Jayjust passed At 8 a.m. on Sept. 27, five days before the first round of the presidential election, an employee of Datafolha, Brazil’s most prominent polling company, stopped pedestrians at a five-way intersection in São Paulo and asked them who they planned to vote for. He chooses his subjects based on age and gender quotas that reflect voters — an old man in a baseball cap, a young woman with a hole in her nose. He asked them to name their candidates on a chart (circular, to reduce bias) and recorded their answers on a tablet.
Your browser does not support
Over the next few days, 440 Datafolha workers in 332 cities plan to speak to some 6,800 voters. Within hours of the last interview, the results will be announced on evening news programs watched by a third of Brazilian households.
While in-person polls have long been considered the best way to gauge voting intentions in a vast and unequal country, changes in society and opinion polls are testing that belief. The increasing use of mobile phones (now in 95% of households) and the Internet (90%) has brought about new polls and technologies. The pandemic has accelerated these trends. As a result, electoral courts registered 974 polls this election season, twice as many as during the previous presidential campaign in 2018. The percentage of on-site surveys dropped from 70% to 60%.
Brazil’s right-wing populist president Jair Bolsonaro is trailing his rival, left-wing former president Luiz Inácio Lula da Silva, in almost every opinion poll. Inácio Lula da Silva) at least 6 percentage points. Mr Bolsonaro said any outcome other than his re-election would be a “fraud”. He denounced major pollsters like Datafolha and Ipec, which showed Lula’s lead so wide (14 and 17 percentage points respectively) that he could win in the first round.
The broad divide in the polls has sparked debates over methodology that echo those taking place in the United States and elsewhere. The stakes are high. If Lula’s victory is as big as Datafolha suggests, Mr Bolsonaro’s bogus allegations of fraud will convince fewer people.
Both Datafolha, who polls the streets, and Ibope, the scion who goes door-to-door, have correctly predicted every presidential race since 1989. 2018 in-person and telephone polls. But the latter is considered a “subspecies,” said Jota analyst Daniel Marcelino. That’s because they may be underestimating poorer voters who have fewer cell phones.
In-person polls may have the opposite problem. Wealthier Brazilians, who increasingly live in gated communities and work from home, may be underrepresented. Live polls in the right places provide a near-perfect data set, so the belief that weighting is not needed after the data is collected is also called into question by broad estimates of the number of poor voters. While Ipec found that 55% of voters reported household incomes below twice the minimum wage (2,424 reais, or $450 a month), telephone poll Ideia weighted its sample using an official estimate of 39%. Lula, who excels among the poor, is the best performer in face-to-face polls; telephone polls show Mr Bolsonaro faring less badly.
Most companies admit they can better estimate how many people won’t vote. Voting is compulsory in Brazil, but fines are low and abstention rates can be as high as 20%. Felipe Nunes, the boss of Quaest, said that while pollsters knew that people who did not vote tended to be poorer and less educated, which could hurt Lula, they had not yet used the “maybe vote” A Model That Lightens Low-Voting Groups,” a six-year-old firm that conducts face-to-face polls. Most just ask questions in the final week of the campaign to weed out potential abstainers.
There is less consensus on whether Brazil has a problem of “shy voters” — voters who either don’t tell pollsters the truth or don’t talk to them at all. Raphael Nishimura of the University of Michigan thinks Bolsonaro supporters who distrust the polls may be reluctant to participate.
But shy voters may also fear Lula’s supporters: Three people were murdered by supporters of Mr Bolsonaro. “The more polarized society is, the more the impact of human interaction becomes an issue,” said Andre Roman of Atlas Intelligence, a company that conducts online polls. Its latest showed Lula leading by seven percentage points.
Bolsonaro supporters point out that in 2018 Datafolha polls on the eve of the election suggested he would get 40 percent of the vote when he actually got 46 percent. Some experts also believe it is the result of a polling error. Luciana Chong, head of Datafolha, said the snapshot at the time was accurate. “A lot can happen from Saturday to Sunday,” she said. Mr Bolsonaro picked up last-minute votes from undecided Brazilians and supporters of other candidates who voted tactically to defeat Mr Bolsonaro’s leftist rivals.
This year is different. A record proportion of voters said they had made up their minds. Even so, Lula’s campaign is seeking to convince supporters of Ciro Gomes and Simone Tebet, who each have about 5 percent support, to give Lula victory in the first round through tactical voting. It was also a victory for the established pollsters. ■
