Not all disadvantages are equal: Racial/ethnic minority students have largest disadvantage of all demographic groups in both STEM and non-STEM GPA
NNot all disadvantages are equal: Racial/ethnic minority students have largestdisadvantage of all demographic groups in both STEM and non-STEM GPA
Kyle M. Whitcomb and Chandralekha Singh
Department of Physics and Astronomy, University of Pittsburgh, Pittsburgh, PA, 15260 (Dated: March 11, 2020)An analysis of institutional data to understand the outcome of the many obstacles faced bystudents from historically disadvantaged backgrounds is important in order to work towards pro-moting equity and inclusion for all students. We use 10 years of institutional data at a large publicresearch university to investigate the grades earned (both overall and in STEM courses only) bystudents categorized on four demographic characteristics: gender, race/ethnicity, low-income status,and first-generation college student status. We find that on average across all years of study andfor all clusters of majors, underrepresented minority students experience a larger penalty to theirmean overall and STEM GPA than even the most disadvantaged non-URM students. Moreover,the underrepresented minority students with additional disadvantages due to socioeconomic statusor parental education level were even further penalized in their average GPA. Furthermore, we alsofind that while women in all demographic groups had a higher average overall GPA, these genderdifferences are almost completely non-existent in STEM GPA except among the most privilegedstudents. These findings suggest that there is need to provide support to bridge the gaps thatemanate from historical disadvantages to certain groups.
INTRODUCTION AND THEORETICALFRAMEWORK
The importance of evidence-based approaches to im-proving student learning and ensuring that all studentshave the opportunity to excel regardless of their back-ground is becoming increasingly recognized by Science,Technology, Engineering, and Mathematics (STEM) de-partments across the US [1–9]. With advances in digitaltechnology in the past few decades, institutions have beenkeeping increasingly large digital databases of studentrecords. We have now reached the point where there issufficient data available for robust statistical analyses us-ing data analytics that can provide valuable informationuseful for transforming learning for all students [10, 11].This has lead to many recent studies utilizing many yearsof institutional data to perform analyses that were pre-viously limited by statistical power [12–16]. Therefore,here we focus on harnessing institutional data to investi-gate the obstacles faced by students with various disad-vantages who must overcome obstacles in their pursuit ofhigher education.The theoretical framework for this study has two mainfoundations: critical theory and intersectionality. Crit-ical theories of race, gender, etc. identify historicalsources of inequity within society, that is, societal normsthat perpetuate obstacles to the success of certain groupsof disadvantaged people [17–23]. Critical theory tells usthat the dominant group in a society perpetuates thesenorms, which are born out of their interests, and pushesback against support systems that seek to subvert thesenorms [17–19]. These highly problematic societal normsare founded in the historical oppression of various groupsof people, and manifest today in many ways includingeconomic disadvantages, stereotypes about who can suc- ceed in certain career paths, and racist and/or sexist bar-riers to opportunity, including educational advancement.While these norms are, by definition, specific to a partic-ular culture or even country, they are nonetheless perva-sive and oppressive and demand attention to rectify thesehistorical wrongs.Much important work has been done on buildingcritical race and/or gender theories of STEM educa-tion [1, 2, 22–31]. In one study, Bancroft (2018) laysout a “critical capital theory,” using varying forms ofcapital (economic, social, and cultural) to examine per-sistence through graduation in STEM doctoral programsand to contextualize the mechanisms behind racial in-equities in STEM education [32]. The idea that race,gender, or another demographic characteristic alone can-not fully explain the intricacies of the obstacles that stu-dents face is rooted in the framework of intersectional-ity [33–37]. In particular, the combination of differentaspects of an individual’s social identity (e.g., gender,race, first-generation college status, and socioeconomicstatus) leads to unique levels of disadvantages that can-not be explained by simply adding together the effects ofthe individual components of identity [33]. For example,according to the framework of intersectionality, in manySTEM disciplines where the societal norm expects thatstudents are white men, the experience of a black womanis not a simple sum of the experiences of white womenand black men [36, 37].With an eye toward this intersectional approach tocritical theory, we seek to understand the relationshipbetween four different aspects of student identity thatcan lead to obstacles in STEM education: race/ethnicity,gender, low-income status, and first-generation collegestudent status. The students disadvantaged by low-income or first-generation status are likely to experi- a r X i v : . [ phy s i c s . e d - ph ] M a r ence a lack of resources relative to their more privilegedpeers [38–40]. Women and underrepresented minoritystudents are susceptible to additional stress and anxi-ety from stereotype threat (i.e., the fear of confirmingstereotypes pertaining to their identity) which is not ex-perienced by their majority group peers [1, 25, 29–31, 41–50]. In summary, the