how many students drop each major,we answer
RQ3 by plotting in Fig. 5 where those droppedmajors ended up. In particular, the major indicated inthe legends of Fig. 5a and 5b shows which major wasdropped, while the plot shows the percentage of thosewho dropped that major and ultimately earned a degreein each of the majors on the horizontal axis, includingthe case when“no degree” was earned. For example in ll l ll l l ll l
Major T e r m l Average Add TermPeak Add Term (a) All Students ll l ll l l ll l
Major T e r m l Average Add TermPeak Add Term (b) Female Students ll l ll l l ll l
Major T e r m l Average Add TermPeak Add Term (c) Male Students
FIG. 3. For each major, the term with the peak number of students adding the major in that term is plotted (triangles) aswell as the average term in which students add that major (circles). This is done separately for (a) all students, (b) femalestudents, and (c) male students.
Fig. 5a, we see that among the students that drop thephysics major (indicated by the line color in the legend),roughly 15% of them end up earning a degree in mathe-matics (by looking at this line’s value above “Math” onthe horizontal axis). The figure also shows that the twomost common destinations for those who drop any majoris either no degree or a degree in non-STEM, except fornon-STEM majors who are most likely to earn no degreeor a degree in psychology.Apart from that main signal of dropped STEM ma- jors later earning degrees in non-STEM or leaving theuniversity without a degree, we see a few other interest-ing spikes. For instance, those who drop a physics ma-jor are likely to earn a degree in mathematics (Fig. 5a)and those who drop chemistry and physics (Fig. 5a) aswell as biological science (Fig. 5b) are likely to earn en-gineering degrees. Further, those who drop from the“other STEM” category (geology and environmental sci-ence, neuroscience, and statistics) are likely to major ineconomics and biology (Fig. 5b). While all students who ll l ll l l ll l
Major P e r c en t age o f S t uden t s t ha t D r op (a) All Students ll l ll l l ll l Major P e r c en t age o f S t uden t s t ha t D r op (b) Female Students ll l ll l l ll l Major P e r c en t age o f S t uden t s t ha t D r op (c) Male Students ll l ll l l ll lll l ll l l ll l Major P e r c en t age o f S t uden t s t ha t D r op Gender ll FM (d) Comparison By Gender FIG. 4. For each major, the percentage of students who declared the major but subsequently dropped the major is plottedalong with its standard error. This is done separately for (a) all students, (b) female students, and (c) male students. The plotsfor these men and women are combined into a single plot (d) with the error bars omitted for visibility and ease of comparison,along with lines connecting different points as guides to the eye. drop any major are very likely to earn no degree, the per-centage of dropped majors in this category exceeds 50%for computer science (Fig. 5a), non-STEM, and psychol-ogy.In order to further answer
RQ3 , Fig. 6 plots thesesame proportions of degrees earned by students who dropa major separately for men (Figs. 6a and 6b) and women(Figs. 6c and 6d). We see for the most part very similarpatterns between men and women, with a few notable dif-ferences. For example, among students who drop a math-ematics degree, we see that roughly 23% of the womeneventually earn a degree in economics (Fig. 6c) comparedwith roughly 10% of the men (Fig. 6a). We see a similarpattern with the roles reversed among those students whodrop a major in the Other STEM category (see Table I), with roughly 18% of the men earning a degree in eco-nomics (Fig. 6b) compared with only 5% of the women(Fig. 6d).A few other examples of gender differences in the tra-jectories of those who drop a major are that men are morelikely than women to earn computer science degrees af-ter dropping a chemistry major (Figs. 6a and 6c), andsimilarly men are more likely than women to earn engi-neering degrees after dropping a biological science major(Figs. 6b and 6d). Finally, we note that across all ofFig. 6 in every major except psychology, the women whodrop that major are more likely than the men to earn adegree in another major rather than leaving the univer-sity (that is, the women have a lower rate of earning “NoDegree”). llllllllll lllll llllllllll lllll lllll llllllllll lllll lllll
Earned Degree P e r c en t age o f D r opped M a j o r s DroppedMajor l l ll l
Chem CS EngrMath Phys (a) Students that Drop a Chemistry, Computer Science, Engineering,Mathematics, or Physics Major llllllllll lllll llllllllll lllll lllll llllllllll lllll lllll
Earned Degree P e r c en t age o f D r opped M a j o r s DroppedMajor l l ll l
Bio Econ Non−STEMOtherSTEM Psych (b) Students that Drop a Biology, Economics, Other STEM,Non−STEM or Psychology Major
FIG. 5. Among the students that drop each STEM major as well as psychology and non-STEM majors, the fractions ofstudents that go on to earn a degree in other majors, or who do not earn a degree at all, are plotted along with their standarderror. Dropped majors are grouped into (a) chemistry, computer science, engineering, mathematics, and physics and astronomymajors, and (b) biological science, economics, other STEM, psychology, and non-STEM majors.
A more detailed accounting of the degrees earned bystudents who drop each major is provided in Tables VIII,X, and IX in Appendix C.
Degree-Earning Rates
In order to answer
RQ4 , we investigated how manystudents successfully earn a degree in each major. Fig-ure 7a shows these degree-earning rates for all studentsin each major, while Fig. 7b shows these rates for femalestudents and Fig. 7c for male students. While these arebroadly similar to an inverse of the drop rates in Fig. 4,since some students drop a major and subsequently de-clare the same major again, these degree-earning ratesare a more direct measurement of persistence in a major.Looking first at the overall rates in Fig. 7a, there arefairly wide differences across majors, from the lowest ratein physics of about 66% to the highest in psychologyand non-STEM, each at about 94%. The highest degree-earning rate in STEM occurs in computer science, withabout 88% of declared computer science majors complet-ing the degree requirements. As in Fig. 4, this can beat least partially explained by the requirements prior todeclaring the major, which causes only students who havealready progressed through a significant portion of thecomputer science curriculum to have declared a computerscience major.Considering then the differences for women (Fig. 7b) and men (Fig. 7c), we see relatively few gender differ-ences in these degree-earning rates. While the slightlyhigher completion rate of women in non-STEM and psy-chology or men in chemistry may turn out to be statis-tically significant, the effect sizes appear to be small dueto differences of only 4-5%. As in Fig. 4, the largest dif-ference between men and women seen here appears to bein physics, with 75% of female physics majors earning aphysics degree compared to 63% of male physics majors.However, the large error on these proportions, driven bythe low sample size in physics shown in Fig. 1, makes itdifficult to draw any conclusions from this gender differ-ence in physics degree-earning rates. Similarly, womenare more likely to complete a degree in economics, butagain the size of the standard error prevents any conclu-sive statements about this difference.Across all of Fig. 7, we note that since we have com-bined many majors for the “non-STEM” category, thisis only a measure of the number of non-STEM majorswho successfully earn a degree in any non-STEM major.That is, a student who drops one non-STEM major butearns a degree in a different non-STEM major will stillbe counted as having successfully earned a non-STEMdegree. The same is true for the “other STEM” and “en-gineering” categories which also combine several majors.The high “success rates” of computer science and psy-chology may be due in part to the structure of their pro-gram encouraging students to declare slightly later thanother disciplines, and so this measure may not be captur-0 llllllllll lllll llllllllll lllll lllll llllllllll lllll lllll
Earned Degree P e r c en t age o f D r opped M a j o r s DroppedMajor l l ll l
Chem CS EngrMath Phys (a) Men that Drop a Chemistry, Computer Science, Engineering,Mathematics, or Physics Major llllllllll lllll llllllllll lllll lllll llllllllll lllll lllll
Earned Degree P e r c en t age o f D r opped M a j o r s DroppedMajor l l ll l
Bio Econ Non−STEMOtherSTEM Psych (b) Men that Drop a Biology, Economics, Other STEM,Non−STEM or Psychology Major llllllllll lllll llllllllll lllll lllll llllllllll lllll lllll
Earned Degree P e r c en t age o f D r opped M a j o r s DroppedMajor l l ll l
Chem CS EngrMath Phys (c) Women that Drop a Chemistry, Computer Science, Engineering,Mathematics, or Physics Major llllllllll lllll llllllllll lllll lllll llllllllll lllll lllll
Earned Degree P e r c en t age o f D r opped M a j o r s DroppedMajor l l ll l
Bio Econ Non−STEMOtherSTEM Psych (d) Women that Drop a Biology, Economics, Other STEM,Non−STEM or Psychology Major
FIG. 6. Among the men and women that drop each STEM major as well as psychology and other non-SEM majors, thepercentages of men and women that go on to earn a degree in other majors, or who do not earn a degree at all, are plottedalong with their standard error. Dropped majors are grouped into chemistry, computer science, engineering, mathematics, andphysics and astronomy majors who are (a) men and (c) women, and biological science, economics, other STEM, and psychologymajors who are (b) men and (d) women. ing attrition that happens prior to an official declarationof major (e.g., a student intending to major in a disci-pline decides against it before ever declaring that major).On the other hand, since all students enrolled in the engi-neering school are considered “undeclared engineering,”the relatively low degree-earning rate of engineering re-flects attrition even from the first to the second term,which is not captured for many other majors in which most students have not yet formally declared a major intheir first term. Thus, each reported degree-earning ratehere is a ceiling on the true rate that would include thosestudents who intended to major but never declared.1 ll l ll l l ll l
Major P e r c en t age o f D e c l a r ed M a j o r s T ha t E a r n a D eg r ee i n t ha t M a j o r (a) All Students ll l ll l l ll l Major P e r c en t age o f D e c l a r ed M a j o r s T ha t E a r n a D eg r ee i n t ha t M a j o r (b) Female Students ll l ll l l ll l Major P e r c en t age o f D e c l a r ed M a j o r s T ha t E a r n a D eg r ee i n t ha t M a j o r (c) Male Students FIG. 7. For each major listed on the horizontal axis, the percentages of (a) all students, (b) female students, and (c) malestudents who declare that major and then earn a degree in that major are plotted along with the standard error.
