Systemic inequalities for LGBTQ professionals in STEM – Science Advances

Career opportunities and resources

Figure 1 summarizes results testing the hypothesis that LGBTQ STEM professionals enjoy fewer career opportunities than their otherwise similar non-LGBTQ peers (H1). As shown by both the difference in means (indicated by the height of the bars and nonoverlapping CIs) and the significance levels of the difference between LGBTQ and non-LGBTQ professionals holding constant variation by other demographic, discipline, and job-related factors (indicated by asterisks), LGBTQ STEM professionals were significantly less likely to report they had opportunities to develop their skills than their non-LGBTQ peers and were less likely to report that they had access to the resources they needed to do their jobs well. LGBTQ STEM professionals were also less likely than their non-LGBTQ peers to be confident that they could whistleblow without fear of retaliation. See table S2 for multivariate regression models for each outcome.

Professional devaluation

A second possible dimension of LGBTQ disadvantage is the likelihood that respondents report that their colleagues devalued or discounted their STEM expertise (H2). The professional devaluation scale captures, for example, whether respondents reported being treated as less skilled professionals than their colleagues and whether they were held to higher standards of productivity (see scale measure operationalization below). As Fig. 2 indicates, LGBTQ respondents were significantly more likely to experience professional devaluation than their non-LGBTQ peers who were otherwise similar along demographic, work experience, education level, subfield, and employment sector characteristics. (See table S2 for multivariate models.) Represented another way, LGBTQ professionals were 20.2% more likely than their non-LGBTQ peers to agree that they had experienced at least one of the five facets of devaluation measured in the professional devaluation scale.

Social exclusion

A third potential dimension of LGBTQ inequality is social marginalization (H3). Again, compared to their non-LGBTQ peers, LGBTQ professionals were significantly more likely to experience exclusion by their colleagues (e.g., to not feel like they “fit in,” to be excluded from invitations to after-work social gatherings). See Fig. 3 and measure operationalization below. A third (32.9%) of LGBTQ STEM professionals experienced one or more of these types of social exclusion, compared to less than a quarter (22.7%) of their non-LGBTQ peers. LGBTQ STEM professionals were also about 30% more likely than their non-LGBTQ peers to have experienced harassment at work in the past year (Fig. 4). See table S2 for multivariate models.

Health and wellness difficulties

A fourth dimension of inequality we considered is whether the frequency of experiencing health and wellness difficulties differed by LGBTQ status. Figure 5 presents the average responses on several health and wellness measures for LGBTQ and non-LGBTQ respondents. Controlling for variation by demographic and job-related factors, LGBTQ STEM professionals experienced minor health problems, insomnia, stress, and depressive symptoms more frequently in the past year than their peers (supporting H4). Compared to their non-LGBTQ counterparts, LGBTQ persons were 27% more likely to have experienced minor health problems at least some of the time in the past year, 41% more likely to have experienced insomnia, 22% more likely to have felt nervous or stressed from work, and 30% more likely to have experienced one or more depressive symptoms in the past year.

Mediation analysis with structural equation modeling (SEM) found significant indirect effects of LGBTQ status on these health difficulties through the career limitations, professional devaluation, and social marginalization measures discussed above. (See table S2 for regression models and table S3 for focal direct and indirect coefficients from the SEMs.) These significant indirect effects suggest that LGBTQ STEM professionals’ greater likelihood of experiencing career limitations, devaluation, and marginalization at work helps account for why they experience these health and wellness difficulties more frequently than their peers.

Intentions to leave

Fifth, we examined whether there were differences in respondents’ intentions to leave STEM by LGBTQ status. As hypothesized (H5), LGBTQ STEM professionals thought about leaving their job more frequently and were more likely to have considered leaving their STEM field entirely, compared to their non-LGBTQ peers (controlling for other factors). See Fig. 6. Represented more concretely, 22% of LGBTQ professionals, versus 15% of their non-LGBTQ peers, had thought about leaving their STEM job at least once in the last month, and 12% of LGBTQ respondents (versus 8% of non-LGBTQ respondents) planned to leave their STEM profession within the next 5 years. As with the health outcomes, mediation analysis showed that LGBTQ professionals’ experience with career limitations, devaluation, and marginalization helps account for why they had higher intentions to leave than their non-LGBTQ peers. (See table S2 for regression models and table S4 for SEM coefficient estimates.)

