Global S&T Development Trend Analysis Platform of Resources and Environment
| DOI | 10.1126/science.aba7377 |
| Understanding persistent gender gaps in STEM | |
| Joseph R. Cimpian; Taek H. Kim; Zachary T. McDermott | |
| 2020-06-19 | |
| 发表期刊 | Science
 |
| 出版年 | 2020 |
| 英文摘要 | Gender gaps in science, technology, engineering, and math (STEM) college majors receive considerable attention, and it is increasingly recognized that not all STEM majors are equal in terms of gender disparities ([ 1 ][1]–[ 4 ][2]). For example, the male-to-female ratio among U.S. college majors in biology, chemistry, mathematics, and many other STEM fields is now about 1-to-1 ([ 2 ][3], [ 5 ][4]), whereas in physics, engineering, and computer science (PECS), the ratio appears to have plateaued at about 4-to-1 ([ 2 ][3], [ 4 ][2], [ 5 ][4]). Here, we make two important contributions, showing (i) how gender relates to pursuit of a PECS degree throughout the achievement distribution and (ii) that student characteristics that predict PECS pursuit in the literature are not equally predictive of the gender gap throughout the achievement distribution. We find that a surprisingly large number of low-achieving men are majoring in PECS, relative to women, and this cannot be explained by an extensive set of student-level factors proposed in the prior literature ([ 6 ][5]–[ 10 ][6]). We can, however, explain the gender gap among high-achieving students. These patterns suggest that interventions to close the gender gap may work to attract high-achieving women; yet, something beyond these student factors may be attracting low-achieving men and repelling average- and low-achieving women, and without addressing those factors, it is unlikely that the PECS gender gap will fully close.
Despite historical emphasis on measuring whether gender gaps in STEM achievement exist and if there is greater male variability in achievement, there is surprisingly little research linking gender differences in achievement to pursuit of a college major [see supplementary materials (SM), section 1]. Work that links these factors accounts for average differences in STEM achievement and suggests that controlling for them does little to reduce the gender gaps in PECS, but it does not tell us where in the achievement distribution the PECS majors are coming from or if this differs for men and women.
We use new data from the U.S. Department of Education's High School Longitudinal Study of 2009 (HSLS:09), a nationally representative, longitudinal cohort of students who were in ninth grade during the 2009–2010 school year and were followed for 7 years. For several reasons, the HSLS:09 is the ideal dataset for examining who majors in PECS throughout the achievement distribution. First, it allows us to study a critical 7-year period, from the beginning of high school through the first few years of college when students declare majors ([ 3 ][7], [ 11 ][8]). Second, its multiple measures of STEM achievement during high school allow for more comprehensive and robust measurement than most datasets. We combine these measures of STEM achievement into a composite STEM achievement variable with very high reliability (Cronbach's α = 0.86; moreover, each individual measure yields a similar pattern of gender gaps; see fig. S1). Third, surveys conducted during the summer and fall of 2013 (after high school graduation for most) asked students about their college major intentions; then, they were asked 3 years later which major they were ultimately pursuing, allowing us to examine persistence in interest in a major and newfound interest in a major.
The final analytic dataset contains 5960 students, with complete data on majors and STEM achievement, and multiply-imputed data for any missing covariates. All analyses use attrition-adjusted longitudinal sampling weights to ensure national representativeness and to adjust standard errors to account for the multistage survey sampling design. Detailed methods are in the SM. Although the data do not capture degree completion, they are likely a reasonable proxy, because research suggests that persistence to STEM degree is similar for both genders (see SM, section 2). The data reflect the United States, not international contexts.
Of all U.S. students attending a 4-year college, 23.8% of men and 5.5% of women pursued a PECS major, representing a substantial and significant gender gap on average ( p < 0.001). When ranked from lowest to highest achieving, we see that males at and below the 1st percentile are majoring in PECS at the same rate as females at the 80th percentile of STEM achievement (see the figure, top). At every point in the STEM achievement distribution, men majored in PECS at higher rates than did women (all values of p < 0.004). The raw percentage-point gap in the top third of the achievement distribution is about twice the size of the gap in the bottom third (23 versus 12%, all values of p < 0.001; see Model 1 in the table). The size of these gaps is worth noting when thinking about the scale of the disparity throughout the distribution; however, percentage-point gaps do not factor in base rates, and thus, the gap represented in ratios is critical when examining representation. Importantly, the ratio at the bottom of the distribution is much larger than at the top. Overall, the male-to-female ratio in PECS is 4-to-1, whereas at the bottom decile, it is greater than 10-to-1 and at the top decile less than 2-to-1 (see the figure, top). This point is critical for considering retention in the PECS “pipeline” and should fundamentally shift how progress toward gender equity is evaluated.
