Dado Bruto = “raw data” in Portuguese. Independent data and news about Brazil
Elections 2026 · Runoff · Polls
Brazil runoff polls: all 20 pollsters underestimated Lula
In each pollster's final survey, Lula had between 47.0% and 50.6% of the valid votes. He got 51.79%. In 16 of 20, the miss was larger than the pollster's own stated margin of error. In the first round, the polls had missed in the other direction.
51 of 51polls released between the two rounds gave Lula less than he got
16 of 20final polls missed by more than their own margin of error
−2.7 ptaverage miss of those 20 polls on Lula's share
3 of 20had Lula ahead in their final poll
Lula, the incumbent, won Brazil's Oct. 25 presidential runoff with 51.79% of the valid votes, against 48.21% for Flávio Bolsonaro. None of the 51 national polls released between the first round and the runoff, from 20 pollsters, reached that number: the closest had him at 51.0%. In each pollster's final survey, Flávio was ahead in 17 of 20.
This is not just about calling the winner. In 16 of the 20 final polls, the gap between Lula's number and the result was larger than the margin of error the pollster itself registered. If the only error were sampling error, which is what the margin measures, you would expect roughly 1 in 20 outside it, and on both sides.
Each pollster's final survey: Lula's share of valid votes and the stated margin of error
miss larger than the marginmiss within the marginlight band: the number ± the stated margin
Source: polls registered with Brazil's Superior Electoral Court (TSE), compiled by Dado Bruto; result: TSE, 100% of polling stations. Valid votes; when a pollster released only total votes, we computed the valid share.
What the margin of error measures, and what it doesn't
The margin of error is arithmetic about sample size. It says that if the poll were repeated many times with samples of the same size, 95 of 100 results would fall within that range. It does not cover errors that don't come from the draw: how the sample is designed, how answers are weighted, who decides at the last minute, who stays home.
When nearly every poll misses in the same direction, and most of them by more than the margin, the error is not behaving like chance. Counting all 51 polls in the period, not just each pollster's last one, 41 fell outside their stated margin. Public data do not show which of those sources of error weighed most.
The 51 polls between the two rounds: Lula's share of valid votes, by release date
Each dot is a poll. Dark line: average of polls released in the previous three days. Average of the 15 released from Oct. 22 on: 49.4%. Source: polls registered with the TSE; result: TSE.
In the first round, the miss went the other way
On Oct. 4, Flávio Bolsonaro got 47.03% of the valid votes and Lula 45.16%. The final first-round polls from 8 pollsters all gave Flávio less than he got, and 5 of 8 put Lula ahead. Three pollsters had Flávio ahead in the first round (Futura, Gerp and Palver); in the runoff, all three put him ahead again.
First round: gap between the final poll and the result, in percentage points
Flávio BolsonaroLula
Bar left of zero: the poll gave the candidate less than he got. Valid votes. Source: polls registered with the TSE (fieldwork Sept. 25 to Oct. 3); result: TSE.
The same pollsters in both rounds (valid votes; Flávio × Lula in the first round, Lula × Flávio in the runoff)
Pollster
First round
Ahead
Runoff
Ahead
Datafolha
42.0 × 45.0
Lula
49.0 × 51.0
Flávio
MDA/CNT
42.1 × 47.8
Lula
49.3 × 50.7
Flávio
PoderData
43.0 × 45.0
Lula
49.0 × 51.0
Flávio
AtlasIntel
44.1 × 47.0
Lula
49.2 × 50.8
Flávio
Futura
44.8 × 42.6
Flávio
49.9 × 50.1
Flávio
Quaest
45.0 × 46.0
Lula
49.0 × 51.0
Flávio
Gerp
46.0 × 44.0
Flávio
47.0 × 53.0
Flávio
Palver
47.0 × 43.0
Flávio
49.5 × 50.5
Flávio
Result: first round, Flávio 47.03% × Lula 45.16%; runoff, Lula 51.79% × Flávio 48.21%. MDA/CNT and CNT/MDA are the same partnership; PoderData and PoderData/Aya were treated as one pollster.
