Hostile and Benevolent Sexism and College Women’s STEM Outcomes

Ambivalent Sexism Theory

Glick and Fiske (1996, 2001) proposed that gender relations are characterized by a combination of hostile and benevolent attitudes and behaviors toward women. Hostile sexism consists of overtly negative, angry attitudes and behavior toward women who occupy masculine domains or push for social change; it often takes the form of disparaging attitudes about women’s incompetence relative to men’s or suspicion and resentment directed at women who are perceived as insubordinate. Benevolent sexism consists of affectively positive but condescending attitudes and reactions to women who embrace traditional gender roles; it can take the form of protective paternalism (protectiveness toward women who are seen as weaker than men), complementary gender differentiation (flattering beliefs about women’s morality and sweetness), or heterosexual intimacy (the belief that men are incomplete without a woman’s love).

Women’s responses to hostilely sexist treatment include anger, disgust, stress, and heightened cardiovascular reactivity (Barreto & Ellemers, 2005; Bosson, Pinel, & Vandello, 2009; Salomon, Burgess, & Bosson, 2015; Schneider, Tomaka, & Palacios, 2001). In the workplace, women exposed to hostilely sexist treatment experience lower job satisfaction and increased absenteeism, depression, and physical illness symptoms (Fitzgerald, 1993). In contrast, exposure to benevolent sexism elicits a different set of responses including self-objectification, body shame (Calogero & Jost, 2011), and prolonged cardiovascular recovery (Salomon et al., 2015). Benevolent sexism also impairs women’s cognitive performance more than hostile sexism does (Dardenne, Dumont, & Bollier, 2007; Vescio, Gervais, Snyder, & Hoover, 2005) because it reduces working memory capacity by increasing intrusive thoughts about incompetence and by eliciting attempts to suppress these thoughts (Dardenne et al., 2007; Dardenne et al., 2013).

Stereotypes and Ambivalent Sexism in STEM Environments

Women’s underrepresentation in STEM fields reflects deeply embedded stereotypes that link men and masculinity more strongly than women and femininity with STEM competence (Nosek, Banaji, & Greenwald, 2002). To examine the role of these stereotypes in women’s STEM experiences, researchers often measure people’s endorsement of decontextualized statements about women’s STEM abilities (e.g., “Men are just better at science than women”; Lane, Goh, & Driver-Linn, 2012). Research illuminates the pervasiveness of the gender stereotypes that women face in STEM environments (Gunderson, Ramirez, Levine, & Beilock, 2012; Hartman & Hartman, 2008) and demonstrates that gender-based stereotypes can undermine women’s STEM interest (Cundiff et al., 2013), reduce their felt belonging in STEM environments (Cheryan, Plaut, Davies, & Steele, 2009), and undermine their STEM motivation (Fogliati & Bussey, 2013). Frequent stereotype exposure can also lead women to internalize the stereotypes, which can further degrade their interest in pursuing STEM careers (Schmader, Johns, & Barquissau, 2004).

One question that remains unaddressed, however, is how such gender stereotypes are delivered to women in STEM environments. Do women perceive that these stereotypes are delivered in a hostile manner, a benevolent manner, or both? Which delivery of gender STEM stereotype is perceived more frequently, and do they predict different outcomes for women? Hostile deliveries of STEM stereotypes may involve insulting jokes about women’s inferiority in STEM fields or a tendency to ignore women’s contributions to STEM tasks, whereas benevolent deliveries may involve treating women as if they cannot handle difficult STEM challenges (Bernstein & Russo, 2008; King et al., 2012) or have communal personalities that are ill-suited for competitive STEM environments. In other words, stereotypes about women’s relative lack of STEM competence may arrive via either hostile or benevolent sexism.

There is reason to assume that gender STEM stereotypes assume both hostile and benevolent forms, although with different frequency. Women perceived as non-traditional—which likely characterizes many women who inhabit STEM environments—are more frequent targets of hostile sexism than are traditional women (Sibley & Wilson, 2004). And yet, because overt expressions of gender bias are normatively eschewed and legally sanctioned (Dovidio & Gaertner, 1998; Pettigrew & Meertens, 2006), even people who hold negative gender STEM stereotypes may be disinclined to communicate them overtly. This may be especially true in a university environment in which students and faculty alike receive regular messages regarding gender sensitivity. Moreover, across cultures, men tend to endorse benevolent sexism more strongly than hostile sexism (Glick & Fiske, 2001), suggesting that their behaviors may reflect more benevolence than hostility toward women. Thus, we expected women to report more benevolent than hostile sexism experiences in their university STEM courses.

We also expected benevolent sexism to predict negative STEM outcomes for women. As noted, benevolent sexism impairs women’s cognitive performance, mediated by mental intrusions about competence and intrusive thought suppression (Dardenne et al., 2007; Dardenne et al., 2013). Similarly, subtle signs of bias—such as benevolent sexism—may produce cognitive disruption because their ambiguity elicits uncertainty and hypervigilance in targets (Murphy, Richeson, Shelton, Rheinschmidt, & Bergsieker, 2012). Moreover, benevolent sexism exposure elicits prolonged, low-level stress responses (Salomon et al., 2015), indicating that its stressful effects last a relatively long time. Finally, benevolent sexism can have a pacifying effect on women, reducing their motivation for social change (Becker & Wright, 2011) and increasing their tendencies to critically surveil and objectify themselves (Calogero & Jost, 2011). Thus, we expected perceived benevolent sexism to predict decreases in women’s STEM major intentions, self-efficacy, and performance.

