Brazil Election Polls Now Disagree by Nine Points on the Same Race
Brazil · Analysis
Key Facts
- —The race in question Brazil picks a president on 4 October. The two front-runners are Lula and Flávio Bolsonaro.
- —Who they are Lula is the sitting president. Flávio is a senator and the son of the jailed Jair Bolsonaro.
- —What the polls say One firm has Lula ahead by 7.3 points. Another has him behind by 2. Both surveyed the same week.
- —The spread That is a nine-point gap on an identical question, far wider than any stated margin of error.
- —Where it comes from The firms leave between 7 and 18 percent of voters unassigned, which moves every margin.
- —What they agree on Every survey puts the contest close, and disapproval of the government above approval.
Five Brazilian polling firms went into the field in the same week and came back with five different contests. The disagreement is not really about Lula or Flávio Bolsonaro. It is about what each firm does with the voters who will not answer.
Brazil election polls published this week do not describe the same country. Four national surveys went into the field between 9 and 13 September.
They disagree by nine points on one head-to-head question. Every one of them states a margin of error of about two points.
The Same Four Days, Four Different Races
CNT/MDA interviewed 2,002 people face to face and found Lula beating Flávio Bolsonaro by 47.3 to 40 in a run-off. Nexus, working for BTG Pactual by telephone, found 47 to 46.
Indexa, surveying for the Estadão group, found 43 to 42. Quaest, working face to face for Grupo Globo, found Flávio ahead by 42 to 40.
Those four fieldwork windows overlap almost completely. Each firm states a margin of error of about two points, and each pair of numbers sits well outside the others.
Datafolha, in the field slightly earlier, landed in the middle at 46 to 44 for Lula. The sensible reading is that the middle is where the race actually is.
Where the Difference Actually Lives
The gap is not in the votes the firms report. It is in the votes they do not report at all.
Add up the blank ballots, the refusals and the undecided in each run-off table. Nexus leaves 7 percent of voters unassigned, CNT/MDA leaves 12.7, Indexa leaves 15 and Quaest leaves 18.
That is an eleven-point spread in how much of the electorate each firm sets aside. A margin calculated on 82 percent of voters cannot be compared with one calculated on 93 percent.
None of this is a mistake. It is a design choice about how hard to press someone who says they have not made up their mind.
Method Explains Less Than People Assume
The easy story is that telephone polls favour one side and doorstep polls the other. Brazil election polls this month refuse to behave that way.
CNT/MDA and Quaest both interviewed people face to face, and they are 9.3 points apart on the run-off margin. Telephone surveys sit on both sides of the divide.
Daniel Marcelino, who analyses survey data for the legal news service JOTA, put the point plainly earlier this year. No method is universally better, he said, and reliability depends more on execution than on the technique itself.
He rates doorstep interviewing as generally producing the most representative samples. Telephone work is faster and cheaper but misses people who ignore unknown numbers.
The Question Wording Moves Numbers Too
Brazilian pollsters run two kinds of question. A spontaneous question asks who you would vote for; a stimulated one hands you a list of names.
Real Time Big Data published both from the same national sample at the start of September. Spontaneously, Lula led 35 to 25; with a card of names, 38 to 29.
That is a ten-point swing in the size of the field from question design alone. Most September releases do not say which version their run-off figure uses.
A State Poll That Travelled Badly
One number circulating this week does not survive contact with the poll it came from. Real Time Big Data surveyed Rio de Janeiro, Brazil’s third-largest electorate, between 9 and 12 September.
The result was Flávio 40 and Lula 33 in a first round, and 47 to 41 in a run-off. A widely repeated thirteen-point lead appears in a headline rather than in the tables.
Rio matters on its own terms. It holds 12.86 million voters, behind only São Paulo and Minas Gerais.
What the Surveys Do Agree On
Strip out the horse race and a consistent picture appears. Disapproval of the Lula government runs ahead of approval in every September reading.
Nexus put disapproval at 49 against 47 approval. Quaest put it at 50 against 43, and its director Felipe Nunes called that level unusual for an incumbent seeking re-election.
Both leading candidates also carry rejection near 55 percent. A contest between two widely disliked names tends to be decided late and by small margins.
Augusto Cury, a psychiatrist and best-selling author running for Avante, is the only other candidate consistently above 5 percent. He polls between 6 and 8.
How to Read the Next Three Weeks
Every poll in Brazil must be registered with the electoral court before publication, with its sample size, method and margin. That registry is the first place to check any figure.
The useful habit is to read the residual before the headline. A firm that parks 18 percent of voters is telling you how much of the race it cannot yet see.
Brazil election polls agree the contest is close and disagree about which side of a coin toss it sits on. Voting is on 4 October, with a run-off on 25 October if nobody clears half.
More: Brazil coverage, every day from The Rio Times.
Frequently Asked Questions
Why do Brazilian polls disagree so much this month?
Mostly because of the voters they leave out. Between 7 and 18 percent of respondents are recorded as undecided, blank or refusing, depending on the firm, and that residual mechanically widens or narrows every reported margin. Interview method matters less than people assume: two firms using the same doorstep technique are 9.3 points apart.
Who is Flávio Bolsonaro?
He is a senator for Rio de Janeiro and the eldest son of Jair Bolsonaro, the former president now serving a sentence of 27 years and three months over the plot to overturn the 2022 election. Flávio was confirmed as the Liberal Party candidate in July 2026 and registered with the electoral court in August.
Can Lula legally run again?
Yes. Brazil’s constitution bars more than one consecutive re-election, and his 2003 to 2010 terms were not consecutive with the current one. The electoral court approved his registration unanimously on 2 September 2026.
Which poll should a reader trust?
No single one. The defensible approach is to read them as a range, note where each firm sits relative to the middle, and check the residual and the registration number before quoting a figure. Datafolha’s 46 to 44 currently sits closest to the centre of the field.
Sources: CartaCapital on the CNT/MDA survey, Gazeta do Povo on the Indexa survey for Broadcast, Poder360 on the Nexus survey for BTG Pactual, Poder360 on the Quaest survey, CartaCapital on the Real Time Big Data survey in Rio de Janeiro, Gazeta do Povo on the Datafolha survey, ConJur on how Brazil’s electoral courts are ruling on polling methods, Superior Electoral Court, 2026 electoral calendar
This article was produced by The Rio Times’ automated newsroom system. How we use AI · Report an error
Read More from The Rio Times