Everyone expected a certain result in 2026. It did not happen. I wanted to understand why, and what started as one afternoon of checking turned into three notebooks. Every explanation I had in my head came apart a little once I looked at the numbers.
One warning: I work with municipalities, not people. Everything here says "where", never "who".
Abstention
The first explanation I heard, and the one I half believed, was that Lula's voters stayed home. The series does not help it. First-round abstention went from 20.95% in 2022 to 21.08% in 2026, barely a tenth of a point. The real movement is older: 16.75% in 2006, then a steady climb in every election since. If abstention decided 2026, it would have to be a twenty-year trend that suddenly started choosing sides.
The national rate can hide a lot, so I split the municipalities into ten groups by how much they voted for Lula in 2022. If his voters had stayed home, abstention would rise most in the top groups. It does not. In the most Lula-leaning decile it fell 0.4 point, in the next two it rose less than 0.4, and the largest increases are in the middle (deciles 4 and 5, about 1.2 points). Where it fell the most, by more than 2 points, is in the least Lula-leaning municipalities.
Meanwhile, Lula lost between 3 and 4.4 points of the valid vote in every decile, slightly more where he was already weaker. A loss that shows up with the same size whether abstention rose or fell is hard to blame on abstention.
The raw accounts tell the same story in votes. From 2022 to 2026 Lula lost 3.4 million votes and the right gained 5.0 million, more than his entire loss. Abstention grew by 0.7 million, about what you would expect from an electorate that grew by 2.3 million. Votes moved; they did not disappear. And because this is aggregate data, I cannot say that the same people switched: a mix of new voters, people who changed their minds and different turnout can produce the same totals.
Still, I wanted to give the theory its best shot. Adding up only the municipalities where abstention rose, and ignoring every place where it fell, gives about 1.58 million "extra" abstainers. The margin to overturn was about 2.24 million. Even if every one of them had shown up and voted for Lula, that covers about 71% of the gap. "Every" is doing heroic work here: in any realistic split the number falls fast, as the slider below shows.
The fairer version of the exercise goes the other way. If I apply each municipality's 2022 turnout to its 2026 vote, the margin does not shrink: it grows, from 2.24 to about 2.45 million, because turnout went up precisely in the most Lula-leaning places.
So who abstains? Mostly the oldest. Among voters aged 70 and over, for whom voting is optional, abstention is 58.9%, almost three times the rate of any other group. Together with 16 and 17 year olds, the other optional group, they are about 13% of the electorate and account for roughly a third of all abstentions. Part of the slow twenty-year climb is probably just an aging electorate, which says much more about demography than about 2026.
Mapping the shift
If abstention does not explain it, the next question is where the votes moved. I took the 4,100 municipalities that leaned left in at least 3 of the first 4 presidential elections I analyzed (2002–2014). Together they cast 83% of the 2022 votes, so this is not a regional curiosity: it is most of the country.
Up to 2014, these municipalities split between left (62–71%) and center, with the right practically absent. In 2018 the center collapsed, from 29% to 6%, and the right went from 1% to 47% in a single election. The left also fell, from 70% to 47%. The center and the left each lost about 22 to 23 points, which together roughly add up to what the right gained, so the new right was built on the space of both, not only on the center's. 2022 recovered part of it (55% left, 41% right) and 2026 brought it back to almost a tie, 49% to 46%.
At the municipality level, 39% of them (1,608) have the right ahead in 2026. When I look at the first year each one flipped, 1,238 had already done it in 2018, only 42 first flipped in 2022 and 328 are new in 2026. Many of the 2018 flips went back to the left in 2022 (the share with the right ahead fell from 31% to 22%) and returned to the right now. So 2026 is less a rupture than a 2018 that repeated itself and widened a bit.
The advance was very uneven. In Santa Catarina, Rio de Janeiro, Goiás, Mato Grosso do Sul, São Paulo and Rio Grande do Sul, around 90% of the historically left municipalities now have the right ahead (the 100% in DF and Rondônia covers very few units, so I would not read much into it). In the Northeast it is the opposite: in Maranhão, Paraíba, Bahia, Rio Grande do Norte, Ceará and Pernambuco, less than 3% flipped, and none in Piauí and Sergipe; Alagoas, at 15%, is the exception in the region. Minas Gerais sits in the middle, at 41%. The country did not move in one block: the historic left almost vanished in the South, Southeast and Midwest and held almost intact in the Northeast.