But which manifestation of benevolent sexism should predict negative STEM outcomes? As noted earlier, benevolent sexism comprises three subjectively positive stances toward women: protective paternalism, complementary gender differentiation, and heterosexual intimacy (Glick & Fiske, 1996). In the context of STEM classes, protective paternalism might involve treating women as if they are weaker or need more assistance than men, and complementary gender differentiation might involve treating women as if their virtuous and sweet personalities render them unsuited for competitive STEM fields. Because protective paternalism implies that women cannot handle STEM tasks without men’s assistance, we expected it to impair women’s cognitive abilities and undermine their confidence and motivation. In contrast, we reasoned that complementary gender differentiation does not directly communicate negative stereotypes about women’s lack of STEM competence because it pertains to women’s moral character rather than their intelligence. And yet, given cross-cultural stereotypes of women as higher in warmth and lower in competence than men (Fiske, Cuddy, Glick, & Xu, 2002), reminders of stereotypes about their warmth may also communicate to women—albeit indirectly—the assumption that they lack competence. If so, then complementary gender differentiation may predict negative STEM outcomes for women. We therefore treated complementary gender differentiation as an exploratory predictor of women’s STEM outcomes in analyses. Finally, given that heterosexual intimacy involves the belief that men are incomplete without the romantic love of a woman (e.g., “Every man ought to have a woman he adores”; Glick & Fiske, 1996), we did not consider it here because of our exclusive focus on treatment that occurs within academic environments and communicates stereotypes about women’s STEM competence.

Our expectations regarding the associations of hostile sexism with women’s STEM outcomes were unclear. Although hostile sexism exposure predicts anger, negative affect, and stress in women, it also elicits approach-oriented motivations including desires to confront the source of sexism and engage in social change efforts (Becker & Wright, 2011; Chaudoir & Quinn, 2010; Schneider et al., 2001). Moreover, women’s angry reactions to hostile sexism tend to fade quickly (Salomon et al., 2015), suggesting a capacity to cope promptly with hostile episodes and direct cognitive resources elsewhere (rather than perseverating). This may be because hostile sexism explicitly conveys animosity and is thus readily perceived as a form of prejudice (Dardenne et al., 2007). Recognizing sexism as prejudice is valuable for ego protection because attributing negative outcomes to prejudice can buffer targets’ self-esteem (Major, Quinton, Kaiser, & McCoy, 2003). Thus, in the short-term time line of this study, it was not clear to us whether perceiving hostilely sexist treatment would predict negative STEM outcomes for women. We therefore treated hostile sexism as an exploratory predictor in analyses.

Perceived Sexism and STEM Identity

We expected the links between perceived benevolent sexism and STEM outcomes to be moderated by the strength of women’s STEM identity, defined as the importance of STEM domains to women’s self-concept (Leach et al., 2008; Luhtanen & Crocker, 1992). Specifically, based on motivational intensity theory (Brehm & Self, 1989; Gendolla & Richter, 2010), we expected the associations of benevolent sexism and STEM outcomes to be especially strong among women low in STEM identity. Conversely, women with a strong STEM identity should experience relatively weak links between benevolent sexism and STEM outcomes.

Motivational intensity theory suggests that as investment in a given domain increases, motivation to perform well in the domain also increases, and this leads people to exert more effort during difficult tasks (Brehm & Self, 1989; Gendolla & Richter, 2010). Thus, individuals with a stronger domain identity should persist at higher levels of difficulty, whereas those with a weaker domain identity should lose motivation and withdraw effort sooner. Effort is maintained when the benefits of success (e.g., self-efficacy, achievement) outweigh the costs (e.g., exertion) and, conversely, effort is withheld when the costs outweigh the benefits. For women high in STEM identity, who are strongly invested in STEM domains (e.g., Steele, 1997; Turner, Hogg, Oakes, Reicher, & Wetherell, 1987), the benefits of success in the face of protective paternalism (i.e., reminders that the task is likely “too difficult” for them) should outweigh the costs of effort. Conversely, for women low in STEM identity, the effort needed to perform tasks that are presented as “too difficult” may outweigh the potential benefits of task success. As one illustration of this phenomenon, African American adolescents who identified more strongly with academics were buffered against the negative effects of discrimination on academic performance, while those who identified more weakly showed impaired performance after experiencing discrimination (Thompson & Gregory, 2011).

At first glance, this prediction may seem inconsistent with findings from the stereotype threat literature. Theoretically, stereotype threat should most strongly affect those who care the most about doing well in the threatened domain—in other words, those with high domain identification (Steele, 1997). For example, some studies find that women who identify more strongly with math tend to suffer the largest performance decrements following manipulated reminders of gender math stereotypes (Keller, 2007; Steinberg, Okun, & Aiken, 2012). However, the literature is inconsistent regarding the role of domain identification in stereotype threat effects. While one meta-analysis found that the harmful effects of stereotype threat on performance were stronger among domain-identified targets than non-identified targets (Walton & Cohen, 2003), another meta-analysis found that stereotype threat effects were smaller for highly math-identified women than moderately low math-identified women (Nguyen & Ryan, 2008). As Nguyen and Ryan noted, only a small number of stereotype threat experiments include domain identification as a moderator, and the findings from these studies are mixed.

Further, stereotype threat manipulations tend to involve the presence or absence of contextual cues that activate stereotypes and have immediate effects on testing outcomes. In contrast, our interest is in different deliveries (i.e., hostile vs. benevolent) of the same stereotype (i.e., “women lack STEM competence”) that occur in everyday, naturalistic contexts. The stereotype threat literature is largely silent on the question of whether stereotype encounters that involve differential behavioral treatment—that is, overt insult versus patronizing protectiveness—are differently impactful among targets who vary in domain identification. Finally, stereotype threat is only one of multiple theorized pathways by which gender stereotypes may reduce women’s performance in negatively stereotyped domains. While stereotype threat has acute effects on performance, specifically in testing contexts when women are concerned about confirming the stereotype, stereotypes may also more generally decrease women’s interest and engagement in certain domains by lowering their self-confidence and the value that they place on the domain (Baron, Schmader, Cvencek, & Meltzoff, 2014). Little research has examined how domain identification moderates the effects of stereotypes on motivation and interest, two of the main goals of the current study. The current focus on the moderating role of STEM identity should add to our understanding of when and why domain identification matters among targets of negative stereotypes.