Mean GPA of Degree-Earners vs. Major-Droppers
In order to further our understanding of why studentsmay have dropped a given major and answer
RQ5 , Fig. 8plots the mean GPA of students who declared differentsets of majors and then either earned a degree withinthat set of majors or dropped those majors. Note thatstudents who dropped a major could have gone on toearn a degree with a different major or left the universitywithout a degree. Both overall GPA (Figs. 8a, 8c, and 8e) and STEM GPA (Figs. 8b, 8d, and 8f) are plotted.Across all of Fig. 8, the large drop in sample size fromyear four to five and again from five to six is primarilydue to students graduating.We observe that in general, the students who drop thatmajor have a lower GPA and STEM GPA than studentswho earned a degree in that major. However, the dif-ference between the two groups varies based on whichcluster of majors we consider. For chemistry, computerscience, engineering, mathematics, and physics and as-tronomy majors (Figs. 8a and 8b), those that earned a2 l l l l l ll l l l l l
755 670 587 526 241 933305 3299 3276 3272 1830 405
Year M ean G PA Degree
Dropped MajorEarned Degree in Major (a) Overall GPA: Chemistry, Computer Science, Engineering, Mathematics, and Physics Majors l l l l l ll l l l l l
754 633 448 314 132 493293 3273 3037 2844 1155 185
Year M ean S T E M G PA Degree
Dropped MajorEarned Degree in Major (b) STEM GPA: Chemistry, Computer Science, Engineering, Mathematics, and Physics Majors l l l l l ll l l l l l
Year M ean G PA Degree
Dropped MajorEarned Degree in Major (c) Overall GPA: Biology, Economics, Geology, Neuroscience, and Statistics Majors l l l l l ll l l l l l
Year M ean S T E M G PA Degree
Dropped MajorEarned Degree in Major (d) STEM GPA: Biology, Economics, Geology, Neuroscience, and Statistics Majors l l l l l ll l l l l l
442 429 361 288 146 537624 7593 7509 7479 2510 1072
Year M ean G PA Degree
Dropped MajorEarned Degree in Major (e) Overall GPA: Non−STEM and Psychology Majors l l l l l ll l l l l l
387 328 227 179 70 286939 6078 3358 2545 512 118
Year M ean S T E M G PA Degree
Dropped MajorEarned Degree in Major (f) STEM GPA: Non−STEM and Psychology Majors
FIG. 8. GPA and STEM GPA over time. Each GPA is calculated yearly, not cumulatively. Majors are divided into threegroupings: (a) and (b) chemistry, computer science, engineering, mathematics, and physics; (c) and (d) biology, economics,geology, neuroscience, and statistics; and (e) and (f) non-STEM including psychology. GPA in all courses – (a), (c), and (e) –and in only STEM courses – (b), (d), and (f) – are calculated separately for two categories of students that declared at leastone of the majors in each group: those that ultimately earned a degree in that group of majors and those that dropped fromthat group of majors. For each group, the mean GPA is plotted along with its standard error, with the sample size listed aboveeach point and guides to the eye connecting the points. − is 0.5 grade points. Further, thenumber of students dropping from this set of majors isroughly 19% of the total.For biological science, economics, geology and envi-ronmental science, neuroscience, and statistics majors(Fig. 8c and 8d), those that earned a degree in this setof majors have a GPA of roughly 0.2 grade points higherthan those that dropped, with roughly 26% of majorsdropping. As with the first set of majors, this is consis-tent between overall GPA and STEM GPA, and acrossthe first four years of study.Finally, for non-STEM majors including psychology(Fig. 8e and 8f), the overall GPA disparity widens overtime from roughly 0.4 grade points in the first year toroughly 1.2 grade points in the fourth year, while inSTEM courses the GPA disparity rises from roughly 0.4grade points in the first year to roughly 0.7 grade pointsin the fourth year. Notably, a much smaller fraction ofstudents are dropping from this set – about 5% of thetotal number of students – which could be due in part tothe wide net of considering all non-STEM majors.We further consider the same measures separately formen and women in Fig. 9, with the same subfigure struc-ture as Fig. 8. That is, overall GPA is plotted in Figs. 9a,9c, and 9e and STEM GPA in Figs. 9b, 9d, and 9f. Andstudents belonging to different clusters of majors are con-sidered separately: chemistry, computer science, engi-neering, mathematics, and physics and astronomy majorsare included in Figs. 9a and 9b; biological science, eco-nomics, geology and environmental science, neuroscience,and statistics majors in Figs. 9c and 9d; and non-STEMand psychology majors in Figs. 9e and 9f.In all cases, the main finding is that women are earninghigher grades on average than men, both among thosewho drop a given major and those who earn a degreein that major. However, the grade differences betweenmen and women who dropped the major and men andwomen who earned a degree in the major are inconsis-tent. In particular, we see that during the first four yearsin Fig. 9b that among those who drop one of the listedSTEM majors, women earn 0.4 to 0.6 grade points higherthan men on average. More specifically, these womenhave roughly a B − average while the men have a C+average. Comparing this to those who earn a degree inthose majors (Fig. 9b), where women earn only 0.1 to 0.2grade points higher than men on average, we see a trendwhere, on average, men will drop a major if they have sig-nificantly lower grades than their peers while women aredropping their major with only somewhat lower grades than their peers.This trend is also reflected in the overall GPA of thesame population (Fig. 9a) as well as the overall andSTEM GPA of the remainder of STEM majors (Figs. 9cand 9d) and non-STEM majors including psychology(Figs. 9e and 9f). The trend is even more pronouncedamong students in the second cluster of STEM majorswhere in both overall GPA (Fig. 9c) and STEM GPA(Fig. 9d), the women who drop those majors have, onaverage, the same GPA as the men who earn a degree inthose majors.In order to test these differences, we report in Ta-bles IIa and IIb the results from four comparisons. Wecompare the STEM GPA of the women who drop eachset of STEM majors (Figs. 9b and 9d) with the STEMGPA of men who both drop (Table IIa) and earn a degree(Table IIb) in the same majors. These comparisons areperformed with two measures: the p -value from a two-tailed t -test and the effect size from Cohen’s d [64, 67].These tests were not performed for non-STEM majorssince such a low percentage of the students, roughly 5%,dropped from the non-STEM majors (including psychol-ogy) altogether (Fig. 8).Considering the STEM GPA of the first set of STEMmajors in Table IIa, namely chemistry, computer sci-ence, engineering, mathematics, and physics majors (cor-responding to Fig. 9b), we see again that the women whodrop these majors are earning statistically significantlyhigher grades on average than the men who drop thesemajors ( p < . d = 0 .