Intersectional patterns and variation by subfield and sector

Table S5 presents supplemental analyses examining the extent to which these patterns of LGBTQ disadvantage vary intersectionally by other demographic factors (gender identity, race/ethnicity, and age) and whether they vary systematically by STEM subfield or employment sector (51, 52). Consistent with literature on the particular challenges they face in the workforce (19), transgender and gender nonbinary respondents reported experiencing minor health problems, stress, and depressive symptoms more frequently than their cisgender sexual minority peers and non-LGBTQ respondents and were more likely to have considered leaving their STEM jobs. Other disadvantages were amplified for LGBTQ professionals of color and LGBTQ-identifying women: LGBTQ-identifying women and racial/ethnic minorities were more likely than white and men LGBTQ STEM professionals, respectively, to experience professional devaluation and harassment at work. These intersectional patterns are echoed in the LGBTQ-only models in table S7. Overall, however, we find that the dimensions of disadvantage documented above were present for persons across specific identities within the LGBTQ umbrella and among LGBTQ persons across other axes of demographic variation (e.g., age and race). In addition, the interaction analyses in table S7 show little variation in the LGBTQ status disadvantages by specific STEM disciplines or employment sectors. This suggests that these LGBTQ disadvantages are not isolated to certain STEM fields or to certain employment sectors but may be operating across the U.S. STEM workforce.

Alternative explanations

Last, we address two possible alternative explanations for the results represented in the figures above. One potential alternative explanation is that these patterns are driven by benign supply-side differences between LGBTQ and non-LGBTQ professionals’ average education level, dedication, and/or work ethic. To test this explanation, we compared LGBTQ STEM professionals’ highest degree, hours worked, work dedication, and likelihood of having a “core technical” primary job responsibility (basic research, design, or computer programming) to that of their non-LGBTQ peers, holding constant their subfield, employment sector, employer size, and other demographic characteristics. Results from these analyses refute this supply-side explanation: Compared to their non-LGBTQ peers, LGBTQ STEM professionals were equally highly educated and just as likely to have core technical primary job responsibilities. They worked just as many hours, on average, as non-LGBTQ professionals and were just as personally committed to their work. (See table S8.)

A second possible alternative explanation is that LGBTQ professionals have uniformly more negative views of their jobs that are unrelated to LGBTQ-specific mistreatment. To test this, we reran all models with a control for respondents’ job satisfaction. The LGBTQ differences documented above were generally robust to this control for job satisfaction, although job satisfaction is strongly negatively correlated with the devaluation and marginalization measures tested in the models and thus reduces the predictive power of LGBTQ status in those models. (See table S9.) These analyses offer counterevidence to contentions that the outcomes above are the result of LGBTQ workers’ weaker qualifications or work commitment or their propensity to have more negative assessments of their jobs overall.

Data and participants

We conducted the analyses above using survey data from representative samples of the United States–based membership of 21 STEM-related professional societies. These societies span the physical and life sciences, mathematics, and engineering. They encompass 8 U.S. national flagship disciplinary societies in the natural, life, and physical sciences and mathematics; 5 U.S. national flagship disciplinary societies in engineering; 4 interdisciplinary STEM societies; 2 STEM teaching-focused societies; and 2 demographic-focused professional societies. We do not specify the names of these societies to protect respondent confidentiality.

Our analysis uses the 25,324 United States–based full-time STEM professionals who participated in the survey, including 1006 who identified as LGBTQ. We excluded from our analysis those respondents who were retired, out of work, or full-time students at the time of the survey. A third of the sample (33%) worked in the for-profit industry sector; 39.9% worked in higher education; and the remaining 26% work in government, nonprofit, and K-12 sectors. The study was approved by the human subjects board at each author’s institution. Potential respondents were informed about the study on the landing page of the online survey; only those who gave their consent were allowed to continue with the survey. See the Supplementary Materials for additional details on survey collection procedures and validity and reliability testing.

