Patterns of Gendered Performance Differences in Large Introductory Courses at Five Research Universities

Abstract

Significant gendered performance differences are signals of systemic inequity in higher education. Understanding of these inequities has been hampered by the local nature of prior studies; consistent measures of performance disparity across many disciplines and institutions have not been available. Here, we report the first wide-ranging, multi-institution measures of gendered performance difference, examining more than a million student enrollments in hundreds of courses at five universities. After controlling for factors that relate to academic performance using optimal matching, we identify patterns of gendered performance difference that are consistent across these universities. Biology, chemistry, physics, accounting, and economics lecture courses regularly exhibit gendered performance differences that are statistically and materially significant, whereas lab courses in the same subjects do not. These results reinforce the importance of broad investigation of performance disparities across higher education. They also help focus equity research on the structure and evaluative schemes of these lecture courses.

In a recent review of research on gendered performance disparities in undergraduate science, technology, engineering, and mathematics (STEM) courses, Eddy and Brownell (2016) describe a confused research landscape: Some courses favor men, some favor women, and some show little bias. Their review calls specifically for systematic measurement of performance gaps across an array of disciplines and institutions, all accounting for prior academic performance, in the hope that emergent patterns might inform our understanding of “the relative contributions of different factors to performance and/or persistence in STEM.” In this study, we answer this call, analyzing data on more than a million student enrollments in hundreds of courses drawn from five research-intensive public universities in the Big Ten Academic Alliance.

We find evidence of statistically significant, persistent gendered performance differences (GPDs) in some large, introductory courses, differences that are also materially significant. In particular, men earned relatively higher grades than women in biology, chemistry, physics, accounting, and economics lecture courses, even after accounting for the influence of some measures of prior academic achievements on performance. These results are remarkably consistent across all five universities, which together enroll more than 150,000 undergraduate students in any given year. These patterns confirm the importance of conducting systematic studies of performance equity, provide the impetus for extending this work to other sectors of higher education, and focus research attention on the structure and evaluative schemes of these lecture courses.

Background

Women have achieved parity with men on many indicators of educational outcomes; indeed, women now outpace men in terms of college enrollment and overall attainment of bachelor’s and higher-level degrees (Snyder & Dillow, 2015). Still, significant gaps in enrollment and degree attainment remain in engineering, mathematics, computer science, and physical science disciplines (DiPrete & Buchmann, 2013). Even in the life sciences, where women now dominate numerically in terms of awarded bachelor’s degrees (Mann & DiPrete, 2013), gendered differences that favor men have been identified in exam performance, participation in whole-class discussions, and who is viewed as most knowledgeable about course content (Eddy, Brownell, & Wenderoth, 2014; Grunspan et al., 2016). These gaps undermine the national priority that all students have the opportunity to fully participate in STEM fields (President’s Council of Advisors on Science and Technology, 2012). Because sex is a legally protected class, disparate educational outcomes for male and female students also raise important questions of equity.

Girls and boys pursue science and mathematics courses in primary and secondary school in roughly equal proportion, but by the time they are freshmen in college, men are more likely to choose a science or mathematics major (Hill, Corbett, & St. Rose, 2010), and the underrepresentation of women in these disciplines carries all the way through to the professoriate (Urry, 2015). Research from diverse academic disciplines shows that a variety of factors affect gendered differences in STEM major selection, degree attainment, and careers (Blickenstaff, 2005). When considering performance in undergraduate STEM courses (the level of interest in this study), prior academic performance, engagement, and affective variables are all considered relevant constructs for investigating and explaining gendered differences (Eddy & Brownell, 2016). Here, we briefly review a range of important factors that influence the decisions of women and men about pursuing undergraduate STEM courses and degree programs.

