Polls rarely hit or miss, and the first round between Luiz Inácio Lula da Silva and the current president of Brazil was no exception. The average expected Lula to be 46.7%, and he ended up with 48.4%: just 1.7 points difference, or a few million votes in a country of more than 200 million souls. Now Bolsonaro was ahead of him with 35.3% and he finished with 43.2%; eight points higher. What explains this asymmetric inaccuracy? There are at least three hypotheses, all of which are plausible and not mutually exclusive: last-minute decision, awkward voting, or sampling error.
The election in Brazil is proving to be one of the clearest polarizations of recent times: the ideological and affective intensity of the Lula-Bolsonaro split corresponds to that of Boric-Kast, Biden-Trump, Macron-Le Pen or Arauz-Lasso. It also matched and even surpassed the previous ones, with the exception of the USA, in the excellent visibility of the two candidates that concentrated most of the odds of winning. So it was somewhat surprising that one of the two candidates stayed so close to the 50% threshold while the other missed. The final results confirmed that this was indeed not the case. However, it must be taken into account that the surveys that served as a reference for creating the asymmetric forecast were conducted up to 96 hours before the election. They fixed the idea of a Lula just before winning the first round, thereby encouraging anti-left mobilization and the concentration of votes around the most viable alternative to the former president.
A first indication of this first hypothesis would be the fact that the voting of the polls regarding the final results improves slightly when compared with the voting of the valid overall result, thereby eliminating the ambiguity from the basis of calculation. This is equivalent to assuming they stayed at home. If instead we assume (but we don’t have data to do the specific simulation) that a majority voted in favor of Bolsonaro, the fit would improve even further.
In the same vein, the overestimation of the third candidate (Ciro Gomes, who would actually finish fourth) and also the fourth (who, despite beating Gomes, would finish worse than expected).
But eight points is too many to explain solely by last-minute decisions, so it’s worth considering the additional explanations that would address the existence of an a priori incorrectly measured blind vote.
Subscribe to EL PAÍS to follow all the news and read without limits.
Subscribe to
When about half of the ballots had been counted on election night, political scientist Pedro Magalhães shared on Twitter the cover of a political science classic: The Spiral of Silence, by Elisabeth Noelle Neuman. This theory is based on the premise that a (vague, but significant) portion of the citizenry expresses opinions that are closer to what they perceive to be the consensus around them than to what they actually think. In Brazil, without necessarily having a consensus as such (in fact, the terms “polarization” and “echo chambers” would better define the form of current public debate in the country), it could be argued that the social norm has fallen little more on the side of Lula as Bolsonaro, for at least two reasons: the current president’s low rating and the general climate of victory suggested by the same polls that surrounded the returnee’s campaign. We don’t have the tools today to measure it, but it doesn’t seem far-fetched to think that it played a quantitatively relevant role in those eight extra points for Bolsonaro, perhaps interacting with the last-minute moves outlined above: in the end , it’s very difficult to distinguish, even internally, between a decision we tell ourselves when the moment of truth approaches and one we would always make.
Finally, it cannot be ruled out that this hidden far-right voice was hidden not because of silence but because of invisibility in the samples of some pollsters. This was the origin of Trump’s measurement error in 2016. Certainly, the differences in proximity or distance in Bolsonaro’s estimate between recent polls in Brazil suggest that the methodological differences played a role.
Now all pollsters, with few exceptions, agreed that Lula would be ahead of Bolsonaro, that both would dominate the election over the rest of the candidates and that a second round would not be avoided. It’s tempting, and to some extent improvable, to look at what hasn’t been properly captured. But we must not lose sight of the fact that the polls got it right once more as they painted the broad outlines of what would ultimately happen in the elections.
Follow all international information on Facebook and Twitteror in our weekly newsletter.
Subscribe to continue reading
read limitless