different mechanisms by which stu-dents belonging to each demographic characteristic canbe disadvantaged are as follows. • Race/Ethnicity: Students belonging to underrep-resented minority (URM) groups may experiencestereotype threat that causes anxiety and robs thestudents of their cognitive resources, particularlyduring high-stakes testing. • Gender: There are pervasive societal biases againstwomen succeeding in many STEM disciplines whichcan result in stereotype threat. • Low-Income Status: Low-Income (LI) students aremore likely to need to work to support themselves,reducing their time and energy available to devoteto their studies, in addition to anxiety due to the fi-nancial burden of attending college. These burdensare in addition to other factors that low-income stu-dents may be more likely to face, such as lowerquality preparation for college. • First-Generation Status: First-Generation (FG)students may lack the resources of encouragement,advice, and support that are available more readilyto students with degree-holding parents. This lackof resources can make FG students more suscepti-ble to the stress of the unknown in college.All of these mechanisms can produce an inequitablelearning environment wherein students belonging to anyof these groups are forced to work against obstacles thattheir peers do not have. The framework of intersectional-ity asserts that for students that belong to more than oneof these groups, complex interactions between these dif-ferent obstacles can result in compounded disadvantagesthat are not a simple sum of the individual effects [33–37].In order to measure the long-term effects of these sys-temic disadvantages, we will investigate the academicachievement of students belonging to these various de-mographic groups over the course of their studies at onelarge public research university using 10 years of institu-tional data. By grouping students according to their de-mographic background, we will be able to investigate howdifferent combinations of obstacles affect student gradepoint averages.
RESEARCH QUESTIONS
Our research questions regarding the intersectional re-lationships between demographic characteristics and aca- demic achievement are as follows.
RQ1.
Are there differences in the overall or STEM gradesearned by students belonging to different demo-graphic groups (i.e., underrepresented minority,low-income status, and first-generation college stu-dent status)?
RQ2.
Do any patterns observed in RQ1 differ for menand women?
RQ3.
Do grades earned in STEM courses alone exhibitsimilar demographic patterns as grades earned inall courses?
RQ4.
What are the trends over time in the mean GPAof these different demographic groups among differ-ent clusters of majors (i.e., computer science, engi-neering, mathematics, and physical science majors,other STEM majors, and non-STEM majors)?
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 [51].The university provided for analysis the de-identifiedinstitutional data records of students with InstitutionalReview Board approval. In this study, we examinedthese records for N = 24 ,
567 undergraduate studentsenrolled in three colleges within the university: the col-leges of Arts and Sciences, Computing and Information,and Engineering. This sample of students includes allof those from ten cohorts who met several selection cri-teria, namely that the student had first enrolled at theuniversity in a Fall semester from Fall 2005 to Fall 2014,inclusive, and the institutional data on the student wasnot missing or unspecified for any of the following mea-sures: gender, race/ethnicity, parental education level,and family income. This sample of students is 50% femaleand had the following race/ethnicities: 79% White, 9%Asian, 7% Black, 3% Hispanic, and 2% other or multira-cial. Further, this sample is 16% first-generation collegestudents and 21% “low-income” students (to be definedin the following section).We acknowledge that gender is not a binary construct,however in self-reporting their gender to the universitystudents were given the options of “male” or “female”and so those are the two self-reported genders that weare able to analyze. There were 39 students who had metall other selection criteria but who had not indicated anygender on the survey, these students were removed fromthe sample and are not included in the reported samplesize or any analyses.
Measures
Demographic Characteristics
Four primary measures are the demographic character-istics mentioned in the previous section, namely gender,race/ethnicity, parental education level, and family in-come. All of these were converted into binary categoriesintended to distinguish between the most and least priv-ileged students on each measure. • Gender . Gender was reported as a binary categoryto begin with (either “male” or “female”), thereforeno further steps were required. • First-generation . Students for whom both parentshad a highest completed level of education of highschool or lower were grouped together as “first-generation” (FG) college students and correspond-ingly students for whom at least one parent hadearned a college degree were labeled non-FG. • Low-income . Students whose reported family Ad-justed Gross Income (AGI) was at or below 200%of the federal U.S. poverty line were categorized as“low-income” (LI), and those above 200% of thepoverty line as non-LI [52, 53]. • Underrepresented minority . All students who iden-tified as any race or ethnicity other than White orAsian were grouped together as “underrepresentedminority” (URM) students, including multiracialstudents who selected White and/or Asian in ad-dition to another demographic option. Studentswho only identified as White and/or Asian studentswere categorized as non-URM students.
