Perceived Centrality in Social Networks Increases Women’s Expectations of Confronting Sexism

There are several theoretical reasons why perceived centrality could shape perceptions of the situation when a woman is faced with a sexist comment in an interaction with a coworker. First, we propose that centrality in advice networks could afford a greater understanding of whether network members disagree with the comment and would support confrontation. Through their interactions, central individuals acquire more information, both about their jobs and organizations, and about their team members’ thoughts and beliefs, than those who are peripheral (Sparrowe et al., 2001). These greater informational resources may give central (vs. peripheral) individuals a more accurate sense of the attitudes their network members hold. If central women believe or know that their network members would be offended by a gender-biased comment, they might be more likely to speak out to disagree with it than peripheral women who have less information about how their network members would perceive the comment. We test this possibility in Studies 3, 5, and 6. Second, we propose that central individuals might perceive less social and professional risk to confronting, relative to peripheral women. Individuals who are central in advice networks are, by definition, less dependent on any single person compared with individuals who are peripheral in the network (Brass & Burkhardt, 1993). Taking action is less risky for those who have more social options: If one of their contacts is offended by their actions and severs their tie, central individuals still have many other exchange partners (Gargiulo & Benassi, 2000). Given that people are more likely to confront when perceived risks are low (Kaiser & Miller, 2001), central women might be more likely to speak out than peripheral women. We test this possibility in Studies 2 to 6.

Another perspective would suggest the opposite prediction: that perceived centrality will be associated with reduced intentions to confront because centrality might be associated with viewing the situation as more public than private. As people who are more central have more exchange partners in their advice network, they may perceive themselves as being “on display” and thus see the situation as more public than those with fewer connections (Burt, 2005). Given that individuals are less likely to confront in public than in private (Stangor et al., 2002), this raises the alternate possibility that central women would be less likely to confront sexism relative to peripheral women. We test this in Studies 2 to 4.

Network cognition documents that central individuals evoke different person schemas from peripheral individuals (Brands et al., 2015). Because extroverted, knowledgeable, and popular people are more likely to attain central positions in advice networks (Fang et al., 2015), people learn this association and come to expect central individuals to have these qualities more than peripheral individuals. To test this possibility, we measured whether central women are seen as more competent, confident, knowledgeable, extroverted, and popular than peripheral individuals, and whether this difference would explain differences in anticipated confronting (Study 6). Furthermore, women who are central in their advice network are inherently more practiced at speaking their mind and sharing their perspective with others than women who are peripheral. As such, in Study 6, we test whether central individuals might be seen as more skilled at confronting, and therefore more likely to do so.

We focus our investigation of whether perceived centrality shapes intergroup dynamics on the workplace context because the expression of sexism in everyday interactions between coworkers continues to be prevalent (Ely et al., 2006; Swim et al., 2001). We focus on women’s desire to confront a sexist comment because women must want to engage in this action before actually confronting (Ashburn-Nardo et al., 2008; Lindsey et al., 2015). Past research has focused on how individual characteristics (e.g., optimism, Kaiser & Miller, 2001, 2004; hardiness, Foster & Dion, 2004; trait activism, Hyers, 2007; Swim & Hyers, 1999) and situational factors (e.g., formal power, Ashburn-Nardo et al., 2014; Woodzicka & LaFrance, 2005; salient costs, Shelton & Stewart, 2004; public contexts, Stangor et al., 2002; risk to standing, Kaiser & Miller, 2001, 2004) can restrict confronting. However, relatively less research has identified situational factors that heighten women’s desire to speak out (but see Rattan & Dweck, 2010), despite the well-documented benefits to both targets of bias and those who express bias that follow from confrontation (e.g., Czopp et al., 2006; Shelton et al., 2006, but see Kaiser & Miller, 2001).

The core contribution of the current research to the study of prejudice confrontation is to uncover a never before studied factor that shapes how much women who receive gender-biased comments feel able to confront. This investigation also advances the study of social networks by moving beyond the link between network cognition and networking behavior (i.e., the formation of new ties, Janicik & Larrick, 2005; Shea & Fitzsimons, 2016) to investigate social network cognition as a resource for women in the face of a social threat.

