The country of bubbles

How far apart do Lula voters and Flávio Bolsonaro voters live, street by street? We measured the neighborhood of each polling place in the country across three presidential elections, using the method that Brown and Enos applied to 180 million American voters. The result runs against the intuition of a country divided into bubbles: the separation between neighbors has fallen since 2018, almost all of it is regional, and the 2026 comeback did not stop at the border of the most closed-off neighborhoods. What separates Brazilians within cities is less the distance and more the wall.

Daniel Pessini Sobreira · October 29, 2026 · TSE polling-station votes and polling places (2018, 2022 and 2026), electorate profile by polling station, IBGE population

1. 15.9 → 9.7points of isolation between the PT voter and the voter of the right, from the 2018 runoff to the 2026 one.
2. 85%of the separation is between states and between cities. Only 15% is between neighbors in the same city.
3. −1.0point of swing in a closed-off Lula neighborhood, with the same profile and the same starting point. On the scale that discounts the ceiling, nothing.

In the 2026 runoff, Lula took 77.5% of the votes cast for the two candidates at the Nossa Senhora de Fátima municipal school, in Bom Jesus, in Porto Alegre. 1.8 kilometers away, at Colégio Farroupilha, in Três Figueiras, he took 30.1%. In São Paulo, CEU Paraisópolis gave Lula 74.2%; a school in Jardim Morumbi, 1.2 kilometers away, gave him 37.3%. In Salvador, 82.4% at Santa Cruz and 47.2% at Itaigara, 1.5 kilometers away. Scenes like these feed the idea of a country split into bubbles that never touch. This analysis asks how true that is, and whether it has gotten worse.

Why the neighborhood matters

Political opinion is not formed in a vacuum. Since the studies by Huckfeldt and Sprague (1995) in South Bend, it has been known that voters absorb information and preferences from conversations with the people around them, and that the neighborhood shapes those conversations. Mutz (2006) showed the other side of the coin: those who live alongside people who think differently are more tolerant, but also hesitate more and participate less. Allport's (1954) contact hypothesis, tested in more than 500 studies by Pettigrew and Tropp (2006), holds that contact between groups reduces prejudice. If Lula voters and Flávio voters live in separate worlds, each of these mechanisms works in favor of polarization.

In the United States, this concern got a name, the big sort: the thesis of Bishop and Cushing (2008) that Americans are moving closer to people who vote as they do. The thesis has been contested. Mummolo and Nall (2017) showed that, when people move, they weigh price, safety and commuting time far more than neighborhood politics. Martin and Webster (2020) calculated that partisan preference in choosing a home is about five times too small to produce the separation observed, which suggests that neighbors resemble each other because they influence each other, and not only because they choose each other. Even so, the separation exists and has grown: Johnston, Manley and Jones (2016) document it across counties from 1992 to 2012, and Kaplan, Spenkuch and Sullivan (2022) show a sharp increase since the mid-twentieth century.

In Brazil, the geography of the presidential vote has been studied mostly across municipalities and states. Soares and Terron (2008) and Zucco (2008) showed how Lula's reelection in 2006 shifted the PT's base to the poorest municipalities of the North and Northeast; Terron and Soares (2010) called the split between Lula's territories and the PT's a "divorce." Alkmim and Terron (2022) concluded that the country is electorally polarized, with marked territorial patterns. What no one had measured is the scale of the street: with whom each voter shares the neighborhood. That is what the TSE data now allow, because the register of polling places carries the latitude and longitude of almost every school where people vote.

This text moves through four scales, from largest to smallest: the five regions, the 27 states, the 5,570 municipalities and, within cities, the polling places. At each one the question is the same, how close the voter on one side lives to the other, and the tool changes with the scale. It is a way of dealing with an old warning from geography: the separation you see depends on the lens you look through (OPENSHAW, 1984; REARDON; O'SULLIVAN, 2004).