Is Sexism “in the Air”?

Thus far, we have focused our discussion primarily on women’s perceptions of hostile and benevolent sexism in STEM environments. But how do we know whether perceptions of sexism reflect real sexism experiences? For instance, women differ in the extent to which they expect to be gender stereotyped by others, or their stigma consciousness (Pinel, 1999), and those higher in stigma consciousness may report more sexism due to their perceptions rather than actual experiences. To bolster our confidence that the sexism reported by women in this study was actually “in the air,” as opposed to merely “in their heads,” we took two steps. First, we measured and controlled for individual differences in women’s stigma consciousness to remove variance in our measure of perceived sexism that reflects women’s chronic expectations of being stereotyped. Second, we measured the hostile and benevolent sexist attitudes of male students who were recruited from the same STEM courses as the female students. While a measure of privately held attitudes cannot reveal men’s behavioral enactment of hostile and benevolent sexism toward women in their classes, it can at least offer insight into the extent to which the STEM environments that surround our female participants are occupied by male students who endorse hostile and benevolent sexist beliefs. Consistent with cross-cultural patterns (Glick & Fiske, 2001), we expected male STEM students to report higher levels of benevolent than hostile sexism.

The Present Study

The harmful effects of gender stereotypes on women’s STEM outcomes are well-established, but we do not yet know whether such stereotypes are typically conveyed to women in STEM environments via primarily hostile or benevolent treatment. Further, while most research on sexism’s effects on women in STEM examines hostile sexism (Faulkner, 2009; Settles, Cortina, Buchanan, & Miner, 2013), there is reason to assume that benevolent sexism predicts negative STEM outcomes for women. Because no existing measure of women’s personal experiences of sexism distinguishes between hostile and benevolent treatment, we created measures of these constructs for use in the current study. Our goal was not to fully develop and validate a scale but instead to create measures that (1) allowed us to assess women’s self-reported sexism experiences with adequate internal reliability and convergent validity and (2) clearly differentiated between hostile and benevolent treatment. This latter goal allowed us to examine the unique roles of hostile and benevolent treatment in predicting STEM outcomes (STEM major intentions, STEM self-efficacy, and STEM GPA), moderated by women’s STEM identity. We also examined whether women’s hostile and benevolent sexism reports coincided with men’s reports of their hostile and benevolent sexism attitudes.

Participants were women and men undergraduates currently enrolled in large, lower-level engineering and health science courses, as well as psychology students who had taken at least one undergraduate STEM course. We predicted that female STEM students would report more benevolent than hostile sexism experiences in their STEM courses (Hypothesis 1) and that male STEM students would endorse benevolent sexism attitudes more strongly than hostile sexism attitudes (Hypothesis 2). Next, we expected that protective paternalism would predict lower STEM major intentions, STEM self-efficacy, and STEM GPA among women in STEM courses (Hypothesis 3). Finally, we predicted that the associations between protective paternalism exposure and women’s STEM outcomes would be moderated by the strength of women’s STEM identity. That is, we expected protective paternalism to more strongly predict negative STEM outcomes for women who were weakly STEM-identified than women who were strongly STEM-identified (Hypothesis 4). We treated as exploratory analyses in which complementary gender differentiation and hostile sexism predicted women’s STEM outcomes.

Constructing Measures of Hostile and Benevolent Experiences

We wrote items to measure specific acts of hostile and benevolent sexism, based on ambivalent sexism theory (Glick & Fiske, 1996), and in consultation with instructors of university STEM courses (the last two authors) to ensure that they captured realistic, relevant experiences. To measure specific behaviors without imposing any interpretation on those behaviors, we did not use the word “sexism” in any items. Fourteen items assessed experiences with hostile sexism (e.g., “How often have you been harassed or threatened”), 8 items assessed experiences with protective paternalism (e.g., “How often have you been treated as if you need a man’s help”), and 7 items assessed experiences with complementary gender differentiation (e.g., “How often have you been treated as if you are a ‘nice girl’”). Respondents indicated how frequently, in their daily life, they experienced each type of treatment on 6-point scales. Scale labels were modified from the Schedule of Sexist Events (SSE; Klonoff & Landrine, 1995), which uses a scale in which 1 = the event has never happened, 2 = the event happened once in a while (less than 10% of the time), 3 = the event happened sometimes (10–25% of the time), 4 = the event happened a lot (26–49% of the time), 5 = the event happened most of the time (50–70% of the time), and 6 = the event happened almost all of the time (more than 70% of the time). Because we wanted objective frequency estimates of recent and current sexism episodes, we converted the percentages from the SSE Scale to days per month and used the following scale: 1 = the event has never happened to you, 2 = the event happens rarely (2–3 times per month), 3 = the event happens sometimes (4–7 times per month), 4 = the event happens often (8–14 times per month), 5 = the event happens frequently (15–21 times per month), and 6 = the event happens very frequently (more than 21 times per month).

