Did Brazilians Vote for Jair Bolsonaro Because They Share his Most Controversial Views?

The similarity of Jair Bolsonaro’s and Donald Trump’s divisive views on a variety of controversial issues has led many critics to argue that Brazilians elected a ‘Tropical Trump’ in 2018. Research on Trump’s election shows that authoritarian, racist, and sexist voters were essential to his narrow victory; however, it is an open question whether Trump’s pathway to power is the norm or the exception among right-wing nationalists. Even though candidates espousing controversial ideas about democracy and prejudice are gaining much electoral support, we know little about the extent to which their voters hold similar views. This study confirms that many Brazilians share Bolsonaro’s ambivalence about democracy as well as his attitudes denigrating women and sexual minorities; however, the degree of congruence between his supporters’ and his own views on these topics played a minor role at most in shaping voter choice. As in previous elections, ideology and partisanship – specifically, attitudes about Brazil’s Workers’ Party – largely explain whether a voter supported him. This finding largely holds across gender and racial boundaries, although white Brazilians appear to have been modestly more inclined than Afro-Brazilians to vote for Bolsonaro if they shared his divisive views.


Do voters supporting right-wing nationalists share their most controversial views?
It is not yet clear whether Bolsonaro's most divisive views helped or hurt him vis-à-vis the Brazilian electorate as a whole. In the U.S., political scientists have examined closely the central role that Trump's appeals to authoritarianism and prejudice played in his winning electoral coalition. Trump entered the general election with the support of the demographic and religious groups that typically back Republican nominees, but it was authoritarian, racist, and sexist voters who tipped a very close election in his favor (e.g., MAcWILLIAMS, 2016;SCHAFFNER, MAcWILLIAMS, and NTETA, 2018;SETZLER and YANUS, 2018;VALENTINO, WAYNE, and OCENO, 2018). In his Republican party primary contests, after controlling for other factors, highly authoritarian voters were five times as likely as non-authoritarian voters to back him, and experts have demonstrated that authoritarian voters were essential to Trump's nomination (MAcWILLIAMS, 2016).
Racist and sexist voters were even more important than authoritarians in Since Trump took office, several unconventional, right-wing candidates for their nations' highest offices, including Bolsonaro, have been compared to Trump due to their ambivalence toward democracy and appeals to voters' outgroup prejudices and nationalism (ROODUIJN, BRUG, and LANGE, 2016 The item was recoded into four levels that distinguish respondents who disagree that democracy is best (i.e., were in the top bottom of the scale before recoding) from those who agree to varying degrees that it is the best system. To explore individuals' openness to military intervention in society, I combined two items that asked respondents whether a military coup would (or would not) be justified in a period of high crime (half the sample) or corruption (the other half).
The survey also includes a standard measure of sexism, asking respondents how much they disagree/agree that 'men, generally speaking, make better political leaders than women'. Respondents were coded into three levels of sexism, depending on whether they strongly disagreed, disagreed, or agreed with the statement. And finally, the measure of homophobia is a question asking respondents how much they approve/disapprove (self-placement on a 10-point scale) with the liberty of 'homosexuals' to run for public office. These variables were all rescaled to ______________________________________________________________________________________________ 2 Because of its high-quality, face-to-face administration, AmericasBarometer surveys frequently are used to analyze voters' attitudes and vote choice, and the survey has been used previously to examine the extent to which right-wing populists in Latin America are disproportionately supported by authoritarian voters (COHEN and SMITH, 2016). As noted by one of the anonymous reviewers, ideally the survey would have been administered as close to the election as possible, with vote-choice data being collected immediately afterward. Had this been the case, there would be less chance that some respondents who voted for/against Bolsonaro may have shifted their views after the election to better match/oppose his opinions. However, it is unlikely that this was a widespread issue. Bolsonaro had been in office only around a month when most of the AmericasBarometer survey interviews were conducted, and his most controversial views were so well known in advance of the election that they had sparked the national #EleNão movement. Moreover, my results suggest that the correlation between voting for/against Bolsonaro and sharing/rejecting his controversial views is weak at best.  Bolsonaro's controversial views all appear related to his support. Notably, in the absence of controls that separately measure the influence of each variable, the most sexist of Brazilians were statistically no more likely to vote for Bolsonaro than the least sexist. This also is the case among voters who fully diverged over the question of whether democracy is the best form of government.
Isolating the effect of each factor to better understand its influence on vote choice requires multivariate analysis. The first logistic regression model summarized in Table 01 predicts the likelihood of voting for Bolsonaro as each independent variable is increased across its full range (i.e., going from zero to one),  Notes: Coefficients are unstandardized logistic regression coefficients; standard errors in parentheses. + p<.1, * p<.05, ** p<.01, *** p<.001, two-tailed. Pseudo R 2 (Cox & Snell) measures of fit were estimated separately with standard logistic regression, with pweights and robust clustering options applied; they are approximate. Multicollinearity tests revealed no concerns (the highest VIF score for any variable in any model is 1.31; the highest condition index for any model is 12). The calculations take into account the survey's complex survey design and weights as described in the methodology section.
In the full model, neither sexism nor homophobia is a significant predictor of vote choice, but attitudes toward democracy and openness to coups are.
Nevertheless, the most powerful predictor of voting for Bolsonaro was hostility toward the PT, closely followed by being more conservative or evangelical. Consider the hypothetical case of two individuals who are alike in every way, except that one of them has the sexism, homophobia, militarism, and support for democracy scores typical of a Bolsonaro voter while the other individual has the average score for nonsupporters on each measure. The full model estimates that the probability that each of these voters voted for Bolsonaro differs by just three percentage points (56% versus 53%). In contrast, consider the effect of diverging partisanship on two hypothetical individuals who differ only in their level of 'antipetismo', with the first voter having the level typical of a Bolsonaro supporter and the other the average 'antipetismo' score of his non-supporters. The full model estimates that the first voter would be 21 percentage points more likely to have voted for Bolsonaro (64% versus 43%). In short, partisanship and ideology were much more powerful determinants of vote choice than sharing Bolsonaro's controversial views.
Do these findings hold up if we look at groups that were especially supportive or not of Bolsonaro? Since unstandardized logistic regression coefficients are difficult to interpret without mathematical transformation and additional context, I computed marginal effects for the last four models in Table 01.
In Figure   Similarly, while the popular press stressed the role that evangelical Pentecostals played in Bolsonaro's victory, that source of support, too, was largely restricted to white adherents.  Table 01; they show how much going from the lowest to the highest value of each independent variable changes the probability that a person voted for Bolsonaro. The lines on the bars denote 90% confidence intervals. The estimates take into account the survey's complex survey design and weights.
Revised by Fraser Robinson Submitted on February 12, 2020 Accepted on July 01, 2020