We used a standard cognitive social network assessment (Brands et al., 2015). Respondents listed the initials of up to 10 members of their work team and then indicated ages, genders, ethnicities, and formal leadership responsibilities of each, as well as whether they were the leader. Participants were randomly assigned to complete a matrix measure to report on either the network of advice relations (n = 145) or communication relations (n = 149) in their team. After reading instructions for how to use the matrix, participants in the Advice Network condition were asked, “Who would you go to for advice on work-related matters?” followed by a list of their coworkers’ names. Respondents checked the names of those coworkers they sought advice from. Then, for each team member in turn, respondents also checked the names of coworkers whom that team member went to for advice. Thus, each respondent provided a complete network map concerning her perceptions of who shared advice relations with whom in the team.¹ Participants in the Communication Network condition completed the same procedure, except they were asked who communicates with whom in the team.

Participants were presented with a scenario that described a work interaction in which a new male employee, John, makes a biased statement to them (“I’m really surprised at the types of people who are working here. When you get to the top level—a company like this—you expect only the best people here. I mean, I think they must be hiring associates just for diversity reasons. With all the women here, I wonder how long this company will stay on top.” See Appendix A in Supplemental Online Materials for full scenario). Previous research shows that this statement is reliably perceived as explicit bias (adapted from Rattan & Dweck, 2010, 2018).

An advice (or communication) tie was said to be perceived by a respondent between person i and person j if the respondent indicated that person i asked person j for advice (or communicated with person j; Wasserman & Faust, 1994). We calculated how central each respondent perceived themselves to be by counting the number of informal advice (communication) ties that respondents perceived themselves to receive from other members of their team.

After reading the scenario, participants completed the two-item confronting measure: how likely they would be to calmly but firmly express their disagreement to John, and how likely they would be to not express any disagreement to John (1 = very unlikely to 7 = very likely; Rattan & Dweck, 2010). Responses to the latter question were reverse-scored and, given their high correlation (r = .57, p < .0001), averaged to form mean likelihood of confronting (M = 5.80, SD = 1.21).

Larger teams entail a larger potential audience after the fact, and women may be more likely to confront when their network is composed of more women than men. We thus calculated and controlled for team size (a count of the number of people in the team) and gender composition of the team (proportion of the team who were women).

Centrality is considered distinct from other types of status imbued in demographic characteristics or the power imbued in formal roles. To empirically support this perspective, we measured (in a standard demographics form) social status: education level (less than high school, high school, some university, 3-year university degree, 4-year university degree, master’s degree, doctoral degree, professional degree), combined annual household income (1 = below £20,000; 15 = £150,000+), and subjective social status on a status ladder (1–9, Adler et al., 2000).

We ran a regression entering network type and centrality in Step 1 and their interaction term in Step 2. In support of the hypothesis, there was a significant Network Type × Centrality interaction on anticipated confronting, B = 0.16, SE = .07, p = .02. The more women perceived themselves to be central in their team advice network, the more they reported wanting to confront the sexist statement, B = 0.11, SE = .05, p = .04. Women’s communication network centrality was unrelated to anticipated confronting, B = −0.05, SE = .05, p = .26.

We repeated the analysis, controlling for team size and the proportion of the team who were women. Controlling for these team characteristics, which did not predict anticipated confronting, there was still a significant Network Type × Centrality interaction on anticipated confronting, B = 0.15, SE = .07, p = .03. Advice network centrality was a significant predictor of anticipated confronting, B = 0.11, SE = .05, p = .04, whereas communication network centrality was not, B = −0.04, SE = .05, p = .45.

Controlling for education, income, subjective social status, and formal leadership role, which did not themselves predict confronting, there was again (marginal) support for the Network Type × Centrality interaction on anticipated confronting, B = 0.14, SE = .07, p = .06. Perceived advice network centrality was a nonsignificant but marginal predictor of anticipated confronting, B = 0.09, SE = .09, p = .09, whereas communication network centrality was not, B = −0.04, SE = .07, p = .32.

Participants were 121 North American women recruited from Amazon’s Mechanical Turk (paid $2). Participants reported an average age of 33.71 years (SD = 9.18); 94 self-identified as White American, 12 African American, six Asian American, six Latina American, two Native American, one Indian American, and 17 did not specify their ethnicity.

To enter the survey, participants had to indicate that they were a woman, a member of a team, and employed (either full- or part-time), and provide informed consent.

In support of the hypothesis and replicating Study 1, the more women perceived themselves to be central in their team advice network, the more likely they were to report that they would confront the sexist statement, r = .25, p = .01.

Regressing team size and proportion of the team who were women in Step 1, which did not themselves predict anticipated confronting, advice network centrality (in Step 2) remained a significant predictor of anticipated confronting, B = 0.16, SE = .06, p = .01.

