When sexism is persuasive: agreement with hostile and benevolent sexism as a function of source gender

Abstract

Expressions of prejudice are considered more severe when consistent with prototypical features. Two studies examine how speaker gender interacts with the type of sexism expressed (hostile vs. benevolent) to influence participants’ agreement with sexism. Participants (N = 806) evaluated a male or female target whose opinions reflected hostile or benevolent sexism toward women (Study 1) or men (Study 2). Participants were more likely to agree with benevolent sexist statements expressed by speakers praising their out-groups and were more likely to agree with hostile sexist statements expressed by speakers denigrating their in-groups. Mediation analyses suggested that participants’ agreement increased as their perceptions of the speaker as prejudiced decreased. These findings are discussed in the context of the justification of sexism.

Keywords:

• ambivalent sexism
• self-interest
• persuasion
• anti-women sexism
• prejudice

Among the many examples of sexism that women report experiencing are situations in which they are not taken seriously by men or when their previously ignored comments are considered only after a man repeats them (e.g., Swim, Hyers, Cohen, & Ferguson, 2001). Indeed, although there are some exceptions, men are generally more persuasive and influential than women (Carli, 2001). In this study, we consider how the nature of a speaker’s sexist comments may moderate and reverse typically observed male advantages in influence. Men who express hostile forms of sexism are likely to be evaluated unfavorably and this negativity may inhibit their influence (Mae & Carlston, 2005). However, a woman’s identical expression of hostile anti-women sexism may be received more favorably. The goal of this research is to compare the perception and influence of male vs. female speakers who express hostile or benevolent sexism.

Hostile and benevolent sexism

Glick and Fiske (1996) outlined two forms of sexism: hostile sexism and benevolent sexism. Hostile sexism is traditional prejudice, targeting women with blatantly negative stereotypes or attitudes. Hostile sexism operates to disenfranchise women by characterizing them as weak and incapable of providing for themselves, while implying or directly stating that men are superior both in terms of physical prowess and ability to acquire power. Benevolent sexism is the seemingly positive attitude that women have a special and admired place that is ultimately under the authority of men and reinforces the ideas that women must be cared for and protected (Glick & Fiske, 2001).

Benevolent sexism is often a difficult prejudice to detect because it appears to confer advantages to women over men. A man who compliments women as especially gentle and caring may sound egalitarian, and others may form positive evaluations of him (Kilianski & Rudman, 1998). Benevolent sexism may also go undetected as prejudice due to its inconsistency with perceivers’ prototype of what prejudice should be like (Inman & Baron, 1996). When a speaker makes a sexist statement, perceivers compare the statement to a mental representation of what they believe a sexist statement is. If the statement does not fit the mental representation (the prototype), then it is less likely to be considered as sexist. Consistent with this approach, Barreto and Ellemers (2005) found that participants were less likely to consider benevolent sexist statements as reflecting prejudice compared to hostile sexist statements, and the former elicited less anger than the latter.

Speaker gender and persuasion

In addition to the perceived valence of expressed sexism, the gender of the source who expresses such attitudes is likely to influence perceptions of and reactions to sexism. People’s prototypes of sexism likely include characteristics about the perpetrator and target of prejudice. That is, just as the non-prototypical favorability of benevolent sexist speakers renders them more palatable, the same sexist comments may be evaluated differently depending on who makes them. Inman and Baron (1996) demonstrated that negative anti-women statements made by men were seen as more sexist than the same statements made by women and suggested that this was because men are perceived as prototypical sexists. Just as perceivers compare the content of potentially sexist statements to a prototype and judge accordingly, sexist speakers are judged against a mental representation of a sexist person. If the speaker does not match the representation, the perceiver may not judge the speaker (and the speaker’s statements) as sexist. Men are more likely to be considered sexist because our internal working model for who is more likely to be sexist tells us that men are more likely to be anti-female sexist than women. Due to this, when given the same sexist statement endorsed by men or women, individuals are more likely to evaluate the man as more sexist than the woman.