Sample 3

Undergraduate women completed the sexism items as part of a large-scale survey required of all students in the psychology department’s participant pool (N = 935). For the three-factor solution, the data yielded an excellent model fit, comparative fit index = .97, root mean square error of approximation = .05, standardized root mean residual = .04; (Hu & Bentler, 1999). The χ² test was significant, indicating bad fit. Yet, because this significance test is oversensitive to sample size, we would expect it to be significant with this sample: χ²(101) = 363.37, p < .01.

Table 1 contains the final 16 sexism items, their factor loadings, and their communalities. Supporting convergent validity of the sexism measures, correlations with the SSE (collected in Sample 1) were significant or marginally significant for hostile sexism (SSE-year: r = .69, SSE-life: r = .71; ps < .001), protective paternalism (SSE-year: r = .50, SSE-life: r = .57; ps < .001), and complementary gender differentiation (SSE-year: r = .15, p < .08, SSE-life: r = .20, p < .02). Moreover, correlations with the SCQ (collected in Sample 2) were significant for hostile sexism (r = .47, p < .001), protective paternalism (r = .48, p < .001), and complementary gender differentiation (r = .27, p < .001). This indicates that the three sexism measures correlated meaningfully both with women’s experiences of past sexism and their expectations of encountering gender-based stereotypes.

OPEN IN VIEWER

Table 1.

Final Items From the Experiences With Ambivalent Sexism Scale.

	Factor Loadings			Communalities
Item (“How often have you…”)	HS	PP	CGD	Initial	Extraction
Been subjected to offensive comments	.84	.06	.10	.70	.70
Been treated in an insulting manner	.74	−.001	.03	.68	.64
Been harassed or threatened	.80	−.10	.01	.57	.54
Been the target of obscene sexual comments or gestures	.73	.11	.11	.60	.56
Been the target of insulting jokes	.68	.12	.03	.57	.54
Been accused of using your gender to your advantage	.58	−.01	−.03	.43	.39
Been treated in an angry manner	.55	−.14	−.02	.54	.42
Been treated as if you need more assistance than men do	−.07	.86	.05	.70	.71
Been treated as if you are weaker than men	−.05	.79	.03	.63	.63
Been treated as if you need a man’s help	−.02	.77	.06	.67	.66
Been treated as if you are more vulnerable than men	.06	.55	.18	.61	.60
Been treated as if you cannot take care of yourself	.10	.45	−.02	.54	.50
Been treated as if you are a “nice girl”	−.008	.01	.83	.53	.62
Been treated as if you are innocent	−.06	−.05	.77	.49	.55
Been treated as if others expected you to be sweet and friendly	.03	.11	.63	.48	48
Been treated as if you are more morally pure than men	.12	.01	.60	.40	.41

Note. All scale items were preceded with the phrase “How often have you” and ended with the phrase “in your STEM courses and course-related activities.” Bolded values indicate loadings on primary target factors. Items were retained if they loaded ≥.45 on the primary target factor, <.39 on all other factors, and the difference between loadings on the primary target factor and other factors was at least .10. STEM = science, technology, engineering, and mathematics; HS = hostile sexism; PP = protective paternalism; CGD = complementary gender differentiation.

Participants

Given the novelty of the constructs under investigation, the use of a new scale, and our desire to reach a wide sample of university STEM environments, we decided a priori to seek as large a sample as possible. Thus, while a power analysis conducted using G*Power indicated that a sample of n = 109 was sufficient to detect an interaction for a moderate effect size in the women’s data, we continued collecting data for the entire semester. Participants were 755 undergraduates (592 women and 163 men) at a large, Southeastern public university. Women were recruited from engineering or health science courses (n = 407) or from psychology courses (n = 185). For the purposes of this study, we did not define psychology or other social sciences as STEM disciplines; therefore, to be eligible for participation, women recruited from psychology courses had to indicate that they had taken at least one university-level STEM course (which we defined for them). Men were recruited from the same engineering and health science courses as women participants, to ensure comparability of men’s and women’s reports of ambivalent sexism (see Table 2 for participant demographics). Participants recruited from psychology received course credit for their time, those recruited from one engineering course received a chance to win a US$100 gift card, and those recruited from other engineering courses or health science courses received extra course credit. No data were excluded from analyses, although dfs differ across analyses because some participants provided incomplete data (see below on missing data).

OPEN IN VIEWER

Table 2.

Participant Demographics and Means and Standard Deviations for Study Variables by Recruitment Sample.

Variables	Women in STEM Courses	Women in Psychology Pool	Men in STEM Courses
	n = 410	n = 185	n = 163
Race
White	244 (60.0%)	108 (58.4%)	99 (60.7%)
Black	70 (17.2%)	22 (11.9%)	28 (17.2%)
Asian	28 (6.9%)	22 (11.9%)	9 (5.5%)
Native American	3 (0.7%)	0	0
Middle Eastern	12 (2.9%)	7 (3.8%)	5 (3.1%)
Biracial	26 (6.4%)	15 (8.1%)	10 (6.1%)
Other	19 (4.7%)	6 (3.2%)	7 (4.3%)
Age	20.85 (2.72)	20.04 (2.61)	21.60 (3.72)
Class rank
First year	32 (7.9%)	55 (29.7%)	9 (5.5%)
Second year	110 (27.0%)	44 (23.8%)	33 (20.2%)
Third year	129 (31.7%)	42 (22.7%)	2 (1.2%)
Fourth year	97 (23.8%)	37 (20.0%)	54 (33.1%)
Fifth year +	38 (9.3%)	7 (3.8)	19 (11.7%)
STEM identity	3.52 (0.85)	3.21 (1.15)	3.52 (0.77)
HS experiences	1.49 (0.77)	1.52 (0.82)
CGD experiences	2.39 (1.11)	2.33 (1.01)
PP experiences	1.86 (1.02)	1.95 (1.12)
Number of STEM courses	12.67 (3.15)	10.52 (3.42)	13.52 (3.86)

Note. Age is presented as mean years of age with standard deviations in parentheses. STEM = science, technology, engineering, and mathematics; HS = hostile sexism; PP = protective paternalism; CGD = complementary gender differentiation.