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 . 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. In this study we considerup through the students’ sixth year of study or the endof their enrollment at the studied institution, whichevercomes first.
Analysis
The primary method by which we grouped studentsin this analysis was by their set of binary demographiccategories. This grouping was performed in two differentways. First, use of all four binary categories (gender, FG,LI, URM) resulted in sixteen mutually exclusive groups(e.g., “female, FG+URM” or “male, LI”). Second, useof all categories except gender resulted in eight mutuallyexclusive categories.We calculated each student’s yearly (i.e., not cumula-tive) grade point average (GPA) across courses taken ineach year of study from the first to sixth years. In ad-dition, we calculated the student’s yearly STEM GPA,that is, the GPA in STEM courses alone. Then, usingthe aforementioned grouping schemes, we computed themean GPA in each demographic group as well as thestandard error of the mean separately for each year ofstudy [54]. Further, in the case of grouping by gender, wecomputed the effect size of the gender differences withineach demographic group using Cohen’s d , which is typi-cally interpreted using minimum cutoff values for “small”( d = 0 . d = 0 . d = 0 . tidyverse [59] for data manipulation andplotting. RESULTSGPA Trends by Demographic Group: “DinosaurPlots”
In order to answer
RQ1 , we plotted in Fig. 1 the meanGPA earned by students in each demographic group, in-cluding gender as a grouping characteristic. We startwith overall GPA, rather than STEM GPA alone, in or-der to provide context for the results in STEM GPA andidentify trends that may or may not be present whenviewing STEM grades alone. Groups are ordered fromleft to right first by the ascending number of selectedcharacteristics and then alphabetically. Mean GPA isplotted separately (i.e., not cumulatively) for each yearof study from the first to sixth year. Setting aside thegender differences for a moment, we note that the gen-eral GPA trends by demographic group in Fig. 1 followa shape resembling the neck, back, and tail of a sauro-pod, and so accordingly we refer to the plots in Fig. 1 as“dinosaur plots.” This shape is clearest in the plots forthe first through fourth years, as the sample size dropssignificantly in the fifth year as the majority of studentsgraduate.Looking more closely at Fig. 1, particularly the firstfour years, we see that the “neck” is consistently com-prised of the group of students with the most privileges,namely those students that are non-FG, non-LI, and non-URM. Following this, the “back” is relatively flat acrossthe next four groups, namely students that are FG only,LI only, URM only, or FG and LI. Notably, the URMgroup of students typically have the lowest mean GPAwithin this set of demographic groups. Finally, the “tail”consists of the final three groups, FG+URM, LI+URM,and FG+LI+URM. The mean GPA in this set of groupstends to decrease from left to right in the plots. Notably,the four groups that contain URM students are consis-tently in the lowest four or five mean GPAs.
Intersectionality with Gender
We now turn our attention to the differences betweenmen and women in Fig. 1 in order to answer
RQ2 . Wenote in particular that across all demographic groupswomen’s mean GPA is roughly 0.2 grade points higherthan men’s. The effect sizes (Cohen’s d ) of this differ-ence range from small to medium [55]. This differencein mean GPA earned is substantial enough to indicate a change in letter grade, given that the grading systemat the studied university uses increments of 0.25 gradepoints for letter grades containing “+” or “ − .” Further,this trend holds in the fifth year (Fig. 1e) and sixth year(Fig. 1f), with some exceptions in demographic groupswith particularly low sample sizes after the fourth year. STEM GPA Trends
In order to answer