Because women are not seen as prototypical sexists, the consequences of their anti-female sexist remarks may be different than that of men making the same remarks. Although Barreto and Ellemers (2005) compared male and female sources of hostile and benevolent sexist comments, they did not report any results related to their manipulation of source gender. Furthermore, their study focused on participants’ evaluation of the likability of the speaker and whether the comments should be considered prejudiced. In the current research, we focus on the extent to which participants agree with hostile and benevolent sexist statements made by a male or female source. This is an important distinction because the recognition that a speaker’s statement reflects prejudice does not necessarily signify disagreement with the corresponding beliefs of the statement. For people who privately agree with such attitudes, there may be underlying motivations to express such prejudice that are constrained and suppressed by public norms of non-prejudice. Indeed, Mae and Carlston (2005) demonstrated that speakers who endorse negative forms of prejudice are evaluated as less likable and as less socially acceptable, even by those who internally agreed with the bigoted statements. According to Crandall and Eshleman (2003), expressions of prejudice reflect not only our attitudes, but also our ability to suppress or justify our held beliefs. When it is inappropriate to express prejudice, such as when expressions would lead to negative external evaluations from others or contradict one’s own internal self-image, people attempt to suppress such expressions so as to not appear prejudiced to oneself or others. However, people often look for or otherwise find ways to justify their suppressed prejudice. In situations where they can justify their prejudice, it will override the need to suppress and they will express the prejudice through justified means. Justifications take several forms, such as attempting to cover opinions by masking it as something else (e.g., humor) or looking for excuses to support or explain their views. For example, upon hearing a woman express anti-female hostile sexism (e.g., that women are not as smart as men), a sexist perceiver could justify agreeing with the speaker because such a context (a woman expressing anti-women sentiments) is inconsistent with prototypes and therefore not considered sexist. The perceiver is free to express sexism by agreeing with the speaker without feeling the action is “risky.” Through this justification–suppression model, we expect that people will agree with sexist statements when the context (i.e., speaker gender, type of sexism) allows them to justify expressions of prejudice.

We examined the extent to which participants agreed with benevolent sexist or hostile sexist statements expressed by either a man or woman. Because benevolent sexism appears complimentary rather than prototypically prejudicial, men and women endorsing such statements may be seen as speaking in support of women. However, participants may be better able to justify their agreement with benevolent sexist statements when expressed by a man rather than a woman, because a man is likely to be perceived as “complimenting” women but a woman is likely to be perceived as demonstrating in-group favoritism. Thus, among speakers who express statements consistent with benevolent sexism, we hypothesized that participants would more strongly agree with a male speaker than a female speaker. Hostile sexism, however, is viewed as reflecting prototypical prejudice against women, especially when expressed by a man. Because a woman who expresses negative prejudice toward other women is non-prototypical (and thus not consistent with conceptualizations of sexism), participants would be able to justify any suppressed sexist attitudes by agreeing with a woman’s hostile sexism. Thus, among speakers who express hostile sexism, we hypothesize that participants would more strongly agree with a female speaker than a male speaker.

Study 1

Method

Participants

Two hundred eighty-eight undergraduate students (105 males, 183 females) from a large university participated in this study for course credit. The sample had an age range of 19–59 (M = 19.54), and was 64.6% White. All participants provided informed consent before participating.

Procedure

Participants were recruited for a study that they believed was studying perceptions of individuals voicing opinions on social issues. Participants were randomly assigned to one of four conditions: a male or female speaker expressing hostile or benevolent sexism. Afterwards, participants were debriefed and received course credit.

Materials

The critical opinion included a picture of an adult White target who was either a man or a woman and a brief statement supposedly expressed by the target endorsing benevolent sexist beliefs or hostile sexist beliefs in response to the prompt, “What are your attitudes toward women?” The benevolent sexist statements included, “Women really are the head of the household, and I think a lot of husbands would be lost without their wives.” The hostile sexist statements included, “I admit that I think some women try to use their gender as an excuse to get what they want, and when they don’t get it, they just complain about it.”

