The comeback the polls didn't see

Lula goes from 2 points behind in the first round to beating Flávio Bolsonaro by 3.6 points. The vote shifted in 97% of municipalities, most in the Southeast and where Flávio was strongest. Polls, betting markets and Placar itself, which gave Lula 18%, all missed together.

Daniel Pessini Sobreira · October 25, 2026, 10 p.m. · updated October 26 with the TSE count of 100% of polling stations

Luiz Inácio Lula da Silva was elected to a fourth term with 64.3 million votes, 51.8% of the valid votes, against 59.8 million (48.2%) for Flávio Bolsonaro. The gap, 4.4 million votes and 3.6 percentage points, is not tight by recent Brazilian standards, and it is wider than in 2022. What makes the result remarkable is how it got there. Three weeks ago, in the first round, Flávio had 47.0% and Lula 45.2%; counting only the votes for the two of them, Flávio was 2 points ahead. On the eve of the vote, the poll average still showed him slightly ahead, the betting market gave Lula a one-in-five chance and this Placar, 18%.

Three weeks, 5.6 points

The margin moved 5.6 points toward Lula between the two rounds. Lula added 10.4 million votes to what he had on October 4; Flávio, 3.7 million. The other first-round candidates had drawn 9.3 million, and turnout rose from 78.9% to 80.8%, bringing in nearly 3 million more voters. The math only works if most of those votes went to Lula. A municipality-level regression suggests something like two-thirds of the other candidates' vote went to him, but this is ecological inference, done with totals by place rather than with voters: it gives an order of magnitude, not a portrait.

Where the votes came from

Lula's net gain, 6.7 million votes in the gap between the two, came from two regions. The Southeast accounted for 43% of it and the Northeast for 38%.

Margin = Lula − Flávio, in percentage points of the votes cast for the two. Lula's gain = net change in (Lula's votes − Flávio's votes) between the rounds. Placar 2026 calculation from TSE data, unit by unit.

São Paulo alone contributed 1.4 million votes to that gap: Flávio won the state, but his lead shrank from 15.2 to 7.7 points. Minas Gerais, which gave Flávio a 5.4-point edge in the first round, ended with Lula 0.9 point ahead, and once again the state went with the national winner. Rio de Janeiro had the largest swing in the country, 8.5 points, and the state capital flipped from Flávio to Lula. The same switch happened in cities on the metropolitan outskirts, such as Osasco, Cotia and São Vicente in São Paulo, Ribeirão das Neves, Santa Luzia and Ibirité in Greater Belo Horizonte, Alvorada and Sapucaia do Sul in Greater Porto Alegre, and Ananindeua in Greater Belém.

In the Northeast, the swing was smaller, 3.9 points, because Lula was already starting from margins above 30 points. There the gain also came from turnout, which rose 3.3 points in Bahia and 3.0 in Pernambuco. The North, which Flávio had won by 4.8 points, ended in a tie, with turnout falling, especially in Amapá, Acre and Roraima.

Vote switching, not just turnout

Lula's margin improved in 97% of the 5,571 municipalities. He flipped 211 cities that had gone to Flávio in the first round, against a single one in the opposite direction, and went from 2,663 to 2,875 municipalities won. The shift was strongest precisely where Flávio was strongest. In municipalities Flávio had won by more than 30 points, the margin moved 7.1 points toward Lula; in strongholds where Lula had more than 30 points, it moved 3.0. It was also larger in mid-size and large cities (between 6 and 7 points in those with more than 50,000 voters) than in small ones (4.3 points in those with fewer than 10,000).

That pattern matters for reading the result. Had the comeback come from mobilizing his own base, it would show up mostly in Lula's strongholds and would track the rise in turnout. That is not what happened: the change in margin barely relates to the change in turnout across municipalities and electoral zones (correlation of 0.11). Lula's victory came mainly from voters who switched sides or who, having voted for another candidate on October 4, chose Lula, including in hostile territory.

What the numbers missed

None of the tools that track the election anticipated this scale. The poll average on the eve had Flávio 1.1 points ahead, Polymarket gave Lula 20.5% on Sunday morning and this Placar's main model, 18%, with a median of Flávio by 2.1 points. The result came in above the upper bound of the model's 95% interval, which was Lula by 2.0. This was not bad luck within the interval; it was an error.

Part of that error is Placar's own. The adjustments the model applies to the polls (each pollster's known deviation and the discount on final waves) shifted the estimate another 1.1 points toward Flávio. The lesson of the first round, in which the polls underestimated Flávio, was read as a persistent bias. The analysis done after October 4 already pointed to another explanation, tactical voting by supporters of smaller candidates migrating to Flávio in the home stretch, and so the model stopped carrying the first-round error over to the runoff. Even so, the direction of the corrections stayed the same, and in the runoff the error switched sides: the polls underestimated Lula by about 4.7 points of margin. The model's revision has to start there: polling errors in Brazil do not have a fixed sign from one election to the next, nor from one round to the next.

Counting night

The night followed an unusual script. A court ruling extended the count's start until 7 p.m., and many of the polling stations already tallied held off on transmitting. At 5:06 p.m. the TSE published 0.25% of polling stations, and the count only really got moving after 7 p.m. From then on, results arrived in batches by state, with the Federal District, Paraná and states in the North and Center-West ahead and São Paulo, Rio and most of the Northeast behind.

1. 7:05 p.m.1.5% counted; Lula ahead in the count, by 4.5 points.
2. 7:29 p.m.Placar's projection reaches its high point, Lula by 6.1, with 5.4% counted.
3. 7:44 p.m.The model calls the election for Lula, with 13.4% of polling stations.
4. 7:45 p.m.Flávio takes the lead in the count.
5. 8:03 p.m.Flávio's largest lead in the count, 1.0 point, with 25% counted; the projection stands at Lula by 4.5.
6. 8:25 p.m.Lula retakes the lead in the count, with 48.4% counted, and never loses it again.
7. 9:10 p.m.99.7% counted: count at Lula by 3.5 and projection at 3.6.
8. 11:26 p.m.Count of 100% of polling stations: Lula by 3.57 points, 4,430,191 votes ahead.

For 40 minutes, anyone watching only the count saw Flávio ahead while the projection pointed to Lula. There was a reason for the difference: the states that came in first were more favorable to Flávio than the country as a whole, and the model compares each place with its own first-round result. But the projection also overshot early on. The arrival in batches by state, which a test run during the night showed to be the hardest scenario for the model, inflated Lula's margin in the first cycles, and the call, with 13% of polling stations, came earlier than prudence would advise. From 30% counted onward, the projection never missed the final margin by more than 0.63 point, and the result fell within the 80% interval in 96% of updates.

What remains

Lula wins clearly, but in a country that remains divided. Flávio won the South and the Center-West comfortably, tied the North and got almost 60 million votes. What decided the election was a slice of voters in the urban Southeast, the metropolitan outskirts and mid-size cities, plus those who had voted for other candidates, who moved as a bloc in three weeks without the polls registering the scale of the shift. Understanding who these voters are and why the polls did not capture them is the task that remains, for the pollsters and for this Placar.

The figures in this editorial come from the TSE count of 100% of polling stations released at 11:26 p.m. on October 25; the first version, published at 10 p.m., used 99.9% of polling stations. Margins and swings by region, state and municipality were calculated by Placar 2026 from the data for each municipality and electoral zone in both rounds. The estimates of the other candidates' vote and of turnout come from data aggregated by place and indicate trends, not the behavior of individual voters.

↓ The results dashboard: map, states and electorate profile · The live count and projection · Home page