Maybe He Is Relatable Too: Encouraging Women to Identify With Male Scientists by Highlighting Bias Against Fathers

The Benefits of Guides and Role Models in STEM

The overrepresentation of men in many STEM domains (NSF, National Center for Science and Engineering Statistics, 2017) results in people generally stereotyping scientists as possessing masculine traits (e.g., agentic, assertive) rather than feminine traits (e.g., communal, warm; Carli et al., 2016), which in turn can decrease women’s attraction to STEM (Diekman et al., 2011). However, multiple theories posit that ingroup scientist role models can help mitigate the harmful influence of these stereotypes. For instance, Dasgupta’s (2011) stereotype inoculation model posits that when women identify with a successful female scientist, the scientist becomes a role model who “inoculates” or protects against harmful gender stereotypes. The motivational theory of role modeling (Morgenroth et al., 2015) further suggests that when individuals identify with a successful role model, that role model acts as inspiration (i.e., fosters valuing a field) and represents what is possible (i.e., promotes confidence or self-efficacy in a field). Both theories suggest that women do not need to interact with a role model, for the role model to encourage belonging and positive perceptions (i.e., valuing, feelings of self-efficacy) about a given discipline (Dasgupta, 2011; Morgenroth et al., 2015).

In contrast to role models, guides provide women with advice and opportunities to learn new skills (Gibson, 2004). As a result, women must be comfortable approaching a scientist for career opportunities and advice, for the scientist to function as a guide. Having helpful guides is essential for promoting women’s success both in college (Downing et al., 2005; Marra et al., 2009) and in their careers (Eby et al., 2008; Misra et al., 2017). Moreover, guides can provide valuable research opportunities in STEM laboratories, which encourage students, particularly those from marginalized groups, to both enter and persist in STEM majors (Jones et al., 2010; Villa et al., 2013).

Dasgupta’s (2011) stereotype inoculation model posits that for an exemplar to act as a role model and inoculate against threatening stereotypes, it is critical that the exemplar has a matching identity. In contrast, Morgenroth and colleagues’ (2015) motivational theory of role modeling does not postulate that a role model must be a member of the ingroup. Rather, this model posits that exemplars serve as the best role models when individuals can identify with the exemplars. Because women tend to identify more with ingroup than outgroup members, both theories suggest that female scientists would function as better role models than male scientists for women (cf. Cheryan et al., 2011; Dennehy & Dasgupta, 2017; Stout et al., 2011). In contrast to role model research, previous work finds that guides are beneficial even when they possess outgroup identities (Fuesting & Diekman, 2017). Male guides can successfully promote women’s career advancement in STEM (Huston et al., 2019), and male guides are equally effective as female guides for encouraging female college students’ academic achievement (Blake-Beard et al., 2011; Downing et al., 2005).

At the same time, individuals feel more comfortable interacting with scientists when they feel similar to the scientist (Cheryan et al., 2011; Pietri, Johnson, Ozgumus, & Young, 2018), and hence, women also may feel more comfortable approaching potential female scientist guides than male scientist guides (Dennehy & Dasgupta, 2017; Stout et al., 2011). Indeed, female STEM majors indicate wanting female mentors, and students in general feel more comfortable approaching and asking favors (e.g., help on assignments) from female professors than male professors (Blake-Beard et al., 2011; El-Alayli et al., 2018). However, in certain STEM fields with stark gender disparities (e.g., computer science, engineering; NSF, National Center for Science and Engineering Statistics, 2017), it may not be feasible to ensure women have access to female scientist guides and role models. Of importance, requiring female scientists to act as guides (e.g., mentoring many students) and role models (e.g., serving on panels, giving lectures) may create a service burden for these women and hurt their research productivity (Guarino & Borden, 2017). Thus, it is imperative to develop new approaches to promote women’s identification with male scientists and to encourage women to view male scientists as potential guides and inspirational role models (Akcinar et al., 2011).

Enhancing Identification With Scientists via Perceptions of Identity-Based Adversity

Highlighting how a male scientist has experienced adversity stemming from a devalued identity may address this need because research shows that believing people or groups have encountered similar experiences with discrimination is a powerful way to foster connections with those people or groups (Cortland et al., 2017; Craig et al., 2012). For instance, in a series of experiments, researchers found that relative to control information (i.e., about endangered animals or about the importance of women affinity groups), teaching women about pervasive sexism in STEM (e.g., Eaton et al., 2020; Moss-Racusin et al., 2012) enhanced their perceptions that a female scientist had faced bias and had similar experiences with adversity. These beliefs in turn predicted enhanced identification with a successful female biologist, female computer scientist, and female scientist who embodied masculine STEM stereotypes (e.g., had no hobbies and only cared about research; Pietri, Johnson, Ozgumus, & Young, 2018). Additional past work has demonstrated that because Latinas tend to be more sensitive to discrimination due to their ethnicity than their gender (Levin et al., 2002), they believed a Latino male scientist had more similar experiences with discrimination than a White female scientist. Thus, Latinas identified more strongly with a Latino male scientist than a White female scientist (Pietri, Drawbaugh, et al., 2019). In this example, Latinas felt similar to a male scientist with an overlapping ethnicity because they believed this male scientist had faced comparable identity-based (i.e., ethnic) adversity.

Knowing a male scientist has dealt with past identity-based adversity may therefore foster identification with that male scientist. Even though gender stereotypes are particularly damaging for women’s success in STEM (Eaton et al., 2019; Moss-Racusin et al., 2012), gender stereotypes also can constrain men. For example, male applicants are disliked when they act modest (Moss-Racusin et al., 2010) and when they display sadness after receiving negative feedback (Motro & Ellis, 2017). It also is important to note that male scientists possess multiple identities beyond those related to gender and profession and may face adversity due to one of these additional identities (Abrams & Hogg, 1999). For instance, in addition to being a scientist, a man may be a devoted father. Because proscriptive stereotypes dictate that men are not supposed to act communal, men who are primary caregivers or report experiencing work–family conflict tend to be disliked (Brescoll & Uhlman, 2005; Butler & Skattebo, 2004). Thus, a male scientist, who is also a father, might be positively stereotyped and high status in one domain (e.g., in STEM; Eaton et al., 2020; Moss-Racusin et al., 2012) but might encounter negative stereotypes in another domain (e.g., fatherhood; Brescoll & Uhlman, 2005). Indeed, both women and fathers are perceived as feminine and nurturing (Banchefsky & Park, 2016; Fiske et al., 1999), which is viewed as incompatible with STEM (Carli et al., 2016). Encouraging women’s awareness of bias against fathers may promote their felt similarity with a male scientist, who has the additional and often devalued identity as a father.

Past research has specifically tested whether believing a scientist has had similar discrimination experiences encourages identification with that scientist, and as a result, in these previous studies, the scientist always had at least one matching identity (i.e., ethnicity or gender; Pietri, Drawbaugh, et al., 2019; Pietri, Johnson, Ozgumus, & Young, 2018). Of importance, women may struggle to believe a man who holds dual roles as both a scientist and father (i.e., a father scientist) has had similar experiences with discrimination. Although both women and fathers may be marginalized in STEM contexts, stereotypes harm women and fathers in different ways. That is, people view fathers negatively for diverging from masculine stereotypes and acting nurturing rather than assertive (Brescoll & Uhlman, 2005), whereas individuals dislike women when they act assertive and perceive women as less competent than men (Fiske et al., 1999; Rudman & Glick, 1999). In this instance, women do not possess the same marginalized identity as father scientists and fathers are not part of the female ingroup but rather are part of a dominant and high status outgroup in STEM (i.e., men; Eaton et al., 2020; Moss-Racusin et al., 2012). An important empirical question is whether women will feel similar to a father scientist when women believe that father has dealt with different identity-based unfair treatment.

To answer this question, it is important to note that two processes can help explain the relationship between facing similar identity-based adversity and identification—empathic concern and shared life experiences. People typically feel empathic concern or other-oriented emotions (e.g., concern, sympathy) when hearing about another’s challenges (Batson, 1991; Cialdini et al., 1997), and these feelings of empathy can lead to perceived similarity with the distressed other (Cialdini et al., 1997). Individuals also tend to identify with successful others who have had similar life experiences generally (e.g., going to the same university; Asgari et al, 2012). Consequently, past work has demonstrated that both feelings of empathy and perceptions of common experiences underscore the link between shared adversity and identification with a successful scientist. In particular, Pietri, Johnson, Ozgumus, and Young (2018) found that believing a female scientist had encountered past discrimination led to feelings of empathic concern for the scientist and believing the scientist had common life experiences, and these feelings and beliefs ultimately predicted enhanced identification with the scientist.

Learning about harmful stereotypes against fathers may not enhance perceptions that a father scientist has had similar life experiences generally because stereotypes impact men and women differently (Brescoll & Uhlman, 2005; Rudman & Glick, 1999). However, knowing a father scientist has experienced identity-based adversity may, at a minimum, stimulate feelings of empathic concern for the father scientist, which in turn may promote identification with this scientist (Cialdini et al., 1997; Pietri, Johnson, Ozgumus, & Young, 2018). Indeed, individuals can experience empathy for outgroup members who are facing adversity, and there is a large body of work demonstrating that feelings of empathic concern for an outgroup member can promote positive attitudes toward that member’s whole group (Batson et al., 2002; Vescio et al., 2003; see also Dovidio et al., 2010, for review). Thus, we explored whether beyond promoting positive outgroup attitudes, feelings of empathy can encourage women’s identification with an outgroup (i.e., a father) scientist. Differing from past research (i.e., Pietri, Johnson, Ozgumus, & Young, 2018; Pietri, Drawbaugh, et al., 2019), we examined whether knowing a scientist has a different devalued identity in STEM would encourage identification with that scientist.

