Data

We employ data from a national survey of Brazilian respondents to conduct this assessment. The sample of 2,015 respondents was recruited by Opinion Box between 22 September and 04 October 2022Footnote ¹ and mirrors the Brazilian adult population in terms of gender, age, region, and socioeconomic status.

We employ two sets of questions to compare the salience of football team identity with those of other established collective identities. The first set asks respondents whether they identify with any 1) political party, 2) social movement, 3) religion, and 4) football team. These questions yield four binary variables in which 0 indicates no identification, and 1 indicates identification with the above collective enterprises.

Those who declared identification with any of the four types of collective enterprise were subsequently asked to rate the importance of that political party, social movement, religion, and/or football team to their life. From these questions, we create four binary variables, for which 0 corresponds to “Not at all important” and 1 corresponds to “Somewhat or very important”.

Results

Given that, at this stage, we are concerned with how the salience of football team identity compares with others, we report descriptive statistics of the share of respondents who identify with each of the four types of collective enterprises and their respective importance in respondents’ lives.Footnote ²

As Figure 1 (panel A) indicates, 74.6% of respondents declare identification with a football team. In other words, a substantially larger share of the Brazilian population identifies with a football team than with a political party or social movement. Of the four sources of identity we examine, the share of people who identify with a football team is only lower than those who identify with a religion.

Figure 1. The share of respondents who identify with a football team and consider it important in their life compared to other types of established identities (a) Identification (b) Importance.

Note: The unit of analysis is the respondent (N = 2,015).

Perhaps more critically, football teams also seem central to people’s lives. As shown in Figure 1 (panel B), 48.6% of respondents consider their football team important; this may seem low when compared to the 84.7% of those who deem their religion important, but it is considerably higher than the share of respondents who consider their social movements (34.5% or political parties (31.6%) important.

These analyses suggest that identification with a football team – largely non-value-oriented organizationsFootnote ³ – may be a relevant (and overlooked) source of collective identity. Next, we examine whether this identity can also be leveraged for changing expressed attitudes.

Experimental setting

We test our expectations using data from an original online survey experiment conducted in collaboration with Bahia Football Club (Esporte Clube Bahia, ECB).Footnote ⁶ Among the social initiatives, the Club has promoted since 2018 is a campaign against sexism in football – an issue that remains prevalent in Brazil and beyond (Gentile, Boca and Giammusso, Reference Gentile, Boca and Giammusso.2018). Our survey aimed to evaluate the impact of this campaign.Footnote ⁷ See Appendix Q for information about the correspondence between the manuscript and the pre-registration document.

Based in the northeastern state of Bahia, ECB competes in Brazil’s top football league and has a supporter base of approximately 2 million fans (Murito and Zarko, Reference Murito and Zarko.2023). As shown in Appendix A, ECB fans, like those of other Brazilian football teams, stem from across the left-right ideological spectrum and are plausibly similarly responsive to the campaign.

Fielded online from November 14 to 30, 2020, we recruited participants via the Club’s mailing list of about 40,000 members and its official Twitter profile, which had 1.4 million followers. Our sample includes 2,143 valid responses. In line with the general profile of Brazilian football fans (Appendix J), the sample is 90.5% male. Appendix D includes further information on sample characteristics.

Our main objective was to determine whether identifying with the campaign’s sponsor (the Club) made individuals more receptive to its message, thereby testing if football team identity can be leveraged to change expressed socio-political attitudes. To this end, the experimental design varied the campaign’s sponsor while keeping its content constant. Specifically, participants were randomly assigned to initiatives from either a non-membership-based organization, the São Paulo Football Federation, our control group, or their own club, the ECB, our treatment group.Footnote ⁸

To ensure realism and ethics, we based our campaign primes on real initiatives. To design comparable primes, we highlighted common aspects of these real campaigns. To increase attention, primes also included campaign images. Figure 2 depicts the randomization set-up and the text and images employed. As balance tests reported in Appendix E indicate, there is no evidence of randomization failure.

Figure 2. Randomization set-up condition does not produce substantively meaningful effects in any of our outcome variables.