58 indi-cates a medium effect size (i.e., | d | ≥ .
50) [67], show-ing that the women who drop are earning meaningfullyhigher grades on average than the men who drop. Ta-ble IIb compares the women who drop with the men whopersist and earn degrees in the majors, and we again seea statistically significant gender difference ( p < . d = − .
65) [67], now withthe men having higher grades on average. Notably, theeffect sizes are similar in magnitude in both tables, whichis consistent with these women who drop the major hav-ing an average STEM GPA roughly midway between themen who drop and the men who persist.We turn then to the gendered STEM GPA differencesof the other group of majors in Tables IIa and IIb,namely biology, economics, geology, neuroscience, andstatistics majors (corresponding to Fig. 9d). We see asimilar pattern among the students who drop these ma-jors (Table IIa) as in the other set of majors: the womenearn statistically significantly higher grades on average( p < . d = 0 . p = 0 .
047 indicating only marginal statistical signif-icance for an effect size ( d = − .
09) that does not reach4 l l l l l ll l l l l ll l l l l ll l l l l l
527 460 400 355 177 64228 210 187 171 64 292425 2420 2410 2400 1372 306880 879 866 872 458 99
Year M ean G PA Degree
Men: DroppedMajor Men: EarnedDegree in MajorWomen: DroppedMajor Women: EarnedDegree in Major (a) Overall GPA: Chemistry, Computer Science, Engineering, Mathematics, and Physics Majors l l l l l ll l l l l ll l l l l ll l l l l l
526 432 304 214 103 41228 201 144 100 29 82418 2399 2230 2096 874 149875 874 807 748 281 36
Year M ean S T E M G PA Degree
Men: DroppedMajor Men: EarnedDegree in MajorWomen: DroppedMajor Women: EarnedDegree in Major (b) STEM GPA: Chemistry, Computer Science, Engineering, Mathematics, and Physics Majors l l l l l ll l l l l ll l l l l ll l l l l l
Year M ean G PA Degree
Men: DroppedMajor Men: EarnedDegree in MajorWomen: DroppedMajor Women: EarnedDegree in Major (c) Overall GPA: Biology, Economics, Geology, Neuroscience, and Statistics Majors l l l l l ll l l l l ll l l l l ll l l l l l
994 890 740 563 195 653234 711 622 458 102 293198 2952 2745 1009 1917371657 1642 1539 1449 369 54
Year M ean S T E M G PA Degree
Men: DroppedMajor Men: EarnedDegree in MajorWomen: DroppedMajor Women: EarnedDegree in Major (d) STEM GPA: Biology, Economics, Geology, Neuroscience, and Statistics Majors l l l l l ll l l l l ll l l l l ll l l l l l
252 247 212 171 91 2033190 182 149 117 552845 2826 2799 2788 1047 3834779 4767 4710 4691 1463 689
Year M ean G PA Degree
Men: DroppedMajor Men: EarnedDegree in MajorWomen: DroppedMajor Women: EarnedDegree in Major (e) Overall GPA: Non−STEM and Psychology Majors l l l l l ll l l l l ll l l l l ll l l l l l
221 184 128 105 45 20166 144 99 74 25 82582 2249 1343 1060 269 63554357 3829 2015 1485 243
Year M ean S T E M G PA Degree
Men: DroppedMajor Men: EarnedDegree in MajorWomen: DroppedMajor Women: EarnedDegree in Major (f) STEM GPA: Non−STEM and Psychology Majors
FIG. 9. GPA and STEM GPA over time by gender. Each GPA is calculated yearly, not cumulatively. Majors are dividedinto three groupings: (a) and (b) chemistry, computer science, engineering, mathematics, and physics; (c) and (d) biology,economics, geology, neuroscience, and statistics; and (e) and (f) non-STEM including psychology. GPA in all courses – (a), (c),and (e) – and in only STEM courses – (b), (d), and (f) – are calculated separately for four categories of students that declaredat least one of the majors in each group: men and women that ultimately earned a degree in that group of majors and menand women that dropped from that group of majors. For each group, the mean GPA is plotted along with its standard error,with the sample size listed above each point and guides to the eye connecting the points. (a) Women who Dropped Men who Dropped Statisticalthe Major the Major ComparisonsFig.
N M SD N M SD p d
9b 228 2.73 0.73 527 2.29 0.81 < .
001 0.589d 744 3.10 0.64 1004 2.70 0.83 < .
001 0.54 (b)
Women who Dropped Men who Earned Statisticalthe Major the Degree ComparisonsFig.
N M SD N M SD p d
9b 228 2.73 0.73 2425 3.14 0.52 < .
001 -0.659d 744 3.10 0.64 3257 3.15 0.52 0.047 -0.09TABLE II. A comparison of the cumulative STEM GPA of women who drop STEM majors with (a) the men who drop thesame STEM majors and (b) the men who persist and earn a degree in the same STEM majors. Along with the number ofstudents ( N ), we report the mean ( M ) and standard deviation ( SD ) of cumulative STEM GPA for each group. The p -valuefrom a two-tailed t -test is reported comparing the women and men, along with Cohen’s d measuring the effect size of thegender difference (the sign of d matches the sign of the mean GPA for women minus the mean GPA for men). The comparisonis performed separately for two clusters of STEM majors corresponding to the indicated figure, i.e., for Fig. 9b we considerchemistry, computer science, engineering, mathematics, and physics majors; for Fig. 9d we consider biology, economics, geology,neuroscience, and statistics majors. the threshold for “small” effect sizes (i.e., | d | ≥ .
20) [67].Thus, the women who are dropping these majors have,on average, the same grades as the men who persist andstatistically significantly higher grades than the men whodrop.
DISCUSSION
In this section, we will begin by discussing the gen-eral trends (i.e., setting aside the gender differences), andthen follow up with a discussion of the gender differences.
General Enrollment Patterns
Despite large differences in the number of students en-rolling in different STEM disciplines at the studied uni-versity (Fig. 1a), there are broadly similar patterns ofwhen those students declare the major (Fig. 3a), withsome some exceptions (i.e., engineering and computerscience due to the constraints on when a student can de-clare a major). However, there are notable differencesin the attrition of students from the different majors(Fig. 4a), and the corresponding degree-earning rates(Fig. 7a). Notably, a few STEM disciplines stand outas having particularly high rates of attrition (or lowrates of degree completion for students who declaredthose majors),e.g., mathematics and physics. This isconsistent with the Leslie et al. study which identifiesmathematics and physics as the two STEM disciplineswith the highest “ability belief” (i.e., emphasis on bril-liance) [42]. This trend of high attrition rates is partic-ularly problematic for mathematics and physics, sincethese disciplines recruit very few students in the firstplace (Fig. 1). Moreover, mathematics and physics are also two disciplines with deeply hierarchical knowledgestructures, which could influence student decision mak-ing, e.g., whether to leave the discipline after unsatisfac-tory experiences in earlier courses.