Following the passage were 11 Likert scale items (1 = very little, 7 = very much) intended to measure participants’ evaluation of the speaker. Speaker likability was measured using five items (Cronbach’s α = .90) such as, “How friendly is this person?” Speaker prejudice was also measured using five items (α = .84) including items such as “How biased is this person?” Participants then responded to the item, “Using the scale below, please indicate the extent to which you personally agree with the statements this person has made” (emphasis included in survey question) using an 11-point scale (0 = completely disagree, 10 = completely agree).

Results and discussion

Overview of analyses

Initially, the three dependent variables (speaker likability, speaker prejudice, and agreement with speaker) were examined using a 2 (speaker gender: male or female) × 2 (type of sexism: hostile or benevolent) × 2 (participant gender: men or women) multivariate analysis of variance. This analysis revealed a statistically significant main effect of type of sexism, F(3,237) = 13.71, p < .001, η_p² = .15, and a statistically significant interaction between speaker gender and type of sexism, F(3,237) = 25.99, p < .001, η_p² = .25. These analyses are examined at the univariate level below and summarized in Table 1 (ANOVA) and Table 2 (means as a function of speaker gender and type of sexism). Although there was no main effect of participant gender, participant gender did interact with type of sexism.¹ Importantly, there was no three-way interaction indicating that participant gender did not qualify any of the analyses described below.

Table 1. Study 1: F values and probabilities.

	Likability			Prejudice			Agreement
	F(1,239)	p	ηp^2	F(1,239)	p	ηp^2	F(1,239)	p	ηp^2
Type of sexism	40.49	<.001	.15	14.26	<.001	.06	25.37	<.001	.10
Speaker gender	.912	.341	.004	1.26	.262	.005	.586	.445	.002
Sexism by gender	31.88	<.001	.12	71.86	<.001	.23	51.07	<.001	.18

Table 2. Study 1: Means (and standard deviations) for speaker ratings by speaker gender and type of sexism.

	Hostile sexism			Benevolent sexism
	Female	Male	F(1,239)	Female	Male	F(1,239)
Likability	4.70 (1.16)a	3.87 (1.11)b	21.22^***	4.94 (.95)a	5.51 (.83)c	11.43^**
Prejudice	3.49 (1.25)a	4.46 (1.17)b	26.07^***	4.07 (1.06)c	2.64 (1.03)d	47.86^***
Agreement	6.31 (2.71)b	4.08 (2.69)a	30.17^***	6.17 (2.20)b	7.90 (1.87)c	21.15^***

Note: Means in the same row with different subscripts differ at p < .05;

^** p <.01;

^*** p < .001.

Univariate analyses

Speaker likability

There was a statistically significant main effect of type of sexism, but this effect was qualified by the speaker gender by type of sexism interaction. Participants rated the hostile sexist female speaker as statistically significantly more likable than the hostile sexist male speaker. Participants rated the benevolent sexist male speaker, however, as statistically significantly more likable than the benevolent sexist female speaker.

Speaker prejudice

There was a statistically significant main effect of type of sexism, but this effect was qualified by the speaker gender by type of sexism interaction. Participants evaluated the hostile sexist female speaker as statistically significantly less prejudiced than the hostile sexist male speaker. In contrast, participants evaluated the benevolent sexist male speaker as statistically significantly less prejudiced than the benevolent sexist female speaker.

Agreement with speaker

There was a statistically significant main effect of type of sexism, but this effect was qualified by the speaker gender by type of sexism interaction. Participants agreed with the hostile sexist female speaker statistically significantly more than the hostile sexist male speaker. Conversely, participants agreed with the benevolent sexist male speaker statistically significantly more than the benevolent sexist female.