To ensure comparability of the samples of women recruited from STEM courses versus from psychology, we conducted independent samples t-tests on women’s STEM identity and their reports of the sexism that they experienced in their STEM courses. Women recruited from STEM versus psychology courses did not differ in their reports of hostile sexism, protective paternalism, or complementary gender differentiation (ps > .32). However, women recruited from STEM courses reported a stronger STEM identity than those recruited from psychology courses, t(593) = 3.64, p < .01.

Procedure

After giving their informed consent, participants were introduced to an online survey. The first question asked them to report their gender and then routed women and men to different surveys. Women first read a definition of “STEM fields” accompanied by a detailed list of STEM disciplines and then rated the strength of their STEM identity. We placed the STEM identity scale first in the survey, given its role as the individual difference moderator in our theoretical model. Next, women reported their experiences with hostile and benevolent sexism, which served as our primary predictor variables. Finally, women completed measures of the three outcome variables: STEM major intentions, STEM self-efficacy, and STEM GPA. Women were instructed to keep their college STEM courses in mind as they completed these last measures. In contrast, men simply completed the Ambivalent Sexism Inventory (ASI; Glick & Fiske, 1996). All data and materials are available from the first author via email.

Measures

Preliminary Analyses

Table 3 shows correlations and descriptive statistics for all study variables. Because the number of STEM courses taken correlated with our dependent measures, we controlled for it in analyses on STEM outcomes. We also covaried STEM GPA in analyses on STEM major intentions and STEM self-efficacy, to control for the possibility that low-performing women report more sexism because they do, in fact, elicit more hostile treatment or receive more paternalistic help. As noted, we also controlled for stigma consciousness in all analyses on women’s data. Age and class year also correlated with one dependent measure each, but the present analyses do not include these as covariates. However, all findings reported here remained the same when we entered these two variables as covariates.

OPEN IN VIEWER

Table 3.

Bivariate Correlations, Means, and Standard Deviations for All Variables.

Variables	1	2	3	4	5	6	7	8	9	10	11
1. HS
2. PP	.80***
3. CGD	.55***	.71***
4. STEM identity	.14**	.13**	.19***
5. Major intentions	−.08	−.09*	.01	.49***
6. STEM GPA	.01	−.01	.04	.09*	−.003
7. Self-efficacy	−.09*	−.09*	.05	.29***	.38***	.26***
8. SCQ	.23***	.32***	.26***	−.05	−.03	.08*	.05
9. STEM #	.10*	.08*	.15***	.23***	.28***	.06	.15***	.02
10. Age	−.08	−.04	−.08	−.04	−.01	−.10*	−.02	−.06	.37***
11. Class year	−.03	−.002	.05	.01	.06	−.01	.09*	.01	.62***	.59***
12. Race	.06	.03	.002	.05	−.02	−.08	−.06	.03	−.002	−.01	−.01
Mean	1.50	1.89	2.37	3.46	4.07	3.37	0.007	3.23	12.28	20.81	2.90
Standard deviation	0.78	1.05	1.10	0.93	0.95	0.51	0.62	0.64	3.55	3.01	1.16

Note. Race is coded 0 = White, 1 = non-White. HS = hostile sexism; PP = protective paternalism; CGD = complementary gender differentiation; SCQ = Stigma Consciousness Questionnaire; STEM = science, technology, engineering, and mathematics; STEM # = number of STEM courses taken.

*p < .05. **p < .01. ***p < .001.

Missing data analyses were conducted on all of the primary variables. In the women’s data, less than 0.7% of all items for all cases were missing, and 66.7% of variables were not missing data for any case. Overall, 95.4% of women participants had no missing data, and no item had more than 5% of missing values. Men had no missing data.

Female STEM Students’ Perceptions of Hostile and Benevolent Sexism

To test Hypothesis 1, that women would perceive more frequent experiences of benevolent sexism than hostile sexism, we ran a one-way repeated measures analysis of variance (ANOVA) comparing women’s perceptions of the three sexism types (hostile sexism vs. protective paternalism vs. complementary gender differentiation). This analysis revealed a significant main effect, F(2, 1188) = 343.31, p < .001, η p 2 = 0.37. Consistent with Hypothesis 1, women reported more complementary gender differentiation (M = 2.37, SE = .04) and protective paternalism (M = 1.89, SE = .04) than hostile sexism (M = 1.50, SE = .03). The average frequency of perceived complementary gender differentiation (2.36) corresponded to three to four episodes per month, the average frequency of perceived protective paternalism (1.89) corresponded to two to three episodes per month, and the average frequency of perceived hostile sexism (1.56) corresponded to zero to two episodes per month. Post-hoc least significant difference tests revealed that complementary gender differentiation was perceived more frequently than protective paternalism, p < .001, 95% CI [.41, .55], and protective paternalism was perceived more frequently than hostile sexism, p < .001, 95% CI [.34, .44]. This analysis remained significant when controlling for women’s stigma consciousness, F(2, 1184) = 8.92, p < .001, η p 2 = 0.02, as did the pattern of means and each of the post-hoc tests, ps < .001.