Gendered Enrollment Patterns
The most notable example of gender differences in en-rollment patterns observed in our analysis is in Fig. 2.In biological science, geological and environmental sci-ence, neuroscience, and statistics we see a balanced rep-resentation of men and women. However, we see an un-derrepresentation of women in chemistry, computer sci-ence, engineering, mathematics, physics, and economics,and a corresponding overrepresentation of women in non-STEM including psychology (Fig. 2a). Again, the resultsobserved here are roughly consistent with those observedby Leslie et al. [42]. The gender imbalance in these STEMdisciplines itself plays a pernicious role in recruitmentand retention of women who do not have many role mod-els and also affects the performance of women, who areconstantly forced to prove themselves and counter thesocietal stereotypes working against them in these fields.Furthermore, despite these differences in the number ofmen and women in these STEM disciplines, we see veryfew differences in the remainder of the enrollment mea-sures, including the time of major-declaration (Figs. 3band 3c), rates of attrition (Figs. 4b and 4c), and ratesof degree-attainment (Figs. 7b and 7c). There are somehints towards differing rates of attrition in physics andeconomics but these differences suffer from large standarderror, particularly in physics, due to a low sample size.Moreover, we hypothesize that the higher attrition rate ofmen in physics and economics may be due to the pressurethat women experience not to enroll in these disciplines6in the first place, which would have occurred prior to thedeclaration of majors and instead manifests in the starkunderrepresentation of women in these disciplines notedearlier (Fig. 2a). There have been many studies thatfind highly problematic gender inequities in introductoryphysics and mathematics that could at least partly con-tribute to this underrepresentation [2, 24, 68, 69].
Trajectories of Students Who Drop a Major
As with the attrition and degree-earning rates, we seebroadly similar patterns between men and women whodrop the various STEM majors (Fig. 6). One notableexceptions to this are that women who drop a mathe-matics major are more likely to instead earn a degreein economics than men who drop a mathematics major.Moreover, we find that across all disciplines (includingnon-STEM majors) except psychology, women who dropa major are more likely to subsequently earn a degree ina different major at the same university than men whodrop a major (see the “No Degree” entries in Fig. 6), whoare more likely to leave the university altogether, eitherby dropping out of college completely or transferring toanother university. This may be easier to see in Tables IXand X in Appendix C.
Gendered GPA Differences
We find pervasive and deeply troubling genderedtrends in the overall GPA and STEM GPA of those stu-dents who drop different STEM majors. In particular,Fig. 9 shows that the women who drop a STEM majorhave a higher average GPA than the men who drop aSTEM major, and these differences are shown to be sta-tistically significant as shown in Table IIa. This impliesthat on average, women are more likely to drop thesemajors with a significantly better GPA than men. Inchemistry, computer science, engineering, mathematics,and physics and astronomy (Fig. 9a and 9b), the womendropping these majors have an average GPA (B − to B)that is halfway between the men who drop these majors(C+ to B − ) and the men who earn degrees with thesemajors (B to B+). This trend is even worse in biology,economics, geological and environmental science, neuro-science, and statistics (Fig. 9c and 9d), where the womenwho drop these majors have the same GPA as the menwho earn degrees in these majors, both around B to B+(see Table IIb). Thus, on average, among students withthe same GPA, the women in all STEM majors are morelikely to drop the majors than the men, who are morelikely to persist. It is important to note that this is eventrue among the STEM majors with gender imbalancesin the population as well as those majors without imbal-ances. While it is true that women at the studied universityon average earn higher grades than men in most STEMcourses, that does not explain why women are choosingto drop STEM majors with the same grades as men whopersist. The difference must then be coming from anothersource, for example inequitable and non-inclusive learn-ing environments and lack of mentoring and support forwomen who may have lower self-efficacy [50–55]. Womenmay also have a lower sense of belonging and value per-taining to remaining in these disciplines [26, 47–49] iflearning environments are not equitable and inclusive,especially because they must bear the high cost of man-aging the burden of societal stereotypes and the ensuingstereotype threats. We hypothesize that the brilliance at-tributions of STEM disciplines [42] and who is likely toexcel in them could influence women away from a disci-pline in which they could have succeeded. Thus, withoutexplicit effort to improve the learning environments inthese disciplines, the culture and stereotypes surround-ing STEM in general may be creating an environment inwhich women are being unfairly driven out of these fieldsin which they could have thrived while their male coun-terparts are not subjected to these same pressures andare persisting with worse performance. Limitations and Implications
One limitation of this study is that physics which hasthe most consistently problematic gender differences alsohas the lowest number of students. Future studies canmake use of more data (either data available further backin time or as more data continues to accumulate) to ex-plore the gender differences in physics better. This studyalso limits its considerations to gender. Other studieswith larger data sets could investigate how other under-served populations are being left behind in STEM, suchas underrepresented minority students, first-generationcollege students, or low-income students.A critically important extension of this work wouldbe for other institutions of different types and sizes todo similar analyses in order to broaden the wealth ofknowledge available and continue to work towards thegoal of equitable and inclusive education. Other institu-tions noting similar highly problematic trends can helppinpoint common sources of inequities, while institutionsthat do not observe these trends may be able to identifyhow they have structured their programs to avoid theseinequitable trends. Studies such as this can thus pro-vide a framework for other institutions to perform similaranalyses, and for particular departments to understandhow their own trends differ from those of other depart-ments at their own university. For instance, our findingshere for, e.g., physics and mathematics drop out ratesindicate that there is substantial room to improve theirsupport of their intended majors so they do not drop7out. Focus on increasing equity and inclusion in learn-ing is especially important in the early courses for thesemajors, since they are fraught with problematic genderdifferences and may be contributing to the underrepre-sentation of women in these majors in the first place.Similar steps should be taken for the other majors thathave low representation of women, especially computerscience and engineering, and to a lesser extent chemistry.The other STEM majors (biological sciences, eco-nomics, geology and environmental science, neuroscience,and statistics) should also take a closer look at whichstudents are choosing to leave those disciplines, and whywomen choosing to leave have similar GPAs as the menwho persist and earn degrees in those disciplines. Theseare signs of inequity even in majors in which women arenot underrepresented. All of these issues should be ad-dressed since they are critical for improving equity andinclusion in higher education STEM learning environ-ments.