Similarly, when we controlled for the demographic status and power variables measured (education, income, subjective social status, and formal leadership role) in Step 1, which did not themselves predict confronting, advice network centrality was still a significant predictor of anticipated confronting, B = 0.15, SE = .06, p = .02.

Perceived advice network centrality did not predict differences in the degree to which women thought the situation was public versus private, r = −.10, p = .29, as we predicted.

As predicted, the more central women perceived themselves to be, the less risk they perceived, r = −.2, p = .03. Given this pattern, we explored an indirect effect. Using the PROCESS macro (model 4; Hayes, 2012), we entered perceived advice network centrality as the predictor X, perceived risk as the mediator M, and confronting as the outcome Y. To the extent that women perceived themselves as central in their team advice network, they perceived less professional and social risk associated with confronting (Β = −0.10, SE = .05, p = .03). Furthermore, to the extent that respondents perceived confronting the biased statement to be risky, they were less likely to do so (Β = −0.31, SE = .11, p = .01). Based on a bootstrap sample of 5,000 iterations, the 95% confidence interval (CI) for the indirect effect was [.0003, .0903], supporting an indirect effect of advice network centrality through perceptions of risk on anticipated confronting. The direct effect of advice network centrality on anticipated confronting remained significant, Β = 0.13, SE = .06, p = .02.

The preregistered target sample size was 850 employed women who worked in teams of at least four or more. We recruited a turkprime.com panel (paid $2). Although 875 people started the survey, 21 identified as men and 13 did not report gender so they were excluded prior to analysis as they did not meet the preregistered inclusion criteria. This left 841 women: 85 self-identified as African American/Black, 35 Asian American/Asian, 581 European American/White, 46 Hispanic/Chicana/Latina American, three South Asian Indian, two Native American, three Pacific Islander, 16 other, and 69 unreported: M_age = 36.35 (SD = 10.26) years. A power analysis based on the observed effect size of the main result in Study 2 found that a total sample size of 90 would be sufficient to detect an effect with power = .8 and alpha error probability = .05. Given the two between-subjects factors in this study (context and order), our sample size afforded sufficient power to detect the key relationship of interest.

After giving their informed consent, participants were randomly assigned to complete the networks assessment either before or after reading the scenario (Order Condition).

The 2 order (before, after) by 2 context (work, party) by centrality regression on anticipated confronting yielded no main effect of display order, B = −0.06, SE = .09, t = −0.70, p = .48, and no main effect of context, B = 0.13, SE = .09, t = 1.54, p = .12, but a significant main effect of centrality, B = 0.08, SE = .02, t = 3.66, p < .001. The Context × Order interaction was not significant, B = 0.02, SE = .09, t = 0.20, p = .84, nor was the Centrality × Order interaction, B = 0.001, SE = .02, t = 0.06, p = .96, the Context × Centrality interaction, B = −0.02, SE = .02, t = −0.92, p = .36, or the three-way interaction, B = −0.01, SE = .02, t = −0.52, p = .60. As expected, the order of the measures did not affect anticipated confrontation. Supporting our hypothesis, women who reported that they were central in their workplace advice networks were more likely to anticipate confronting a sexist comment. However, contrary to our preregistered hypothesis, the two-way centrality by context interaction was nonsignificant. This suggests that women who believe they are central in workplace advice networks are more likely to feel they can confront bias, even outside the workplace.

There were no significant main effects, two-way interactions, or three-way interaction on participants’ ratings of the setting as public versus private, ts < 1.72, ps > .08.

The 2 order (before vs. after) by 2 context (work vs. party) by centrality regression on perceived risk yielded only a significant main effect of centrality, B = −0.05, SE = .02, t = −3.03, p = .003, as did the same analysis on network members’ attitudes, B = 0.05, SE = .02, t = 3.12, p = .002. Given this result, we explored both of these variables as potential mechanisms. Using Hayes (2012) PROCESS macro (model 4), we entered centrality as the predictor X, confrontation as the outcome Y, network attitudes as the mediator M1, and perceived risk as the mediator M2 with 5,000 iterations. Both indirect effects for risk and network attitudes were supported: the total indirect effect was B = 0.04, SE = .01, 95% CI = [0.02, 0.06], the indirect effect of network attitudes was B = 0.02, SE = .01, 95% CI = [0.01, 0.04], and the perceived risk indirect effect was B = 0.02, SE = .01, 95% CI = [0.01, 0.03], although the direct effect of centrality was still significant, B = 0.04, SE = .02, t = 2.12, p = .03, 95% CI = [0.003, 0.07].