Who majors in PECS can be disaggregated further by intention just after high school graduation to major in PECS: students who intended to major in PECS versus those who had different intentions (see the figure, middle and bottom). More males intended to major in PECS than did females (23.7 versus 5.2%, p < 0.001). Although persistence rates from post–high school intention to college major are similar overall (74 versus 75% for men and women, respectively; p = 0.87), these rates mask a substantial gender differential throughout the distribution, where females and males persisted at similar rates among the highest achievers but males persisted more among the lowest achievers (see the figure, middle). Indeed, among PECS-intenders who persisted, females averaged almost nine percentiles higher on the STEM composite measure ( p = 0.004; table S1), because the male average was brought down by low achievers who persisted. This pattern of men's persistence being less sensitive to feedback on their actual performance is consistent with a general pattern found in other recent work (see SM, section 3). Importantly, males not previously intending to major in PECS join these majors at significantly higher rates than females at all points in the distribution (all values of p < 0.008; see the figure, bottom). Among non-PECS-intenders, males scoring at and below the 1st percentile were at least as likely to join PECS as females scoring above the 99th percentile who did not initially intend to major in PECS.
Supplemental analyses disaggregated PECS into physics only, engineering only, and computer science only (figs. S2 to S4), and common (although not identical) patterns were found, supporting the combining of majors for this main analysis.
Are these gaps explained by other differences between men and women? Prior research suggests that women may choose PECS (and STEM, more generally) less often than men because they have lower confidence in their math abilities ([ 6 ][5]), have other options owing to a comparative advantage in English and/or reading ([ 7 ][9]), value societal goals and a work-life balance more than salary ([ 8 ][10]), take different high school STEM courses ([ 9 ][11]), and have different career aspirations ([ 10 ][6]).
Our ability to statistically account for the PECS gender gap with these and other variables depends on whether we look at low-, average-, or high-achieving students (see the table and fig. S5). Overall, our model that simultaneously adjusts for a range of student attributes explains almost all of the gender gap among high achievers, but not a substantial amount among low achievers. Among the top third of achievers, the average male-female gap goes from 23.0% ( p < 0.001) in the unadjusted model to just 0.3% and no longer significant ( p = 0.84) in the final adjusted model, for a total reduction of 98.6% [i.e., equal to (23.0% − 0.3%)/23.0%]. The rich set of covariates, however, does not do as well predicting the gaps among average- or low-achieving students. For example, the gap among the bottom third reduced only from 11.7 to 7.7% and remained significant ( p < 0.001) after all adjustments (see the table). All patterns are robust to a wide range of sensitivity checks, including mean replacement for missing data, not using sampling weights, using cross-sectional weights, nonparametric models, and restricting the sample to students in colleges with engineering programs (figs. S6 and S7).
![Figure][12]
Who majors in PECS?
The shaded areas represent one standard error above and below the estimated means.
GRAPHIC: X. LIU/ SCIENCE
We examine gender gaps in other STEM majors and race gaps in STEM as a point of comparison for our PECS-focused analyses (fig. S5 and tables S2 and S6 to S9. We find no significant gender gaps on average or throughout the achievement distribution in other STEM majors (table S11) after adjusting for all covariates. Similarly, there are no significant race gaps [Asian or white versus underrepresented racial minorities (URM)] in either PECS or other STEM, either on average or throughout the distribution after an extensive set of controls. It is also worth exploring gender-by-race patterns in PECS ([ 12 ][13]). The data revealed notable consistencies in gender gaps when looking just among Asian or white individuals or just among URM race groupings (figs. S8 and S9).
This new work reveals the critical importance of assessing the gender balance in PECS throughout the achievement distribution, rather than averages. Here, we discuss how this distributional approach (i) can transform how the field evaluates gender equity in the STEM pipeline and (ii) differentially affects how to target interventions to high versus average and low achievers. Ultimately, this approach suggests that student-level factors are complex and can vary across achievement levels. We urge the field to continue to look beyond student-level factors—toward discipline and societal ones—to further reduce the gender gap and raise the overall achievement of students in PECS.