ACKNOWLEDGMENTS
This research is supported by the National ScienceFoundation Grant DUE-1524575 and the Sloan Founda-tion Grant G-2018-11183. [1] A. Johnson, The mathematics of sex: How biology andsociety conspire to limit talented women and girls, Sci-ence Education , 960 (2012).[2] A. Johnson, M. Ong, L. T. Ko, J. Smith, and A. Hodari,Common challenges faced by women of color in physics,and actions faculty can take to minimize those challenges,The Physics Teacher , 356 (2017).[3] H. Metcalf, D. Russell, and C. Hill, Broadening the sci-ence of broadening participation in stem through crit-ical mixed methodologies and intersectionality frame-works, American Behavioral Scientist , 580 (2018),https://doi.org/10.1177/0002764218768872.[4] B. King, Does postsecondary persistence in stem varyby gender?, AERA Open , 2332858416669709 (2016),https://doi.org/10.1177/2332858416669709.[5] A. V. Maltese and R. H. Tai, Pipeline persistence: Ex-amining the association of educational experiences withearned degrees in STEM among US students, Science Ed-ucation , 877 (2011).[6] A. V. Maltese and C. S. Cooper, Stem path-ways: Do men and women differ in why they enterand exit?, AERA Open , 2332858417727276 (2017),https://doi.org/10.1177/2332858417727276.[7] B. Means, H. Wang, X. Wei, E. Iwatani, andV. Peters, Broadening participation in stem collegemajors: Effects of attending a stem-focused highschool, AERA Open , 2332858418806305 (2018),https://doi.org/10.1177/2332858418806305.[8] M. Borrego, R. A. Streveler, R. L. Miller, and K. A.Smith, A new paradigm for a new field: Communicating representations of engineering education research, Jour-nal of Engineering Education , 147 (2008).[9] M. Borrego and J. Bernhard, The emergence of engineer-ing education research as an internationally connectedfield of inquiry, Journal of Engineering Education ,14 (2011).[10] M. Borrego and C. Henderson, Increasing the use ofevidence-based teaching in STEM higher education: Acomparison of eight change strategies, Journal of Engi-neering Education , 220 (2014).[11] C. Henderson and M. H. Dancy, Physics faculty and ed-ucational researchers: Divergent expectations as barri-ers to the diffusion of innovations, American Journal ofPhysics , 79 (2008).[12] M. Dancy and C. Henderson, Pedagogical practices andinstructional change of physics faculty, American Journalof Physics , 1056 (2010).[13] C. Henderson, M. Dancy, and M. Niewiadomska-Bugaj,Use of research-based instructional strategies in intro-ductory physics: Where do faculty leave the innovation-decision process?, Phys. Rev. ST Phys. Educ. Res. ,020104 (2012).[14] National Student Clearinghouse Research Center, Sci-ence & Engineering Degree Attainment: 2004 –2014 , (National Student Clearinghouse, 2015) avail-able at https://nscresearchcenter.org/snapshotreport-degreeattainment15/.[15] National Science Board,
Science and EngineeringIndicators 2018
Learning Analytics (Springer,2014) pp. 61–75.[17] Z. Papamitsiou and A. A. Economides, Learning analyt-ics and educational data mining in practice: A systematicliterature review of empirical evidence, Journal of Edu-cational Technology & Society , 49 (2014).[18] M. W. Ohland, S. D. Sheppard, G. Lichtenstein, O. Eris,D. Chachra, and R. A. Layton, Persistence, engagement,and migration in engineering programs, Journal of Engi-neering Education , 259 (2008).[19] S. M. Lord, M. M. Camacho, R. A. Layton, R. A. Long,M. W. Ohland, and M. H. Wasburn, Who’s persistingin engineering? A comparative analysis of female andmale Asian, black, Hispanic, Native American, and whitestudents, Journal of Women and Minorities in Scienceand Engineering , 167 (2009).[20] O. Eris, D. Chachra, H. L. Chen, S. Sheppard, L. Ludlow,C. Rosca, T. Bailey, and G. Toye, Outcomes of a longi-tudinal administration of the persistence in engineeringsurvey, Journal of Engineering Education , 371 (2010).[21] Y. Min, G. Zhang, R. A. Long, T. J. Anderson, andM. W. Ohland, Nonparametric survival analysis of theloss rate of undergraduate engineering students, Journalof Engineering Education , 349 (2011).[22] S. M. Lord, R. A. Layton, and M. W. Ohland, Multi-institution study of student demographics and outcomesin electrical and computer engineering in the USA, IEEETransactions on Education , 141 (2015).[23] M. W. Ohland and R. A. Long, The multiple-institutiondatabase for investigating engineering longitudinal devel-opment: An experiential case study of data sharing andreuse, Advances in Engineering Education , 1 (2016). [24] R. L. Matz, B. P. Koester, S. Fiorini, G. Grom, L. Shep-ard, C. G. Stangor, B. Weiner, and T. A. McKay, Pat-terns of gendered performance differences in large intro-ductory courses at five research universities, AERA Open , 1 (2017).[25] E. B. Witherspoon and C. D. Schunn, Locating and un-derstanding the largest gender differences in pathways toscience degrees, Science Education , 144 (2019).[26] N. Safavian, What makes them persist? expectancy-value beliefs and the math participation, per-formance, and preparedness of hispanic youth,AERA Open , 2332858419869342 (2019),https://doi.org/10.1177/2332858419869342.[27] M. Gopalan and A. A. Nelson, Understand-ing the racial discipline gap in schools,AERA Open , 2332858419844613 (2019),https://doi.org/10.1177/2332858419844613.[28] K. Crenshaw, N. Gotanda, G. Peller, and K. Thomas, Critical Race Theory: The Key Writings that Formed theMovement (The New Press, 1995).[29] D. Kellner, Toward a critical theory of education, Democ-racy & Nature , 51 (2003).[30] T. J. Yosso, Whose culture has capital? A critical racetheory discussion of community cultural wealth, RaceEthnicity and Education , 69 (2005).[31] R. Guti´errez, Framing equity: Helping students “play thegame” and “change the game”, Teaching for Excellenceand Equity in Mathematics , 4 (2009).[32] E. Taylor, D. Gillborn, and G. Ladson-Billings, Foun-dations of critical race theory in education (Routledge,2009).[33] S. Tolbert, A. Schindel, and A. J. Rodriguez, Relevanceand relational responsibility in justice-oriented scienceeducation research, Science Education , 796 (2018).[34] K. Schenkel and A. Calabrese Barton, Critical scienceagency and power hierarchies: Restructuring powerwithin groups to address injustice beyond them, ScienceEducation (2020).[35] M. Bang and D. Medin, Cultural processes in scienceeducation: Supporting the navigation of multiple episte-mologies, Science Education , 1008 (2010).[36] M. Estrada, A. Eroy-Reveles, and J. Matsui, The influ-ence of affirming kindness and community on broadeningparticipation in STEM career pathways, Social Issues andPolicy Review , 258 (2018).