Operationalization of outcome measures

We capture respondents’ career opportunities and resources with a scale measure and two single-question variables. Career opportunities is a scale measure (α = 0.633) that averages two questions that asked respondents’ level of agreement, based on their experiences in their current STEM job, that “I have been given opportunities to take on a leadership role” and “I have limited opportunities to develop my skills” [Reverse Coded] [1 = strongly disagree (SD) to 5 = strongly agree (SA)]. Sufficient resources is a measure of respondents’ level of agreement that “I have sufficient resources to get my job done” (1 = SD to 5 = SA). Comfort with whistleblowing was measured with a question that asked respondents’ level of agreement that “I can disclose a suspected violation of any law, rule, or regulation without fear of reprisal” (1 = SD to 5 = SA).

Professional devaluation is a scale measure (α = 0.753) averaging respondents’ agreement with the following five statements: “I have to work harder than my colleagues to be perceived as a legitimate professional,” “My colleagues sometimes think I am less productive than I actually am,” “In my workplace, my work is respected” [Reverse Coded], “My colleagues treat me as an equally skilled professional” [Reverse Coded], and “I am held to the same standards as others for promotion or advancement” [Reverse Coded] (1 = SD to 5 = SA).

Social exclusion is a scale measure (α = 0.734) of respondents’ agreement with the following three statements: “Overall, I feel I ‘fit in’ with the other people in my workplace” [Reverse Coded], “I feel included in casual conversations among my colleagues” [Reverse Coded], and “when my co-workers get together socially at lunch or after work I am usually included in the invitation” [Reverse Coded] (1 = SD to 5 = SA). The harassment measure is a dichotomous item that asked whether respondents were “harassed verbally or in writing on the job” once or more during the past 12 months (1 = yes, 0 = no).

We measure the frequency of experiences of health and wellness difficulties with three single question items that asked “how often, if at all, have you experienced the following in the last year:” “been bothered by minor health problems such as headaches or stomach upset?” “Had trouble sleeping to the point that it affected your performance on and off the job?” and “Felt nervous or stressed?” (1 = never to 5 = very often). In addition, reflecting standard survey measures used to detect symptoms of depression (54), we used the following two markers (averaged together): How often in the past year they “felt that you were unable to control the important things in life?” and “felt that difficulties were piling up so high that you could not overcome them?” (1 = never to 5 = very often).

We measure intentions to leave STEM with two items: one that asked how frequently in the last year they had “thought about leaving my current job?” (1 = never to 5 = almost every day) and another that asked how long they planned to “stay in your current profession (even if you change jobs)?” (5 = less than 5 years; 4 = 5 to 9 years; 3 = 10 to 15 years; 2 = 16 = 20 years; 1 = the rest of my career). These two measures capture both respondents’ intentions to leave their current STEM job and their intentions to leave their STEM field altogether.

Where appropriate, we reverse-coded items to maintain consistent positive-to-negative or negative-to-positive numerical scaling on items within the same index or hypothesis. Reverse coding was used for interpretive ease only and did not alter the strength of the statistical relationships between dependent and independent variables.

LGBTQ status

We designed our LGBTQ status measures in accordance with best practice guidelines from the Williams Institute, the Center for Disease Control, and the American Sociological Association’s Committee on the Status of LGBTQ People in Sociology. We measured LGBTQ status with a set of indicators that asked separately about respondents’ sexual identity and gender identity. First, respondents were asked, “please mark the category that best matches your sexual identity,” and they could choose between the following options: “Straight or heterosexual,” “Gay or Lesbian,” “Bisexual,” “Queer,” “Something else (please specify)” or “I do not know how to answer.” Anyone who marked “Gay or Lesbian,” “Bisexual,” or “Queer” was included in the LGBTQ category. Respondents who marked “do not know” or “something else” were asked in a follow-up question, “Do you identify as part of the LGBTQ community?” Those who answered yes to this question were included in the LGBTQ category in our analysis.