Psychological and environmental factors have been shown to contribute to observed gendered gaps (Murphy, Steele, & Gross, 2007), such as the perpetuation of a “fixed mind-set” model that tends to favor men (Good, Rattan, & Dweck, 2012) as well as stereotype threat, which has been shown to reduce the performance of female students in mathematics when gendered stereotypes are invoked (H. Johnson, Barnard-Brak, Saxon, & Johnson, 2012). Microaggressions, brief and often subtle messages based on membership in a group (Sue, 2010), have been shown to act as a barrier to participation in STEM (Grossman & Porche, 2014). Unconscious bias plays a role as well. For example, biology, chemistry, and physics faculty members, regardless of their own gender, have been shown to view a male undergraduate job candidate as more competent and employable than an identical (excepting the name) female candidate (Moss-Racusin, Dovidio, Brescoll, Graham, & Handelsman, 2012). The affective dimensions of confidence and interest have also been linked to gendered differences that can impact course performance; women have reported feeling less confident than men in their calculus and engineering abilities (Ellis, Fosdick, & Rasmussen, 2016; Micari, Pazor, & Hartmann, 2007) and have provided more pessimistic self-reports of performance on a science assessment and reported subsequent diminished interest in scientific activities (Ehrlinger & Dunning, 2003). Hazari, Sonnert, Sadler, and Shanahan (2010) demonstrated that explicitly discussing gendered gaps in science positively impacted physics identity for women, which in turn strongly predicted decisions about pursuing a physics career.

The culture in undergraduate STEM courses—broadly including pedagogy, curriculum, assessment, instruction, and interaction between students and faculty—has been a major point of study with respect to gendered differences. Women in engineering programs have been shown to more frequently perceive gendered discrimination than men (Vogt, Hocevar, & Hagedorn, 2007), and some women in science courses have described being discouraged in attending large introductory classes where they felt anonymous, responding to and posing questions in class, and engaging with faculty in research (A. Johnson, 2007). Regular interaction with faculty, which is certainly easier to facilitate in smaller courses, has been shown to positively influence STEM degree completion rates for all students but especially so for women (Gayles & Ampaw, 2014).

Although experiences in college classrooms are no doubt meaningful for students, Ceci, Ginther, Kahn, and Williams (2014) argue that the roots of gendered differences in mathematically intensive fields are solely in precollege experiences that, among other outcomes, influence the likelihood of men and women pursuing different degree programs. Other studies find that showing an initial interest in STEM fields at the middle or high school level is indeed predictive of STEM degree completion but that demonstrating interest in college is still a significant factor on the pathway to a STEM degree (Maltese, Melki, & Wiebke, 2014; Maltese & Tai, 2011).

In this work, we focus on one important part of the pathway to STEM degree completion: large, foundational university courses from a range of disciplines. The gatekeeping nature of these courses is widely acknowledged—indeed, they are often collectively described as “gateway courses.” To measure GPDs, we use data likely to be available on every college campus: grades in each course compared to expectations formed from an array of prior performance information, including grades in other courses and standardized admissions test scores.

Grades have been widely criticized as poor measures of learning. Nevertheless, grades remain the only measure of academic achievement that all institutions reliably record and value. They are taken seriously by institutions, used as a threshold for passage of courses, to select students for academic awards and honors, and even to dismiss students from campus. As a result, good grades are aggressively pursued by students, sometimes to the detriment of learning (Pulfrey, Buchs, & Butera, 2011). Inflation in average grades over time has been widely reported (D. Freeman, 1999; Jewell & McPherson, 2012; Kostal, Kuncel, & Sackett, 2016; Rojstaczer & Healy, 2012). Although in some contexts this might raise concerns about their utility for comparing student performance, it is worth noting that grade inflation has been relatively small in the foundational courses we study here (Achen & Courant, 2009).

Grades constitute the only universally accessible performance feedback provided to students. Student performance in a course, particularly performance relative to that in other courses, plays an important role in shaping major and career choices (Ost, 2010). For all these reasons, understanding and ultimately addressing the GPDs we report here is essential for ensuring equitable access to participation in STEM careers.

Method

This cross-institutional study is based on administrative data, so we restricted ourselves to covariates that are readily available, complete, and similarly defined from campus to campus. Following a method described by Huberth, Chen, Tritz, and McKay (2015), we used a measure called grade point average in other courses (GPAO) because it is a powerful predictor of students’ final course grades. GPAO is the cumulative GPA for a student calculated across all semesters, including the current semester, excluding only the course enrollment being analyzed. As such, GPAO is a property of a given course enrollment but does not exist when a student has taken only a single course.

In a population largely similar to the students represented in these data, and across a similar range of STEM, social science, and humanities courses, prior research (Huberth et al., 2015; Koester, Grom, & McKay, 2016) has shown GPAO to exhibit the strongest correlation with course grades out of measures regularly collected in administrative databases (e.g., high school GPA and standardized test scores). As the most important predictor of course performance in these studies, GPAO was shown to independently account for 32% of the variance in final physics course grades. Further, traditional cumulative GPAs are known to be good predictors of college outcomes (Creech & Sweeder, 2012; S. Freeman et al., 2007; Gershenfeld, Hood, & Zhan, 2016). Koester et al. (2016) found that additional covariates recorded in administrative data, such as estimated gross income and college of admission, correlate with course grades but explain negligible additional variation in grades.