The Current Research

To explore this possibility and add to previous research, we employed a similar paradigm as Pietri, Johnson, Ozgumus, and Young (2018) and provided female participants with information about the negative and constraining nature of masculine stereotypes for men in caretaker roles to manipulate perceptions that a specific father scientist had faced hardship. Across five experiments, we tested whether teaching women about these detrimental masculine stereotypes encouraged women to believe a father scientist has faced bias, to feel empathic concern for the scientist, and, ultimately, to identify with the scientist. In Experiment 2, we additionally examined whether learning of a father scientist’s specific experiences with unfair treatment also promoted identification with that scientist. Thus, we first predicted (Hypothesis 1a) that relative to control information, learning about harmful masculine stereotypes would increase perceptions that a father scientist has faced bias, empathy for that scientist, and identification with the scientist. We also ran a mediation model (Hypothesis 1b) to test whether believing a father scientist has faced bias would stimulate feelings of empathy for the scientist, which would ultimately relate to identification with the scientist. However, because stereotypes hurt men and women in different ways, we also expected (Hypothesis 1c) that information about masculine stereotypes/bias would not increase perceptions that a father scientist has had similar encounters with adversity or has had similar experiences generally.

Of importance, we examined whether identifying with the father scientist would encourage women to view him as a potential guide and/or role model. To explore whether women perceived the scientist as a possible guide, we measured interest in working with the father scientist (i.e., taking classes with the scientist, doing research with the scientist; Experiments 1b, 3a, and 3b) and attraction to the scientist’s company or school (Experiment 3a and 3b). Moreover, to examine whether the father scientist was acting as a role model to inspire positive perceptions of STEM generally, we assessed belonging and trust in STEM environments (Experiments 1a–3b) and positive beliefs about computer science research (i.e., self-efficacy, valuing, and interest in computer science research; Experiment 3b). We anticipated that increasing identification with a father scientist would help turn him into a potential guide (Hypothesis 2) and role model (Hypothesis 3). To test Hypotheses 2 and 3, we look at whether teaching women about constraining masculine stereotypes directly affected the guide and role model-related outcomes and also indirectly influenced these outcomes via identification. That is, we ran mediation models to test whether our manipulation enhanced felt similarity with the father scientist and whether identification in turn related to perceiving the father scientist as a possible guide and to positive beliefs about STEM.

Participants

We recruited 249 female participants from the United States using Amazon’s Mechanical Turk (MTurk) website, who were compensated US$1.00. To ensure the data quality, we utilized consistent attention checks throughout the current and subsequent experiments and employed the same exclusion criterion as Pietri, Johnson, Ozgumus, and Young (2018). Example attention check items are available in Table 1. Moreover, in the current and following experiments, we ran our MTurk study through the “TurkPrime” website and employed a variety of screeners to enhance data quality. First, through TurkPrime, we only advertised our study to participants who identified as women. Second, we ensured that participants were from the United States and prevented participants from “suspicious” geolocations and IP addresses from taking the study. That is, TurkPrime created a list of suspicious geolocations and IP addresses that were tied to repeated poor responses, which can be excluded from a study (see Moss & Litman, 2018).

OPEN IN VIEWER

Table 1.

Attention Checks and Excluded Participants Across All Experiments.

Attention Check	Attention Check Item	Participants Excluded in Experiment 1a	Participants Excluded in Experiment 1b	Participants Excluded in Experiment 2	Participants Excluded in Experiment 3a	Participants Excluded in Experiment 3b
Module attention check (three items total, participants had to get two of three correct, which was the same cut-off from Pietri, Johnson et al., 2018)	Example item: According to the module, when men act modest: (a) They are liked a lot, (b) they are perceived as better workers than women, (c) they are rewarded with many job opportunities, and (d) they are disliked (correct answer)	2 (0.8%)	18 (6.5%)	27 (6.4%)	13 (4.3%)	11 (3.7%)
Article attention check (one item in Experiments 1a, 1b, 3a, and 3b, one to two items in Experiment 2)	The article talked about (a) Brandon, who works with small children; (b) maintaining a healthy lifestyle with a busy work schedule; (c) Brandon, who is a successful computer scientist (correct answer); and (d) Jill, who is the CEO of a large tech company	5 (2.0%)	12 (4.3%)	51 (12.0%)	8 (2.8%)	7 (2.4%)
General attention check (one item)	Select the correct answer: (a) do not choose this answer, (b) do not choose this answer, (c) choose this answer (correct answer), and (d) do not choose this answer	0 (0%)	8 (2.9%)	0 (0%)	3 (1.1%)	0 (0.0%)

We excluded seven participants for failing attention checks embedded throughout the study, who did not vary across information condition, χ²(1, N = 249) = 0.109, p = .522. We had a final sample of 242 female participants. The specific racial/ethnic demographics of participants was 196 White (81.3%), 20 Black or African American (8.3%), eight Asian (3.3%), six Latino (2.5%), 11 Other (4.5%), and one unreported (0.4%). With regard to educational level of participants, one had completed less than high school education (0.4%), 64 had a high school degree/GED (26.4%), 50 had a 2-year college degree (20.7%), 86 had a 4-year college degree (35.5%), 32 had a master’s degree (13.2%), four had a doctorate degree (1.7%), and five had a professional degree (2.1%). Finally, 31 (12.8%) participants were currently working in a STEM field. Because of a coding error, we did not collect participants’ age in Experiment 1a.

Procedure

All the manipulations and measures for the current and subsequent experiments are available in Online Supplemental Materials. Experiment 1 was advertised as a study examining participants’ general impressions of an article and module. The current and following experiments were all administered through Qualtrics online software. Upon beginning the experiment, we employed Qualtrics to randomly assign participants to watch one of the two informational modules on either masculinity bias or giant pandas (i.e., control information). Both modules presented facts and information about the given topic in a similar manner as a PowerPoint presentation. All information was shown in a video format, and thus we controlled how quickly the information was presented. However, participants could pause and rewind the video if they missed information. Prior to beginning the modules, we informed participants that we would test their memory of the modules to encourage them to carefully attend to the video.

The masculinity bias module was modeled after the gender bias in STEM module employed in Pietri, Johnson, Ozgumus, and Young (2018). This module began by discussing how men are disliked and punished when they act modest and express emotions and presented study results supporting these assertions. The module next highlighted research demonstrating the penalties men face when they value fatherhood and caretaker roles. Finally, the module featured quotes about men’s difficult experiences balancing fatherhood at competitive technology companies. All information presented in the module was based on research (e.g., Brescoll & Uhlman, 2005) and the real experiences of men in computer science and technology companies.

Similar to Pietri, Johnson, Ozgumus, and Young (2018), we employed a control module featuring endangered giant pandas because information on harmful stereotypes and endangered animals is mildly sad and relies on scientific facts and research. We kept the format of the giant panda module as consistent as possible with the masculinity bias module. This control module began by providing general facts about giant pandas and the uncertain future for the giant panda species. The module continued by discussing the mating difficulties of giant pandas and ended with quotes from giant panda experts, who discussed their opinions about giant pandas’ precarious situation. All the information featured in both the masculinity bias module and giant panda control module is available in Supplemental Materials.

Immediately following the modules, participants completed three easy attention check questions related to their respective module (see Table 1 example item and Supplemental Materials for the full list of questions). Relying on the same exclusion criteria as Pietri, Johnson, Ozgumus, and Young (2018), we excluded participants who did not answer at least two module attention check question correctly. Previous research found that the panda control module and a gender bias in STEM module were well matched (see Pietri, Johnson, Ozgumus, & Young, 2018). In Experiment 1a and b, participants completed measures assessing their impressions that the modules were informative, sad, entertaining, and engaging. In both the current experiment and Experiment 1b, we found that the modules were viewed as equally engaging and sad but not equivalently informative and entertaining. As detailed in the Supplemental Materials, participants perceived the panda control module as more informative than the masculinity bias module across Experiment 1a and b. Relative to the masculinity bias module, participants also viewed the panda control module as more entertaining in Experiment 1a but as less entertaining Experiment 1b. Nevertheless, when we controlled for these variables, our results did not change (see Supplemental Material for these items and analyses).

After the modules and attention check questions, all participants read an article about a successful male computer scientist (Dr. Brandon Evans), whose research focused on “making innovative advances in artificial intelligence.” This article was modeled after a female computer scientist article employed in Pietri, Johnson, Ozgumus, and Young (2018, Experiment 2), and the full article is available in Supplemental Materials. This article began by discussing Brandon’s background and how he became interested in computer science. The article next described Brandon’s life outside of work, and in this section, Brandon notes that he values spending time with his young and growing family. Specifically, he writes

I have two children, a little girl and boy, and they are at great ages now: 2 and 5. I love spending time with them, playing with them, watching them grow. They are so curious and adventurous, and just turning into interesting little people.

