Pollsters point to technical tie in Brazil election run-off

A “technical tie” does not mean the candidates have exactly equal support. It means the reported difference is too small to establish a reliable leader because it falls within the poll’s margin of error. That makes the election effectively too close to call.

PoderData showed Bolsonaro ahead of Lula by 49% to 48%, with a 1.6-percentage-point margin of error. The one-point lead was therefore not statistically decisive. Quaest similarly showed Bolsonaro at 45% and Lula at 44%, also within its margin of error. BTG/Nexus reported 46.5% against 45.9%, another very narrow gap.

These results show why small polling leads should not be treated as predictions of victory. Minor changes in turnout, undecided voters, blank ballots, spoiled ballots, or interview responses could alter the measured lead. The final result could therefore differ from the apparent ranking in the polls.

The latest polling showed Bolsonaro narrowly ahead, but not decisively. PoderData reported 49% for Bolsonaro and 48% for Lula, with a 1.6-percentage-point margin of error. Quaest found 45% and 44%, respectively, within its margin of error. Atlas Intel measured 49.8% against 47.6%.

Other surveys were similarly tight. Real Time Big Data gave Bolsonaro 48% and Lula 45%, with a three-percentage-point margin of error. BTG/Nexus reported 46.5% for Bolsonaro and 45.9% for Lula, well inside its margin of error. When only valid votes were counted, several polls showed Bolsonaro’s lead widening slightly.

The common pattern was a narrow challenger advantage rather than a clear winner. PoderData’s valid-vote projection was 51% to 49%, while Atlas Intel’s was 51.1% to 48.9%. Because these gaps were small relative to polling uncertainty, the runoff remained difficult to call.

The runoff was especially difficult to predict because first-round polls had already missed the shape of the contest. Most pollsters expected Lula to finish first by a narrow margin. Instead, Bolsonaro led the first round with 47.03% of valid votes, while Lula received 45.15%.

That unexpected result changed the campaign’s starting point. Pollsters had underestimated Bolsonaro’s support, then had to adjust their models for a smaller field containing only two candidates. They also had to account for high volatility, meaning voters’ reported choices could shift during the campaign.

Turnout added another challenge. More than 33 million citizens did not vote, producing a record abstention rate of 21.08% of Brazil’s electorate. The final debate could also move voter intentions. Together, the polling error, changing models, uncertain turnout, and possible debate effects made the second round unusually difficult to forecast.

The most important potential decision-makers were the more than 33 million citizens who did not vote in the first round. Their abstention represented 21.08% of Brazil’s electorate, a record figure. Since they have not yet expressed a choice in the election, both campaigns sought to attract them.

This group matters because the first round was extremely close between Bolsonaro and Lula. Bolsonaro received 56,104,503 valid votes, while Lula received 53,879,538. A substantial number of previously absent voters could therefore affect the final balance if they participate in the runoff.

The campaigns had to persuade people who had not voted rather than simply defend their first-round totals. Pollsters also needed to estimate whether these citizens would remain absent or become active voters. That turnout question added uncertainty to already narrow surveys, especially when projected leads were often only one or two percentage points.

If the polls were wrong again, public expectations could be sharply disrupted because nearly all major surveys presented a technical tie. A candidate shown slightly ahead might lose, while a narrow lead could prove meaningless. The result would underline how difficult it is to measure the nation’s mood in a volatile election.

Campaign strategy could also change. The article shows both candidates already targeting more than 33 million people who abstained in the first round. A polling error would make turnout assumptions even more important. Campaigns would have stronger reason to keep pursuing nonvoters and monitor late movement instead of relying on small polling advantages.

The final debate was another possible source of change. The article notes that either candidate’s performance could move the dial. If polls missed such movement, the runoff would demonstrate that late events and participation levels can matter more than a narrow survey lead.

Polling methods differ mainly in how researchers reach voters and collect answers. Telephone interviews can gather responses across many locations quickly. Face-to-face interviews can reach people directly in selected municipalities. In both cases, pollsters use a sample to estimate the wider electorate rather than asking everyone.

PoderData interviewed 4,000 voters by telephone across 865 municipalities. Its survey showed Bolsonaro at 49% and Lula at 48%. Quaest conducted face-to-face interviews with 2,154 people across 124 municipalities. It found Bolsonaro at 45% and Lula at 44%. The different methods produced similarly close results.

Neither method guarantees a precise forecast. Results depend on who is reached, who responds, how voters report their intentions, and how pollsters model turnout and valid votes. The article says Brazilian pollsters adjusted their models after underestimating Bolsonaro, while also accounting for the runoff’s smaller candidate field and high volatility.

Pollsters do not need to question every voter to estimate national preferences. They interview a sample intended to reflect the broader electorate, then calculate the share supporting each candidate. Larger and geographically distributed samples can provide substantial information, but the result remains an estimate rather than a guaranteed count.

For example, PoderData interviewed 4,000 voters across 865 municipalities and reported Bolsonaro at 49% and Lula at 48%. Quaest interviewed 2,154 people across 124 municipalities and found 45% for Bolsonaro and 44% for Lula. Atlas Intel used almost 5,000 respondents and also found a narrow Bolsonaro lead.

The margin of error describes the uncertainty around the reported percentage. PoderData’s was 1.6 percentage points, while Real Time Big Data’s was three points. A one-point lead inside that range is not decisive. It indicates that the candidates may effectively be tied in the wider electorate.