Outcome variables

As specified in our pre-analysis plan, our main goal is to examine whether the Club’s campaign can effectively reduce fans’ display of misogynistic attitudes. To measure attitudes toward women in football, our primary outcome of interest, we employ a summative index based on respondents’ agreement with six statements about the abilities of women players, coaches, and referees. The scale of agreement with each statement ranges from 1 to 7, where 1 corresponds to “strongly disagree” and 7 to “strongly agree”. Adding the six items, we derive an index that ranges from 6 to 42 and has a mean of 16.592. In all statements, men players/coaches/referees are portrayed as superior to their women counterparts. As such, higher values of the index indicate more sexist attitudes.

Albeit not the main goal of our study, we also conduct exploratory analyses of the campaign’s effect on respondents’ perceptions of the initiative and the Club’s responsiveness to fans – which we interpret as proxies for institutional trust and belonging. We measure the first of these variables with a summative index composed of respondents’ answers to two questions: 1) “How do you evaluate this campaign promoted by the São Paulo Football Federation/the ECB?” 2) “Through this initiative, the São Paulo Football Federation/the ECB made a public commitment against gender discrimination. What do you think about the São Paulo Football Federation/the ECB making a public commitment to this issue?” The variable ranges from 2 to 10 and has a high mean of 9.510.

Finally, we measure perceptions of the Club’s responsiveness with a summative index of respondents’ agreement with three statements, such as “The ECB board is interested in what fans think”. Adding the three items, we derive an index that ranges from 3 to 21 and has a mean of 16.026. Appendix D provides descriptive statistics for all our outcome variables for the whole sample, as well as for each treatment condition.

Our discussion focuses on the heterogeneous treatment effects of the assigned prime, specifically examining how the effects differ between female and male respondents in the treatment and control groups. These effects are captured using ordinary least squares regression models, with and without pretreatment controls. We report average treatment effects in Appendices F, K, and M.

Results

We begin by examining the impact of the Club’s campaign on fans’ sexist attitudes. Our results suggest that football identity can be leveraged to change prejudicial attitudinal expression, albeit not necessarily in the hypothesized direction. As shown in Figure 3, exposure to their Club’s campaign indeed moves the expressed preferences of some respondents. However, contrary to our expectations, the displayed preferences of women primed to their Club’s anti-gender discrimination campaign remain unchanged, while the expressed preferences of men are more sexist than those in the control group (panel A: coefficient = 0.141, p-value = 0.013)Footnote ⁹ . These results are robust to the inclusion of pretreatment controls (panel B).Footnote ¹⁰

Figure 3. The effect of ECB’s campaign on fans’ sexist attitudes (a) DV: Sexist attitudes (index) (b) DV: Sexist attitudes (index) with controls (c) DV: Sexist attitudes (PCA) (d) DV: Sexist attitudes (PCA) with controls.

Note: The unit of analysis is the respondent (N = 1,420). We run OLS models with robust standard errors and report confidence intervals at the 90% and 95% levels. The X-axis is expressed in standard deviations. Across panels (A–D), higher values of our outcome variables indicate attitudes that are more sexist. We interpret our results in terms of their deviation from the mean (i.e., zero). For example, a coefficient of 0.1 means the treatment has increased sexist attitudes by 0.1 of the standard deviation. Appendix I reports tables of results with raw coefficients.

Since the six items that compose our index are likely correlated, we employ a principal component analysis (PCA) to produce a factor resuming the common variation in the dimensions used to derive the original index (Appendix H). We use this as an alternative outcome variable. As reported in panels C and D, our results remain robust, regardless of whether or not we include pretreatment controls. We interpret backlash to their Club’s anti-gender discrimination campaign as men’s effort to protect their space.

Male respondents’ attitudes toward their Club’s social role in other areas further reinforce our interpretation. Figure 4 displays estimates from our analyses of the impact of our treatment on female and male respondents’ levels of support for their Club carrying out social campaigns in six areas. In addition to displaying reduced support for future initiatives to tackle gender inequality (coefficient = −0.114, p-value = 0.047), men primed to their Club’s anti-sexism campaign are also less supportive of their Club’s involvement in other areas, namely racial discrimination (coefficient = −0.145, p-value = 0.012), LGBT rights (coefficient = −0.131, p-value = 0.023), and environmental protection (coefficient = −0.147, p-value = 0.010).

Figure 4. The effect of ECB’s anti-sexism campaign on fans’ support for the Club’s social role (a) DV: Racial Equality (b) DV: Gender Equality (c) DV: Sexual Harassment (d) DV: LGBT-phobia (e) DV: Social Inclusion (f) DV: Environmental Protection.