Measures

For all of the following indices, we averaged responses to the items in the scales, with higher scores indicating more of the measured constructs. All measures are available in Supplemental Materials. After reading the article and completing the article attention check, participants reported the degree from 1 (not at all) to 7 (extremely) to which they felt six other-oriented empathetic emotions (softhearted, compassionate, warm, sympathetic, moved, and tender) “while reading the article and thinking about Brandon” (Batson, 1991; six items; M = 4.43, SD = 1.39, α = .95). Participants then completed measures assessing their impressions of Brandon, which were presented in random order. Participants rated their agreement from 1 (strongly disagree) to 5 (strongly agree), with statements assessing the extent to which they believed that Brandon had faced bias (e.g., “Most likely, Brandon has experienced adversity”; four items; Pietri, Johnson, Ozgumus, & Young, 2018; M = 2.61, SD = 0.93, α = .92), indexing their perceptions that Brandon is relatable (e.g., “Brandon seems similar to me”; seven items; Pietri, Johnson, Ozgumus, & Young, 2018; M = 3.46, SD = 0.75, α = .91), and their sense of self-other overlap with Brandon. The Self-Other Overlap Scale was comprised of eight items that measured the extent from 1 (not at all) to 7 (very much) to which participants felt similar and connected to Brandon (e.g., “To what extent do you feel you are similar to Brandon?”). The eighth item asked participants to indicate which of seven Venn diagrams featuring different degrees of overlap best symbolized their relationship with Brandon (1 = two non-overlapping circles to 7 = two nearly completely overlapping circles; Goldstein et al., 2014; M = 3.26, SD = 1.58, α = .97).

Participants additionally completed indices assessing the extent from 1 (not at all) to 7 (a lot) to which they believed Brandon had comparable past experiences (e.g., “To what extent do you feel you and Brandon have had similar experiences?”; two items; Pietri, Johnson, Ozgumus, & Young, 2018; M = 3.05, SD = 1.35, α = .88) and had similar past occurrences with unfair treatment (e.g., “To what extent do you think you and Brandon have faced similar adversity in the past?”; four items; Pietri, Johnson, Ozgumus, & Young, 2018; M = 2.52, SD = 1.26, α = .94).

To assess women’s belonging and trust in STEM, participants were shown the website for a fictitious technology and science company (“CompTech”) and were told to imagine they were employed by this company. Participants rated their agreement from 1 (strongly disagree) to 5 (strongly agree) with items measuring sense of belonging (e.g., “People at the company = would like me”; eight items taken from Good et al., 2012, and Walton & Cohen, 2007; M = 3.05, SD = 0.76, α = .90) and trust and comfort (e.g., “I think I would be treated fairly by my colleagues at the company”; four items; Purdie-Vaughns et al., 2008; M = 3.10, SD = 0.86, α = .90) at the company. Of note, this was a different company than the company where the male scientist supposedly worked, and the male scientist was unrelated to this company. Finally, participants indicated their agreement from 1 (strongly disagree) to 5 (strongly agree) with statements assessing their awareness of masculinity bias (e.g., “In my opinion, men often face negative reactions for being modest”; “In my opinion, men face career penalties when they take time off for family reasons”; six items) modeled from awareness of gender bias in STEM Scale (Pietri et al., 2017; M = 3.33, SD = 0.62, α = .62) and their awareness of gender bias in STEM (e.g., “In my opinion, women in science fields often face negative reactions for being assertive”; eight items; Pietri et al., 2017; M = 3.72, SD = 0.73, α = .88).

We included three additional exploratory measures assessing perceptions of the male scientist—the male scientist was competent and warm, the male scientist cares about women, and the male scientist can take participants’ perspective (see Supplemental Material for analyses with these measures). Critically, our manipulations did not harm perceptions of the male scientist’s competency.

Awareness of Bias Outcomes

We first found that compared to participants who saw the control information (n = 123; M = 3.17, SD = 0.59), participants who viewed the masculinity bias information (n = 119; M = 3.48, SD = 0.61) reported significantly higher awareness of masculinity bias, t(240) = 4.06, p < .001, d = 0.52, mean difference = 0.31, SE = 0.08. We also found that participants in the masculinity bias information condition (M = 3.82, SD = 0.63) reported significantly higher awareness of gender bias in science than participants in the control information condition (M = 3.63, SD = 0.82), t(240) = 2.07, p = .040, d = 0.27, mean difference = 0.19, SE = 0.09.

Perceptions of the Scientist

In line with Hypothesis 1a, relative to participants in the control information condition (M = 4.20, SD = 1.54), those in the masculinity bias condition (M = 4.65, SD = 1.20) indicated feeling significantly more empathy for Brandon, t(240) = 2.51, p = .013, d = 0.32, mean difference = 0.44, SE = 0.18. Further supporting Hypothesis 1a, compared to those in the control condition (M = 2.29, SD = 0.88), participants in the masculinity bias condition (M = 2.91, SD = 0.88) had higher perceptions that Brandon had faced bias, t(240) = 5.46, p < .001, d = 0.70, mean difference = 0.62, SE = 0.11. Participants in the masculinity bias condition (M = 0.18, SD = 0.92) also reported a stronger shared identity with Brandon than participants in the control condition (M = −0.19, SD = 0.94), t(240) = 3.07, p = .002, d = 0.40, mean difference = 0.37. However, supporting Hypothesis 1b, there were no significant effects of information condition on perceptions that Brandon had experienced similar past situations, t(240) = 0.01 p = .989, d < 0.001, mean difference < 0.001, SE = 0.17 (control condition: M = 3.05, SD = 1.31; masculinity bias condition: M = 3.05, SD = 1.40), or that Brandon had encountered similar past unfair treatment, t(240) = 1.16, p = .249, d = 0.15, mean difference = 0.19, SE = 0.16 (control condition: M = 2.61, SD = 1.30; masculinity bias condition: M = 2.42, SD = 1.22). Thus, participants indicated increased identification with Brandon, despite not believing that Brandon had similar life experiences or encounters with discrimination.

We next ran a mediation analysis to test Hypothesis 1b and to examine whether believing the father scientist faced bias was related to feelings of empathy and whether empathy in turn predicted feeling a shared identity with the scientist. We employed a serial mediation model to test our entire hypothesized model (i.e., masculinity bias information → Brandon faced gender bias → empathy → shared identity), and we ran this model with Hayes’s (2018) PROCESS Macro Model 6 and 10,000 bootstrap resamples. Although we found the predicted relationships between the variables (see Tables 2 and 3 for relationships between variables), the 95% confidence interval crossed 0 (indirect effect = 0.05, 95% CI [−0.003, 0.11]; see Table 3 for mediation model). Thus, Hypothesis 1b was not supported in this first study.

OPEN IN VIEWER

Table 3.

Mediation Analyses Across Experiment 1a and b.

Experiment and Tested Effect	Effect (Effect Controlling for Mediators)	SE (SE Controlling for Mediators)	95% CI (95% CI Controlling for Mediators)
Experiment 1a
Condition → faced bias	.62	.11	[.40, .84]
Condition → empathy	.44 (.32)	.18 (.19)	[.10, .79] ([−.04, .69])
Condition → shared identity	.38 (.15)	.12 (.10)	[.15, .61] ([−.04, .34])
Faced bias → empathy	.19	.10	[−.01, .39]
Faced bias → shared identity	.16 (.08)	.07 (.05)	[.03, .29] ([−.02, .19])
Empathy → shared identity	.40	.03	[.33, .46]
Indirect effect for serial mediation	.05	.03	[−.003, .11]
Condition → shared identity	.38	.12	[.15, .61]
Condition → belonging and trust	−.10 (−.25)	.13 (.12)	[−.36, .15] ([−.50, −.01])
Shared identity → belonging and trust	.40	.07	[.27, .53]
Indirect effect for belonging mediation	.15	.06	[.05, .27]
Experiment 1b
Condition → faced bias	.35	.11	[.14, .56]
Condition → empathy	.61 (.51)	.18 (.18)	[.26, .96] ([.16, .87])
Condition → shared identity	.32 (.03)	.12 (.31)	[.09, .52] ([−.16, .22])
Faced bias → empathy	.28	.11	[.07, .49]
Faced bias → shared identity	.32 (.22)	.07 (.06)	[.19, .45] ([.11, .34])
Empathy → shared identity	.34	.04	[.27, .41]
Indirect effect for serial mediation	.03	.02	[.004, .07]
Condition → shared identity	.32	.12	[.09, .52]
Condition → belonging and trust	.06 (−.07)	.12 (.11)	[−.17, .30] ([−.29, .15])
Shared identity → belonging and trust	.40	.06	[.28, .52]
Indirect effect for belonging mediation	.13	.05	[.03, .23]
Condition → interest in working with Brandon	.19 (.05)	.09 (.08)	[.01, .37] ([−.10, .20])
Shared identity → interest in working with Brandon	.43	.04	[.35, .52]
Indirect effect for working mediation	.14	.05	[.03, .25]

Note. SE = standard error; CI = confidence interval.

Role Model Outcome

There was no significant difference between conditions for reported belonging and trust at the technology company, t(240) = 0.88, p = .379, d = 0.11, mean difference = 0.11, SE = 0.12 (control condition: M = 0.05, SD = 1.01; masculinity bias condition: M = −0.05, SD = 0.88). However, in partial support of Hypothesis 3, there was a significant indirect effect of condition via identification on belonging and trust (indirect effect = 0.15, 95% CI [0.05, 0.27]; see Table 3 for mediation model). Participants felt a greater shared identity with the scientist in the masculinity bias condition relative to the control condition and identifying with the scientist correlated with higher belonging and trust at the tech company.