Gender identity was measured with a sequence of three questions. The first question asked “what sex were you assigned at birth?” “Male” or “Female.” The second question asked, “How do you currently describe yourself?”: “Male,” “Female,” “Transgender Male (FTM),” “Transgender Female (MTF),” “Something else,” or “I do not know how to answer.” Respondents whose answers on the first and second questions were different received the following confirmation question: “Just to confirm, you were assigned a different sex at birth than how you currently describe yourself. Is that correct?” “Yes” or “No.” This confirmation question limited the number of false positives on transgender or gender nonbinary status—an important step for appropriately capturing proportionally small populations like non-cisgender persons. Respondents who answered yes to this confirmation question were included in the LGBTQ category. Respondents who marked “something else” or “I do not know” on the current gender identity question were coded as “gender nonbinary” for their current gender identity category and were given the follow-up question, “Do you identify as part of the LGBTQ community?” We pretested this question structure in the pilot survey of STEM professionals and in a survey of engineering students at eight universities (37). We paid careful attention to these measures of LGBTQ identity in the face validity and construct validity testing described in the Supplementary Materials.

Respondents who indicated that their current gender identity was female (whether they identified as cisgender or transgender) were coded as women; respondents who indicated that their current gender identity was male (whether they identified as cis- or transgender) were coded as men. Because of the very small proportion of respondents who identified as gender nonbinary, and our desire to protect the confidentiality of these respondents, we did not provide results for gender nonbinary respondents as a separate category in most regression models. Instead, cis- and transgender women are compared both to cis- and transgender men and gender nonbinary respondents. In the intersectional analyses in tables S5 and S7, we did assess the particular experiences of transgender and gender nonbinary persons vis a vis their cisgender sexual minority peers.

Control measures

We included controls for several important demographic characteristics in each model. We controlled for the racial/ethnic categories with which respondents identified (they could choose multiple): Hispanic/Latinx, Black, Asian, Native American/Asian Pacific Islander, white, and other racial/ethnic category (1 = yes, 0 = no); respondent age (in years), their highest degree (1 = high school or less to 8 = PhD), and the highest education attained by a parent (1 = high school degree or less to 8 = PhD).

We included controls for respondents’ STEM field (life science, physical science, computer science and mathematics, engineering, or other STEM field); social science fields were excluded as they have distinctive demographic and cultural contexts (1). We also controlled for other employment factors: employment sector (university/college, nonprofit, government, K-12 education, private sector, or other employment sector); employer size (1 = less than 10 employees to 8 = more than 25,000 employees); whether they had supervisory responsibilities (1 = yes, 0 = no); and whether their primary work responsibility was a core technical task of basic research, design, or computer programming (1 = yes, 0 = no). Last, each model included dichotomous indicators for the STEM professional societies from which respondents were recruited.

Analytic strategy

Figures 1 to 6 present predicted values for LGBTQ (shaded bars) and non-LGBTQ (unshaded bars) STEM professionals, holding constant variation by demographics, subfield, employment sector, and other job-related factors. To produce these values, we ran ordinary least squares (OLS) and logistic regression models (as appropriate) in Stata predicting each outcome measure with LGBTQ status and controls. Using the resulting regression equations, we calculated the marginal values when LGBTQ = 1 and LGBTQ = 0, holding all other controls at their mean. The asterisks above each pair of bars represent the two-tailed significance level of LGBTQ status produced by the regression analyses presented in table S2. While the raw means on each outcome measure show the same patterns of variation and significance levels by LGBTQ status as the predicted means in Figs. 1 to 6, raw means are more difficult to interpret given the possibility of variation by other confounding variables (e.g., respondent age). Hence, we opted to present predicted means that control for this variation. Where appropriate, we also present percentage differences in the proportion of LGBTQ and non-LGBTQ respondents who reported having experienced the career limitations, devaluation, and marginalization outcomes in question.

Robustness checks and supplementary analysis

To test the robustness of these findings to alternative analytic approaches, we reran the analyses using three additional modeling strategies. First, instead of including individual dichotomous controls for each professional society, as we did above, we reran the analyses as hierarchical linear models with respondents nested in professional societies using the MIXED command in Stata. Second, instead of using multiple imputation to handle missing data, as is standard practice, we ran each model with listwise deletion. Last, we created a weighting variable that adjusted the distribution of respondents in our sample by major demographic and employment characteristics (gender, race/ethnicity, STEM field, and sector) to match the distribution that exists in the U.S. STEM workforce nationally [as represented by the National Science Foundation (NSF) data in table S1] and reran the analyses with this weight. The patterns presented above and in the Supplementary Materials were replicated using each of these three alternative modeling strategies, indicating that the results are robust to this variation in analytic approach.