Extending the work of Barreto and Ellemers (2005) and consistent with Inman and Baron’s (1996) prototype approach to prejudice, speakers who expressed beliefs that did not fit with the prototype of prejudice (i.e., a man endorsing benevolent sexism or a woman endorsing hostile sexism) were rated as more likable, less prejudiced and elicited more agreement compared to a context that more closely matched prototypes of prejudice (i.e., a man endorsing hostile sexism or a woman endorsing benevolent sexism). To test our argument that the reduction in perceived prejudice in non-prototypical contexts may serve as justification for agreeing with such (prejudiced) statements (cf. Crandall & Eshleman, 2003), we conducted a moderated mediation analysis.

We used PROCESS (Model 7; Hayes, 2013) for SPSS, with speaker gender (coded female speaker = 0, male speaker = 1) predicting agreement, mediated by perceived speaker prejudice, and with type of sexist comment moderating the link between speaker gender and speaker prejudice. This model is shown with final unstandardized path weights in Figure 1. The effect of speaker gender on perceptions of speaker prejudice was statistically significant such that male speakers who expressed benevolent sexism were rated as less prejudiced than female speakers, a_bs = −1.43, t = −7.07, p < .001; but female speakers who expressed hostile sexism were rated as less prejudiced than male speakers, a_hs = .97, t = 4.74, p < .001. The effect of speaker prejudice on agreement was statistically significant, b = −1.42, t = −14.87, p < .001. The direct effect of speaker gender on agreement was statistically significant and remained statistically significant when accounting for the indirect effect of speaker prejudice (moderated by speaker gender), c = −.58, t = −2.30, p = .02. Because the indirect effect of speaker gender on agreement, mediated by prejudice, is potentially moderated by the type of sexism, there is no one coefficient of the indirect effect. Instead, the indirect effect must be evaluated at each level of the moderator. When the sexism expressed was benevolent, the indirect effect was positive, ab_bs = 2.05, 95% confidence interval (CI) = 1.47–2.71. When the sexism is hostile, the indirect effect was negative, ab_hs = −1.39, 95% CI = −2.05 to −.75.

Figure 1. Unstandardized path weights of the moderated mediation model (Study 1).

*p < .05; ***p < .001.

Thus, a female speaker expressing benevolent sexism decreased agreement (via increased perceptions of her as prejudiced), but a male speaker making the same benevolent comments increased agreement (via decreased perceptions of him as prejudiced). Conversely, a male speaker expressing hostile sexism decreased agreement (via increased perceptions of him as prejudiced), but a female speaker making the same hostile comments increased agreement (via decreased perceptions of her as prejudiced). Because the 95% CI for the index of moderated mediation (Hayes, 2015) does not include zero (2.24–4.11), we can conclude that the two indirect conditional effects calculated earlier were statistically significantly different from one another, and therefore the overall model of moderated mediation was supported.

These findings are consistent with the justification–suppression model of prejudice (Crandall & Eshleman, 2003). Participants who were more likely to agree with a speaker may have done so because they were able to justify their views as non-prejudiced because they evaluated the (non-prototypical) speaker as similarly non-prejudiced. However, our findings are also consistent with persuasion processes related to the credibility of sources who argue against their interests. Specifically, Petty, Fleming, Priester, and Feinstein (2001) suggested that when speakers argue against their group’s interest they are granted a degree of credibility by the audience who receives their messages. In their study, White participants rated a Black author that argued against minority scholarships (a violation of Blacks’ group interest) as more trustworthy and the statements were more persuasive than when the same speaker argued in favor of minority scholarships (to the benefit of Blacks’ group interest). Applying this process to our current research, because hostile sexism is blatantly anti-women in nature, a woman who endorses hostile sexism may be seen as arguing against her own group interest. Likewise, a man who “praises” women via benevolent sexism may be perceived as suggesting women are better than men and therefore arguing against his group as well. Thus, the corresponding increase in participants’ agreement with the statements of the hostile sexist woman and the benevolent sexist man may have been the result of the credibility associated with sources arguing against their group’s interests. Study 2 was designed to examine such a process and compare it to the justification–suppression approach of Study 1.

