As Brazil prepares for its presidential run-off this weekend, Anna Petherick, Associate Professor of Public Policy at the Blavatnik School of Government, and researcher João Areal of the University of Amsterdam, draw on new survey data to explore what is shaping voters' decisions.
This coming Sunday, Brazil will go to the polls to decide who among President Luiz Inácio Lula da Silva and Senator Flávio Bolsonaro, a son of imprisoned ex-president Jair, will lead the country for the next four years.
Our latest model suggests that it is close, with Lula relatively unlikely to win, even though the odds had started to look a little rosier for him earlier this week. These predictions have tracked well with leading opinion polls .
Our analysis draws on surveys of Brazilian voters, exploring their political views, identities and attitudes towards democracy. The findings point to some less obvious predictors of Sunday's outcome, and broad support for liberal democracy is not among them. Affective polarisation - emotions felt towards political allies and opponents - and views on the legitimacy of Jair Bolsonaro's imprisonment matter much more.
This is hardly what President Lula would wish for. Lula, who is seeking a fourth non-consecutive term, claims that nothing less is at stake on Sunday than the struggle to protect Brazilian democracy itself and defend the country's sovereignty against foreign interference, particularly from the United States.
Yet the stakes had already shifted during the left's first-round disappointments. In elections on 4 October, the lower house and two-thirds of the senate were elected, and more gubernatorial positions than ever before this century were decided without the need for a run-off, with many going to the political right. This means that even if he beats the odds, Lula would face a huge challenge in assembling a workable coalition to overcome political gridlock, implement his agenda and govern effectively, while a victorious Flávio would enter office with a fairly like-minded congress
What our election modelling reveals
The way we model the election outcomes is tricky, but intuitive. We have mainly drawn upon responses coming in daily between the two rounds of voting, comparing them with earlier responses from approximately 2,500 of the same respondents surveyed in September. This allows us to assess and calibrate how stable voting intentions are across different groups.
Similarly to most of the polls, our surveys predicted who would vote for Lula relatively accurately in the first round, but were less successful at pinpointing the split between those who voted for Flávio versus other candidates on the right. Our data also reveals an 11th hour strategic shift within the right, with three out of every four voters who had previously said they would vote for another candidate in the first round switching to Flávio rather than Lula.
Another curiosity is that even though voting is compulsory in Brazil, 21% of registered voters stayed home on 4 October and only 6% of our respondents said they did not vote. This disparity is common in surveys; those people less inclined to vote are often less inclined to fill out surveys, meaning their behaviours can be missed. And where voting is legally required, as is the case for literate 18- to 70-year-olds in Brazil, people are understandably reluctant to admit they stayed home.
We accounted for these differences, carefully weighted our second-round prediction model with regional voting results and then tested our model's accuracy against state-level voting results from the first round.
The first thing to note is that essentially nobody changes their voting intention between the two main candidates, who captured 92.2% of valid votes in the first round. When we examine what drove voters into either of the two main crystallised camps, the usual suspect emerges as a prominent predictor: how people evaluated government performance during Jair Bolsonaro's term in office, and Lula's most recent term. This is also a strong predictor of the group we call 'in-play voters' - those whose second-round choices are less firmly established.
If, as our data suggests, 92.2% of valid votes are effectively decided, that leaves less than 8% of first-round valid votes up for grabs. Also available are the 4.8% of votes cast that were invalid, either because voters spoiled their ballot or left it blank in the first round. Those who did not turn out before are harder to model.
Identity, not democracy
Our data shows that what matters most after government evaluations is political identity, affective polarisation (measured by liking assessments and emotions felt towards your political ingroup and outgroup) and one key issue: how people perceive the legitimacy of the 27-year prison term that was handed down to Jair Bolsonaro.
He received this sentence in September 2025, for plotting a coup to overthrow the 2022 elections, alongside four other crimes , including the destruction of public property that occurred when his supporters stormed government buildings on 8 January 2023. Flávio has vowed to get his father out of jail.
We found that young voters, aged 18 to 29, are more likely to perceive Jair's imprisonment to be fair, and that this variable is a weaker predictor of voting intentions for them than for older age cohorts.
But that does not mean that young people care more about liberal democracy in general. In fact, attitudes to democracy are barely better than a coin toss at predicting the election results overall. Lula's first-round supporters show remarkably similar support for five out of seven measured aspects of liberal democracy compared to Flávio supporters.
The two aspects of democracy that reveal some separation are whether government can ignore the courts (which tracks with views on Jair's imprisonment, decided by the Supreme Court) and questioning universal suffrage - perhaps because of a longstanding right-wing narrative suggesting that votes from recipients of Bolsa Família, particularly people in the northeast (the political left's heartland) are less legitimate, as well as some recent tropes questioning women's right to vote in favour of ' family voting '. So Lula going on about democracy in the abstract when he steps up for tonight's televised debate is unlikely to tug at the heartstrings of in-play voters - or even his established supporters, for that matter.
For non-Brazilians around the world, these elections will shape Brazil's place in the international order and will determine the trajectory of deforestation in the Amazon. A Lula victory would maintain downward pressure on deforestation rates, buck the trend of a conservative tide sweeping across Latin America, and would draw Brazil towards greater collaboration with middle powers. Flávio, conversely, would favour agribusiness and align Brazil with the White House.