Using MTurk, we recruited 201 U.S. women who received $2 each for participating. Of these, seven participants were excluded on the a priori criterion of previous participation in our studies. This left 194 participants in the final sample (190 women, four men) who self-identified as 24 African American/Black, 11 Asian American/Asian, 156 European American/White, eight Hispanic/Chicana/Latina American, one East Indian, one Native American, and one no response, M_age = 37.22 (SD = 10.39) years.

After informed consent, participants read about Erica, a member of the digital media team within a large professional services organization. They were randomly assigned to centrality condition and rank condition.

The 2 (advice network centrality: central vs. peripheral) × 2 (rank: same vs. higher) analysis of variance (ANOVA) on participants’ expectations for Erica’s confronting yielded significant main effects of advice network centrality, F(193) = 19.47, p < .001, ${\eta }_{\mathrm{p}}^2$ = .09, and rank, F(193) = 5.42, p = .02, ${\eta }_{\mathrm{p}}^2$ = .03. Participants who read that Erica was central in the advice network (M = 3.81, SD = 0.1) expected her to confront more than those who read that she was peripheral (M = 3.12, SD = 0.11). Participants who read about Erica interacting with a manager (M = 3.29, SD = 0.11) expected her to confront less than participants who read that she interacted with a peer (M = 3.65, SD = 0.11). This was qualified by a significant interaction, F(193) = 4.32, p = .04, ${\eta }_{\mathrm{p}}^2$ = .02; see Figure 1. As predicted, when participants read that Erica held a central network position, they expected her to confront John to an equal degree, regardless of whether he was higher rank (M = 3.79, SD = 0.15, N = 52) or at the same rank (M = 3.83, SD = 0.15, N = 52), F(190) = 0.03, p = .85. In contrast, when participants read that Erica was peripheral, they thought she would be more likely to confront John when he was at her level (M = 3.47, SD = 0.16, N = 47) than when he was a manager (M = 2.79, SD = 0.16, N = 43), F(190) = 9.05, p = .003.

Participants saw confrontation as less risky when Erica was depicted as central (M = 5.3, SD = 0.46) rather than peripheral (M = 6.84, SD = 0.5) in the advice network, F(190) = 5.07, p = .03. There was no difference in perceived risk when the perpetrator of the biased statement was a manager (M = 5.1, SD = 0.48) versus the same rank (M = 6.64, SD = 0.49) as Erica, F(190) = 2.7, p = .10. The interaction between network role and rank was not significant, F(193) = 0.2, p = .66, ${\eta }_{\mathrm{p}}^2$ = .001.

Using Hayes (2012) PROCESS macro (model 5), we entered network role condition as the predictor X, rank of perpetrator as W the moderator of the direct effect, and perceived risk as the mediator M. When Erica was depicted as peripheral, participants saw confronting as riskier, Β = 1.51, SE = .69, p = .03, and risk, in turn, predicted lower anticipated confronting, Β = −.1, SE = .01, p < .0001. Based on a bootstrap sample of 5,000 iterations, the 95% CI for the indirect effect was [−0.31, −0.28], suggesting support for the indirect effect. The direct effect of network role, moderated by the status of the perpetrator, remained significant, CI = [−1.12, −0.04].

Only women who indicated they were a member of a team and working full- or part-time were able to enter the survey. Respondents were 402 North American women employees recruited from MTurk (paid $2): five self-identified as African American, three Asian American, 86 White American, one Latino American, one Native American, 216 other, and 90 did not report: M_age = 35.77 (SD = 10.05) years.

Sixty-nine individuals said they could recall an incident of direct sexism, amounting to 53.49% of women who reported experiencing any type of direct sexism (see Table 4 for means, standard deviations, and correlations among the study variables for this sample). This represents the final sample for this study—all analyses were on this final sample (one self-identified as African American, 16 White American, one Native American, 36 other, and 15 did not report: M_age = 35.09 years, SD = 9.12).

Women who reported being central in their advice network were more likely to report having confronted an expression of gender bias, r = .34, p < .004.

We ran a regression with team size and the proportion of the team who were women in Step 1, advice network centrality in Step 2, with reported confronting as the dependent variable. We found that controlling for these team characteristics, neither of which significantly predicted confronting, advice network centrality remained a significant predictor of anticipated confronting, B = 0.06, SE = .03, t = 2.43, p = .02.