As such, GPAO helps account for many potential confounding variables that influence student achievement, reducing both systematic and random sources of error. Because each grade is considered relative only to other courses from that institution, the GPAO measure facilitates cross-institutional comparisons even when courses at different institutions may be subject to different grading practices or degrees of grade inflation. Although GPAO is clearly sensitive to the mix of courses each student takes, when comparing all students in a given course of interest, we find empirically that enrolled women and men have taken other courses with similar levels of difficulty (see online Supplementary Materials Section 2.6). Differences in GPAO for female and male students emerge from differences in performance in their other courses rather than from difference in overall grading patterns in those courses.

Six years of student record data for introductory courses were collected at each of five large, public research universities. These student-level data were locally maintained and analyzed separately at each institution using common code written in R (detailed methods are available in the online Supplementary Materials). The overall data set includes 1,122,586 course enrollments across 249 courses in 13 disciplines. The courses are primarily from STEM (i.e., biology, chemistry, engineering, mathematics, physics, and statistics) and social science (i.e., communication, economics, political science, psychology, and sociology) disciplines; accounting and writing courses are included from the business and humanities disciplines, respectively. Selection criteria for both the disciplines and comparable courses are provided in the online Supplementary Materials Section 2.4. In addition to disciplines and generalized course names, we provide classifications related to course structure by labeling each course as either lecture, lab, or mixed. Mixed courses are usually worth four or five credits and contain elements of both a lecture and a lab.

For each course, we focus on two measures: the average grade anomaly (AGA) and the GPD. AGA compares students’ performance in this course to their other courses; it is simply the difference between final course grade and GPAO averaged across all student enrollments in the course. A positive AGA for a course indicates that, overall, students’ final grades in the selected course tended to be higher than their GPAO. We call a positive AGA a grade bonus. In contrast, a negative course AGA indicates that, overall, students’ final grades in the selected course tended to be lower than their GPAO. We call a negative AGA like this a grade penalty. In general, AGA measures average student performance relative to expectation.

GPD compares the AGA between women and men, that is, it measures the gendered difference in performance relative to expectation. A GPD can result from differences between men and women in final course grade, GPAO, or both. For example, a GPD that favors men could result from either (a) men and women having similar GPAOs, with men earning higher grades in the targeted course; (b) women having a higher GPAO than men, with women and men earning similar course grades; or (c) a combination of these two scenarios. Our convention in this work is that a positive GPD favors women and a negative GPD favors men.

Although GPAO is by far the strongest predictor of student performance (Huberth et al., 2015), it is possible that other factors might account for observed GPDs. To test this possibility, the GPD in each course is additionally calculated while accounting for a combination of GPAO, SAT, or ACT Mathematics and English subscores (converting when necessary) and individual course/term factors. Two methods were used: multiple linear regression and optimal matching. Each analysis offers a strength. The regression method is a familiar way to correct for confounding factors and in comparison to the matching method is more precise (i.e., has a smaller standard error) but is also less accurate because of founding assumptions (e.g., that all predictors are linearly independent). The matching method is often noisy and less precise than the regression method, but it is more accurate.

With course grade as the dependent variable, the following covariates for the regression model were selected based on the LASSO (least absolute shrinkage and selection operator) method (Hastie, Tibshirani, & Friedman, 2009) as well as the restrictions inherent in comparing multiple institutions: gender, GPAO, ACT Mathematics and English subscores, and term. Term was included as a categorical variable to account for term-to-term variation in instructors and the time of year that the course was offered, as differences between “on-” and “off-semester” student populations are common.

The matching model included the same factors as the regression model and relies on propensity scores for matching cases and controls (Hansen, 2004, 2007). Matching was performed on a term-by-term basis so long as in each term the course contained more than 50 students; nine courses would otherwise have been included in the data but did not meet this criterion. The differences in GPDs obtained by the matching and regression methods are marginal at best (see online Supplementary Figure S4 and Supplementary Table S6), with the regression and matching methods resulting in the highest and lowest GPD, respectively. Over multiple iterations of this work, we found similar results even with small changes to the baseline predictive model (i.e., to the covariates) using both methods. Analyses were performed using a custom R code (see online Supplementary Materials Section 2.4), and figures were developed using Excel and Tableau, an interactive data visualization program.