Study 2

Study 2 had two goals. First, we attempted to replicate the findings of Study 1 but this time in the context of hostile and benevolent sexism toward men. According to Glick and Fiske (1999), these constructs are similar to their corresponding attitudes toward women in that hostility toward men represents blatantly negative attitudes toward men (as controlling, aggressive, and childish), and benevolence toward men represents subjectively favorable attitudes that reinforce traditional gender roles between men and women. Thus, in a conceptual replication of Study 1, we expected that non-prototypical contexts (i.e., a male speaker who endorses anti-male hostility and a female speaker who endorses “pro-male” benevolence) would elicit stronger agreement than the corresponding prototypical contexts (i.e., a male speaker who endorses “pro-male” benevolence and a female speaker who endorses anti-male hostility).

The second goal was to examine more closely the potential mediating processes for these effects. Consistent with Study 1, we sought to test the possible justifying effects of the (reduced) perceived prejudice of the speaker. In addition, we examined the possible mediating effect of a non-prototypical speaker’s group interest violation (i.e., a male speaker who denigrates his in-group and a female speaker who praises the superiority of her out-group are both arguing against their corresponding group’s interests).

Method

Participants

Five hundred eighteen undergraduate students (179 males, 339 females) from several mid-size universities participated in this study for course credit. The sample had an age range of 19–54 (M = 19.97), and was 85.1% White. All participants provided informed consent before participating.

Procedure and materials

The procedure for Study 2 was identical to that of Study 1. The critical opinion included a picture of an adult White target who was either a man or a woman and a brief statement supposedly expressed by the target endorsing benevolent sexist beliefs or hostile sexist beliefs in response to the prompt, “What are your attitudes toward men?” The benevolent sexist statements included, “Men really are the head of the household, and I think a lot of wives would be lost without their husbands.” The hostile sexist statements included, “I think a lot of men can be so egotistical and think they’re God’s gift to the world when really they’re just idiots.”

Following the passage were 11 Likert scale items (0 = not at all, 10 = very much) intended to measure participants’ evaluation of the speaker. Speaker likability was measured using four items (α = .86) including, “How friendly is this person?” Speaker prejudice was measured using four items (α = .78) including “How biased is this person?” Speaker group interest violation was measured with two items (α = .57) such as “How much is this person arguing against his/her group’s interests?” Finally, participants responded to the item, “How much do you personally agree with the statements the person has made?” (emphasis included in survey question) using an 11-point scale (−5 = completely disagree, 5 = agree strongly).

Results and discussion

Overview of analyses

The four dependent variables (speaker likability, speaker prejudice, group interest violation, and agreement with speaker) were examined using a 2 (speaker gender: male or female) × 2 (type of sexism: hostile or benevolent) × 2 (participant gender: men or women) multivariate analysis of variance. Consistent with Study 1, there was a statistically significant interaction between type of sexism and speaker gender, F(4,504) = 138.00, p < .001, η_p² = .523. Although there were statistically significant two-way interactions between participant gender and both type of sexism and speaker gender,² there was no three-way interaction, indicating that participant gender does not qualify the analyses reported below at the univariate level and summarized in Table 3 (ANOVA) and Table 4 (means as a function of speaker gender and type of sexism).

Table 3. Study 2: F values and probabilities.

	Likability			Prejudice			Group interest violation			Agreement
	F(1,514)	p	ηp^2	F(1,514)	p	ηp^2	F(1,514)	p	ηp^2	F(1,514)	p	ηp^2
Type of sexism	22.94	<.001	.04	8.77	.003	.02	5.28	.022	.01	5.73	.017	.01
Speaker gender	40.31	<.001	.07	1.52	.218	.003	6.33	.012	.012	.007	.933	.000
Sexism by gender	50.79	<.001	.09	83.87	<.001	.14	382.61	<.001	.43	48.12	<.001	.09

Table 4. Study 2: Means (and standard deviations) for speaker ratings by speaker gender and type of sexism.