Male STEM Students’ Sexism Attitudes

Hypothesis 2 stated that the male STEM classmates of the female participants would endorse benevolent sexism attitudes more strongly than hostile sexism attitudes. A paired samples t-test revealed that the men endorsed benevolent sexism (M = 3.67, SD = 0.73) more strongly than hostile sexism (M = 3.46, SD = 0.80), t(162) = 3.02, p < .01. To examine more closely whether men’s attitudes reflected the specific pattern of sexism that women reported experiencing, we ran a one-way repeated measures ANOVA comparing men’s endorsement of the three sexism types (hostile sexism vs. protective paternalism vs. complementary gender differentiation). This analysis revealed a significant main effect of sexism type, F(2, 159) = 35.45, p < .001, η p 2 = 0.31, and post-hoc tests revealed that men reported stronger endorsement of protective paternalism attitudes (M = 4.03, SE = .07) than hostile sexism attitudes (M = 3.46, SE = .06), p < .01, 95% CI [.40, .74]. However, men’s endorsement of complementary gender differentiation attitudes (M = 3.32, SE = .08) did not differ significantly from their hostile sexism attitudes, p = .09. Thus, consistent with Hypotheses 1 and 2, women reported experiencing more protective paternalism than hostile sexism in their STEM courses, and men in these same courses endorsed protective paternalism more strongly than hostile sexism attitudes.

Predicting STEM Outcomes

We used hierarchical multiple regression analyses and Hayes’s (2013) PROCESS macro to test Hypotheses 3 and 4. In each model, we predicted one of the outcome variables from sexism type (protective paternalism, complementary gender differentiation, or hostile sexism), STEM identity, and their interaction. Simple slope analyses were conducted at ±1 standard deviation from the mean on each variable. Covariates were the number of STEM courses taken, stigma consciousness, and the two sexism types that were not treated as predictors (to examine unique effects of each sexism type). In analyses on major intentions and self-efficacy, we controlled for STEM GPA to rule out the possibility that low-achieving women drove the findings. Results for Hypotheses 3 and 4 without covariates remained largely unchanged.

Sexism direct effects

Hypothesis 3 stated that protective paternalism would predict lower STEM major intentions, STEM self-efficacy, and STEM GPA among women; in contrast, we treated hostile sexism and complementary gender differentiation as exploratory predictors of STEM outcomes. Model 2 in Tables 4 and 5 shows tests of direct effects of protective paternalism, complementary gender differentiation, and hostile sexism on the outcome variables. As shown in Tables 4 and 5, protective paternalism negatively predicted STEM self-efficacy (p = .03, 95% CI [−.18, −.007], and hostile sexism negatively predicted intentions to major in STEM (p = .02, 95% CI [−.31, −.02], but no other direct effects emerged for protective paternalism or hostile sexism on any outcome variable. Stronger STEM identity predicted major intentions, self-efficacy, and GPA (ps < .05). Moreover, because no interactive effects of complementary gender differentiation emerged on any outcome variables (all ts < 1.58, ps > .11), we did not include a separate table displaying the output from analyses in which complementary gender differentiation was treated as a primary predictor. However, complementary gender differentiation was a significant positive predictor of STEM self-efficacy in Model 2, t = 2.39, p = .02, 95% CI [.01, .13]. Therefore, Hypothesis 3 was partially supported.

OPEN IN VIEWER

Table 4.

Summary of Regression Analyses for Protective Paternalism Predicting STEM Outcomes.

		Model 1				Model 2				Model 3
Outcomes	Variable	B	SE(B)	β	t	B	SE(B)	β	t	B	SE(B)	β	t
Intentions to major in STEM	STEM #	0.09	0.01	.29	7.08**	0.06	0.01	.19	5.32**	0.06	0.01	.20	5.48**
	GPA	−0.04	0.08	−.02	−0.46	−0.12	0.07	−.06	−1.77	−0.14	0.07	−.07	−2.03*
	SCQ	−0.05	0.07	−.03	−0.82	0.05	0.06	.03	0.79	0.05	0.06	.03	0.78
	CGD	0.05	0.04	.05	1.06	0.01	0.05	.01	0.26	0.01	0.05	.01	0.32
	HS	−0.18	0.06	−.15	−3.05**	−0.16	0.08	−.13	−2.27*	−0.18	0.08	−.14	−2.50*
	STEM ID					0.49	0.04	.49	13.14**	0.33	0.04	.32	4.71**
	PP					−0.08	0.07	−.09	−1.26	−.09	0.07	−.10	−1.40
	ID × PP									0.09	0.03	.19	2.71**
	R 2 change		.10				.22				.01
	F for R 2 change		12.03**				87.43**				7.34**
STEM self-efficacy	STEM #	0.03	0.01	.14	3.59**	0.02	0.01	.09	2.23*	0.02	0.01	.09	2.33*
	GPA	0.29	0.05	.24	6.07**	0.26	0.05	.22	5.54**	0.25	0.05	.21	5.31**
	SCQ	0.04	0.04	.04	0.96	0.09	0.04	.09	2.12*	0.09	0.04	.09	2.12*
	CGD	0.06	0.03	.13	2.29*	0.08	0.03	.12	2.39*	0.08	0.03	.13	2.44*
	HS	−0.13	0.04	−.17	−3.54**	−0.10	0.06	−.10	−1.54	−0.08	0.06	−.11	−1.70
	STEM ID					0.18	0.03	.27	6.81**	0.09	0.03	.14	1.91
	PP					−0.09	0.05	−.16	−2.13*	−0.10	0.04	−.17	−2.24*
	ID × PP									0.05	0.02	.15	2.02*
	R 2 change		.11				.08				.006
	F for R 2 change		13.65**				25.75**				4.09*
STEM GPA	STEM #	0.01	0.01	.06	1.42	0.01	0.01	.04	0.90	0.01	0.01	.04	1.01
	SCQ	0.07	0.04	.09	1.91	0.09	0.04	.11	2.38*	0.08	0.04	.10	2.35*
	CGD	0.02	0.03	.04	0.68	0.04	0.03	.07	1.23	0.04	0.03	.07	1.28
	HS	−0.02	0.03	−.04	−0.69	0.02	0.05	.04	0.58	0.02	0.05	.03	0.37
	STEM ID					0.05	0.02	.09	2.03*	0.04	0.04	.08	−0.98*
	PP					−0.07	0.04	−.14	−1.71	−0.07	0.04	−.15	−1.82
	ID × PP									0.05	0.02	.19	2.41*
	R 2 change		.01				.01				.01
	F for R 2 change		1.69				3.58*				5.81*