The neighborhood of each polling place

The unit is the building where people vote: 91,810 polling places in 2026, with the votes summed across all the polling stations that operate there. For each one, we define the neighborhood as the nearest places, not counting its own, until they add up to 10,000 votes cast for Lula or Flávio (or for Haddad and Bolsonaro in 2018, and Lula and Bolsonaro in 2022). In a state capital, that is a radius of one or two kilometers; in sparsely populated countryside, it can exceed 20. The median radius in 2026 was 5.9 km.

The main measure is what Brown and Enos (2021) call isolation, and what urban sociology knows as the difference between exposure indices (MASSEY; DENTON, 1988): the PT's share of votes in the neighborhood of the typical PT voter, minus the same share in the neighborhood of the typical voter of the right. If the two lived mixed together at random, the difference would be zero. If they lived in entirely separate worlds, it would be 100.

Every segregation measure depends on the size of the unit chosen, Openshaw's (1984) modifiable areal unit problem. So we repeated the calculation with neighborhoods of 2,500, 25,000 and 100,000 votes and with fixed radii of 1, 2 and 5 km, following the recommendation of Reardon and O'Sullivan (2004) to measure segregation at several scales. The levels change little, and the order among the elections does not change in any of them.

Isolation in points: the PT's share in the neighborhood of the typical PT voter minus that of the typical voter of the right, counting only votes for the two candidates.

Two tools complete the measure. The first decomposes the separation: the variation in the vote across polling places (the η², the share of the variance that is due to place) can be split between the part that exists because states vote differently, the part that exists because municipalities within the same state vote differently, and what remains among the polling places of a single municipality. The second is spatial correlation. Tobler's (1970) "first law of geography" says that everything is related to everything else, but near things are more related than distant things. Moran's I (MORAN, 1950) measures how much: it runs from −1 to 1, and is positive when neighboring places vote alike. Local indicators of spatial association, Anselin's (1995) LISA, say where: they point out the clusters of polling places with a strong PT vote surrounded by similar places (high-high), those with a strong vote for the right (low-low), and the exceptions, places that vote the opposite of their neighbors.

The country has become less divided between neighbors

In the 2018 runoff, the typical Haddad voter lived in a neighborhood where 52.3% of the votes went to the PT; the typical Bolsonaro voter, in one where only 36.4% did. The gap was 15.9 points. In 2022 it fell to 10.0, rose slightly in the 2026 first round (10.8) and reached 9.7 in the runoff, the lowest value in the series.

The direction is the opposite of the American one. While there geographic separation grew for decades (JOHNSTON; MANLEY; JONES, 2016; KAPLAN; SPENKUCH; SULLIVAN, 2022), here it receded over eight years of closely fought elections. The explanation lies less in the street than on the map: in 2018, the Southeast and the South voted as a bloc for Bolsonaro, and the Northeast as a bloc for Haddad. In 2022 and 2026, São Paulo, Minas Gerais and Rio de Janeiro became contested, and a Lula voter in Campinas or Juiz de Fora stopped being a rarity.

This does not mean that Brazilians became less polarized. Fuks and Marques (2022) found affective polarization, antipathy toward the other side, to be far more visible in 2018, and Fuks, Ribeiro and Borba (2021) showed how anti-PT sentiment became a broader antipartisanship. What the data show is that this antipathy does not translate into greater physical separation. Lula supporters and Bolsonaro supporters keep voting in the same schools, and increasingly so.

The divide is between regions, not between neighbors

All of the separation can be broken down into three parts: the part that exists because states vote differently, the part that exists because cities in the same state vote differently, and the part that exists between places in the same city. In the 2026 runoff, 56% of the separation was between states, 29% between municipalities in the same state and only 15% within municipalities.

