The Victory of Jair Bolsonaro According to the Brazilian Electoral Study of 2018*

The 2018 elections surprised most analysts and political scientists in Brazil and abroad. The competitive dynamic that opposed the PT and the PSDB in presidential elections since 1994 has been disrupted, resulting in a novelty for the Brazilian party system. The purpose of this research article is to identify the determinants of votes for Jair Bolsonaro in the two rounds of the 2018 presidential election from data collected by the Brazilian Electoral Study. By means of multivariate analyses of the two rounds of the presidential election, we found that flourishing 'antipetismo' (i.e. anti-Workers' Party sentiment), growing numbers of voters self-identifying as right wing and increased importance of variables linked to voters' political identification all underpinned the victory of Jair Bolsonaro.


he 2018 elections surprised most analysts and political scientists in
Brazil and abroad. The biggest prize, the Presidency of the Republic, was won by a Congressional Deputy with a modest career record: retired army captain Jair Bolsonaro, who headed a coalition comprising two low-profile conservative parties, the Social Liberal Party (PSL) and the Brazilian Labor Renewal Party (PRTB).
The 2018 election may represent a watershed for the Brazilian party system, by having broken the competitive logic that had set the Workers' Party (PT) against the Brazilian Social Democracy Party (PSDB) in presidential disputes since 1994. The purpose of this research article is to present and discuss the determinants of votes for Bolsonaro in 2018 with reference to data collected by the Brazilian Electoral Study (Eseb) and to point out some possible routes for future analyses.
The selection of variables that we analyzed was based on similar studies of the last four Brazilian presidential elections, such as those by Amaral and Ribeiro (2015), Bohn (2007), Carreirão (2007aCarreirão ( , 2007bCarreirão ( , 2002, Nicolau (2014Nicolau ( , 2007, Peixoto and Rennó (2011), Borba (2016, 2011), Samuels and Zucco (2018) and Zucco (2013). The text shows strong growth in antipetismo and the number of voters considered to be right wing, as well as the importance for Bolsonaro's victory of variables linked to the political identification of voters.

The 2018 Eseb and votes for Jair Bolsonaro
The Eseb 2018 was conducted by the Center for Public Opinion Studies (Cesop) of the State University of Campinas (Unicamp). In 2018, the Eseb was nationally sampled and interviews were conducted of 2506 voters in 172 cities across the country, including all state capitals. The survey was held shortly after the elections, between November 10 and 24, with home interviews.
Its margin of error is two percentage points and the confidence index is 95% 1 .
For the analysis in this research article, dependent variables were constructed by including voters who declared that they voted for Jair Bolsonaro ______________________________________________________________________________________________ Ideological Self-Placement). In addition to these two blocks, sociodemographic variables and a contextual variable were included in the analysis.
Retrospective evaluation of government performance, especially in the economic arena, is a classic approach to electoral behavior, as attested by the work of Downs (1999), Key (1966) and Fiorina (1981), among others. As Rennó (2007) points out, votes based on a retrospective evaluation can be multidimensional, that is, they can take into account various aspects of government actions, not just issues related exclusively to the economy. Thus, the very idea of retrospective voting brings us closer to the possibility of vertical accountability, that is, the ability to punish or reward representatives according to the evaluation of their performance.
The relationship between political identification and electoral behavior has a long history in political science. Both authors linked to the Michigan School (CAMPBELL et al., 1960;CONVERSE, 1964) and those of the Rational Choice persuasion (DOWNS, 1999;FIORINA, 1981), in spite of differing assumptions, have pointed out the importance of political identification, especially party identification, in voter decisions.
The contextual variable inserted in the model was Region. The choice of this variable was due to the fact that authors found a positive effect in multivariate analysis between voters in the Northeast region and Dilma Rousseff's candidacy in both the 2010 and 2014 elections (AMARAL and RIBEIRO, 2015;NICOLAU, 2014).

Results and discussion
The analysis of the vote in the first round was made by multinomial logistic regressions. As mentioned, those who voted for Ciro Gomes, Jair Bolsonaro The results show that the six models were statistically significant.
However, some important differences between them should be highlighted. In both the first and second rounds, the political identification models presented more robust results than those that included retrospective evaluation, including explanatory capacity close to the general models. In other words, the impact of these variables was stronger than that of the retrospective evaluation variables to explain votes in the 2018 presidential election 2 . In this article we will discuss the results of the general models only.
Individually, regarding retrospective evaluation, we found consistent effects on three variables in both rounds. Evaluating the Temer Government as great/good and average, compared to those who evaluated it as bad/terrible, increased chances of voting for Bolsonaro rather than Haddad. Evaluating the country's economic situation as equal or better than in the previous 12 months ______________________________________________________________________________________________ 2 The model including only the retrospective evaluation variables had an R2 Nagelkerke of .269 (p <.001) in the first round and of .204 (p <.001) in the second round. The model including the political identification variables had a R2 Nagelkerke of .512 (p <.001) in the first round and of .539 (p <.001) in the second round. Repeated corruption scandals, President Dilma Rousseff's impeachment and an acute economic crisis certainly contributed to this result, indicating that some groups of supporters may be more susceptible to short-term stimuli and may therefore alter their preferences, as Samuels and Zucco argue (2018, p. 102). In short, in 2018, the PT had a significantly smaller stock of votes to count on than in the previous four presidential elections.
The results of the multivariate analyses show that, as in previous elections, party identification remains one of the strongest predictors of votes in Brazil. Identification with the PT, in both rounds, increased the chance of voting for Haddad rather than Bolsonaro by more than nine times, with the reference category being those who did not prefer any party. Similarly, identification with the PSL increased the chance of voting for Bolsonaro by more than four times in the first round and by more than seven in the second (Tables 01 and 02). Rejection of the PT also proved to be a strong predictor of vote. A voter who rejected the PT in both rounds was ten times more likely to vote for Bolsonaro than a voter who did not reject any party. As Ribeiro, Carreirão and Borba (2016) and Samuels and Zucco (2018)    The third novelty regarding political identification variables is linked to the distribution of voters along the left-right spectrum. As can be seen in Graph 01, respondents who positioned themselves on the right jumped from 27% in 2014 to 43% in 2018. Those who positioned themselves on the left went from 08% to 15%. Conversely, those who did not know what left and right were or could not position themselves on the spectrum dropped from 41% to 18%.
Regarding multivariate analysis, it is possible to observe that ideological self-placement was a good predictor of the presidential vote in 2018. Being on the right increased the chances of choosing Bolsonaro over those on the left and center in both rounds. The stronger correlation with ideological self-placement seen in 2018 relative to previous elections may also be linked to ideological polarization.
According to , the more polarized