In what follows, we report results using GPDs measured by the optimal matching method. On average, this method returns the most conservative measures of GPD of the three approaches, accounting as thoroughly as possible for each student’s prior academic performance.

Results

Comparison across STEM disciplines reveals two trends when each course is characterized as a lecture, lab, or mixed (Figure 1). First, the majority of lecture (74%) and mixed (93%) courses yield a grade penalty (negative AGA), and the majority of lab courses (64%) yield a grade bonus (positive AGA) for students (chi-square p < .001, Cramer’s V = .33). Second, the lecture and mixed courses that yield grade penalties tend to favor men (negative GPD), meaning men have smaller grade penalties in these courses than women. The average GPDs across lecture and mixed courses are –.07 and –.10 grade points, respectively. The lab courses that yield grade bonuses tend to be more equitable, with an average GPD across all lab courses of .01 grade points.

Figure 1.

Gendered performance differences in science, technology, engineering, and mathematics (STEM) courses. Gendered performance differences based on a matching method versus average grade anomaly for 172 introductory STEM courses across five universities, representing 677,949 course enrollments, including lectures (gray), labs (orange), and mixed courses (blue). Crosshairs indicate the average standard error on the mean.

Separating these data by discipline shows that both trends are apparent across biology, chemistry, engineering, and physics courses (Figure 2). It is worth noting that both AGAs and GPDs are especially large and negative for large general chemistry courses—the first STEM courses encountered by many college students. Mathematics and statistics courses exhibit somewhat different patterns. Although the majority of these courses yield a grade penalty, overall they appear to favor neither men nor women, with an average GPD across all courses of –.03 grade points. This result is not unexpected as the ACT Mathematics score covariate, although still second to GPAO, may reasonably explain more of the grade variation in mathematics and statistics courses than it does in biology, chemistry, and physics.

Figure 2.

Gendered performance differences in science, technology, engineering, and mathematics courses by discipline. Gendered performance differences based on matching versus average grade anomaly parsed by discipline for 28 (n = 115,066), 41 (n = 192,487), 53 (n = 222,498), and 47 (n = 130,908) introductory courses across five universities in (a) biology, (b) chemistry, (c) mathematics and statistics, and (d) physics, respectively, including enrollments in lectures (gray), labs (orange), and mixed courses (blue). Crosshairs indicate the average standard error on the mean. Engineering courses are not pictured due to a small sample size; these data are included in the online Supplementary Materials.

Comparison among non-STEM courses in these data (Figure 3) shows that the majority of introductory accounting and economics courses produce grade penalties that favor men (average GPD = –.14), exhibiting a pattern similar to STEM lecture and mixed courses. Conversely, writing courses yield grade bonuses that slightly but significantly favor women (average GPD = .06). Overall, social science courses exhibit little to no GPDs (average GPD = .01). We note again that women tend to slightly outperform men overall in college (Keiser, Sackett, Kuncel, & Brothen, 2016); the lecture courses with significant GPDs favoring men are unusual within the college landscape.

Figure 3.

Gendered performance differences in non–science, technology, engineering, and mathematics courses by discipline. Gendered performance differences based on matching versus average grade anomaly parsed by discipline for 22 (n = 160,828), 39 (n = 172,345), and 16 (n = 111,464) introductory lecture courses across five universities in accounting and economics (gray); communication, political science, psychology, and sociology (blue); and writing (orange). Crosshairs indicate the average standard error on the mean.

At the individual course level, the final course grade and GPAO contribute differently to the observed GPDs. In some cases, the GPDs that favor men tend to result from women having higher GPAOs than men yet earning similar or slightly lower grades. In other cases, we find small GPAO differences but substantial final grade differences such that most of the GPD can be attributed to the difference in grades. Regardless, we find that these differences are stable over time at the individual course level, robust to changes in instructors and the time of year the course was taken (Figure 4).

Figure 4.

Variation in grade and grade point average in other courses (GPAO) in similar courses across institutions. Differences between average GPAO (arrowtail) and average grade (arrowhead) for men (blue) and women (red) by academic year (e.g., 2008 indicates fall 2008 and spring 2009) in selected introductory chemistry, economics, and physics lecture and mixed (M) lecture and laboratory courses. This figure shows that gendered performance differences (GPDs) emerge from a combination of differences in GPAO, average grade, or both. Despite this, the GPDs are consistent from term to term, across changes in both the groups of students and instructors.