It is the picture of the realignment that Soares and Terron (2008) and Zucco (2008) described starting in 2006, when the PT's base migrated to the poorest municipalities and to the Northeast. It also helps explain why Brazil does not repeat the urban-rural fracture that Rodden (2019) and Chen and Rodden (2013) show in the United States, where Democrats are concentrated in urban centers. Here the PT wins in the outskirts of the metropolises, in the sertão and in the riverside Amazon all at once, and loses in the agribusiness of the Center-West and in a good part of the interior of the South. The line that separates the two electorates runs more between states and regions than between neighborhoods.

The five regions

Looking inside each region, the separation fell in all of them since 2018 and takes different forms. The North is the most internally divided region (11.6 in 2026, against 15.8 in 2018), with half of the separation between municipalities of the same state. The Northeast, as Lula-leaning as it is homogeneous, fell from 11.1 to 6.2. The Southeast is the most mixed region (4.6), and the Center-West is where the separation most takes place within municipalities (46%), partly because the Federal District is a single municipality.

How far the similarity extends

The correlogram shows Moran's I at increasing distances: between each polling place and its nearest neighbors, and between municipalities that are farther and farther apart. The similarity is very high up close (a polling place and its nearest neighbor have an I of 0.89), declines slowly and is still strong between municipalities 400 or 800 km from each other (0.58). It vanishes between 800 and 1,600 km (0.07) and reverses beyond that (−0.59): that is the distance separating the Lula-leaning Northeast from the South and the Center-West, where the right wins. Tobler's first law holds here on a continental scale.

The two curves tell different stories. Between polling places, the similarity to neighbors fell at every distance from 2018 to 2026: for the 101st to the 500th neighbor, which lie at a median of 52 km, the I went from 0.75 to 0.69. Between municipalities, it fell in 2022 and in 2026 returned to the 2018 level (from 0.92 to 0.88 and 0.93 among municipalities less than 50 km apart). The Brazil of 2026 has regional blocs as sharp as those of 2018, but more mixing within them.

The states: less divided, each in its own way

Within each state, the separation between polling places fell in 24 of the 27 from 2018 to 2026. The largest drops were in Amazonas and Alagoas (−8.7 and −7.8 points). The separation rose only in the Federal District, Rio de Janeiro and Rio Grande do Sul, and only slightly: 0.6 point at most. Between states, the similarity of the vote also declined: Moran's I between neighboring states went from 0.75 in 2018 to 0.69 in 2026.

The map also shows where the separation lives. In Alagoas and Paraíba, 88% and 85% of it is between municipalities: entire cities vote differently from their neighbors, and within each one the electorate is alike. At the other extreme, in Mato Grosso do Sul and Amapá, most of it (74% and 68%) is within municipalities, largely because the electorate is concentrated in a large, diverse capital. The Federal District, a single municipality, has all of its separation, which is small, within it.

The municipalities: sharp blocs, and some split from within

At the municipal scale, the vote is one of the most geographically organized phenomena one can measure. Moran's I of the PT's share across the 5,570 municipalities was 0.90 in 2026 (0.90 in 2018 and 0.88 in 2022). LISA identifies 1,908 municipalities in a cluster of strong PT vote, almost all in the Northeast and in Minas Gerais, and 1,999 in a cluster of strong vote for the right, in São Paulo, the South, Goiás, Mato Grosso, Rondônia and part of Minas Gerais. The exceptions, municipalities that vote the opposite of their neighbors, number only 36.

The separation within municipalities also has a geography, a weaker one (Moran's I of 0.28). Internally divided municipalities cluster together: 241 are in a high-separation cluster, mainly in Maranhão (47), Pará (40), Rio Grande do Sul (34), Mato Grosso do Sul (22) and Mato Grosso (19). The change in separation from 2018 to 2026 has almost no spatial pattern (I of 0.09): the mixing advanced diffusely across the country.

Weights of the k nearest neighbors, row-standardized. All values have a permutation pseudo-p ≤ 0.01 (999 permutations for states and municipalities, 99 for polling places). Polling places with 100 or more votes for the two candidates.