In another regression, we entered the power and status variables measured (formal leadership role, education, income, and subjective social status) as predictors in Step 1, advice network centrality in Step 2, with reported confrontation as the dependent variable. Controlling for these status characteristics, of which only formal leadership position was a significant predictor, centrality was still a significant predictor of confronting, B = 0.06, SE = .02, t = 2.78, p = .007.

Controlling for whether other people witnessed the statement, centrality remained a significant predictor of confronting, B = 0.07, SE = .02, t = 2.94, p = .004.

Using Hayes (2012) PROCESS macro (model 4), we entered centrality as the predictor X, self-reported confronting as the outcome Y, and retrospective risk as the mediator M with 5,000 iterations. The model was not supported (95% CI = [−0.02, 0.23]) because centrality did not predict retrospective risk, B = −0.09, SE = .06, t = −1.52, p = .13.

The same model with network attitudes as the mediator M was not supported (95% CI = [−0.02, 0.14]), although centrality predicted women’s perceptions of their network members’ attitudes, B = 0.12, SE = .05, t = 2.19, p = .03.

Participants were 304 U.K. adults recruited from Prolific Academic (paid £2). Of these, six participants were excluded on the a priori criterion of failing the attention checks. This left 298 participants in the final sample (296 women, two men) who self-identified as five Black, 14 Asian, 265 White, and 14 no response: M_age = 35.28 (SD = 11.39) years.

After informed consent, participants read the Erica scenario from Study 4 and were randomly assigned to advice network condition.

An independent-samples t test indicated a significant effect of advice network centrality on confronting, F(296) = 12.66, p < .0001. Participants in the central condition (M = 4.21, SD = 0.81) expected Erica to confront more than those in the peripheral condition (M = 3.66, SD = 1.03).

No differences in perceived risk emerged: central (M = 2.73, SD = 1.25), peripheral (M = 2.83, SD = 1.83), F(296) = 1.41, p = .24.

No differences in skill at confronting emerged, central (M = 5.11, SD = 1.03), peripheral (M = 4.18, SD = 1.1), F(296) = 0.09, p = .77.

No differences emerged for ratings of competence, central (M = 5.19, SD = 1.02), peripheral (M = 4.04, SD = 0.91), F(296) = 2.08, p = .15, or extroverted, central (M = 4.21, SD = 1.16), peripheral (M = 2.93, SD = 1.25), F(296) = 2.76, p = .10. However, participants in the central condition perceived Erica as more confident (M = 5.05, SD = 0.89), knowledgeable (M = 5.37, SD = 0.8), and popular (M = 4.79, SD = 0.95) than those in the peripheral condition (confident, M = 3.72, SD = 1.22; knowledgeable, M = 3.95, SD = 1.16; popular, M = 2.85, SD = 1.2), confident: F(296) = 28.84, p < .0001, knowledgeable: F(296) = 12.64, p < .0001, popular: F(296) = 11.08, p< .001.

Participants in the central condition (M = 4.96, SD = 0.8) perceived Erica’s network members’ attitudes more positively than those in the peripheral condition (M = 4.56, SD = 0.92), F(296) = 4.78, p = .03.

Using Hayes (2012) PROCESS macro (model 4), we entered network role condition as the predictor X, confidence, knowledgeable, popular (included because the main effect was marginal), extroverted, and network attitudes as the mediators M. The model was supported. When Erica was depicted as central, participants saw her as more confident (Β = 1.33, SE = .13, p < .0001), knowledgeable (Β = 1.42, SE = .12, p < .0001), popular (Β = 1.94, SE = .13, p < .0001), and extroverted (Β = 1.28, SE = .14, p < .0001) than when she was peripheral, and they thought her network members would want her to confront more (Β = 0.4, SE = .1, p < .0001). Confidence (Β = 0.31, SE = .08, p < .0001) and network member attitudes (Β = 0.28, SE = .06, p < .0001), in turn, predicted anticipated confronting, whereas knowledgeable (Β = −0.08, SE = .08, p = .29), popularity (Β = −0.002, SE = .06, p = .97) and extroverted (Β = 0.01, SE = .05, p = .26) did not. Based on a bootstrap sample of 5,000 iterations, the 95% CI for the indirect effects was confident [0.17, 0.7], network member attitudes [0.04, 0.21], knowledgeable (−0.35, 0.11], popular [−0.22, 0.2], and extroverted [−0.11, 0.15]. The direct effect of network role was not significant, CI = [−0.14, 0.38]. Perceived confidence and network attitudes account for the effect of network position on expectations of confronting.