Discussion

In these data, we find evidence that GPDs in many courses, although modest in size, are statistically significant and reliably present from term to term; that GPDs in biology, chemistry, and physics lecture courses tend to favor men, whereas those in corresponding lab courses tend to be more equitable; that writing and social science courses (with the exception of economics) do not tend to yield substantial GPDs; and that these results are consistent across five relatively similar universities and six academic years. These patterns mirror those observed in a precursor study local to University of Michigan even though that study accounted for high school GPA, which has a small but unique amount of power in predicting grades (Koester et al., 2016).

We do not focus on precise measurement of the magnitudes of these GPDs. Rather, we stress that, for some courses (e.g., biology, chemistry, physics, accounting, and economics lectures), they are materially significant. Scholarships, university honors, and even employment decisions rely heavily on GPA, often turning on tenth-of-a-point distinctions. Students respond individually to grade signals they receive as well, and prior research suggests that response to these signals may be gendered, compounding the potential impact of modest performance differences (Rask & Teifenthaler, 2008).

Further empirical research is required to ascertain what magnitude of GPD is meaningful to students in which contexts and to what extent the differences might accumulate throughout a student’s degree program. Nevertheless, the presence of statistically and materially significant GPDs in an array of courses creates significant equity concerns for these institutions. It is also important to use parallel measures of performance equity to explore other aspects of identity and background that might intersect with gender, including race and ethnicity, first-generation status, and socioeconomic status. Although the data available for this study do not enable this analysis, we encourage others to pursue this work and provide some first insights from analyses at two institutions (see online Supplementary Materials Section 3).

AGAs themselves raise a different set of questions. That some courses are graded more harshly than others, and that these courses cluster by discipline, is well known and has been true since the adoption of letter grades (Goldman, Schmidt, Hewitt, & Fisher, 1974; King, 2015; Meyer, 1908). Still, this practice perpetuates a system in which it is normal to earn low grades in introductory STEM lecture courses, the starting point for most students who will eventually pursue a STEM major. The comparatively low grades received by students in these courses result from decisions about grading practices made by instructors rather than student ability. Indeed, there is clear evidence that students who take these low-graded courses are, by other measures, especially strong students (Koester, Fogel, Murdock, Grom, & McKay, 2017).

The GPDs we identify here in introductory biology, chemistry, physics, accounting, and economics lecture courses are surprising and clearly not predicted by students’ prior performance. Although previous studies have identified GPDs in particular disciplines at particular institutions (Creech & Sweeder, 2012; Eddy et al., 2014; Kost, Pollock, & Finkelstein, 2009; Lauer et al., 2013; Rauschenberger & Sweeder, 2010), the results presented here provide the first comprehensive, cross-disciplinary picture of how consistent these trends are across an array of similar institutions.

Conflated characteristics of the courses studied here may contribute to patterns in the results. For example, large lecture courses most typically employ high-stakes, timed, and often multiple-choice exams to assess students, whereas lab courses are more often graded through written reports, projects, and lower-stakes quizzes. Although some conflicting evidence exists (Federer, Nehm, & Pearl, 2016; C. Wright et al., 2016), men tend to outperform women on multiple-choice items, and women tend to outperform men on constructed-response exercises (Garner & Engelhard, 1999; Madsen, McKagan, & Sayre, 2013; Weaver & Raptis, 2001). Particular cases in the data appear to support this claim. For instance, a reformed introductory biology course at Institution D that makes use only of constructed-response assessments shows a GPD half that of the prerequisite course in the introductory sequence, although it is also true that the reformed course draws from a subset of the student population in the traditional, prerequisite course.

Another characteristic that may be related to the patterns in GPDs is whether course work tends to be more competitive or collaborative and, relatedly, whether class sizes are large or small, respectively. Especially at large universities like those studied here, lecture courses usually enroll hundreds of students per section, and lab courses usually enroll a few dozen students per section, making collaborative work easier to implement. Although, again, some conflicting evidence exists (Hazari et al., 2010; Micari et al., 2007; Pollock, Finkelstein, & Kost, 2007), women have often been shown to benefit from small-group work and small course sizes more than men (A. Johnson, 2007; Kokkelenberg, Dillon, & Christy, 2008; Lorenzo, Crouch, & Mazur, 2006; Rodger, Murray, & Cummings, 2007; Stump, Hilpert, Husman, Chung, & Kim, 2011), particularly in terms of student attitudes (Springer, Donovan, & Stanne, 1999).