Within cities: little distance, many walls

In the 338 cities with more than 100,000 inhabitants, we measured the separation only among the places within each city. It is small: the median isolation per place was 1.1 points in 2026, against 1.6 in 2018. Between 2018 and 2022 the separation fell in 281 of the 330 cities with data in all three elections, and between 2018 and 2026 in 274.

Spatial correlation within cities confirms the picture. Moran's I among the polling places of each city, with the 8 nearest places as neighbors, has a median of 0.30 in 2026 (0.31 in 2018): neighboring polling places vote alike, but much less so than neighboring municipalities (0.90). Nationwide, LISA within each municipality finds 7,244 polling places in clusters of strong PT vote, 12,009 in clusters of strong vote for the right and 3,282 exceptions, polling places that vote the opposite of their neighborhood, like those at the beginning of this text. The map in section 8 shows each one of them.

The most divided are of two kinds. The first is the city with a large rural zone or an agribusiness hub, where the urban center votes one way and the districts another: Santarém (PA), Luís Eduardo Magalhães (BA), Sorriso (MT), Dourados (MS), Paragominas (PA). The second is the city of gated communities. Nova Lima, in Greater Belo Horizonte, is the extreme case and the only one that closed in continuously: isolation went from 1.8 in 2018 to 5.7 in 2022 and 7.1 in 2026, the biggest rise in the country. In Vila da Serra, Lula got 19.4% of the votes for the two; in Alphaville, 22.5%; in Galo Novo, the city's old neighborhood, 64.1%.

Nova Lima is an almost literal illustration of what Teresa Caldeira (2000) called fortified enclaves: the elite who leave the center for walled communities, close to the poor on the map and far from them in everything else. Villaça (1998) showed that segregation in Brazilian metropolises is driven by the movement of the elites, and Marques and Torres (2005) documented, in São Paulo, the coexistence of physical proximity and social distance. The scenes at the start of this text, Paraisópolis next to Morumbi, Bom Jesus next to Três Figueiras, are this pattern. A neighborhood measure of 10,000 votes mixes both sides of the wall and therefore registers little separation. Brazilian separation within cities is one of walls, not of kilometers.

The state capitals

Among the state capitals, Rio de Janeiro is the most divided (3.3), followed by Florianópolis, Teresina, São Paulo and Porto Alegre. Teresina and São Paulo mixed the most: from 5.7 and 5.0 in 2018 to 2.9 and 2.8 in 2026. Maceió, Rio Branco and Manaus have the lowest internal separation.

Isolation between polling places in the city (η² × 100, corrected for chance). Dissimilarity: the share of voters on one side who would have to vote at a different polling place for all polling places to have the same composition (DUNCAN; DUNCAN, 1955). Share explained by profile: how much of the separation is predicted by the age, schooling and sex of each polling place's electorate (section 11). The rank (#) is among the 338 cities.

The most divided

Find your polling place

Choose the state and the city and type the name of the school or neighborhood where you vote. The map shows each polling place, and the card says how closed off your neighborhood is compared with the rest of the country. The map covers all municipalities; the city card, the 338 with more than 100,000 inhabitants. The clusters button shows the LISA of each polling place within its own city: the clusters of strong vote for each side and the exceptions.

The comeback crossed the bubbles

If closed-off neighborhoods work as echo chambers, the shift in votes between the 1st and 2nd rounds of 2026 should have been smaller in them. In the raw calculation, it was: in the 20% of places with the most homogeneous neighborhood, Lula's margin moved +3.8 points; in the 20% most mixed, +6.3.

Looking side by side, however, the pattern is different. The swing was large in Flávio's neighborhoods (+6.1 points where he had more than 75%, +7.3 between 65% and 75%) and small in Lula's neighborhoods (+2.1 where Lula had more than 75%). It is not the bubble that blocks the change; it is two simpler things. The first is the ceiling: where Lula already had three quarters of the votes, few voters were left to convert. The second is composition: Lula's neighborhoods had fewer blank and spoiled votes in the 1st round, and other analyses on this site have already shown that most of the comeback came from that stock of unclaimed votes.