RQ3 , Figure 2 plots students’ meanSTEM GPA in a similar manner to Fig. 1. We note thatthe general “dinosaur” pattern discussed in Fig. 1 alsoholds at least for the first and second years (Figs. 2aand 2b, respectively). In the third year and beyond, thegeneral features of the trend continue to hold, with themost privileged students having the highest mean GPA,followed by those with one disadvantage as well as thefirst-generation and low-income group, followed by theremaining groups of URM students with one or moreadditional disadvantages. However, in these later years,the finer details of the plots noted before fall away infavor of a sharper mean GPA decrease for URM studentswith at least one additional disadvantage in the third year(Fig. 2c) and a more gradual decrease across all groups inthe fourth year (Fig. 2d) and fifth year (Fig. 2e). Whenrestricting the GPA calculations to STEM courses, thesample size becomes too small in the sixth year (Fig. 2f)to draw meaningful conclusions.We further observe a trend of students earning highergrades on average in later years, although the rise fromthe first to the fourth year is somewhat lower in STEMGPA than in overall GPA. Notably, while in overall GPAthis trend seemed to be somewhat universal across de-mographic groups, in Fig. 2 we see a quicker rise in meanSTEM GPA over time for the more privileged studentsthan the less privileged students, particularly comparingthe leftmost and rightmost groups.Regarding gender differences, Fig. 2 shows smaller gen-der differences in STEM GPA than those observed inoverall GPA in Fig. 1. While in overall GPA womenearned roughly 0.2 grade points more than men on aver-age, in STEM GPA that difference is much less consistentand typically ranges from 0 to 0.1 grade points. For manydemographic groups we see no significant differences be-tween men and women’s mean STEM GPA. We do seethat there is still a consistent STEM GPA gender differ-ence, albeit smaller than in Fig. 1, among the group of themost privileged students (i.e., those with “None” of thedisadvantages). There is also a STEM GPA gender dif-ference among first-generation low-income but non-URMstudents, however this difference is less consistent and infact briefly vanishes in the third year. l l l l l l l l d = 0.33 d = 0.24 d = 0.23 d = 0.30 d = 0.28 d = 0.29 d = 0.27 d = 0.28
Demographic Characteristic Group M ean F i r s t Y ea r G PA Gender l FM (a) l l l l l l l l d = 0.35 d = 0.29 d = 0.27 d = 0.32 d = 0.32 d = 0.45 d = 0.25 d = 0.49 Demographic Characteristic Group M ean S e c ond Y ea r G PA Gender l FM (b) l l l l l l l l d = 0.38 d = 0.29 d = 0.35 d = 0.33 d = 0.39 d = 0.53 d = 0.33 d = 0.52 Demographic Characteristic Group M ean T h i r d Y ea r G PA Gender l FM (c) l l l l l l l l d = 0.43 d = 0.42 d = 0.36 d = 0.35 d = 0.39 d = 0.54 d = 0.27 d = 0.54 Demographic Characteristic Group M ean F ou r t h Y ea r G PA Gender l FM (d) l l l l l l l l d = 0.42 d = 0.28 d = 0.50 d = 0.41 d = 0.35 d = 0.83 d = 0.29 d = 0.38 Demographic Characteristic Group M ean F i ft h Y ea r G PA Gender l FM (e) l l l l l l l l
816 83 136 101 73 15 33 12 742 118 133 91 71 11 29 20 d = 0.44 d = 0.37 d = 0.49 d = 0.27 d = 0.95 d = 1.13 d = 0.17 d = 0.61
Demographic Characteristic Group M ean S i x t h Y ea r G PA Gender l FM (f) FIG. 1. Average GPA of each demographic group. Students are binned into separate demographic groups based on theirstatus as first-generation (FG), low-income (LI), and/or underrepresented minority (URM) students. The men and women ineach demographic group are plotted separately. The mean GPA in all courses taken by students in each demographic groupis plotted along with the standard error on the mean, with a separate plot for each of the (a) first, (b) second, (c) third, (d)fourth, (e) fifth, and (f) sixth years. The sample size is reported by each point, and Cohen’s d [55] measuring the effect size ofthe gender difference in each group is reported. l l l l l l l l d = 0.13 d = 0.04 d = 0.06 d = 0.11 d = 0.14 d = 0.20 d = 0.10 d = 0.10 Demographic Characteristic Group M ean F i r s t Y ea r S T E M G PA Gender l FM (a) l l l l l l l l
68 5441 637 1006 685 464 308 98 5943 737 1078 578 453 203 97 109 d = 0.14 d = 0.05 d = 0.07 d = 0.04 d = 0.21 d = −0.13 d = 0.06 d = 0.14
Demographic Characteristic Group M ean S e c ond Y ea r S T E M G PA Gender l FM (b) l l l l l l l l d = 0.09 d = 0.00 d = 0.07 d = 0.03 d = 0.07 d = 0.12 d = 0.04 d = 0.27 Demographic Characteristic Group M ean T h i r d Y ea r S T E M G PA Gender l FM (c) l l l l l l l l d = 0.11 d = 0.15 d = 0.06 d = 0.04 d = 0.34 d = 0.18 d = −0.03 d = 0.29 Demographic Characteristic Group M ean F ou r t h Y ea r S T E M G PA Gender l FM (d) l l l l l l l l
153 16 53 494 127 90 50 12 1100 224 139 91 17 60 12 43 d = 0.15 d = −0.05 d = 0.11 d = 0.07 d = 0.14 d = −0.10 d = −0.43 d = 0.09
Demographic Characteristic Group M ean F i ft h Y ea r S T E M G PA Gender l FM (e) l l l l l l l
55 33 7 13 93 14 10 227 47 33 17 20 2 9 d = 0.26 d = 0.40 d = −0.22 d = −0.76 d = 0.37 d = −0.14 d = −0.48 d = 4.53