Note. STEM = science, technology, engineering, and mathematics; STEM # = number of STEM courses taken; SCQ = Stigma Consciousness Questionnaire; GPA = grade point average; CDG = complementary gender differentiation; PP = protective paternalism; STEM ID = STEM identity; HS = hostile sexism.

*p < .05. **p < .01.

OPEN IN VIEWER

Table 5.

Summary of Regression Analyses for Hostile Sexism Predicting STEM Outcomes.

		Model 1				Model 2				Model 3
Outcomes	Variable	B	SE(B)	β	t	B	SE(B)	β	t	B	SE(B)	β	t
Intentions to major in STEM	STEM #	.08	0.01	.28	6.91**	.06	0.01	.19	5.32**	0.06	0.01	.19	5.35**
	GPA	−.05	0.08	−.02	−0.59	−.12	0.07	−.06	−1.77	−0.13	0.07	−.07	−1.90
	SCQ	−.03	0.07	−.02	−0.51	.05	0.06	.03	0.78	0.05	0.06	.03	0.77
	CGD	.08	0.06	.09	1.64	.01	0.05	.01	0.03	0.01	0.05	.01	0.28
	PP	−.16	0.05	−.17	−3.06**	−.09	0.07	−.09	−1.26	−0.08	0.07	−.08	−1.20
	STEM ID					.49	0.04	.49	13.14**	0.39	0.04	.39	4.94**
	HS					−.16	.07	−.13	−2.27*	−0.19	0.07	−.15	−2.54*
	ID × HS									0.07	0.05	.11	1.38
	R 2 change		.10				.22				.002
	F for R 2 change		12.04**				87.41**				1.91
STEM self-efficacy	STEM #	.03	0.01	.14	3.40**	.02	0.01	.09	2.23**	0.02	0.01	.09	2.25*
	GPA	.29	0.05	.24	5.89**	.26	0.05	.22	5.53**	0.25	0.05	.21	5.34**
	SCQ	.06	0.04	.06	1.37	.09	0.04	.09	2.12*	0.09	0.05	.09	2.10*
	CGD	.10	0.03	.17	3.13**	.08	0.03	.13	2.39*	0.08	0.03	.13	2.31*
	PP	−.13	0.03	−.23	−4.02**	−.09	0.05	−.16	−2.13*	−0.09	0.05	−.15	−2.08+
	STEM ID					.18	0.03	.27	6.81**	0.18	0.03	.17	2.00*
	HS					−.08	0.05	−.10	−1.54	−0.10	0.06	−.12	−1.82
	ID × HS									0.05	0.03	.12	1.35
	R 2 change		.12				.07				.003
	F for R 2 change		14.47**				23.82**				1.82
STEM GPA	STEM #	.00	0.00	.06	1.37	.00	0.00	.04	0.90	0.006	0.007	.04	0.92
	SCQ	.08	0.04	.10	2.15*	.09	0.04	.11	2.38*	0.08	0.04	.10	2.32*
	CGD	.04	0.03	.08	1.47	.04	0.03	.07	1.23	0.04	0.03	.08	1.26
	PP	−.05	0.03	−.10	−1.75	−.06	0.04	−.14	−1.71	−0.06	0.04	−.13	−1.59
	STEM ID					.05	0.02	.09	2.01*	−0.07	0.05	.13	1.36
	HS					.00	0.05	.04	0.56	−0.003	0.05	−.004	−0.06
	ID × HS									0.08	0.03	.25	2.61**
	R 2 change		.02				.008				.01
	F for R 2 change		2.35*				2.27				6.83*

Note. STEM = science, technology, engineering, and mathematics; STEM # = number of STEM courses taken; SCQ = Stigma Consciousness Questionnaire; GPA = grade point average; CDG = complementary gender differentiation; PP = protective paternalism; STEM ID = STEM identity; HS = hostile sexism.

*p < .05. **p < .01.

Moderation by STEM identity

Hypothesis 4 stated that protective paternalism would more strongly predict negative STEM outcomes for women who were weakly STEM-identified than women who were strongly STEM-identified. Supporting Hypothesis 4, STEM identity moderated the associations of protective paternalism with major intentions, p = .007, 95% CI [.02, .15]; see Figure 1a, such that protective paternalism negatively predicted major intentions among weakly-identified women, B = −.17, SE = .78, t = −2.39, p = .01, 95% CI [−.32, −.03,], but not among strongly-identified women (p > .95). Similarly, STEM identity moderated the associations of protective paternalism with STEM self-efficacy, p = .04, 95% CI [.001, .09]; see Figure 1b. Weakly-identified women who experienced more protective paternalism reported lower STEM self-efficacy, B = −.14, SE = .05, t = −2.84, p = .01, 95% CI [−.24, −.04], but protective paternalism was unrelated to self-efficacy among strongly-identified women (p > .25). Last, STEM identity moderated the associations of protective paternalism with STEM GPA, p = .01, 95% CI [.01, .09]; see Figure 1c. Protective paternalism experiences predicted lower GPA among weakly-identified women (B = −.12, SE = .05, t = −2.64, p < .01, 95% CI [−.21, −.03], but not among strongly-identified women (p > .73).