Some have argued that the STEM pipeline from undergraduate to graduate school has stopped “leaking” because the proportion of females in STEM at the undergraduate level has remained constant to the graduate level ([ 1 ][1], [ 13 ][14]). Our analyses suggest that looking at average persistence rates may obscure differential leaking throughout the achievement distribution. If PECS graduate programs selected from the highest-achieving students solely on the basis of merit, we should expect to see a considerably higher proportion of females in PECS graduate programs than in undergraduate programs. For example, if graduate programs select students on the basis of STEM achievement alone from the top decile, where the male-to-female ratio is less than 2-to-1, we should expect to see more than 33% females in graduate school, not 20%, which is the average during undergraduate school.
Thus, the apparent lack of a leaking pipeline on average found in previous studies ([ 1 ][1], [ 13 ][14]) is not the appropriate standard for evaluating equity because it ignores that female PECS majors were initially higher achieving. Furthermore, evidence of differential leaking throughout the distribution could help reconcile seemingly disparate findings, such as a pipeline that no longer leaks on average ([ 1 ][1], [ 13 ][14]) and male-dominated fields holding beliefs that deter females while sending welcoming signals to lower-achieving males for graduate education ([ 2 ][3], [ 14 ][15]). It may be precisely through a male-favoring culture that graduate PECS programs keep the proportion of females at the same low rates as they are during undergraduate school rather than allowing them to rise, as would be expected given females' aforementioned higher competence.
The differential ability of covariates to explain the PECS gender gap at different points in the achievement distribution has direct implications for how researchers and educators should intervene.
For high-achieving students, our analyses suggest that efforts to boost women's PECS aspirations and intentions (i.e., the strongest set of predictors)—such as role-model interventions and Girls Who Code ([ 4 ][2], [ 15 ][16])—may work well to eliminate the gender gap in pursuing PECS at the top of the distribution. That is, although each set of variables chipped away at the gender gap at the top of the distribution, none of them—alone or collectively—were as strong or consistent a predictor as prior intention to major in PECS and having aspirations for a PECS-related career (see the table, fig. S5A, and table S6). This clearly implies that to close the PECS-pursuit gender gap among the top achievers, the critical piece is in converting all other STEM-positive factors (e.g., confidence and interest) into actual aspirations and intentions.
Unfortunately, for average- and low-achieving students, our analyses also revealed that the field's understanding of how to get average- and low-achieving girls into the field is more limited. The same factors that explained the PECS gender gap at the top of the distribution collectively explained less and less of the gap further down the distribution. Thus, the interventions that may work for high-achieving girls are unlikely to work for lower-achieving ones. Moreover, focusing efforts on attracting high-achieving women to PECS may backfire: Whereas men have male peers to serve as role models throughout the achievement distribution, having only high-achieving role models for women may send signals to average- and low-achieving women that they do not have the necessary STEM skills to succeed (see SM, section 4).
![Figure][12]
Modeling the gender gap in PECS
Disaggregating PECS majors by achievement tiers raises a related question: Should PECS be attracting and retaining low-STEM-achieving students (of any gender), or is there an alternative pathway to meet the PECS labor market demands with more competent students? Students in the lowest third of achievement have mean math SAT scores and high-school STEM grade point averages that are well below average, especially for these math-intensive fields. About 90% of the students majoring in PECS from this tier are men, and we have shown that low-achieving men are both retained in and attracted to these fields during college (see the figure, middle and bottom, and figs. S3 and S4). Thus, the question about whether PECS wants low-achieving students is inextricably intertwined with a different question: Why are men—and not women—with low achievement more drawn to these fields? Although we have shown that this question cannot be answered by gender differences in a wide range of student-level factors—including self-efficacy, math or science identity, and earlier aspirations—it is possible that the masculine culture of these fields and the gender stereotypes attached to PECS ([ 2 ][3], [ 4 ][2], [ 11 ][8], [ 14 ][15], [ 15 ][16]) may lead to retaining less-qualified men over more-qualified women. Thus, it is important for PECS fields to break down the barriers for women to enter the field and, in doing so, raise the overall quality of the students entering these fields.
All of this suggests that, although role-model interventions, coding camps, and fostering strong peer networks are worth pursuing, there are real obstacles to achieving gender parity in PECS that transcend these interventions, which may work only for a select group of high-achieving women. To develop effective interventions for gender equity and raise the overall quality of students in PECS, it is critical to recognize that the gender imbalance varies throughout the achievement distribution and may be perpetuated by male-favoring cultures that disproportionately attract low-achieving men to the field.
[science.sciencemag.org/content/368/6497/1317/suppl/DC1][17]
1. [↵][18]1. S. J. Ceci,
2. D. K. Ginther,
3. S. Kahn,
4. W. M. Williams
, Psychol. Sci. Public Interest 15, 75 (2014).