Demographic Characteristic Group M ean S i x t h Y ea r S T E M G PA Gender l FM (f) FIG. 2. Average STEM GPA of each demographic group. Students are binned into separate demographic groups basedon their status as first-generation (FG), low-income (LI), and/or underrepresented minority (URM) students. The men andwomen in each demographic group are plotted separately. The mean GPA in all courses taken by students in each demographicgroup is plotted along with the standard error on the mean, with a separate plot for each of the (a) first, (b) second, (c) third,and (d) fourth, (e) fifth, and (f) sixth years. The sample size is reported by each point, and Cohen’s d [55] measuring the effectsize of the gender difference in each group is reported. GPA Trends By Major Over Time
In order to better understand the trends over time inboth overall and STEM GPA and answer
RQ4 , we plot-ted the mean GPA by year in Fig. 3 and mean STEMGPA by year in Fig. 4. In these plots, we have not sep-arated men and women and instead focus on the otherdemographic characteristics while further grouping stu-dents into three different groups of majors in order tounderstand if these trends differ for students in differentareas of study. Further, since the sample size becomesquite small in years five and six for many of the demo-graphic groups of interest, we plot only the mean GPAover the first four years. In Figs. 3a and 4a, we plot themean overall and STEM GPA, respectively, of all stu-dents. In the other subfigures, we plot the mean GPAearned by students majoring in different clusters of ma-jors. In particular, we plot the mean GPA of engineering(including computer science), mathematics, and physicalscience (i.e., chemistry and physics) majors in Figs. 3band 4b, the remaining STEM majors in Figs. 3c and 4c,and non-STEM majors in Figs. 3d and 4d.These plots make clearer some of the trends noted ear-lier, especially the rise in mean GPA over time from thefirst to the fourth year. However, we can now see thatthis is not universally true since the first-generation URMstudents have a drop in mean GPA in the second year forphysical science majors (Fig. 3b), and in the third yearfor other STEM majors (Fig. 3c). This trend is evenmore noticeable in STEM GPA (Fig. 4), where the meanSTEM GPA of the group of first-generation URM stu-dents drops in the third year for every subpopulation bymajor.
DISCUSSION
To start, we consider how much the current systemdisadvantages students who are first-generation, low-income, or underrepresented minority but not a com-bination of the two. Discussing these groups first ishelpful in setting the stage for a more complex discus-sion of the intersectionality of these various demographiccharacteristics. We find in Figs. 1 and 2 that not all ofthese disadvantages are equal. In particular, non-URMstudents who have one disadvantage, namely the first-generation (but not low-income) and low-income (but notfirst-generation) students, still earn slightly higher gradesthan even the URM students who are not low-income orfirst-generation. Notably, this trend (the “back” of thedinosaur plots) is similar in both overall grades (Fig. 1)and in STEM grades alone (Fig. 2). The size of this meangrade difference varies from year to year, but in STEMgrades it can reach as high as about 0.25 grade points,which at the studied institution is the difference between,for example, a B and B+ or B − grade. The group with the grades most similar to thesenon-first-generation, non-low-income URM students arethe first-generation, low-income non-URM students, whoearn both overall (Fig. 1) and STEM (Fig. 2) grades sim-ilar to or very slightly higher than the URM students.One explanation could be that the lack of resources avail-able due to being first-generation or low-income is not assevere an obstacle as the stereotype threat experiencedby URM students.Turning then to the “tail” in the dinosaur plots, wefind that consistently the most disadvantaged studentsin both overall grades (Fig. 1) and STEM grades (Fig. 2)are the URM students with at least one additional obsta-cle. In this case, it appears that the intersection of beinglow-income and URM is the most disadvantageous combi-nation, with no notable difference in either Fig. 1 or Fig. 2among these students whether or not they are also first-generation. Meanwhile, the first-generation URM stu-dents who are not low-income sometimes have a slightlyhigher mean GPA than the low-income URM students(Fig. 1).Another avenue to investigate intersectionality is howgender interacts with the other demographic groups. In-terestingly, in overall GPA (Fig. 1), gender appears tohave about the same effect across all demographic groups.That is, there does not appear to be an intersectional ef-fect of gender identity with other identities as measuredby