To separate these effects, we compare places in the same city with the same starting point, the same share of blank and spoiled votes and of third-way votes in the 1st round, and the same age, education and sex profile of the electorate. Once that comparison is made, a closed-off Lula neighborhood (with 75% for him) had a swing −1.0 point smaller than a fifty-fifty neighborhood, and a closed-off Flávio one, −0.7. On the logit scale, which discounts the ceiling, both effects turn slightly positive: the sign reverses. The neighborhood alone explained 2% of the differences in swing within cities; the full model explains 32%, almost all of it through blank and spoiled votes, the starting point and the profile.

Difference in swing, in margin points, between a place with a 75% neighborhood on one side and one with a fifty-fifty neighborhood, with a 95% interval (errors clustered by municipality). Model B adds Lula's share at the place itself in the 1st round; C, the blank and spoiled votes and the third way in the 1st round; D, the age, education, sex and size of the place's electorate. 91,810 polling places.

The result is in dialogue with Martin and Webster (2020): if the similarity among neighbors comes in large part from mutual influence, and not from impassable walls, a shift in votes can spread from one neighborhood to another. The 2026 comeback was, in this sense, a national phenomenon, one that moved through all types of neighborhood more or less in proportion to what there was to change.

Living with the other side moderates, but little

The literature on political discussion networks suggests that those who live between the two sides are exposed to more cross-cutting arguments and so vote in a less aligned way: they hesitate, choose a third option or choose no one (HUCKFELDT; SPRAGUE, 1995; MUTZ, 2006). The 2026 data point in that direction, with small effects. With the same starting point and the same profile, a place in a fifty-fifty neighborhood gave +0.7 point more to the third way in the 1st round (Cury, Renan and Caiado) and had +0.5 point more blank and spoiled votes than a place in a neighborhood that was 75% one side.

It is a context effect, measured across places, and it does not prove that coexistence changes anyone's mind (ROBINSON, 1950; KING, 1997). But it is consistent with the idea that a mixed neighborhood produces the ambivalent voter, and with the finding that it was precisely this voter, the blank, the spoiled and the third-way voter, who decided the runoff. The meta-analysis by Pettigrew and Tropp (2006) shows that contact between groups reduces prejudice even outside the ideal conditions described by Allport (1954); Enos (2017) reminds us that proximity without real contact, that of people who only see each other from afar, can have the opposite effect and heighten hostility. The cities of walls are on that side of the border.

Is the separation about class or about politics?

Part of the political separation between neighborhoods is merely a reflection of social separation: if education predicts the vote, and people live apart by income, then votes also live apart. To measure how much, we compare the actual separation in each city with what would be predicted from the age, education and sex of the electorate of each place alone, using the TSE's electorate profile by polling station. Income is not in the register, and education stands in for it.

In large cities, the profile explains a considerable part, but not the largest part: the median is 40% using the country's average relationship between profile and vote, and 63% using the relationship estimated in each city, which is a ceiling, because it inflates in cities with few places. Within cities, the profile explains 32% of the differences in vote between places. In Teresina and Salvador, with the relationship estimated in the city itself, the profile predicts 88% and 77% of the separation: it is almost entirely social. In Rio de Janeiro, the profile explains only between 2% and 21%: Rocinha and Gávea voted for Lula (77% and 76%), Vila Militar and Barra da Tijuca for Flávio (19% and 30%). The line that cuts through Rio separates the Zona Sul and the Centro from the Zona Oeste and the military areas, and not the rich from the poor.

It is what Layton, Smith, Moseley and Cohen (2021) called demographic polarization in 2018: divisions of gender, religion and race that the Brazilian vote did not have before, and that add to the income cleavage described by Samuels and Zucco (2018) and Nicolau (2020). The geography within cities holds both. Where the city is segregated by class, the vote follows class; where there are strong territorial identities, as in Rio, the vote follows territory.