Participants

We recruited 275 female participants from a large Midwestern university for research credit in a psychology course. Thirty-eight participants were excluded for failing attention checks, and these participants did not differ across information condition, χ²(1, N = 275) = 0.52, p = .490 (see Table 1). We had a final sample of 237 female participants with a mean age of 19.64 (SD = 3.50) and age range of 18–44 years. Participants’ specific racial/ethnicity make-up was 169 White (71.3%), 27 Black or African American (11.4%), 21 Latino (8.9%), 13 Asian (5.5%), four Other (1.7%), two American Indian/Alaskan Native (0.8%), and one unreported (0.4%). One hundred seventy-nine participants were in their first year of university (75.5%), 27 were in their second year (11.4%), 13 were in their third year (5.5%), six were in their fourth year (2.5%), 10 were in fifth year or above (4.2%), and two did not indicate their year in university (0.8%). Forty-nine participants reported having taken no STEM classes in college (20.7%), whereas 188 participants reported having taken at least one STEM class in college (79.3%). Thirty-one participants (12.1%) indicated having a STEM majors based on NSF (NSF, National Center for Science and Engineering Statistics, 2017) classifications (e.g., biology, chemistry, mathematics). An additional 74 (31.2%) participants indicated being in a medical related major (e.g., pre-med, nursing).

Procedure

Experiment 1b had the same general design as Experiment 1a, with four modifications. First, the article described Dr. Brandon Evans as a “well-respected professor in Computer Science” at the students’ university. Second, we shortened the empathy measure to five items and removed “softhearted” from this scale. Third, instead of being presented with a fictional computer science company, participants indicated their belonging and trust and comfort in computer science generally, “for example, in a computer science class, working on a computer science research project.” Finally, participants completed a new measure assessing their interest in working with the scientist (Pietri, Johnson, Ozgumus, & Young, 2018; M = 3.51, SD = 0.71, α = .88). For this measure, participants indicated their agreement from 1 (strongly disagree) to 5 (strongly agree) with statements assessing their interest in taking classes with Dr. Evans (e.g., “I would be excited to take a class with Dr. Evans”), their interest in doing research with Dr. Evans (e.g., “I would be excited to work as a research assistant in Dr. Evans’ lab”), and their general comfort in approaching Dr. Evans for opportunities (“I would feel comfortable asking Dr. Evans if I could work as a research assistant in his lab”). See Table 4 for means, standard deviations, and reliabilities across all measures in Experiment 1b.

OPEN IN VIEWER

Table 4.

Descriptive Statistics and Correlations for Experiment 1b Outcomes.

Measure	M	SD	α	1	2	3	4	5	6	7	8
1. Awareness of masculinity bias	3.34	0.59	.61	—
2. Empathy	3.97	1.41	.91	−.07	—
3. Perception the scientist faced bias	2.92	0.85	.88	14*	.21**	—
4. Shared identity with Brandon (z-score composite)	—	—	—	−.01	.57***	.35***	—
5. Similar past situations	2.77	1.32	.90	.05	.25***	.39***	.57***	—
6. Similar past unfair treatment	2.61	1.37	.95	.13*	.25***	.53***	.51***	.79***	—
7. Belonging and trust (z-score composite)	—	—	—	−.10	.24***	.22***	.43***	.31***	.30***	—
8. Awareness of gender bias in STEM	3.63	0.63	.82	.27***	−.20**	−.08	−.17*	−.16*	−.15*	−.25***	—
9. Interest in working with Brandon	3.51	0.71	.88	.03	.51***	.26***	.60***	.57***	.51***	.43***	−.17*

Note. Self-other overlap (M = 2.44, SD = 1.18, α = .95) and relatable (M = 3.06, SD = 0.64, α = .87), r(235) = .68, p < .001, and belonging (M = 2.44, SD = 0.64, α = .81) and trust comfort (M = 2.40, SD = 0.77, α = .77), r(235) = .68, p < .001, were highly correlated. Thus, we created the same z-score composites (i.e., shared identity and belonging and trust) from Experiment 1a. STEM = science, technology, engineering, and mathematics.

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

Guide Outcome

Extending Experiment 1a and in line with Hypothesis 2, female students in the masculinity bias information condition also were more interested in taking computer science courses and conducting computer science research with the father scientist than those in the control information condition (p = .042). Providing additional support for Hypothesis 2, there was a significant indirect effect of information condition on interest in working with the scientist via identification (indirect effect = 0.14, 95% CI [0.03, 0.25]; see Table 3 for mediation model). Participants identified more with the scientist in the masculinity bias condition than the control condition, and feeling similar to the scientist related to greater interest in working with the father scientist.

Role Model Outcome

Finally, we only found partial support for Hypothesis 3. The effect of condition was not significant for belonging and trust in computer scientist (see Table 5 for this effect). However, we found a significant indirect effect of condition on belonging and trust through identification (indirect effect = 0.13, 95% CI [0.03, 0.23]; see Table 3 for mediation model). Participants felt more similar to the scientist in the masculinity bias condition compared to those in the control condition, and feeling similar to the scientist related to stronger feelings of belonging and trust in computer science.

Participants

We recruited 416 female participants from MTurk for US$1.00 payment. Because participants read two scientist articles, we included article attention check questions for each specific article. For the bias specific condition, we included an additional article attention question to ensure participants detected this scientist had encountered hardship. In total, we excluded 78 participants for failing attention checks. However, and unexpectedly, excluded participants differed across the three conditions, χ²(2, N = 416) = 15.27, p < .001 (excluded participants in baseline control = 22, bias specific = 41, and general masculinity bias = 15). This significant effect was driven by a higher number of participants being excluded in the specific bias condition, in which participants had to correctly answer two article attention questions rather than a single question. Indeed, the 27 participants who were excluded for failing the module attention check were equally distributed across condition, χ²(2, N = 416) = 2.56, p = .278. When we only leave out participants from our analyses based on the module attention check questions, we find very similar results (see Supplemental Materials for analyses). Thus, to maintain consistent exclusion criteria, we excluded participants based on both the module and article attention checks.

We had a final sample of 338 female participants. Thirty-five (10.4%) currently worked in a STEM field. The specific racial/ethnicity demographics of participants were 251 White (74.3%), 41 Black or African American (11.2%), 22 Asian (6.5%), 18 Hispanic of Latino/a American (5.3%), seven Other (2.1%), one American Indian or Alaska Native (0.3%), and one unreported (0.3%). With regard to educational level of participants, 89 had a high school degree or GED (26.3%), 72 had a 2-year college degree (21.1%), 123 had a 4-year college degree (36.4%), 47 had a master’s degree (13.9%), two had a doctorate degree (0.6%), and five had a professional degree (1.5%). Because of a coding error, we did not collect participants’ age in Experiment 2.

Procedure

Upon beginning the experiment, participants were randomly assigned to one of the three conditions—baseline control, Brandon specific bias, or masculinity bias. In the baseline control condition, participants watched the same panda control informational module from Experiment 1a and then completed the three module attention check items for the control module (see Table 1). Following these questions, participants read two articles about successful male computer scientists—Dr. Brandon Evans and Dr. Mark Kappen. The article highlighting Brandon was a slightly shortened version of the article employed in Experiment 1a. The new article featuring Mark began by describing his background and how he became interested in programming and then described Mark’s research on simulation software development. The article ended with a section discussing Mark’s interests outside of work, which featured information about Mark’s devotion to fatherhood. For instance, Mark said, “I probably enjoy downtime with my family the most. My best workout is chasing after my kids. They definitely keep me moving!” Participants first read about Brandon and then Mark, and neither article included information suggesting the scientists had encountered past adversity. Immediately following each article, participants completed an article attention check item to ensure they read each article (see Table 1 for Brandon attention check item and Supplemental Materials for Mark attention check item).

In the Brandon specific bias condition, participants first watched the panda control module and completed the control module attention check questions. Participants then read the article highlighting Brandon; however, this article had a new section entitled “Balancing Fatherhood With Work.” In this section, the article described the challenges Brandon faced when his daughter was born prematurely with complications and his struggle to convince his boss that he deserved time off to be with his family. For instance, Brandon indicated that “When I told the project manager, basically my boss, that I needed to take time off to be with my family he was very dismissive and encouraged me not to take much, if any time off.” Right after reading this article, participants completed the same Brandon article attention check item from the baseline control condition. Because this article featured critical information about Brandon’s unique struggles, we included a second attention check question that specifically captured whether participants read and noticed this information (i.e., “In this article: ‘Brandon talks about being treated unfairly by his colleagues [correct answer],’” see Supplemental Material for full question). Participants next read the article about Mark (i.e., the same article employed in the baseline control condition) and completed the Mark article attention check question.

Finally, those assigned to the masculinity bias condition watched the same masculinity bias module from Experiment 1a and completed the three masculinity bias module attention check questions. Next, participants read the article about Brandon followed by the article about Mark. These were the same articles in the baseline control condition, and as a result, neither article provided information suggesting Brandon or Mark had faced hardships. Immediately following each article, participants completed a Brandon article attention check item and Mark article attention check item. This experiment had a three-condition design (baseline control vs. Brandon specific bias vs. masculinity bias).

Measures

Once participants finished the article about Brandon, using the same measures from Experiment 1a, participants indicated their level of empathetic concern for Brandon, their perceptions that Brandon had faced bias, their perceptions that Brandon is relatable, and their self-other overlap with Brandon. After reading about Mark, participants completed the same four measures with regard to Mark.

Participants next viewed the same “CompTech” computer science company from Experiment 1a and indicated their belonging and trust and comfort at the company. The two scientists were again unaffiliated with this company. Finally, participants reported their awareness of masculinity bias and their awareness of gender bias in STEM (see Table 6 for means, standard deviations, and reliabilities across measures).

OPEN IN VIEWER

Table 6.

Descriptive Statistics and Correlations for Experiment 2 Outcomes.