Additionally, women have been shown to prefer collaborator as compared to leader/explainer roles (Eddy, Brownell, Thummaphan, Lan, & Wenderoth, 2015). Again, particular cases in the data appear to support this notion. For example, although each of the introductory engineering design courses included here is labeled as a lecture course in its respective course catalog, design courses usually center on conceiving and building a product with a group. It is unsurprising, then, that these engineering courses cluster with science labs in terms of yielding a grade bonus and generally favoring women. Further, with one exception, all writing courses, which are usually taught in small sections, in these data exhibit GPDs that favor women, and the two accounting courses that favor women have structural characteristics similar to lab courses. Patterns of GPD related to course structure call for research into equitable course design, raising questions of whether evaluative schemes in large lecture classes might disadvantage women as well as how best to support men in writing courses and group work situations.

The repetition of the observed performance differences on all five of these campuses reinforces the need for broader investigation of these patterns across the landscape of higher education. Are GPDs present at private research institutions; public, primarily undergraduate institutions; and community colleges? Although we reasonably expect these results would generalize to other, similar universities, we make no claims about the findings generalizing to other types of universities. These measurements of GPDs are relatively simple to make, relying on administrative data regularly gathered by every institution of higher education, and we encourage faculty, staff, and administrators involved in postsecondary STEM education to examine their own data. We hope that these results will provide the impetus for widespread equity analyses of this kind. When significant GPDs are found, steps should be taken to investigate and address them.

Social psychological interventions designed to improve student performance provide a potential solution, which is being widely explored. Because they do not require changing the structure or mode of instruction of courses, these relatively simple interventions (e.g., values affirmation or sense-of-belonging writing exercises) are attractive approaches to reducing GPDs. Although they have been found effective in some contexts (Miyake et al., 2010; Unkovic, Sen, & Quinn, 2016; Walton, Logel, Peach, Spencer, & Zanna, 2015), replication has not always been possible (Madsen et al., 2013), and there is much to learn about how to apply these interventions at scale (Paunesku et al., 2015; Yeager & Walton, 2011). Large-scale randomized trials of the impact of values affirmation on GPDs are now in progress at one of our institutions.

These results suggest connections between GPDs and the structure of evaluation in courses. It is possible that modest changes to evaluative schemes might reduce GPDs, for example, reducing the time pressure in exams. It is also important to recognize differences among individual STEM disciplines (Cheryan, Ziegler, Montoya, & Jiang, 2017). The solution, then, is not to broadly prescribe a formulaic ratio of multiple-choice to constructed-response assessment items, suggest strict changes in class size, or ask faculty to send encouraging e-mails to particular groups of male or female students. Indeed, we agree with the assessment of Halpern et al. (2007) that “there are no single or simple answers to the complex questions about sex differences in science and mathematics” (p. 1).

This work should compel those at institutions of higher education to ask, as many are already doing (Elliott, 2016), how we can learn from this information to change practices in whatever ways are appropriate in our local contexts. Understanding student performance in context is an important step in pursuing equity (M. Wright, McKay, Hershock, Miller, & Tritz, 2014). Systems capable of personalizing at scale and responding to differences among students rather than prescribing a single solution for all hold some promise. Huberth et al. (2015), for example, describe a digital mentoring tool that is now being tested for its ability to reduce stereotype threat for women in high-enrollment undergraduate science courses.

Grades are consequential performance measures and clearly impact persistence (King, 2015). It is unclear whether men or women are more sensitive to their STEM grades in persistence decisions (Ost, 2010; Rask, 2010), and these differences may be field dependent. Regardless, grade penalties that are worse for female students than for male students create yet another headwind impeding gender equity in STEM. There is widespread evidence that faculty, especially male STEM faculty, are reluctant to accept evidence of gendered biases in STEM (Handley, Brown, Moss-Racusin, & Smith, 2015; Moss-Racusin, Molenda, & Cramer, 2015). In this light, continued investigation of GPDs, coupled with efforts to understand their correlates and causes, is imperative.

Unexplained GPDs of the kind reported here cannot be ignored or simply allowed to persist.