Tables S8 and S9 present analyses addressing the two alternative explanations described above. Contrary to possible arguments that these LGBTQ differences were driven by key supply side differences in education level, work ethic, or commitment to STEM, we found no significant differences by LGBTQ status on those measures (see table S8).

In addition, we reran the models with a control for respondents’ job satisfaction. Job satisfaction is strongly inversely correlated with workers’ experiences of social exclusion and professional devaluation in the workforce generally (55) and in this sample specifically. Hence, the inclusion of a control for job satisfaction should sharply curtail or eliminate the explanatory power of LGBTQ status on the outcomes if these differences were just a matter of LGBTQ workers’ personal unhappiness with, or greater inclination to “complain” about, their work. Table S9 compares the coefficients and significance levels for LGBTQ status in regression models with and without the job satisfaction measure. As above, all models control for demographic and employment characteristics. In all but one of the models that included the control for job satisfaction (the model predicting sufficient resources), LGBTQ status still significantly predicted career opportunities and resources, professional devaluation, social integration, health and wellness difficulties, and persistence intentions—even when differences in LGBTQ and non-LGBTQ workers’ job satisfaction were accounted for. In other words, while LGBTQ persons were less satisfied with their jobs overall (which is tied to their less positive treatment at work), even among STEM professionals who are equally satisfied with their jobs, LGBTQ persons still report disadvantages in career opportunities, professional valuation, inclusion, and health and wellness. Also, even among STEM professionals with the same level of job satisfaction, LGBTQ workers were more likely to consider leaving their jobs and STEM fields than their non-LGBTQ peers. This suggests that the LGBTQ status disadvantages documented above cannot simply be dismissed as LGBTQ persons’ more uniformly negative assessments of their jobs.

Last, consistent with labor force-wide scholarship that has shown positive effects of nondiscrimination protections on LGBTQ employees, we found that LGBTQ respondents in our sample who reported that their organizations had LGBTQ-inclusive nondiscrimination policies and health care benefits were significantly less likely than LGBTQ respondents working in organizations without these policies to report social exclusion (B = −0.287, P < 0.01) and professional devaluation (B = −0.296, P < 0.001) and were more likely to report having sufficient resources (B = 0.353, P < 0.001).

Limitations

While the STEM Inclusion Study data are unmatched in their inclusion of LGBTQ self-identification measures, sufficient sample sizes of both LGBTQ and non-LGBTQ respondents, and detailed measures of differential work experiences, they have a few limitations of note. First, while the 21 organizations represent a wide array of STEM professionals, they do not cover every STEM subdiscipline or interdisciplinary community. Our sample also includes only those STEM professionals who are members of their professional societies and who volunteered to take the survey. As indicated in table S1, our sample overrepresents STEM professionals employed in academia and in engineering and underrepresents STEM professionals in computer science. However, as noted above, supplemental analyses using weights that adjusted our sample to mirror the proportion of STEM professionals in the NSF data by demographics, subfield, and sector revealed that the patterns of LGBTQ inequality presented in Figs. 1 to 6 remain even when weighted to match the U.S. STEM population. In addition, there was little variation in the LGBTQ status effects across disciplines or employment sectors (see table S6), suggesting that over- or underrepresentation of respondents employed in certain subfields or sectors would not meaningfully affect the central result patterns we report here. Second, some of the focal outcome measures (e.g., harassment) used single-question items, as opposed to multi-item scales. Third, the survey does not include precise measurements of respondents’ salary or detailed organizational advancement histories, so we cannot track LGBTQ differentials in promotion or pay over time. Despite these limitations, our analysis is a vital step forward in assessing LGBTQ inequality in STEM. We encourage future research that expands this investigation, with careful attention to the mechanisms producing these disadvantages.