overall GPA. However, Fig. 2 shows that this is acontext-dependent effect, with the gender gap substan-tially and unevenly reduced across all groups in meanSTEM GPA. For most demographic groups in Fig. 2, thehigher overall GPA earned by women in Fig. 1 has van-ished completely in STEM GPA. This is consistent withstereotype threat being the mechanism of disadvantagefor women, where stereotypes surrounding STEM disci-plines unfairly cause stress and anxiety for women [41–44, 46, 48–50]. Notably, while the gender gap is reducednearly to zero for most groups in Fig. 2, there does re-main a small consistent gender gap favoring women in themost privileged group of students. In other groups thegender gap in Fig. 2 is inconsistent across years. One ex-planation could be that the wealth of resources availableto them may help to alleviate the stereotype threat.Taking a more temporal view of these GPA trends,Fig. 3 (overall GPA) and Fig. 4 (STEM GPA) havegrouped men and women together in order to focus on theother demographic characteristics more closely. In theseplots, the most noteworthy trend is again that, with thesole exception of the first year in Fig. 3b, the four groupswith the lowest mean GPA (Fig. 3) and STEM GPA(Fig. 4) across the first four years are always the fourgroups containing URM students. Notably, this trendis true regardless of which group of majors we investi-gate. The consistency of this result is particularly strik-ing, showing that the most otherwise disadvantaged non-URM students have fewer obstacles to success than even l l l ll l l ll l l ll l l ll l l ll l l ll l l ll l l l Year M ean G PA (a) All Students l l l ll l l ll l l ll l l ll l l ll l l ll l l ll l l l Year M ean G PA (b) Chemistry, Computer Science, Engineering, Mathematics, Physics Majors l l l ll l l ll l l ll l l ll l l ll l l ll l l ll l l l Year M ean G PA (c) Biology, Economics, Geology, Neuroscience, Statistics Majors l l l ll l l ll l l ll l l ll l l ll l l ll l l ll l l l Year M ean G PA Demographic Group l l l ll l l l
None FG LI URMFG+LI FG+URM LI+URM FG+LI+URM (d) Non−STEM Majors
FIG. 3. Students are binned into separate demographic groups as in Fig. 1, but not separated by gender. The mean GPA inall courses of each group is plotted over time from year one to four, along with the standard error of the mean. The plots showthis for four subpopulations: (a) all students; (b) chemistry, computer science, engineering, mathematics, and physics students;(c) biology, economics, geology, neuroscience, and statistics students; and (d) non-STEM students including psychology. the most privileged URM students among all students.Focusing further on the STEM GPA of STEM ma-jors in Figs. 4b and 4c, we see that while non-URMstudents consistently rise in mean GPA over time, thesame is not true for all URM students. In particular, thefirst-generation URM students who major in chemistry,computer science, engineering, mathematics, or physics(Fig. 4b) experience a steady decline in mean STEM GPAfrom year one to two and year two to three. While the standard error of those means is quite large due to arelatively small sample size, that lack of representationfor these students could itself be what is hindering theircoursework by causing a stereotype threat.Based upon the frameworks of critical theory and in-tersectionality, the main implication of these findings isthat many students who come from less privileged back-grounds are not being adequately supported in college inorder to catch up with the privileged students [1, 2, 17– l l l ll l l ll l l ll l l ll l l ll l l ll l l ll l l l
Year M ean S T E M G PA (a) All Students l l l ll l l ll l l ll l l ll l l ll l l ll l l ll l l l Year M ean S T E M G PA (b) Chemistry, Computer Science, Engineering, Mathematics, Physics Majors l l l ll l l ll l l ll l l ll l l ll l l ll l l ll l l l Year M ean S T E M G PA (c) Biology, Economics, Geology, Neuroscience, Statistics Majors l l l ll l l ll l l ll l l ll l l ll l l ll l l ll l l l Year M ean S T E M G PA Demographic Group l l l ll l l l
None FG LI URMFG+LI FG+URM LI+URM FG+LI+URM (d) Non−STEM Majors
FIG. 4. Students are binned into separate demographic groups as in Fig. 2, but not separated by gender. The mean GPAin STEM courses of each group is plotted over time from year one to four along with the standard error of the mean. Theplots show this for four subpopulations: (a) all students; (b) chemistry, computer science, engineering, mathematics, andphysics students; (c) biology economics, geology, neuroscience, and statistics students; and (d) non-STEM students includingpsychology.