[37] M. Ong, J. M. Smith, and L. T. Ko, Counterspaces forwomen of color in STEM higher education: Marginal andcentral spaces for persistence and success, Journal of Re-search in Science Teaching , 206 (2018).[38] C. Seron, S. S. Silbey, E. Cech, and B. Rubineau, Per-sistence is cultural: Professional socialization and the re-production of sex segregation, Work and Occupations ,178 (2016).[39] C. E. Brawner, M. M. Camacho, S. M. Lord, R. A. Long,and M. W. Ohland, Women in industrial engineering:Stereotypes, persistence, and perspectives, Journal of En-gineering Education , 288 (2012).[40] C. E. Brawner, S. M. Lord, R. A. Layton, M. W. Ohland,and R. A. Long, Factors affecting women’s persistence inchemical engineering, International Journal of Engineer-ing Education , 1431 (2015).[41] C. M. Ganley, C. E. George, J. R. Cimpian, andM. B. Makowski, Gender equity in college ma-jors: Looking beyond the stem/non-stem dichotomy for answers regarding female participation, Ameri-can Educational Research Journal , 453 (2018),https://doi.org/10.3102/0002831217740221.[42] S.-J. Leslie, A. Cimpian, M. Meyer, and E. Freeland,Expectations of brilliance underlie gender distributionsacross academic disciplines, Science , 262 (2015).[43] L. Bian, S.-J. Leslie, and A. Cimpian, Gender stereotypesabout intellectual ability emerge early and influence chil-dren’s interests, Science , 389 (2017).[44] L. Bian, The roots of gender gaps: investigating thedevelopment of gender stereotypes about intelligence ,Ph.D. thesis, University of Illinois at Urbana-Champaign(2017).[45] L. Bian, S.-J. Leslie, M. C. Murphy, and A. Cimpian,Messages about brilliance undermine women’s interestin educational and professional opportunities, Journal ofExperimental Social Psychology , 404 (2018).[46] L. Bian, S.-J. Leslie, and A. Cimpian, Evidence of biasagainst girls and women in contexts that emphasize in-tellectual ability, American Psychologist , 1139 (2018).[47] J. Eccles (Parsons), T. Adler, and J. Meece, Sex differ-ences in achievement: A test of alternative theories, Jour-nal of Personality and Social Psychology , 26 (1984).[48] J. S. Eccles, J. E. Jacobs, and R. D. Harold, Genderrole stereotypes, expectancy effects, and parents’ social-ization of gender differences, Journal of Social Issues ,183 (1990).[49] J. S. Eccles, Understanding women’s educational and oc-cupational choices: Applying the Eccles et al. modelof achievement-related choices, Psychology of WomenQuarterly , 585 (1994).[50] A. Bandura, Social cognitive theory of self-regulation,Organizational Behavior and Human Decision Processes , 248 (1991).[51] A. Bandura, Self-efficacy, in Encyclopedia of Psychology ,Vol. 3, edited by R. J. Corsini (Wiley, 1994) pp. 368–369.[52] A. Bandura,
Self-efficacy: The Exercise of Control (Macmillan, 1997).[53] A. Bandura, Social cognitive theory of personality, in
Handbook of Personality , Vol. 2, edited by L. A. Per-vin and O. P. John (Oxford University Press, 1999) pp.154–196.[54] A. Bandura, Social cognitive theory: An agentic perspec-tive, Annual Review of Psychology , 1 (2001).[55] A. Bandura, The evolution of social cognitive theory, in Great Minds in Management , edited by K. G. Smith andM. A. Hitt (Oxford University Press, 2005) pp. 9–35.[56] A. W. Astin,
What Matters in College , Vol. 9 (Jossey-Bass, 1993).[57] R. M. Felder, G. N. Felder, M. Mauney, C. E. Hamrin Jr.,and E. J. Dietz, A longitudinal study of engineering stu-dent performance and retention. III. Gender differencesin student performance and attitudes, Journal of Engi-neering Education , 151 (1995).[58] R. M. Felder, G. N. Felder, and E. J. Dietz, A longitudi-nal study of engineering student performance and reten-tion. V. Comparisons with traditionally-taught students,Journal of Engineering Education , 469 (1998).[59] S. Cheryan, S. A. Ziegler, A. K. Montoya, and L. Jiang,Why are some STEM fields more gender balanced thanothers?, Psychological Bulletin , 1 (2017).[60] J. A. Bianchini, D. J. Whitney, T. D. Breton, and B. A.Hilton-Brown, Toward inclusive science education: Uni-versity scientists’ views of students, instructional prac- tices, and the nature of science, Science Education ,42 (2002).[61] S. L. Britner and F. Pajares, Sources of science self-efficacy beliefs of middle school students, Journal of Re-search in Science Teaching: The Official Journal of theNational Association for Research in Science Teaching , 485 (2006).[62] A. Hilts, R. Part, and M. L. Bernacki, The roles of socialinfluences on student competence, relatedness, achieve-ment, and retention in STEM, Science Education ,744 (2018).[63] Indiana University Center for Postsecondary Research, The Carnegie Classification of Institutions of Higher Ed-ucation , Tech. Rep. (Indiana University Center for Post-secondary Research, Bloomington, IN, 2018).[64] D. Freedman, R. Pisani, and R. Purves,
Statistics , 4thed. (W. W. Norton & Co., 2007).[65] R Core Team,
R: A Language and Environment for Sta-tistical Computing , R Foundation for Statistical Comput-ing, Vienna, Austria (2019).[66] H. Wickham, tidyverse: Easily Install and Load the ‘tidy-verse’ (2017), R package version 1.2.1.[67] J. Cohen,
Statistical Power Analysis for the BehavioralSciences , 2nd ed. (Lawrence Erlbaum Associates, 1988).[68] E. Marshman, Z. Y. Kalender, C. Schunn, T. Nokes-Malach, and C. Singh, A longitudinal analysis ofstudents’ motivational characteristics in introductoryphysics courses: Gender differences, Canadian Journalof Physics , 391 (2017).[69] E. M. Marshman, Z. Y. Kalender, T. Nokes-Malach,C. Schunn, and C. Singh, Female students with A’s havesimilar physics self-efficacy as male students with C’s inintroductory courses: A cause for alarm?, Phys. Rev.Phys. Educ. Res. , 020123 (2018). APPENDIX A: NUMBER OF CURRENT MAJORS AND THOSE WHO ADDED OR DROPPED EACHMAJOR BY TERM (a)