OPEN IN VIEWER

Figure 1.

Interactions of protective paternalism and science, technology, engineering, and mathematics (STEM) identity on STEM major intentions, STEM self-efficacy, and STEM GPA.

We used the Johnson–Neyman technique (Spiller, Fitzsimons, Lynch, & McClelland, 2013) to locate the values of STEM identity at which the associations of protective paternalism with each outcome became statistically significant. Protective paternalism significantly predicted lower STEM major intentions among women with STEM identity scores below 2.96 on a 5-point scale (b _JN = −0.13, SE = .07, p = .05), it predicted lower STEM self-efficacy among women with STEM identity scores below 3.69 (b _JN = −0.09, SE = .04, p = .05), and it predicted lower STEM GPA among women with STEM identity scores below 3.30 (b _JN = −0.08, SE = .04, p = .05). These scores fall at the 21st, 60th, and 38th percentiles of STEM identity strength, respectively. Thus, Hypothesis 4 received strong support.

Next, at Step 3, STEM identity did not moderate the associations of hostile sexism with major intentions (p = .17) or self-efficacy (p = .12), but a significant Hostile Sexism × STEM Identity interaction emerged on STEM GPA (p = .009). Hostile sexism experiences were not significantly related to GPA among either weakly identified women (p = .19) or strongly identified women (p = .12). However, the Johnson–Neyman technique revealed that hostile sexism exposure significantly predicted lower STEM GPA among women with STEM identity scores below 1.28 (b _JN = −0.17, SE = .09, p = .05), which falls at the third percentile of STEM identity strength. Thus, hostile sexism experiences negatively predicted STEM GPA among women who were exceedingly low in STEM identity strength.

Limitations

The conclusions drawn from our findings are limited by the cross-sectional and self-report nature of the data. A more rigorous test of our hypotheses requires longitudinal assessment with sexism exposure in current STEM courses predicting future STEM outcomes. Such a design would also allow for direct tests of proposed mechanisms, including women’s felt belonging in STEM environments, rumination about their STEM competence, and motivation to prove gender stereotypes wrong. Tracking women’s experience with sexism over time would allow for a more direct test of the frequency of hostile and benevolent sexism in STEM environments. Also, as noted above, longitudinal assessment may reveal more negative effects of hostile sexism that accrue over time. Finally, perhaps protective paternalism does undermine performance and outcomes for highly STEM-identified women, but this effect is only detectable over a longer time. However, given that this is the first attempt to measure women’s self-reported exposure to hostile and benevolent sexism separately and examine their associations with outcomes in daily life, we believe that our findings have important implications for understanding and increasing women’s engagement in STEM disciplines.

Our conclusions regarding the lack of findings for complementary gender differentiation predicting negative STEM outcomes may be limited by the composition of our measure of complementary gender differentiation. Only 1 of the 4 items in our Complementary Gender Differentiation subscale makes a direct comparison between women and men (see Table 1). In contrast, items on the ASI (Glick & Fiske, 1996) that assess complementary gender differentiation explicitly compare women to men (e.g., “Women, compared to men, tend to have a superior moral sensibility”). Wording that explicitly contrasts women and men may be necessary in order to measure women’s experiences with sexist, gender-based treatment as opposed to other, more benign forms of treatment. Although women in our pilot study interpreted the entire set of benevolent sexism items as reflecting sexism, we did not measure their perceptions of the protective paternalism and complementary gender differentiation items separately. Therefore, it is possible that the complementary gender differentiation items measure women’s experiences with being treated in a genuinely flattering—rather than condescendingly sexist—manner. This could explain why scores on this scale did not predict negative STEM outcomes for women in the current study. Research currently underway in our lab therefore focuses on fine-tuning these items and further establishing the scales’ validity and reliability.

Future research should also examine how the gender composition of STEM courses influences women’s experiences with hostile and benevolent sexism. On the one hand, unequal gender ratios that favor male students may increase the salience of gender stereotypes about women’s STEM competence, thus increasing the frequency with which women STEM students encounter hostile and benevolent sexism in their courses. On the other hand, sexist treatment that occurs in the context of heavily male-dominated environments may be easier to attribute to discrimination, which may reduce its impact on women’s STEM outcomes. Future research may profit from examining how STEM instructors’ endorsement of sexist attitudes translates into women’s perceptions of sexist treatment. Last, conclusions regarding men’s endorsement of protective paternalism should be interpreted with caution due to the low internal reliability of the scale for this sample. And the results for men’s endorsement of hostile and benevolent sexism should also be interpreted with caution due to social desirability concerns.

Practice Implications

These findings have practical implications for advisors and instructors of STEM courses. While instructors who wish to encourage women in STEM may view it as helpful to offer women extra assistance, the findings presented here suggest that such assistance is correlated with negative outcomes for women. We are reminded of an invitation sent out by the Office of Diversity at a large public university that advertised a “Research Boot Camp for Women” to help “women students develop the skills necessary for success in academia.” Although undoubtedly well-intended, this invitation inadvertently communicated negative stereotypes about women’s relative lack of competence. Moreover, the women who are most likely to lose confidence in the face of such paternalism may be women who are not yet firmly committed to STEM fields. This creates somewhat of a conundrum for STEM instructors: How does one encourage and support undecided female STEM students who are struggling, without simultaneously conveying negative gender stereotypes and sexist attitudes? One way may be to avoid messages that are directed only at women—like the Research Boot Camp described above—and instead convey clearly that help is offered to all struggling students based only on performance. Thus, STEM course instructors should be cognizant of how seemingly pro-social efforts to help advance women in STEM can have unintended consequences.