Measure	M	SD	α	1	2	3	4	5	6	7	8
1. Awareness of masculinity bias	3.34	0.60	.66	—
2. Empathy for Brandon	4.70	1.39	.94	.16**	—
3. Empathy for Mark	4.29	1.44	.96	.18**	.74***	—
4. Perception that Brandon faced bias	3.10	0.96	.89	.25***	.32***	.15**	—
5. Perception that Mark faced bias	2.69	0.83	.92	.19**	.11*	.20***	.56***	—
6. Shared identity (z-score composite) with Brandon	—	—	—	.17**	.60***	.53***	.27***	.18**	—
7. Shared identity (z-score composite) with Mark	—	—	—	.18**	.46***	.66***	.15**	.25***	.73***	—
8. Belonging and trust (z-score composite)	—	—	—	−.12*	.10	.19***	.08	.19***	.30***	.38***	—
9. Awareness of gender bias in STEM	3.88	0.67	.87	.23***	.04	−.04	.04	−.04	−.03	−.11	−21***

Note. Self-other overlap (M = 3.30, SD = 1.47, α = .96) and relatable (M = 3.49, SD = 0.71, α = .89) for Brandon, r(336) = .77, p < .001, were highly correlated, and self-other overlap (M = 2.99, SD = 1.42, α = .97) and relatable (M = 3.39, SD = 0.67, α = .90) for Mark were highly correlated, r(336) = .72, p < .001. Belonging (M = 3.13, SD = 0.71, α = .90) and trust comfort (M = 3.15, SD = 0.78, α = .88), r(336) = .80, p < .001, also were strongly correlated. Thus, we created the same z-score composites (i.e., shared identity with Brandon, shared identity with Mark, and belonging and trust) from Experiment 1a and b. STEM = science, technology, engineering, and mathematics.

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

Awareness of Bias Outcomes

We first found that there was a significant effect of condition on participants’ awareness of masculinity bias, F(2, 334) = 14.19, p < .001, η p 2 = .078. Replicating Experiment 1, compared to the participants in the baseline control condition (n = 120; M = 3.13, SD = 0.59), participants in the masculinity bias condition (n = 102; M = 3.53, SD = 0.65) reported more awareness of masculinity bias, p < .001, d = 0.65, mean difference = 0.40, SE = 0.08. Because in the Brandon specific bias condition, the article discussed how Brandon’s tech company did not believe he should take time off to be with his family, participants (n = 117) in this condition also reported greater awareness of masculinity bias relative to those in the baseline control condition (M = 3.36, SD = 0.48), p = .009, d = 0.43, mean difference = 0.23, SE = 0.08. Similar to Experiment 1b, there was no effect of condition on awareness of gender bias in STEM, F(2, 334) = 2.17, p = .116, η p 2 = .013; baseline control: M = 3.78, SD = 0.67; Brandon specific bias: M = 3.94, SD = 0.66; and masculinity bias: M = 3.93, SD = 0.69.

Perceptions of Brandon

Next looking at perceptions of Brandon, we found a significant effect of condition predicting empathic concern, F(2, 335) = 15.67, p < .001, η p 2 = .086; perceptions that Brandon has faced bias, F(2, 335) = 73.95, p < .001, η p 2 = .306; and shared identity with Brandon, F(2, 335) = 14.80, p < .001, η p 2 = .081. Supporting Hypothesis 1a, compared to participants in baseline control condition (faced bias: M = 2.59, SD = 0.81; shared identity: M = −0.36, SD = 0.83), those in Brandon specific bias condition (faced bias: M = 3.88, SD = 0.63; shared identity: M = 0.20, SD = 0.89) and masculinity bias condition (faced bias: M = 2.93, SD = 0.93; shared identity: M = 0.20, SD = 0.98) reported higher perceptions that Brandon had faced bias (specific bias: p < .001, d = 1.78, mean difference = 1.29, SE = 0.11; masculinity bias: p = .004, d = 0.39, mean difference = 0.34, SE = 0.10) and had a stronger shared identity with Brandon (specific bias: p < .001, d = 0.65, mean difference = 0.56, SE = 0.12; masculinity bias: p < .001, d = 0.61, mean difference = 0.55, SE = 0.12; see Figure 1). With regard to empathy, participants in the Brandon specific bias condition (M = 5.29, SD = 1.24) reported feeling significantly more empathetic concern for Brandon than participants in the baseline control condition (M = 4.30, SD = 1.32), p < .001, d = 0.77, mean difference = 0.99, SE = 0.18. Although relative to those in the baseline control condition, participants in the masculinity bias condition also felt more empathy for Brandon (M = 4.60, SD = 1.41), this difference was not significant, p = .207, d = 0.20, mean difference = 0.29, SE = 0.17.

OPEN IN VIEWER

Figure 1.

Condition effects on shared identity from Experiment 2.

We ran a serial mediation analysis to test Hypothesis 1b with Brandon (i.e., condition → scientist faced bias → empathy → shared identity) using the PROCESS Macro Model 6 and 10,000 bootstrap resamples. Because we had three conditions in this model, we dummy coded condition, with control information condition as the reference group. Thus, we tested two predictor contrasts—control (0) versus Brandon specific bias condition (1) and control (0) versus masculinity bias condition (1). We found significant indirect effects via the serial mediation pathway with both control versus Brandon specific bias condition as the predictor (indirect effect = 0.16, 95% CI [0.07, 0.25]) and control versus masculinity bias condition as the predictor (indirect effect = 0.04, 95% CI [0.01, 0.08]; see Table 7 for mediation model).

OPEN IN VIEWER

Table 7.

Mediation Analyses in Experiment 2.

Tested Effect	Effect (Effect Controlling for Mediators)	SE (SE Controlling for Mediators)	95% CI (95% CI Controlling for Mediators)
Brandon perceptions
Control vs. Brandon specific bias → faced bias	1.29	0.11	[1.08, 1.50]
Control vs. masculinity bias → faced bias	0.34	0.10	[0.13, 0.54]
Control vs. Brandon specific bias → empathy	1.00 (0.57)	0.18 (0.21)	[0.63, 1.34] ([0.16, 0.98])
Control vs. masculinity bias → empathy	0.29 (0.18)	0.17 (0.17)	[−0.05, 0.63] ([−0.15, 0.52])
Control vs. Brandon specific bias → shared identity	0.58 (0.10)	0.12 (0.12)	[0.35, 0.82] ([−0.13, 0.33])
Control vs. masculinity bias → shared identity	0.56 (0.42)	0.12 (0.10)	[0.33, 0.79] ([0.23, 0.61])
Faced bias → empathy	0.32	0.09	[0.15, 0.50]
Faced bias → shared identity	0.21 (0.09)	0.06 (0.05)	[0.09, 0.33] ([−0.04, 0.18])
Empathy → shared identity	0.38	0.03	[0.32, 0.44]
Indirect effect for serial mediation with control vs. Brandon specific bias	0.16	0.05	[0.07, 0.25]
Indirect effect for serial mediation with control vs. masculinity bias	0.04	0.02	[0.01, 0.08]
Control vs. Brandon specific bias → shared identity	0.58	0.12	[0.35, 0.82]
Control vs. masculinity bias → shared identity	0.56	0.12	[0.33, 0.79]
Control vs. Brandon specific bias → belonging/trust	−0.09 (−0.29)	0.13 (0.13)	[−0.34, 0.16] ([−0.54, −0.04])
Control vs. masculinity bias → belonging/trust	0.07 (−0.12)	0.12 (0.12)	[−0.17, 0.31] ([−0.36, 0.12])
Shared identity → belonging and trust	0.34	0.05	[0.24, 0.45]
Indirect effect belonging mediation with control vs. Brandon specific bias	0.20	0.05	[0.11, 0.32]
Indirect effect belonging mediation with control vs. masculinity bias	0.19	0.06	[0.09, 0.31]
Mark perceptions
Control vs. Brandon specific bias → faced bias	0.03	0.07	[−0.18, 0.25]
Control vs. masculinity bias → faced bias	0.34	0.11	[0.13, 0.55]
Control vs. Brandon specific bias → empathy	0.014 (0.13)	0.19 (0.19)	[−0.25, 0.52] ([−0.25, 0.50])
Control vs. masculinity bias → empathy	0.46 (0.36)	0.19 (0.19)	[0.09, 0.83] ([−0.01, 0.72])
Control vs. Brandon specific bias → shared identity	0.10 (0.04)	0.12 (0.09)	[−0.14, 0.34] ([−0.14, 0.23])
Control vs. masculinity bias → shared identity	0.50 (0.28)	0.12 (0.09)	[0.27, 0.73] ([0.10, 0.46])
Faced bias → empathy	0.31	0.09	[0.12, 0.49]
Faced bias → shared identity	0.24 (0.12)	0.06 (0.05)	[0.13, 0.36] ([0.03, 0.21])
Empathy → shared identity	0.39	0.03	[0.34, 0.45]
Indirect effect for serial mediation with control vs. Brandon specific bias	0.004	0.01	[−0.02, 0.03]
Indirect effect for serial mediation with control vs. masculinity bias	0.04	0.02	[0.01, 0.08]
Control vs. Brandon specific bias → shared identity	0.10	0.12	[−0.14, 0.34]
Control vs. masculinity bias → shared identity	0.50	0.12	[0.27, 0.73]
Control vs. Brandon specific bias → belonging/trust	−0.09 (−0.13)	0.13 (0.12)	[−0.34, 0.16] ([−0.36, 0.11])
Control vs. masculinity bias → belonging/trust	0.07 (−0.12)	0.12 (0.12)	[−0.17, 0.31] ([−0.35, 0.11])
Shared identity → belonging and trust	0.38	0.05	[0.28, 0.49]
Indirect effect belonging mediation with control vs. Brandon specific bias	0.04	0.05	[−0.05, 0.13]
Indirect effect belonging mediation with control vs. masculinity bias	0.19	0.06	[0.09, 0.32]

Note. SE = standard error; CI = confidence interval.