Chemistry , N unique = 1090Term Number of Majors [% of N unique ]Current Added Dropped1 15 [1.4] 15 [1.4] 1 [0.1]2 102 [9.4] 87 [8.0] 1 [0.1]3 821 [75.3] 724 [66.4] 10 [0.9]4 912 [83.7] 150 [13.8] 61 [5.6]5 942 [86.4] 71 [6.5] 42 [3.9]6 926 [85.0] 24 [2.2] 41 [3.8]7 910 [83.5] 14 [1.3] 25 [2.3]8 884 [81.1] 1 [0.1] 15 [1.4]9 414 [38.0] 6 [0.6] 10 [0.9]10 161 [14.8] 1 [0.1] 7 [0.6]11 32 [2.9] 0 [0.0] 5 [0.5]12 16 [1.5] 1 [0.1] 4 [0.4](c) Engineering , N unique = 3575Term Number of Majors [% of N unique ]Current Added Dropped1 3228 [90.3] 3228 [90.3] 47 [1.3]2 2980 [83.4] 14 [0.4] 262 [7.3]3 2968 [83.0] 159 [4.4] 171 [4.8]4 2902 [81.2] 55 [1.5] 121 [3.4]5 2902 [81.2] 64 [1.8] 64 [1.8]6 2901 [81.1] 27 [0.8] 28 [0.8]7 2881 [80.6] 21 [0.6] 40 [1.1]8 2847 [79.6] 5 [0.1] 25 [0.7]9 1609 [45.0] 4 [0.1] 15 [0.4]10 636 [17.8] 1 [0.0] 15 [0.4]11 114 [3.2] 0 [0.0] 8 [0.2]12 49 [1.4] 0 [0.0] 7 [0.2] (b) Computer Science , N unique = 621Term Number of Majors [% of N unique ]Current Added Dropped1 1 [0.2] 1 [0.2] 0 [0.0]2 20 [3.2] 19 [3.1] 0 [0.0]3 100 [16.1] 81 [13.0] 1 [0.2]4 214 [34.5] 116 [18.7] 5 [0.8]5 395 [63.6] 188 [30.3] 8 [1.3]6 459 [73.9] 74 [11.9] 10 [1.6]7 502 [80.8] 65 [10.5] 17 [2.7]8 495 [79.7] 27 [4.3] 12 [1.9]9 241 [38.8] 37 [6.0] 10 [1.6]10 147 [23.7] 11 [1.8] 11 [1.8]11 69 [11.1] 2 [0.3] 10 [1.6]12 36 [5.8] 2 [0.3] 2 [0.3](d) Mathematics , N unique = 373Term Number of Majors [% of N unique ]Current Added Dropped1 24 [6.4] 24 [6.4] 0 [0.0]2 91 [24.4] 68 [18.2] 1 [0.3]3 185 [49.6] 102 [27.3] 10 [2.7]4 259 [69.4] 95 [25.5] 23 [6.2]5 270 [72.4] 36 [9.7] 26 [7]6 286 [76.7] 27 [7.2] 13 [3.5]7 278 [74.5] 14 [3.8] 15 [4.0]8 264 [70.8] 4 [1.1] 13 [3.5]9 81 [21.7] 8 [2.1] 9 [2.4]10 61 [16.4] 1 [0.3] 2 [0.5]11 18 [4.8] 1 [0.3] 6 [1.6]12 7 [1.9] 0 [0.0] 4 [1.1](e) Physics & Astronomy , N unique = 186Term Number of Majors [% of N unique ]Current Added Dropped1 17 [9.1] 17 [9.1] 0 [0.0]2 74 [39.8] 58 [31.2] 1 [0.5]3 124 [66.7] 62 [33.3] 14 [7.5]4 151 [81.2] 34 [18.3] 8 [4.3]5 148 [79.6] 8 [4.3] 12 [6.5]6 144 [77.4] 4 [2.2] 8 [4.3]7 134 [72.0] 2 [1.1] 12 [6.5]8 128 [68.8] 1 [0.5] 6 [3.2]9 46 [24.7] 1 [0.5] 4 [2.2]10 38 [20.4] 0 [0.0] 2 [1.1]11 3 [1.6] 0 [0.0] 1 [0.5]12 1 [0.5] 0 [0.0] 2 [1.1]TABLE III. For each term from 1 to 12, the current number of declared majors (“Current”) is shown along with the numberof majors who newly declared in that term (“Added”) and the number of former majors who dropped the major as of thatterm (“Dropped”). In square brackets next to each measure is the percentage of all unique students who declared that major.The five sub-tables show this information for five different majors: (a) chemistry, (b) computer science, (c) engineering, (d)mathematics, and (e) physics and astronomy. For example, in (c) we can see that in term 2, there were 2980 students with adeclared engineering major, which represents 83.4% of all students who ever declared engineering. Further, 14 students (0.4%of all engineering students) who did not declare in term 1 added the major in term 2, and 262 students (7.3% of all engineeringstudents) who were engineering majors in term 1 dropped the major in term 2. (a) Biological Sciences , N unique = 2249Term Number of Majors [% of N unique ]Current Added Dropped1 17 [0.8] 17 [0.8] 0 [0.0]2 69 [3.1] 54 [2.4] 3 [0.1]3 1224 [54.4] 1159 [51.5] 12 [0.5]4 1730 [76.9] 582 [25.9] 88 [3.9]5 1968 [87.5] 311 [13.8] 84 [3.7]6 1960 [87.1] 71 [3.2] 85 [3.8]7 1910 [84.9] 44 [2.0] 76 [3.4]8 1824 [81.1] 11 [0.5] 49 [2.2]9 354 [15.7] 7 [0.3] 34 [1.5]10 225 [10.0] 2 [0.1] 16 [0.7]11 41 [1.8] 1 [0.0] 6 [0.3]12 13 [0.6] 0 [0.0] 5 [0.2](c) Psychology , N unique = 1784Term Number of Majors [% of N unique ]Current Added Dropped1 6 [0.3] 6 [0.3] 0 [0.0]2 62 [3.5] 56 [3.1] 0 [0.0]3 404 [22.6] 343 [19.2] 4 [0.2]4 1011 [56.7] 620 [34.8] 25 [1.4]5 1446 [81.1] 458 [25.7] 23 [1.3]6 1594 [89.3] 174 [9.8] 29 [1.6]7 1657 [92.9] 95 [5.3] 13 [0.7]8 1552 [87.0] 23 [1.3] 28 [1.6]9 265 [14.9] 18 [1.0] 12 [0.7]10 138 [7.7] 4 [0.2] 10 [0.6]11 32 [1.8] 0 [0.0] 6 [0.3]12 16 [0.9] 0 [0.0] 5 [0.3] (b) Economics , N unique = 965Term Number of Majors [% of N unique ]Current Added Dropped1 17 [1.8] 17 [1.8] 0 [0.0]2 104 [10.8] 89 [9.2] 5 [0.5]3 306 [31.7] 210 [21.8] 9 [0.9]4 562 [58.2] 271 [28.1] 22 [2.3]5 708 [73.4] 175 [18.1] 31 [3.2]6 766 [79.4] 89 [9.2] 33 [3.4]7 766 [79.4] 70 [7.3] 53 [5.5]8 717 [74.3] 28 [2.9] 30 [3.1]9 183 [19.0] 12 [1.2] 22 [2.3]10 92 [9.5] 4 [0.4] 15 [1.6]11 34 [3.5] 4 [0.4] 4 [0.4]12 11 [1.1] 0 [0.0] 9 [0.9](d) Other STEM , N unique = 1543Term Number of Majors [% of N unique ]Current Added Dropped1 16 [1.0] 16 [1.0] 0 [0.0]2 143 [9.3] 128 [8.3] 4 [0.3]3 781 [50.6] 648 [42.0] 13 [0.8]4 1212 [78.5] 476 [30.8] 52 [3.4]5 1345 [87.2] 188 [12.2] 58 [3.8]6 1333 [86.4] 42 [2.7] 56 [3.6]7 1284 [83.2] 27 [1.7] 56 [3.6]8 1222 [79.2] 9 [0.6] 23 [1.5]9 263 [17.0] 9 [0.6] 19 [1.2]10 135 [8.7] 0 [0.0] 14 [0.9]11 36 [2.3] 1 [0.1] 6 [0.4]12 12 [0.8] 0 [0.0] 11 [0.7](e) Non-STEM , N unique = 5197Term Number of Majors [% of N unique ]Current Added Dropped1 259 [5.0] 259 [5.0] 1 [0.0]2 1038 [20.0] 811 [15.6] 41 [0.8]3 2502 [48.1] 1507 [29] 51 [1.0]4 3600 [69.3] 1186 [22.8] 110 [2.1]5 4225 [81.3] 759 [14.6] 84 [1.6]6 4437 [85.4] 308 [5.9] 70 [1.3]7 4449 [85.6] 221 [4.3] 113 [2.2]8 4201 [80.8] 109 [2.1] 110 [2.1]9 818 [15.7] 45 [0.9] 68 [1.3]10 454 [8.7] 11 [0.2] 32 [0.6]11 115 [2.2] 5 [0.1] 15 [0.3]12 58 [1.1] 3 [0.1] 12 [0.2]TABLE IV. For each term from 1 to 12, the current number of declared majors (“Current”) is shown along with the numberof majors who newly declared in that term (“Added”) and the number of former majors who dropped the major as of thatterm (“Dropped”). In square brackets next to each measure is the percentage of all unique students who declared that major(or cluster of majors as in (d) and (e)). The five sub-tables show this information for five different majors or clusters of majors:(a) biological sciences, (b) economics, (c) psychology, (d) other STEM disciplines not listed separately in Tables III and IV,and (e) other non-STEM majors excluding psychology. APPENDIX B: SUMMARY TABLES PERTAINING TO DIFFERENT DECLARED MAJORSAll Students
Unique Peak Concurrent Peak Added Peak DroppedMajor Majors Majors [Term] Majors [Term] Majors [Term]Biological Sciences 2249 1968 [5] 1159 [3] 88 [4]Chemistry 1090 942 [5] 724 [3] 61 [4]Computer Science 621 502 [7] 188 [5] 17 [7]Economics 965 766 [7] 271 [4] 53 [7]Engineering 3575 3228 [1] 3228 [1] 262 [2]Mathematics 373 286 [6] 102 [3] 26 [5]Physics and Astronomy 186 151 [4] 62 [3] 14 [3]Psychology 1784 1657 [7] 620 [4] 29 [6]Other STEM 1543 1345 [5] 648 [3] 58 [5]Non-STEM 5197 4449 [7] 1507 [3] 113 [7]TABLE V. Summary counts for all students. For each major, the total number of unique students is listed along with peakconcurrent majors, added majors, and dropped majors, as well as the term in which the peak occurs in brackets. For example,in biological sciences, there were 2249 individual students in the sample who had ever declared the major. 1968 of those studentsdeclared biological science majors in term 5 (peak term for concurrent majors), which is higher than the number of majorsdeclared in any other term. Further, 1159 of those students added the major in term 3 (peak term for adding this major) and88 of those students dropped the major in term 4 (peak term for dropping this major).