Each pollster's final survey before the runoff
Pollster
Fieldwork
Interviews
Margin
Lula
Flávio
Miss on Lula
Margin
Gerp
Oct. 19–21
2,400
±2.0
47.0
53.0
−4.79
outside
Real Time Big Data
Oct. 19–22
2,000
±2.0
48.0
52.0
−3.79
outside
Alfa
Oct. 15–19
2,700
±1.9
48.5
51.5
−3.29
outside
DataTrends
Oct. 15–17
2,000
±2.2
48.5
51.5
−3.29
outside
França
Oct. 21–23
1,537
±2.5
48.6
51.4
−3.19
outside
Ipespe/Abrapel
Oct. 13–15
1,500
±2.6
48.8
51.2
−2.99
outside
Meio/Ideia
Oct. 15–19
1,500
±2.5
48.9
51.1
−2.89
outside
Datafolha
Oct. 23–24
4,050
±2.0
49.0
51.0
−2.79
outside
PoderData
Oct. 23–24
4,000
±1.6
49.0
51.0
−2.79
outside
Quaest
Oct. 23–24
3,702
±2.0
49.0
51.0
−2.79
outside
AtlasIntel
Oct. 18–23
4,990
±1.0
49.2
50.8
−2.59
outside
Veritá
Oct. 15–19
15,554
±1.0
49.3
50.7
−2.49
outside
CNT/MDA
Oct. 21–23
2,032
±2.2
49.3
50.7
−2.49
outside
Indexa/Broadcast
Oct. 20–21
2,000
±2.2
49.4
50.6
−2.39
outside
Nexus/BTG
Oct. 23–24
2,000
±2.2
49.5
50.5
−2.29
outside
Palver
Oct. 22–24
5,000
±2.2
49.5*
50.5*
−2.29
outside
Futura
Oct. 23–24
2,000
±2.2
49.9
50.1
−1.89
within
Colectta
Oct. 23
3,000
±2.0
50.3
49.7
−1.49
within
Data Ranking
Oct. 14–16
2,100
±2.1
50.4
49.6
−1.39
within
Vox Brasil
Oct. 20–22
2,100
±2.1
50.6
49.4
−1.19
within
Valid votes, in %. Miss = poll − result (51.79%). * valid share computed by us from the total votes released.
Sources and method
Polls: the 51 national runoff polls released from Oct. 8 to 25, 2026, and the final polls of 8 pollsters before the first round, each with its TSE registration number, compiled from public releases for our runoff page. We use valid votes (excluding blank and spoiled ballots), as in the official result. When a pollster released only total votes, we computed the valid share and marked it with an asterisk.
Result: TSE, 100% of polling stations, including votes cast abroad. Margin of error: as stated by the pollster in its registration, in percentage points. It refers to total votes; for valid votes the margin is slightly larger. With a margin 10% larger, 13 of the 20 final polls are still outside it.
We compare polls taken on different dates with a single result. A poll measures intention during its fieldwork; it is not a forecast. This analysis does not assess any pollster's methodology; it shows the distance between what was published and what came out of the ballot boxes.
You may reuse this. The charts and data in this analysis may be copied and republished, even commercially, under the CC BY 4.0 license. Just credit “dadobruto.com” and, online, link to this page or to dadobruto.com. The original numbers follow the rules of each cited source. The text of the analysis is outside the license.
How to cite:Dado Bruto. Brazil runoff polls: all 20 pollsters underestimated Lula. Code DB-AN-0013. Published Oct. 28, 2026. https://dadobruto.com/en/analysis/brazil-2026-runoff-polls-accuracy
Lula received 51.8% of the valid votes and Flávio Bolsonaro, 48.2%, with 99.8% of polling stations counted. Seven charts show where the vote changed from the first…
Between the poorest and the richest fifth of Brazil's electorate, the gap in the Workers' Party runoff vote fell from 41 to 23 points from 2018 to 2026. Census profile…