Income, region and the vote
The other easy explanation is income: "the poor vote left, the rich vote right". The history is stranger than that. In 2002, the poorest fifth of municipalities was the least left-leaning (54.5% against 66–69% in the others). In 2006 the order inverted, and by 2014 the poorest gave the left 84% while the richest gave 55%. The gap between them peaked in 2018 at about 40 points.
What happened after that is not that the poor turned right. From 2014 to 2026, the left lost 22 to 25 points in the middle quintiles (Q2 to Q4), 17.5 in the poorest and 15 in the richest. Since 2018 the richest fifth has actually been slightly more left-leaning than the fourth, and in 2026 also than the third. In 2026 every quintile lost between 5 and 7 points, with the poorest still at 66% for the left. And this says where the votes are, not who cast them: I cannot conclude that poor people voted for the right, or that rich people did not.
A good part of what looks like an income effect is region. In 2026, at the same income level, the right gets 42% in Northeastern municipalities of the middle quintile and 60% in Southern ones. In the fourth quintile the gap is 28 points. Since the poorest municipalities are concentrated in the Northeast, any national chart by income is partly a map of the Northeast.
Within each region, the right does grow with income, but it peaks in the fourth quintile and drops in the richest one: 66% to 57% in the Midwest, 60% to 51% in the Southeast, 67% to 62% in the South. My guess is that the richest quintile carries the large cities, but I have not separated that, so it stays a guess.
To separate these things a bit more, I compared municipalities inside the same state and controlled for the variables together.
The factor that tracks the left vote most strongly is the share of Black and mixed-race residents: about 9 to 12 points of left vote per standard deviation, stable in every election since 2006. Evangelical share barely mattered until 2014 (−0.4) and jumped to around −5.5 points from 2018 on, the clearest change of regime in the whole series. Urbanization also became more negative after 2018. Income, once the rest is discounted, weighs between −1 and −5 points, about −2 in the last three elections: real, but much smaller than the debate suggests.
The same municipalities do not vote the same way for every office. In 2022, the gap between the poorest and richest fifth was 24 points for president and practically zero for federal deputy (47.4% against 47.1% for the right). In the poorest municipalities, the right got almost twice as much for deputy as for president. In 2026 the pattern repeats with a twist: the gradient for governor (29 points) became even larger than for president (23), while federal deputy stays the flattest (7). The income polarization is mostly a feature of the executive races, not of how these places choose their representatives.
One last detail that caught my attention while going through candidate registrations. Among right-wing candidates for federal deputy, the share with a security or military title in their ballot name ("Sargento", "Delegado", "Coronel") went from about 1% in 2002 to 7.8% in 2018 and 9.4% in 2026. Religious titles went the other way, from 4.6% to 0.9%, even as the evangelical association with the vote grew. On the left, both stay below 2%. I do not know how to interpret the religious drop yet; it is one of the threads I want to pull next.
What I don't know
- Everything here is an analysis of municipalities. Saying "poor municipalities vote for X" is not saying "poor people vote for X". This is the ecological fallacy, and it is the easiest mistake to make with this kind of data.
- Sorting parties into left, center and right is my own judgment call (PSDB, MDB and PSD are in the center). If you disagree, the numbers change, mostly before 2018, when the center was large.
- The predictors are associations controlled for a handful of census variables, measured in 2022. They do not establish cause, and anything that is not in the census (media, local politics, church networks) is out of the model.
- The TSE data answers where. For the why I would need surveys of people.
What taught me the most was to distrust explanations that fit in a tweet. Abstention, income and "the Northeast" all had a piece of truth, but each one, alone, was mostly geographic coincidence. The TSE and IBGE data are public, and so are my notebooks; if you find a mistake, I want to know.
Sources
- TSE, Open data portal: results by polling section and by municipality/zone, 2002–2026, and candidate registrations.
- IBGE, 2022 Census: income, race, religion, education and urbanization by municipality.
- Party classification into left, center and right: my own, in
partido_ideologia.csv. - Notebooks and scripts: rckbrcls/labs on GitHub.