23, 33–37]. The disadvantages of these less privileged stu-dents manifest as lower mean overall and STEM GPA forthose demographic groups. In order to promote equityand inclusion, it is crucial that these students are pro-vided appropriate mentoring, guidance, scaffolding, andsupport in college so that these obstacles can be clearedfor students who have been put at a disadvantage rel-ative to their peers through no fault of their own [60]. We note that these demographic groups with more dis-advantages are likely to consist of students who had K-12 education from schools with fewer resources and lesswell-prepared teachers than those of the more privilegedstudents, with high school being an especially importanttime for disadvantages related to STEM learning increas-ing [3, 61–65]. Analyses such as those discussed here canhelp inform the allocation of resources to support these0students, with efforts to reduce the classroom stereotypethreat of URM students and creating a low-anxiety envi-ronment in which all students have a high sense of belong-ing and can participate fully without fear of being judgedbeing clear priorities. Additional resources to supportlow-income and/or first-generation students, e.g., finan-cial support and timely advising pertaining to variousacademic and co-curricular opportunities, are also impor-tant in order to level the playing field and work towardsa goal of all students succeeding in college, regardless oftheir race/ethnicity, socioeconomic status, and parentaleducation history.
ACKNOWLEDGMENTS
This research is supported by the National ScienceFoundation Grant DUE-1524575 and the Sloan Founda-tion Grant G-2018-11183. [1] A. Johnson, Science Education , 960 (2012).[2] A. Johnson, M. Ong, L. T. Ko, J. Smith, and A. Hodari,The Physics Teacher , 356 (2017).[3] A. V. Maltese and R. H. Tai, Science Education , 877(2011).[4] M. Borrego, R. A. Streveler, R. L. Miller, and K. A.Smith, Journal of Engineering Education , 147 (2008).[5] M. Borrego and J. Bernhard, Journal of Engineering Ed-ucation , 14 (2011).[6] M. Borrego and C. Henderson, Journal of EngineeringEducation , 220 (2014).[7] C. Henderson and M. H. Dancy, American Journal ofPhysics , 79 (2008).[8] M. Dancy and C. Henderson, American Journal ofPhysics , 1056 (2010).[9] C. Henderson, M. Dancy, and M. Niewiadomska-Bugaj,Phys. Rev. ST Phys. Educ. Res. , 020104 (2012).[10] R. S. Baker and P. S. Inventado, in Learning Analytics (Springer, 2014) pp. 61–75.[11] Z. Papamitsiou and A. A. Economides, Journal of Edu-cational Technology & Society , 49 (2014).[12] S. M. Lord, M. M. Camacho, R. A. Layton, R. A.Long, M. W. Ohland, and M. H. Wasburn, Journal ofWomen and Minorities in Science and Engineering ,167 (2009).[13] S. M. Lord, R. A. Layton, and M. W. Ohland, IEEETransactions on Education , 141 (2015).[14] M. W. Ohland and R. A. Long, Advances in EngineeringEducation , 1 (2016).[15] R. L. Matz, B. P. Koester, S. Fiorini, G. Grom, L. Shep-ard, C. G. Stangor, B. Weiner, and T. A. McKay, AERAOpen , 1 (2017).[16] E. B. Witherspoon and C. D. Schunn, Science Education , 144 (2019).[17] K. Crenshaw, N. Gotanda, G. Peller, and K. Thomas, Critical Race Theory: The Key Writings that Formed theMovement (The New Press, 1995).[18] D. Kellner, Democracy & Nature , 51 (2003). [19] T. J. Yosso, Race Ethnicity and Education , 69 (2005).[20] R. Guti´errez, Teaching for Excellence and Equity inMathematics , 4 (2009).[21] E. Taylor, D. Gillborn, and G. Ladson-Billings, Foun-dations of critical race theory in education (Routledge,2009).[22] S. Tolbert, A. Schindel, and A. J. Rodriguez, ScienceEducation , 796 (2018).[23] K. Schenkel and A. Calabrese Barton, Science Education(2020).[24] D. Solorzano, M. Ceja, and T. Yosso, Journal of NegroEducation , 60 (2000).[25] J. L. Lewis, H. Menzies, E. I. N´ajera, and R. N. Page,Science Education , 961 (2009).