Perceptions of Mark

Next looking at perceptions of Mark, we found a significant effect of condition predicting empathic concern, F(2, 335) = 3.19, p = .042, η p 2 = .019; perceptions that Mark has faced bias, F(2, 335) = 5.94, p = .003, η p 2 = .034; and shared identity with Mark, F(2, 335) = 10.74, p < .001, η p 2 = .060. Providing additional support for Hypothesis 1a, relative to those in baseline control condition (empathy: M = 4.09, SD = 1.37; faced bias: M = 2.56, SD = 0.85; and shared identity: M = −0.20, SD = 0.84), participants in the masculinity bias condition reported higher empathy for Mark (M = 4.56, SD = 1.47), p = .037, d = 0.33, mean difference = 0.46, SE = 0.19; higher perceptions that Mark has faced bias (M = 2.90, SD = 0.89), p = .005, d = 0.39, mean difference = 0.34, SE = 0.11; and a stronger shared identity with Mark (M = 0.31, SD = 0.99), p < .001, d = 0.55, mean difference = 0.51, SE = 0.12 (see Figure 1). In contrast, participants in the Brandon specific bias condition did not differ from those in the baseline control condition on empathy for Mark (M = 4.23, SD = 1.47), p = .761, d = 0.10, mean difference = 0.14, SE = 0.19; perceptions that Mark has faced bias (M = 2.60, SD = 0.69), p = .949, d = 0.06, mean difference = 0.03, SE = 0.11; or shared identity with Mark (M = −0.12, SD = 0.87), p = .808, d = 0.09, mean difference = 0.08, SE = 0.12.

We again ran the serial mediation analysis to test Hypothesis 1b (i.e., condition → scientist faced bias → empathy → shared identity) and dummy coded condition with control information condition as the reference group. Using our Mark-related outcomes, we found a significant indirect effect via this serial mediation pathway with control versus masculinity bias condition as the predictor (indirect effect = 0.04, 95% CI [0.01, 0.08]) but not with control versus Brandon specific bias condition as the predictor (indirect effect = 0.16, 95% CI [0.07, 0.25]; see Table 7 for mediation model).

Role Model Outcomes

There was no significant effect of module condition on belonging and trust, F(2, 335) = 0.76, p = .471, η p 2 = .004 (baseline control: M = 0.002, SD = 0.91; Brandon specific bias: M = −0.09, SD = 0.97; and masculinity bias: M = 0.07, SD = 0.99). However, in partial support of Hypothesis 3, we found significant indirect effects on belonging and trust via identification with Brandon and with Mark. In particular, there were significant indirect effects through identification with Brandon with both control versus Brandon specific bias condition as the predictor (indirect effect = 0.20, 95% CI [0.11, 0.32]) and control versus masculinity bias condition as the predictor (indirect effect = 0.19, 95% CI [0.09, 0.31]). There also was a significant indirect effect via identification with Mark, with control versus masculinity bias condition as the predictor (indirect effect = 0.19, 95% CI [0.09, 0.32]) but not with control versus Brandon specific bias condition as the predictor (indirect effect = 0.04, 95% CI [−0.05, 0.13]; see Table 7 for mediation models).

Participants

We recruited 302 female participants from MTurk, who were compensated US$1.00 for their participation. We excluded 24 participants for failing attention checks, who did not vary across information condition, χ²(1, N = 302) = 1.54, p = .214 (see Table 1). This left a final sample of 278 participants. The specific racial/ethnic demographics of participants were 203 White (73.0%), 31 Black or African American (11.2%), 16 Asian (5.8%), 16 Latino (5.8%), and seven Other (2.5%). With regard to educational level of participants, 76 had a high school degree/GED (27.3%), 67 had a 2-year college degree (24.1%), 106 had a 4-year college degree (38.1%), 25 had a master’s degree (9.0%), one had a doctorate degree (0.4%), and three had a professional degree (1.1%). Because of a coding error, we did not collect participants’ age in Experiment 3a.

Procedure

The procedure was almost identical to Experiment 1a. Participants began by watching either the masculinity or giant panda module; however, prior to reading about the scientist, participants first saw a fictional company’s website (“CompTech,” the same company from Experiment 1a). Participants then were told they would read about a computer scientist working at the company (i.e., Dr. Brandon Evans). Brandon was described as a leader at the company and “the head of software innovations and research development team.”

Measures

After reading about Brandon, using the same measures from Experiment 1a, participants indicated their level of empathetic concern for Brandon, their perceptions that Brandon had faced bias, their perceptions that Brandon is relatable, and their self-other overlap with Brandon. Participants also completed measures assessing their anticipated belonging, trust and comfort at the company; their awareness of masculinity bias; and their awareness of gender bias in STEM using the same measure from Experiment 1a (see Table 8 for means, standard deviations, and reliabilities).

OPEN IN VIEWER

Table 8.

Descriptive Statistics and Correlations for Experiment 3a Outcomes.

Measure	M	SD	α	1	2	3	4	5	6	7
1. Awareness of harmful masculine stereotypes	3.31	0.71	.74	—
2. Empathy	4.69	1.51	.96	.16**	—
3. Faced bias	2.46	0.87	.91	.27***	.16**	—
4. Shared identity (z-score composite)	—	—	—	.20**	.59***	.27***	—
5. Interest in working with Brandon	4.02	0.66	.93	.30***	.56***	.19***	.65***	—	.
6. Attraction to company (z-score composite in Experiment 3a only)	—	—	—	.13*	.38***	.19**	.57***	.59***	—
7. Belonging and trust (z-score composite)	—	—	—	.11	.40***	.19**	.65***	.59***	.81***	—
8. Awareness of gender bias in STEM	3.77	0.69	.85	.09	.05	−.14*	−.13*	.06	−.01	−.13*

Note. Self-other overlap (M = 3.35, SD = 1.56, α = .97) and relatable (M = 3.50, SD = 0.76, α = .92), r(276) = .75, p < .001, and belonging (M = 3.54, SD = 0.72, α = .89) and trust comfort (M = 3.64, SD = 0.74, α = .89), r(276) = .78, p < .001, were highly correlated. Thus, we created the same z-score composites (i.e., shared identity and belonging and trust) from Experiment 1. STEM = science, technology, engineering, and mathematics.

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

New to the current experiment, participants also indicated their agreement from 1 (strongly disagree) to 5 (strongly agree) with statements assessing their interest in working with Brandon and having Brandon as their boss (five items, M = 4.02, SD = 0.66, α = .93). Similar to the measure from Experiment 1b, this index included items assessing their interest in interacting with Brandon (e.g., “I would be excited to work with Brandon”) and their anticipated comfort asking Brandon for opportunities (e.g., “If Brandon were my boss, I would feel comfortable asking him if I could take on a larger role on a project.”). Participants also rated their agreement from 1 (strongly disagree) to 5 (strongly agree) with statements related to their organizational attraction to CompTech (e.g., “A job at Comp Tech would be very appealing to me”; Highhouse et al., 2003; five items, M = 3.63, SD = 0.90, α = .90), and job pursuit intentions for CompTech (“I would accept a job offer from Comp Tech”; Highhouse et al., 2003; five items, M = 3.70, SD = 0.79, α = .91).

Awareness of Bias Outcomes

We first found that participants in the masculine bias information condition (n = 137; M = 3.54, SD = 0.74) reported higher awareness of masculinity bias than participants in the control information condition (n = 137; M = 3.08, SD = 0.61), t(276) = 5.64, p < .001, d = 0.68, mean difference = 0.46, SE = 0.08. In contrast, there was no significant effect of information condition on awareness of gender bias in STEM (control condition: M = 3.75, SD = 0.70; masculinity bias condition: M = 3.78, SD = 0.68), t(276) = 0.37, p = .710, d = 0.04, mean difference = 0.03, SE = 0.08.

Perceptions of Scientist

Replicating our previous results and supporting Hypothesis 1a, we also found that compared to those in the control condition (M = 4.48, SD = 1.62), participants in the masculine bias condition (M = 4.91, SD = 1.36) were more likely to feel empathy for Brandon, t(276) = 2.39, p = .017, d = 0.29, mean difference = 0.43, SE = 0.18. Moreover, relative to participants in the control condition (M = 2.23, SD = 0.80), those in the masculinity bias condition (M = 2.69, SD = 0.89) were more likely to believe that Brandon had faced bias, t(276) = 4.45, p < .001, d = 0.54, mean difference = 0.45, SE = 0.10. Further replicating the previous experiments, participants felt a stronger shared identity with Brandon in the masculinity bias condition (M = 0.14, SD = 0.95) than in the control condition (M = −0.13, SD = 0.91), t(276) = 2.44, p = .015, d = 0.29, mean difference = 0.27, SE = 0.11. Finally, using the PROCESS Macro and 10,000 bootstrap resamples, we found support for Hypothesis 1b and a significant indirect effect of condition on shared identity via the predicted serial mediation pathway identity (condition → Brandon faced gender bias → empathy → shared identity; indirect effect = 0.04, 95% CI [0.002, 0.08], see Table 9 for mediation model).

OPEN IN VIEWER

Table 9.

Mediation Analyses Across Experiment 1a and b.