	Hostile sexism			Benevolent sexism
	Female	Male	F(1,507)	Female	Male	F(1,507)
Likability	5.00 (1.90)a	5.12 (1.70)b	3.26	6.82 (1.64)c	4.76 (1.69)b	82.63^***
Prejudice	6.33 (1.80)a	4.95 (1.75)b	31.21^***	5.47 (2.02)b	7.18 (1.72)c	54.33^***
Group interest violation	3.37 (1.74)a	6.35 (1.87)b	144.37^***	6.56 (2.32)c	2.33 (1.78)d	245.19^***
Agreement	−.07 (2.84)a	1.33 (2.37)b	23.34^***	−.14 (3.12)a	−1.92 (2.73)c	24.79^***

Note: Means in the same row with different subscripts differ at p < .05;

^*** p < .001.

Univariate analyses

Speaker likability

There was a statistically significant main effect of type of sexism on ratings of speaker likability, and a statistically significant main effect of speaker sex, but these effects were qualified by the interaction between speaker sex and type of sexism on participant ratings of speaker likability. Male speakers espousing hostile sexism were seen equally likable as female speakers espousing hostile sexism, while male speakers espousing benevolent sexism were seen as less likable than female speakers endorsing benevolent sexism.

Speaker prejudice

There was a main effect of type of sexism on perceived speaker prejudice, but this effect was qualified by the speaker gender by type of sexism interaction. Male speakers espousing hostile sexism were seen as less prejudiced than female speakers espousing hostile sexism. In contrast, male speakers espousing benevolent sexism were seen as more prejudiced than female speakers espousing benevolent sexism.

Speaker group interest violation

There was a statistically significant main effect of type of sexism on ratings of how much speakers were seen as speaking against their own group’s interests, and a statistically significant main effect of speaker gender, but these effects were qualified by the interaction between the type of sexist comment made and speaker gender. Male speakers espousing hostile sexism were seen as speaking against their own group more than female speakers espousing hostile sexism; male speakers espousing benevolent sexism, however, were seen as speaking against their own group less than female speakers endorsing benevolent sexism.

Agreement with speaker

There was a statistically significant main effect of type of sexist comment made on participants’ agreement with the speaker’s comment, but this effect was qualified by the interaction between sexist comment made and the gender of the speaker. Participants were more likely to agree with a male speaker expressing hostile sexist comments against men than a female speaker expressing the same hostile sexist comments. In contrast, participants were less likely to agree with a male speaker expressing benevolent sexist comments in favor of men than a female speaker expressing the same benevolent sexist comments.

Similar to Study 1, we found that when speakers went against prototypical combinations of prejudice (i.e., a male speaker endorsing hostile anti-male sexism, a female speaker endorsing benevolent anti-male sexism), they were rated as more likable, less prejudiced and were agreed with more strongly than those who conformed to prototypes (i.e., a female speaker endorsing hostile anti-male sexism, a male speaker endorsing benevolent anti-male sexism). The decrease in perceptions of prejudice for non-prototypical patterns is consistent with our application of the justification–suppression model to explain why people may have been more influenced, but agreement may also be driven by the non-prototypical speaker’s violation of their corresponding group’s interests (cf. Petty et al., 2001). To compare empirically these two theoretical approaches, we used structural equation modeling to test how well the two corresponding constructs (speaker prejudice or group interest violation) mediate the interaction between speaker gender and type of sexism predicting agreement with the speaker.

Structural equation model

Using EQS 6.1, we created a structural equation model (see Figure 2) such that speaker gender (coded female = 0, male = 1) predicted both group interest violation (path a) and speaker prejudice (path b); in turn, group interest violation and speaker prejudice predicted agreement with speaker (paths c and d, respectively). A correlation table is shown in Table 5. To test the moderating effect of type of sexism, we conducted a two-group analysis in EQS by splitting the data file into hostile sexism and benevolent sexism subsets and applying the same model to both data-sets simultaneously and testing for invariance by first testing a model that assumes statistical equivalence between the two groups for all paths, and then by testing a model that releases the constraints on our hypothesized paths. To the extent the second model improves fit and results in overall good fit based on pre-established conventions, statistical moderation should be demonstrated.