Note: The unit of analysis is the respondent (N = 1,420). We run OLS models with robust standard errors and report confidence intervals at the 90% and 95% levels. The X-axis is expressed in standard deviations. Across panels (A–F), higher values of our outcome variables indicate greater support for the Club’s social role. We interpret our results in terms of their deviation from the mean (i.e., zero). For example, a coefficient of -0.1 means the treatment has decreased support for the Club’s social role by 0.1 of the standard deviation. Appendix L reports tables of results with raw coefficients.

These analyses corroborate our interpretation that male fans’ reprisal of their Club’s anti-sexism campaign is an effort to maintain the status quo and protect their space within their Club. While anti-sexism campaigns sponsored by football associations (as in the case of our baseline group) would be reasonably unlikely to promote change within any particular club, male fans could perceive the ECB’s efforts to bolster women in football as a credible threat to their status within the Club, such as their experiences in the stadium. These results are also consistent with the wider context of football fan demographics: data from our national sample of Brazilian respondents indicate that women are significantly less likely to have a football team and to consider the sport important (Appendix J); our results suggest that the overwhelmingly male group of Club fans seeks to maintain it this way. Since our survey was conducted in partnership with ECB and disseminated by the Club, male fans may have used the survey to express their discontentment with the Club’s plan to make the stadium more gender inclusive.

This is consistent with the literature on backlash, which considers it a form of “dynamic resistance” employed by dominant groups as a strategy to reverse or delegitimize policy advances when their status is under threat (Mansbridge and Shames, Reference Mansbridge and Shames.2008). As a large scholarship shows, men’s resistance to efforts to challenge gendered hierarchies span various areas, from the enforcement of property rights (Brulé, Reference Brulé2020), to gender quotas (Gatto, Reference Gatto2025; Kim and Kweon, Reference Kim and Kweon.2022), women’s participation in politics and public service (Gottlieb, Reference Gottlieb2016; Krook, Reference Krook2017; Mansell et al., Reference Mansell, Harell, Thomas and Gosselin.2022), and women’s empowerment in society (Coffe et al., Reference Coffe, Marta Fraile, Fortin-Rittberger and Banducci.2023). Our findings are also consistent with studies that reveal football fans’ explicit efforts to protect their teams from outsiders (Nobis et al., Reference Nobis, Gomez-Gonzalez, Nesseler and Dietl.2022) and men’s concerns about football becoming “over-feminized” (Cleland, Pope and Williams, Reference Cleland, Pope and Williams.2020).

The effects of the campaign are not all be negative, however. As shown in Figure 5, while the treatment does not affect men’s attitudes in this dimension, our exploratory analyses suggest women exposed to their club’s campaign better evaluate the anti-sexism initiative (panel B, coefficient = 0.158, p-value = 0.065) and, critically, are more likely to view the Club as responsive to fans’ preferences (panel D, coefficient = 0.466, p-value = 0.006). Unsurprisingly, models with more precision that include controls tend to reach higher levels of statistical significance. These findings align with scholarship suggesting that positive institutional recognition of marginalized groups may foster trust and group belonging (e.g., Allen et al., Reference Allen, Kern, Rozek, McInerney and Slavich.2021).

Figure 5. The effect of ECB’s campaign on fans’ campaign evaluations and perceptions of the Club’s responsiveness (a) DV: Campaign evaluation (index) (b) DV: Campaign evaluation (index) with controls (c) DV: Club’s responsiveness (index) (d) DV: Club’s responsiveness (index) with controls (e) DV: Club’s responsiveness (PCA) (f) DV: Club’s responsiveness (PCA) with controls.

Note: The unit of analysis is the respondent (N = 1,420). We run OLS models with robust standard errors and report confidence intervals at the 90% and 95% levels. The X-axis is expressed in standard deviations. Higher values of our outcome variables indicate attitudes more positive evaluations of the campaign (panels A–B) and higher satisfaction with the Club’s responsiveness (C–F). We interpret our results in terms of their deviation from the mean (i.e., zero). For example, a coefficient of 0.4 means the treatment has increased satisfaction with the Club’s responsiveness by 0.4 of the standard deviation. Appendix N reports tables of results with raw coefficients.