[26] M. Bang and D. Medin, Science Education , 1008(2010).[27] M. Estrada, A. Eroy-Reveles, and J. Matsui, Social Is-sues and Policy Review , 258 (2018).[28] M. Ong, J. M. Smith, and L. T. Ko, Journal of Researchin Science Teaching , 206 (2018).[29] A. M. Green, B. R. Brand, and G. E. Glasson, ScienceEducation , 241 (2019).[30] J. W. Mutegi, B. Sorge, G. A. Fore, and G. S. Gibau,Science Education , 1456 (2019).[31] M. J. Sheth, Science Education , 37 (2019).[32] S. F. Bancroft, Science Education , 1319 (2018).[33] K. Crenshaw, Stan. L. Rev. , 1241 (1990).[34] S. Cho, K. W. Crenshaw, and L. McCall, Signs: Journalof Women in Culture and Society , 785 (2013).[35] J. D. Mitchell, C. Y. Simmons, and L. A. Greyerbiehl, Intersectionality & Higher Education (Peter Lang, 2014).[36] L. Charleston, R. P. Adserias, N. M. Lang, and J. F.Jackson, Journal of Progressive Policy & Practice , 273(2014).[37] T. R. Morton and E. C. Parsons, Science Education ,1363 (2018).[38] P. C. Lam, T. Srivatsan, D. Doverspike, J. Vesalo, andP. R. Mawasha, Journal of STEM Education: Innova-tions and Research , 14 (2005).[39] S. L. Dika and M. M. D’Amico, Journal of Research inScience Teaching , 368 (2016).[40] A. V. Katrevich and M. S. Aruguete, Journal of STEMEducation: Innovations and Research , 40 (2017).[41] A. W. Astin, What Matters in College , Vol. 9 (Jossey-Bass, 1993).[42] K. P. Cross, Journal of Engineering Education , 9(1993).[43] R. M. Felder, G. N. Felder, M. Mauney, C. E. Hamrin Jr.,and E. J. Dietz, Journal of Engineering Education ,151 (1995).[44] R. M. Felder, G. N. Felder, and E. J. Dietz, Journal ofEngineering Education , 469 (1998).[45] J. A. Bianchini, D. J. Whitney, T. D. Breton, and B. A.Hilton-Brown, Science Education , 42 (2002).[46] S. L. Britner and F. Pajares, Journal of Research in Sci-ence Teaching: The Official Journal of the National Asso-ciation for Research in Science Teaching , 485 (2006).[47] J. A. Bianchini, Science Education , 163 (2013).[48] V. Basile and E. Lopez, Science Education , 519(2015).[49] S. Cheryan, S. A. Ziegler, A. K. Montoya, and L. Jiang,Psychological Bulletin , 1 (2017).[50] A. Hilts, R. Part, and M. L. Bernacki, Science Education , 744 (2018). [51] 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).[52] N. K. Cauthen and S. Fass,
Measuring Income andPoverty in the United States , Tech. Rep. (National Cen-ter for Children in Poverty, 2007).[53] Y. Jiang, M. Ekono, and C. Skinner,
Basic facts aboutlow-income children , Tech. Rep. (National Center forChildren in Poverty).[54] D. Freedman, R. Pisani, and R. Purves,
Statistics , 4thed. (W. W. Norton & Co., 2007).[55] J. Cohen,
Statistical Power Analysis for the BehavioralSciences , 2nd ed. (Lawrence Erlbaum Associates, 1988).[56] J. Neter, M. H. Kutner, C. J. Nachtsheim, andW. Wasserman,
Applied Linear Statistical Models , 5thed. (McGraw-Hill/Irwin, 2004).[57] D. C. Montgomery, E. A. Peck, and G. G. Vining,
In-troduction to Linear Regression Analysis , 4th ed. (John Wiley & Sons, 2012).[58] R Core Team,
R: A Language and Environment for Sta-tistical Computing , R Foundation for Statistical Comput-ing, Vienna, Austria (2019).[59] H. Wickham, tidyverse: Easily Install and Load the ‘tidy-verse’ (2017), R package version 1.2.1.[60] J. A. Birt, M. Khajeloo, C. C. Rega-Brodsky, M. A.Siegel, T. S. Hancock, K. Cummings, and P. D. Nguyen,Science Education , 770 (2019).[61] J. A. Bianchini, C. C. Johnston, S. Y. Oram, and L. M.Cavazos, Science Education , 419 (2003).[62] B. Means, H. Wang, X. Wei, S. Lynch, V. Peters,V. Young, and C. Allen, Science Education , 681(2017).[63] M. C. Bottia, E. Stearns, R. A. Mickelson, and S. Moller,Science Education , 85 (2018).[64] S. G. Daley, Science Education , 1306 (2019).[65] R. Dou, Z. Hazari, K. Dabney, G. Sonnert, and P. Sadler,Science Education103