Figure 2. Standardized path weights of the structural equation model (Study 2).

***p < .001.

Table 5. Study 2: Correlations of items for SEM analysis, split by type of sexism.

	Speaker gender	Defend	Against group	Prejudiced	Biased	Sexist	Fair minded	Do you agree
Speaker gender	−	−.74^***	−.51^***	.42^***	.26^***	.39^***	.21^***	−.29^***
Defend	.55^***	−	.53^***	−.25^***	−.13^*	−.28^***	−.04	.13^*
Against group	.45^***	.23^***	−	.006	.004	−.002	.008	−.07
Prejudiced	−.24^***	−.21^***	−.09	−	.51^***	.73^***	.51^***	−.64^***
Biased	−.23^***	−.23^***	−.10	.59^***	−	.47^***	.25^***	−.34^***
Sexist	−.40^***	−.32^***	−.07	.44^***	.46^***	–	.49^***	−.59^***
Fair minded	−.22^***	−.06	.005	.37^***	.39^***	.38^***	−	−.55^***
Do you agree	.26^***	.20^***	.08	−.39^***	−.35^***	−.46^***	−.39^***	−

Note: Correlations for hostile sexism are below the diagonal.

^* p < .05;

^*** p < .001.

The analysis excluded a total of six participants who were deemed either multivariate outliers or who had missing data. All path weights reported are standardized. The first model analyzed was the fully constrained model in which every path weight within the hostile sexism data was set to be statistically equivalent to the corresponding path weight within the benevolent sexism data. As expected, the overall fit of this model was poor; Satorra–Bentler (S–B) χ² (41) = 463.62, p < .001; comparative fit index (CFI) = .69; root mean square error of approximation (RMSEA) = .14 (90% CI .031–.154), suggesting that a model that assumed equivalence between the hostile and benevolent sexism conditions was not a good fit for the data.

In the second model, we released the constraints on the paths between speaker gender and perceptions of speaker bias and group interest violation (paths a and b) to test for moderation of type of sexism. This model displayed good fit; S–B χ² (39) = 99.77, p < .001; CFI = .955; RMSEA = .055 (90% CI .042–.069), and was a statistically significant better fit of the data than the first model; Δ χ² (2) = 363.85, p < .001. Thus, releasing the equality constraint allowed the differences in the direction and magnitude of path a (the effect of speaker gender on group interest violation) and path b (the effect of speaker gender on speaker prejudice) as a function of type of sexist comments to emerge, indicating moderation. In examination of the potential mediating role of group interest violation on the moderated relationship between speaker gender and type of sexism on participants’ agreement with the sexist comments, the effect of group interest violation on agreement was not statistically significant (c = −.05; Z = −.867, p = .38). These findings suggest that the interaction between speaker gender and type of sexism in predicting agreement is not mediated by perceptions of the speaker’s group interest violation. Consistent with the findings of Study 1, the effect of speaker prejudice on agreement with speaker was statistically significant (d = −.75; Z = −13.68, p < .001) such that the less prejudiced participants perceived the speaker to be, the more likely they were to agree with that speaker. Thus, providing conceptual replication of similar processes from Study 1, and consistent with the justification–suppression model, participants were more likely to agree with sexist statements to the extent that they evaluated those speakers as less prejudiced (by virtue of the speaker’s gender and the type of sexism the speaker endorsed).