Experiment and Tested Effect	Effect (Effect Controlling for Mediators)	SE (SE Controlling for Mediators)	95% CI (95% CI Controlling for Mediators)
Experiment 3a
Condition → faced bias	.45	.10	[.25, .65]
Condition → empathy	.43 (.32)	.18 (.19)	[.08, .79] ([−.04, .69])
Condition → shared identity	.27 (.04)	.11 (.09)	[.05, .49] ([−.14, .22])
Faced bias → empathy	.24	.11	[.03, .44]
Faced bias → shared identity	.26 (.18)	.06 (.05)	[.14, .39] ([.08, .29])
Empathy → shared identity	.34	.03	[.28, .40]
Indirect effect for serial mediation	.04	.02	[.002, .08]
Condition → shared identity	.27	.11	[.05, .49]
Condition → belonging/trust	.06 (−.12)	.11 (.09)	[−.16, .28] ([−.29, .05])
Shared identity → belonging/trust	.66	.05	[.57, .76]
Indirect effect for belonging mediation	.18	.08	[.03, .34]
Condition → interest in working with Brandon	.28 (.15)	.08 (.06)	[.12, .43] ([.03, .27])
Shared identity → interest in working with Brandon	.45	.03	[.38, .51]
Indirect effect for working mediation	.12	.05	[.02, .23]
Condition → attraction	.20 (.04)	.12 (.10)	[−.02, .43] ([−.15, .23])
Shared identity → attraction	.59	.05	[.49, .69]
Indirect effect for attraction mediation	.16	.07	[.03, .30]
Experiment 3b
Condition → faced bias	.34	.12	[.10, .57]
Condition → empathy	.19 (.10)	.15 (.15)	[−.10, .49] ([−.19, .39])
Condition → shared identity	.39 (.28)	.11 (.10)	[.17, .60] ([.09, .47])
Faced bias → empathy	.27	.07	[.13, .42]
Faced bias → shared identity	.23 (.13)	.05 (.05)	[.12, .33] ([.03, .23])
Empathy → shared identity	.36	.04	[.28, .43]
Indirect effect for serial mediation	.03	.02	[.01, .07]
Condition → shared identity	.39	.11	[.17, .60]
Condition → interest in working with Brandon	.29 (.11)	.08 (.07)	[.12, .45] ([−.03, .25])
Shared identity → interest in working with Brandon	.45	.04	[.38, .52]
Indirect effect for working mediation	.17	.05	[.07, .28]
Condition → attraction	.20 (.03)	.06 (.08)	[.02, .37] ([−.13, .19])
Shared identity → attraction	.42	.04	[.34, .51]
Indirect effect for attraction mediation	.16	.05	[.07, .27]
Condition → belonging/trust	.12 (−.10)	.11 (.10)	[−.10, .34] ([−.29, .09])
Shared identity → belonging/trust	.57	.05	[.47, .67]
Indirect effect for belonging mediation	.22	.07	[.09, .37]
Condition → CS positive beliefs	.16 (.004)	.10 (.09)	[−.04, .36] ([−.18, .19])
Shared identity → CS positive beliefs	.39	.05	[.30, .49]
Indirect effect for positive beliefs mediation	.15	.05	[.06, .26]

Note. CI = confidence interval; CS = computer science.

Guide and Role Model Outcomes

Because job pursuit intentions and organizational attraction were highly correlated, r(275) = .88, p < .001, and assessing the same construct (i.e., interest/attraction to the organization), we averaged the z-scores of these measures to create a composite measure of company attraction. Participants in the masculinity bias (M = 0.10, SD = 0.93) reported more company attraction than those in the control condition (M = −0.10, SD = 1.00), t(276) = 1.76, p = .080, d = 0.21, mean difference = 0.20, SE = 0.12; however, this difference did not reach significance. In contrast and in support of Hypothesis 2, participants in the masculinity bias condition (M = 4.16, SD = 0.57) were significantly more interested in working with Brandon, the father scientist, compared to those in the control condition (M = 3.88, SD = 0.72), t(276) = 3.54, p < .001, d = 0.43, mean difference = 0.28, SE = 0.08. Further supporting Hypothesis 2, there was a significant indirect effect of condition via identification on interest in working with the father scientist (indirect effect = 0.12, 95% CI [0.02, 0.23]) and attraction to the company (indirect effect = 0.16, 95% CI [0.03, 0.30]; see Table 9 for mediation models). Participants identified more with the scientist in the masculinity bias condition than in the control condition, and identifying with the scientist related to greater interest in working with the father scientist and attraction to the tech company.

Similar to the past experiments, we only found partial support for Hypothesis 3. In particular, there was no significant effect of information condition on anticipated belonging and trust at the company (control condition: M = −0.03, SD = 0.91, masculinity bias condition: M = 0.03, SD = 0.97), t(276) = 0.37, p = .710, d = 0.04, mean difference = 0.03, SE = 0.08. However, there was a significant indirect effect of information condition via identification on belonging and trust (indirect effect = 0.18, 95% CI [0.03, 0.27]; see Table 9 for mediation model). Participants felt a greater shared identity with the scientist in the masculinity bias condition relative to the control condition, and identifying with the scientist related to higher anticipated belonging and trust at the tech company.

Participants

We recruited 301 female students from MTurk, who were compensated US$1.25 for their participation. We excluded 18 participants for failing attention checks, and these participants did not vary across information condition, χ²(1, N = 301) < 0.001, p = .965. This left a final sample of 283 female participants with a mean age of 27.23 (SD = 7.73) and age range of 18–60 years. The specific racial/ethnic demographics of participants were 180 White (63.6%), 44 Black or African American (15.5%), one American Indian/Alaska Native (0.4%), 23 Asian (8.1%), 29 Latino (10.2%), and six Other (2.1%). Three were in high school or receiving GED (1.1%), 51 were at a 2-year college (18.0%), 141 were at a 4-year college (49.8%), 34 were in a graduate/professional school (12.01%), and 54 were other or did not indicate (19.1%). Among participants who were at a 2-year or 4-year college, 27 were in their first year (9.5%), 51 were in the second year (18.0%), 52 were in their third year (18.4%), 48 were in their fourth year (17.0%), and 14 were in their fifth year and above (4.9%). Moreover, among these college students, 15 (0.05%) indicated not taking a STEM class in college, whereas 268 (18%) reported having taken at least one STEM class in college. Among all the participants, 51 (12.1%) noted having a STEM major or being in graduate school for a STEM field based on NSF (NSF, National Center for Science and Engineering Statistics, 2017) classifications (e.g., biology, chemistry, mathematics). An additional 28 (0.09%) participants indicated being in a medical-related major (e.g., pre-med, nursing) or in graduate or professional school for medicine.

Procedure and Measures

The procedure and measures were identical to Experiment 1a with four modifications (see Table 10 for means, standard deviations, and reliabilities across the measures). First, because we had a student sample, we presented participants with a “School of Science and Technology” rather than the fictitious “CompTech” company, and the computer scientist was described as a professor at this school. Second, we did not include the job pursuit intentions measure from Experiment 3a because the wording was more relevant to pursuing employment at a new organization. Rather, participants completed a modified version of the organizational attraction measure focused on interest in attending the school (e.g., “Being a student at this school would be very appealing to me”; five items, M = 3.40, SD = 1.02, α = .92). Third, participants completed the same interest in working with the scientist measure from Experiment 1b (i.e., interest in taking classes with and doing research with Brandon, M = 3.92, SD = 0.73, α = .90). Finally, participants indicated their agreement from 1 (strongly disagree) to 5 (strongly agree) with statements assessing their valuing of the computer scientist’s research (e.g., “I value research focused on creating new coding language”; three items, M = 3.90, SD = 0.77, α = .83; Kosovich et al., 2015), their self-efficacy related to computer scientist’s research (e.g., “I think I can learn how to create new coding language”; three items, M = 3.18, SD = 1.11, α = .90; Kosovich et al., 2015), and their interest in learning more the computer scientist’s research (e.g., “I would be interested in taking a class on coding”; two items, M = 3.56, SD = 1.16, α = .91).

OPEN IN VIEWER

Table 10.

Descriptive Statistics and Correlations for Experiment 3b Outcomes.

Measure	M	SD	α	1	2	3	4	5	6	7	8
1. Awareness of harmful masculine stereotypes	3.41	0.68	.65	—
2. Empathy	4.98	1.26	.93	.01	—
3. Faced bias	2.65	1.01	.91	.14*	.23***	—
4. Shared identity (z-score composite)	—	—	—	.09	.51***	.27***	—
5. Interest in working with Brandon	3.92	0.73	.90	.15*	.44***	.21***	.61***	—
6. Attraction to school	3.40	1.02	.92	.03	.36***	.26***	.53***	.66***	—
7. Belonging and trust (z-score composite)	—	—	—	.05	.38***	.19**	.56***	.58***	.77***	—
8. Computer science positive beliefs (z-score composite)	—	—	—	.01	.28***	.13**	.44***	.63***	.70***	.63***	—
9. Awareness of gender bias in STEM	3.88	0.73	.84	.20**	.01	−.20**	−.08	.03	−.02	−.09	−.02

Note. Self-other overlap (M = 3.35, SD = 1.56, α = .97) and relatable (M = 3.50, SD = 0.76, α = .92), r(281) = .78, p < .001, and belonging (M = 3.54, SD = 0.72, α = .89) and trust comfort (M = 3.64, SD = 0.74, α = .89), r(281) = .80, p < .001, were highly correlated. Thus, we created the same z-score composites (i.e., shared identity and belonging and trust) from Experiment 1. STEM = science, technology, engineering, and mathematics.

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

Awareness of Bias Measures

Replicating the previous experiments, we found that relative to the control information, the masculinity bias information resulted in higher awareness of masculinity bias (p < .001) but did not influence awareness of gender bias (p = .489).