Men Only
Unique Peak Concurrent Peak Added Peak DroppedMajor Majors Majors [Term] Majors [Term] Majors [Term]Biological Sciences 1115 962 [6] 620 [3] 52 [4]Chemistry 647 567 [6] 434 [3] 28 [4]Computer Science 512 413 [7] 156 [5] 14 [7]Economics 661 514 [7] 189 [4] 36 [7]Engineering 2678 2410 [1] 2410 [1] 189 [2]Mathematics 242 185 [6] 59 [4] 13 [5]Physics and Astronomy 150 124 [4] 54 [3] 11 [7]Psychology 490 447 [7] 141 [4] 14 [8]Other STEM 781 678 [5] 320 [3] 38 [7]Non-STEM 2235 1853 [7] 613 [3] 63 [7]TABLE VI. Summary counts for men. For each major, the total number of unique male students is listed along with peakconcurrent majors, added majors, and dropped majors, as well as the term in which the peak occurs in brackets.
Women Only
Unique Peak Concurrent Peak Added Peak DroppedMajor Majors Majors [Term] Majors [Term] Majors [Term]Biological Sciences 1134 1013 [5] 539 [3] 53 [6]Chemistry 443 377 [5] 290 [3] 33 [4]Computer Science 109 89 [7] 32 [5] 3 [7]Economics 304 252 [7] 82 [4] 17 [7]Engineering 897 818 [1] 818 [1] 73 [2]Mathematics 131 101 [6] 47 [3] 13 [5]Physics and Astronomy 36 28 [6] 13 [2] 4 [3]Psychology 1294 1210 [7] 479 [4] 20 [6]Other STEM 762 667 [5] 328 [3] 34 [5]Non-STEM 2962 2596 [7] 894 [3] 56 [4]TABLE VII. Summary counts for women. For each major, the total number of unique female students is listed along withpeak concurrent majors, added majors, and dropped majors, as well as the term in which the peak occurs in brackets. APPENDIX C: DEGREES EARNED BY STUDENTS WHO DROPPED A MAJORAll Students % of N drop in a Given Major That Subsequently Earned Degree in Each MajorOther Non- NoMajor N drop Bio Chem CS Econ Engr Math Phys Psych STEM STEM DegreeBiological Sciences 458 2.6 3.9 3.1 1.5 10.0 2.2 0.2 7.9 8.1 44.1 27.1Chemistry 222 10.8 1.4 4.5 1.8 11.7 2.3 0.9 5.0 7.7 37.4 26.6Computer Science 86 5.8 0.0 1.2 2.3 2.3 3.5 1.2 2.3 5.8 26.7 53.5Economics 233 2.6 1.3 3.4 1.3 2.6 6.9 0.0 4.3 7.3 47.2 31.8Engineering 803 4.1 2.6 8.2 5.4 0.9 2.4 0.7 2.5 4.9 29.6 46.3Mathematics 122 10.7 5.7 4.9 15.6 5.7 4.1 1.6 7.4 19.7 20.5 25.4Physics and Astronomy 70 1.4 4.3 4.3 4.3 11.4 15.7 1.4 2.9 2.9 24.3 37.1Psychology 155 1.9 0.0 3.2 1.3 0.6 0.0 0.6 9.7 2.6 35.5 51.6Other STEM 312 11.5 1.3 3.2 11.2 2.6 5.4 0.0 15.1 1.0 36.2 26.6Non-STEM 707 6.6 1.3 4.4 4.7 2.7 1.6 1.0 12.2 5.1 8.8 59.3TABLE VIII. Trajectory of all students who dropped a major. For each major, the total number of students in the datasetwho dropped that major ( N drop ) is listed along with the percentage of N drop who ultimately earned a degree in each majoror earned no degree. For example, there were 458 students who ever dropped their major in biological sciences. Of those 458students, 2.6% went on to earn a degree in biological sciences (i.e., they later declared that major again after dropping it atan earlier point). Similarly, 3.9% of them earned a degree in chemistry, 3.1% in computer science, 1.5% in economics, 10.0%in engineering, and so forth. Finally, 27.1% of those 458 students that dropped a biological sciences major ultimately did notearn a degree from the university. Men Only % of N drop in a Given Major That Subsequently Earned Degree in Each MajorOther Non- NoMajor N drop Bio Chem CS Econ Engr Math Phys Psych STEM STEM DegreeBiological Sciences 232 3.9 6.0 5.6 0.9 16.4 2.6 0.4 3.4 6.9 36.2 30.6Chemistry 118 8.5 2.5 8.5 2.5 11.0 2.5 0.8 4.2 7.6 31.4 28.8Computer Science 72 6.9 0.0 1.4 2.8 2.8 2.8 1.4 0.0 4.2 27.8 55.6Economics 177 2.3 0.6 4.5 1.7 2.8 6.8 0.0 3.4 6.2 45.8 33.9Engineering 609 2.8 2.0 9.7 5.9 0.5 2.3 0.8 1.3 3.9 28.6 49.3Mathematics 80 10.0 7.5 5.0 11.2 7.5 5.0 2.5 3.8 17.5 20.0 31.2Physics and Astronomy 59 1.7 5.1 5.1 5.1 11.9 15.3 1.7 1.7 3.4 22.0 37.3Psychology 61 0.0 0.0 8.2 3.3 0.0 0.0 1.6 13.1 0.0 34.4 47.5Other STEM 164 9.8 1.2 4.9 17.7 4.3 7.9 0.0 11.0 0.6 26.2 29.3Non-STEM 383 6.5 1.0 6.3 6.0 3.1 1.8 1.3 6.0 4.2 7.3 62.7TABLE IX. Trajectory of all men who dropped a major. For each major, the total number of male students in the datasetwho dropped that major ( N drop ) is listed along with the percentage of N drop who ultimately earned a degree in each major orearned no degree. Women Only % of N drop in a Given Major That Subsequently Earned Degree in Each MajorOther Non- NoMajor N drop Bio Chem CS Econ Engr Math Phys Psych STEM STEM DegreeBiological Sciences 226 1.3 1.8 0.4 2.2 3.5 1.8 0.0 12.4 9.3 52.2 23.5Chemistry 104 13.5 0.0 0.0 1.0 12.5 1.9 1.0 5.8 7.7 44.2 24.0Computer Science 14 0.0 0.0 0.0 0.0 0.0 7.1 0.0 14.3 14.3 21.4 42.9Economics 56 3.6 3.6 0.0 0.0 1.8 7.1 0.0 7.1 10.7 51.8 25.0Engineering 194 8.2 4.6 3.6 3.6 2.1 2.6 0.5 6.2 7.7 33.0 37.1Mathematics 42 11.9 2.4 4.8 23.8 2.4 2.4 0.0 14.3 23.8 21.4 14.3Physics and Astronomy 11 0.0 0.0 0.0 0.0 9.1 18.2 0.0 9.1 0.0 36.4 36.4Psychology 94 3.2 0.0 0.0 0.0 1.1 0.0 0.0 7.4 4.3 36.2 54.3Other STEM 148 13.5 1.4 1.4 4.1 0.7 2.7 0.0 19.6 1.4 47.3 23.6Non-STEM 324 6.8 1.5 2.2 3.1 2.2 1.2 0.6 19.4 6.2 10.5 55.2TABLE X. Trajectory of all women who dropped a major. For each major, the total number of female students in the datasetwho dropped that major ( N drop ) is listed along with the percentage of N dropdrop