General discussion

Across two studies, participants’ agreement with sexist statements depended on the source of those statements (i.e., a man or a woman) and type of sexism (i.e. hostile or benevolent). Speakers who expressed benevolent sexism praising their out-group (i.e. men who praised women, women who praised men) were perceived as more likable, less prejudiced and elicited greater agreement than when the same benevolence was expressed by an in-group member (i.e. women who praised women, men who praised men). Speakers who expressed hostile sexism that derogated their in-group (i.e. women who denigrated fellow women, men who denigrated fellow men) were perceived as more likable, less prejudiced and consequently elicited greater agreement from participants than when the same hostility was expressed by an out-group member (i.e. men denigrating women, women denigrating men). These findings are consistent with Inman and Baron’s (1996) prototype approach to prejudice which suggests that perceivers are more likely to identify prejudice when it conforms to mental representation of “typical” forms of prejudice. Our work also complements and extends the work of Barreto and Ellemers (2005) by examining a novel intergroup context (sexism toward men) and measuring an important outcome variable (agreement with sexism). Furthermore, in our attempt to identify mediating processes for these effects, our data did not support a group interest violation approach for the persuasiveness of sexist speakers. Although denigrating one’s in-group by claiming fellow members are incompetent was perceived to be more of a group interest violation than praising the in-group, the “interest” involved reflects an intangible attitude rather than a specific proposal to take tangible resources away (cf. Petty et al., 2001). As such, our speakers’ group interest violation did not influence participants’ agreement with their comments.

Instead, our findings are consistent with the justification–suppression model of prejudice (Crandall & Eshleman, 2003), in that the reduction in perceptions of a speaker’s prejudice (e.g., hostile anti-women sexism was perceived as less prejudiced when expressed by a woman than from a man) may have, in turn, justified the expression of participants’ underlying hostile sexist attitudes by agreeing with the speaker’s sexist comments. However, it is important to note that despite the consistency of our findings across the two studies, we did not measure or manipulate the justification process directly. Future research that examines other social contexts that promote the justification of prejudice may demonstrate similar “release” of expressions of prejudice. For example, explicit disclaimers of prejudice (e.g., “I’m not sexist, but …”) may represent attempts to justify underlying prejudice.

Nonetheless, our findings are also consistent with the larger justification of intergroup differences in status and power between men and women. According to system justification theory, people are motivated to maintain the status quo which is more likely to be perceived as fair when men and women are perceived to have mixed and balanced traits and outcomes (Jost & Kay, 2005). For example, hostile and benevolent sexism offer an arrangement in which men and women appear complementarily equal: both men and women have their unique strengths (and weaknesses)—despite important differences in the status and social value associated with such competencies that ultimately favor men (Glick & Fiske, 2001). The perceived fairness of such an arrangement may have been enhanced in our study when a speaker offered benevolent sexist praise toward an out-group. That is, by praising the stereotypic virtues of the out-group, the perceived complementarity between genders, and thus the perceived fairness of the status quo, may have been strengthened. Conversely, derogating the out-group with hostile sexism may be perceived as a claim of unfairness in gender relations which challenges the status quo, and such speakers were rated as less likable in our study (cf. Kaiser & Miller, 2001). However, endorsing negative prejudice against one’s in-group may reaffirm the balance in strengths and weaknesses between men and women that may have likewise bolstered participants’ beliefs that the status quo is fair.

Notes

1. The interaction between participant gender and type of sexism was statistically significant at the multivariate level (F(3,237) = 4.50, p = .004, η_p² = .05) and at the univariate level for agreement with speaker (F(1,239) = 12.07, p = .001, η_p² = .05). Means for benevolent sexist speaker were higher (i.e., more agreement) than for the hostile sexist speaker, and this difference was more pronounced among female participants than male participants.

2. The interaction between participant gender and type of sexism was statistically significant at the multivariate level (F(4,504) = 16.67, p < .001, η_p² = .12) and at the univariate level for prejudice (F(1,507) = 16.21, p < .001, η_p² = .03), speaker likability (F(1,507) = 29.40, p < .001, η_p² = .06), and agreement with speaker (F(1,507) = 61.68, p < .001, η_p² = .11). Male participants felt that speakers endorsing benevolent sexism less prejudiced and agreed with them more, than speakers endorsing hostile sexism, and female participants showed the opposite pattern. The interaction between participant gender and speaker gender was statistically significant at the multivariate level (F(4,504) = 5.16, p < .001, η_p² = .22) and at the univariate level for against group violation (F(1,507) = 10.37, p = .001, η_p² = .02). Female participants rated the female speakers as higher in group interest violation than the male speakers, while male participants rated both male speakers equally.