Perceptions of the Scientist

Further replicating the past experiments, we found that relative to the control information, the masculinity bias information led to higher perceptions that the scientist faced bias (p = .005) and a stronger shared identity with the scientist (p = .001). Although the effect of module condition on empathy was in the predicted direction, it was not significant (p = .203; see Table 11). Moreover, using the PROCESS Macro and 10,000 bootstrap resamples, we found a significant indirect effect of condition on shared identity (indirect effect = 0.03, 95% CI [0.01, 0.07]) via the predicted serial mediation pathway (condition → Brandon faced gender bias → empathy → shared identity) supporting Hypothesis 1b (see Table 9 for mediation model).

Guide and Role Model Outcomes

In line with Hypothesis 2, relative to those in the control module condition, participants in the masculinity bias condition reported greater interest in working with the scientist (p = .001) and higher attraction to the School of Science and Technology (p = .027). In addition, there was a significant indirect effect of condition via identification on interest in working with the scientist (indirect effect = 0.17, 95% CI [0.07, 0.28]) and attraction to the school (indirect effect = 0.16, 95% CI [0.07, 0.27]; see Table 9 for mediation models).

Because we had three measures related to perceptions of computer science research and these measures all were positively correlated—valuing and self-efficacy, r(281) = .47; valuing and interest, r(281) = .59; and interest and self-efficacy r(281) = .74—we z-scored these scales and averaged them together to create a composite measure of positive computer science beliefs. Looking at this measure of positive computer science beliefs and at belonging and trust, we again found mixed support for Hypothesis 3. In contrast to our predictions, there was no effect of information condition on belonging and trust (p = .279) or positive computer science beliefs (p = .125). However, in partial support of Hypothesis 3, using Hayes’s (2018) PROCESS macro and 10,000 bootstrap resamples, we found significant indirect effects of information condition via identification for belonging and trust (indirect effect = 0.22, 95% CI [0.09, 0.27]) and for positive computer science beliefs (indirect effect = 0.15, 95% CI [0.06, 0.36]; see Table 9 for mediation models). Thus, participants felt a stronger shared identity with the father scientist after learning about the masculinity bias information compared to learning about control information, and identifying with the scientist related to more belonging and trust and more positive beliefs about computer science.

Practice Implications

Beyond adding theoretically to past work on exemplar identification, the current experiments have practical implications for attracting women to STEM environments. We found that promoting perceptions that a father scientist had faced past adversity enhanced female students’ interest in attending his school of science and technology as well as increased their desire to take classes with the scientist and have the scientist act as their research mentor. Thus, identifying with the male scientist helped turn him into a potential guide for women. This finding is noteworthy because having supportive guides and research experiences are critical for promoting women’s persistence and success in STEM (Downing et al., 2005; Eby et al., 2008; Misra et al., 2017). Moreover, previous work suggests that having female scientist guides and role models is more important for female students’ persistence in STEM-intensive majors than those in non-STEM majors (for discussion, see Dasgupta, 2011; Drury et al., 2011), and thus it is important that our findings replicated with female STEM and medicine majors. The current findings suggest one technique that can help female STEM majors identify with male scientists who also are fathers.

Of note, there are variety of reasons women may believe a successful exemplar has encountered identity-based adversity beyond challenges with fatherhood (e.g., exemplars may have a marginalized racial identity or may be the first in their family to go to college). When women are aware of any identity-based adversity facing a successful exemplar, they may feel empathy for and identify with the exemplar, and this individual has the potential to become a helpful guide. Indeed, Experiment 2 demonstrated that learning about a male scientist’s unique fatherhood struggles enhanced empathy and identification with that scientist. Thus, the current research suggests one strategy that male STEM instructors can employ to be more relatable and approachable for female students—disclosing an identity-based struggle (i.e., discussing fatherhood difficulties, identifying as a first generation college student). We acknowledge that not all male instructors would be comfortable using this technique; nevertheless, this approach may help male STEM instructors function as guides and may remove some of the service burdens from their female colleagues (El-Alayli et al., 2018).

Although teaching women about masculinity bias did not directly influence belonging and trust and positive beliefs about computer science, identifying with the father scientist did relate to these beneficial outcomes. Rather than relying solely on masculinity bias information, it is possible that combining the current manipulations with other established interventions may be more effective for promoting belonging in and positive beliefs about STEM. For example, past experiments have found that information about the pervasive sexism in STEM promotes women’s felt similarity with female scientists (Pietri, Johnson, Ozgumus, & Young, 2018) and encourages positive behaviors to address and confront the unfair treatment of women in STEM (Carnes et al., 2015; Moss-Racusin et al., 2018; Pietri et al., 2017). Consequently, having female and male students undergo an intervention where they learn about harmful stereotypes in STEM, as well as successful female and male scientists, may address students’ biases and create welcoming environments for female students, while also helping students identify with multiple female and male scientists (i.e., potential guides and role models). Indeed, Experiment 2 demonstrated that teaching women about harmful stereotypes can encourage identification with two unrelated scientists simultaneously. Thus, we posit that promoting identification with father scientists will be most beneficial when it is one component of a larger intervention.

Future Directions and Limitations

As noted above, a limitation of the current research is that our manipulation did not directly enhance role model-related outcomes (i.e., belonging and trust and positive beliefs about computer science). Although identifying with the male scientist helped turn him into a guide, identifying with a male scientist most likely did not address all of women’s concerns about STEM fields. In particular, women may still worry that STEM disciplines favor men and masculine traits/values (see Cheryan et al., 2017; Diekman et al., 2017). Female scientists may be uniquely beneficial for addressing these stereotypes and demonstrating that masculine traits are not required for success in STEM (Dasgupta, 2011; Stout et al., 2011). Thus, as mentioned earlier in the discussion, interventions that increase women’s identification with both male and female scientists may be most useful for sparking women’s comfort in and attraction to STEM. It also is important to note that we were teaching women about negative stereotypes in STEM. Although this information focused on how stereotypes harm men, learning about any unfair treatment in these domains may have inadvertently threatened women and undermined their sense of belonging in STEM environments (see Pietri, Hennes, et al., 2019; Pietri, Johnson, Ozgumus, & Young, 2018).

Another limitation of the current research was that we did not test whether women would prefer to work with a male scientist over a female scientist after learning about how stereotypes hurt men in caretaking roles. A practical aim of this research was to develop techniques that help alleviate service burdens for female scientists, and this goal would be undermined if women consistently prefer female guides over male guides. However, past work has found that there are situations where women feel more similar to male scientists than female scientists (i.e., when the male scientists is relatable and counter-stereotypical; Cheryan et al., 2011; Pietri, Johnson, Ozgumus, & Young, 2018), suggesting that women will not indiscriminately choose female guides over male guides. It is also a limitation that we did not utilize a sample of female computer science majors, who would have the most opportunities to interact with a father computer scientist guide. Of note, there is an increased trend for all college students to take computer science courses (Kafka, 2020), which suggests that even non-STEM majors may take classes with male computer science professors. Nevertheless, in future research, it will be important to explore whether information about harmful masculine stereotypes enhances identification with potential father guides among computer science majors and in non-STEM contexts. We started to test this question in the supplemental experiment (see Supplemental Materials for Supplemental Experiment 1). Among a sample of working women, we found that empathy encourages interest in working with and seeking advice from a father executive leader and attraction to the father executive’s consulting company. This additional experiment suggests that our findings may generalize to non-STEM contexts.

In addition to exploring our research questions in non-STEM environments, it will be important to test these questions in disciplines that value the female identity and feminine characteristics. Both women and fathers are marginalized in STEM because their feminine and nurturing traits do not align with stereotypes about scientists (Banchefsky & Park, 2016; Carli et al., 2016; Fiske et al., 1999). Thus, a remaining empirical question is whether women would identify with an outgroup member who is marginalized in a domain where women are positively stereotyped. For instance, would women feel similar to a man after learning that man faced identity-based adversity in a traditionally feminine occupation (e.g., nursing or education; Croft et al., 2015)? In the current research, women were more likely to identify with a father scientist, who they believed had faced bias, even though they perceived that scientist as encountering different types of mistreatment. Thus, women also may feel similar to outgroup members, who have dealt with (different) identity-based mistreatment in traditionally feminine disciplines. However, we did not specifically test this possibility and this will be an important question to answer in future studies.

In order to ensure a potential intervention is helpful for all individuals, future research should also examine whether information about bias increases identification with scientists among men as well as with women who have multiple marginalized identities (Johnson et al., 2019; Pietri, Drawbaugh, et al., 2019; Pietri, Johnson, & Ozgumus, 2018). To begin to address this limitation, we ran a supplemental experiment with male participants, and found that teaching men about masculine stereotypes also encourages their identification with a male scientist (see Supplemental Materials for Supplemental Experiment 2). In addition to testing our research questions with different samples, future work might explore how long the effects from the short masculine stereotypes module persist and whether participants continue to feel similar to male scientists a few hours or days after viewing the module. Although the effects from the 7-minute module may be transient, there are longer interventions that have a lasting influence on awareness of bias (see Becker & Swim, 2011; Carnes et al., 2015; Moss-Racusin et al., 2018). Future research might test whether lengthier interventions promote an enduring change in women’s impressions of male scientists across time and situations. Taken together, there are multiple important avenues for future research to further strengthen this approach for sparking women’s attraction to STEM environments. Nevertheless, the current research represents a critical first step. We found promoting perceptions of past identity-based adversity and empathy for a father scientist can help transform this typically high status outgroup member into a beneficial guide.