[OpenUrl][19][CrossRef][20][PubMed][21][Web of Science][22]
2. [↵][23]1. S. Cheryan,
2. S. A. Ziegler,
3. A. K. Montoya,
4. L. Jiang
, Psychol. Bull. 143, 1 (2017).
[OpenUrl][24][CrossRef][25][PubMed][26]
3. [↵][27]1. C. Riegle-Crumb,
2. B. King,
3. E. Grodsky,
4. C. Muller
, Am. Educ. Res. J. 49, 1048 (2012).
[OpenUrl][28][CrossRef][29]
4. [↵][30]1. C. Hill,
2. C. Corbett,
3. A. St. Rose
, “Why so few? Women in science, technology, engineering, and mathematics” (Report, American Association of University Women, 2010).
5. [↵][31]National Science Board, “2018: Science & engineering indicators” (NSB-2018-1, National Science Foundation, 2018).
6. [↵][32]1. L. J. Sax,
2. M. A. Kanny,
3. T. A. Riggers-Piehl,
4. H. Whang,
5. L. N. Paulson
, Res. High. Educ. 56, 813 (2015).
[OpenUrl][33]
7. [↵][34]1. T. Breda,
2. C. Napp
, Proc. Natl. Acad. Sci. U.S.A. 116, 15435 (2019).
[OpenUrl][35][Abstract/FREE Full Text][36]
8. [↵][37]1. A. B. Diekman,
2. M. Steinberg,
3. E. R. Brown,
4. A. L. Belanger,
5. E. K. Clark
, Pers. Soc. Psychol. Rev. 21, 142 (2017).
[OpenUrl][38]
9. [↵][39]1. N. Friedman-Sokuler,
2. M. Justman
, Econ. Educ. Rev. 53, 230 (2016).
[OpenUrl][40]
10. [↵][41]1. S. L. Morgan,
2. D. Gelbgiser,
3. K. A. Weeden
, Soc. Sci. Res. 42, 989 (2013).
[OpenUrl][42][CrossRef][43][PubMed][44]
11. [↵][45]1. C. M. Ganley,
2. C. E. George,
3. J. R. Cimpian,
4. M. B. Makowski
, Am. Educ. Res. J. 55, 453 (2018).
[OpenUrl][46]
12. [↵][47]The National Academies of Sciences, Engineering, and Medicine, “Promising practices for addressing the underrepresentation of women in science, engineering, and medicine: Opening doors (Consensus study report, The National Academies Press, 2020).
13. [↵][48]1. D. I. Miller,
2. J. Wai
, Front. Psychol. 6, 1 (2015).
[OpenUrl][49][CrossRef][50][PubMed][51]
14. [↵][52]1. S.-J. Leslie,
2. A. Cimpian,
3. M. Meyer,
4. E. Freeland
, Science 347, 262 (2015).
[OpenUrl][53][Abstract/FREE Full Text][54]
15. [↵][55]1. N. Dasgupta,
2. J. G. Stout
, Policy Insights Behav. Brain Sci. 1, 21 (2014).
[OpenUrl][56]
Acknowledgments: We thank A. Cimpian, Y. Copur-Gencturk, S. Lubienski, M. Makowski, M. Perry, L. Stiefel, and J. Timmer for helpful comments. Z.T.M. is funded through an Institute of Education Sciences (IES) Predoctoral Interdisciplinary Research Training grant to New York University (R305B140037). Although HSLS:09 has a public-use version of the dataset, some variables used for the analysis are only available in a restricted-use dataset from the U.S. Department of Education. For information on restr Sicted-use licenses, see |
| 领域 | 气候变化 ; 资源环境 |
| URL | 查看原文 |
| 引用统计 | |
| 文献类型 | 期刊论文 |
| 条目标识符 | http://119.78.100.173/C666/handle/2XK7JSWQ/276680 |
| 专题 | 气候变化 资源环境科学 |
| 推荐引用方式 GB/T 7714 | Joseph R. Cimpian,Taek H. Kim,Zachary T. McDermott. Understanding persistent gender gaps in STEM[J]. Science,2020. |
| APA | Joseph R. Cimpian,Taek H. Kim,&Zachary T. McDermott.(2020).Understanding persistent gender gaps in STEM.Science. |
| MLA | Joseph R. Cimpian,et al."Understanding persistent gender gaps in STEM".Science (2020). |
| 条目包含的文件 | 条目无相关文件。 | |||||
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
修改评论