A country divided on the map, mixed in the street

The image of a Brazil split into bubbles is true at the scale of the country and false at the scale of the street. The typical Lula voter votes at a school whose neighborhood gave 44% of its votes to Flávio, and the typical Flávio voter at one where 46% voted for Lula. The separation that exists is mostly regional, an heir of the 2006 realignment, and has diminished since 2018 as the Southeast became contested again. Within cities, it is small and has fallen in most of them.

The four scales tell the same story from different angles. In the regions and states, the blocs remain sharp, and the correlogram shows that the similarity of the vote extends over hundreds of kilometers. In the municipalities, the vote is organized into large clusters, with very few exceptions. In the cities, the separation is small and the correlation between neighboring polling places moderate. From 2018 to 2026, the mixing advanced at the smaller scales, that of the states internally and that of the polling places, while the larger blocs held.

There are two exceptions worth attention. The first is the gated-community cities, such as Nova Lima, where political separation grows along with the walls. The second is the agribusiness cities, where the center and the districts vote like two countries. In both cases, physical separation adds to social separation, the arrangement that the literature associates with greater hostility between groups (ENOS, 2017).

And the 2026 comeback shows that the bubbles that exist are porous. The shift in votes between the rounds moved through Lula, Bolsonaro and mixed neighborhoods in proportion to what there was to change. If affective polarization has grown, as surveys show, it coexists with an electorate that still shares the same schools on election Sunday.

How the numbers were calculated

• Polling places. TSE votes by polling station (2018 and 2026; in 2022, the ballot-box bulletins), matched to each election's register of polling places by the station code. The place is the building: coordinates rounded to about 10 meters within the municipality, which groups together different zones that vote at the same school. Excluded are pretrial detainees, temporary stations and in-transit voting stations, and coordinates more than 150 km from the municipality's center. In 2018, 18.6% of the stations had no coordinate, which was filled in with the 2022 coordinate of the place with the same name in the same municipality. Final coverage is 92.6% of stations in 2018, 94.0% in 2022 and 98.8% in 2026.
• Neighborhood and isolation. For each place, the nearest neighbors (excluding itself) until they add up to 10,000 votes for the two candidates. Isolation = the average, weighted by PT votes, of the PT's share in the neighborhood minus the same average weighted by the votes of the right (BROWN; ENOS, 2021). With the share of the place itself in lieu of the neighborhood, the measure is η², which breaks down into between states, between municipalities and within the municipality. In cities, η² is corrected by the value expected by chance.
• Cities. The 338 municipalities with 100,000 inhabitants or more in the IBGE population estimate (SIDRA table 6579, 2026), with the neighborhood restricted to the municipality itself.
• Swing. Lula's margin over Flávio in the runoff minus the 1st-round margin, by place, in points and in logit. Weighted least squares by votes, with municipality fixed effects and standard errors clustered by municipality. Profile: voters aged 16 to 34, aged 60 or older, with higher education, with at most primary education, women (electorate profile by polling station, TSE, July 2026), and the logarithm of votes.
• Scales and spatial correlation. The decomposition uses the η² between polling places within each group (Brazil, region, state), corrected for chance, and splits it among states, municipalities and polling places by the average of each level. Global Moran's I with row-standardized weights of the k nearest neighbors (k = 4 for states, 8 for municipalities and polling places) and pseudo-p by permutation. Correlogram: mean product of the standardized values by distance band, between polling places (by neighbor order, in a sample of 20,000 polling places) and between municipalities (by the distance between centroids, all pairs). LISA with conditional permutation (ANSELIN, 1995): 499 permutations for municipalities and 199 for the polling places of each municipality with 10 or more polling places, p < 0.05, without correction for multiple comparisons. The spatial statistics do not weight by votes and use only polling places with 100 or more votes.
• Limits. The polling place is not the voter's home: many people keep voting near an old address, which tends to underestimate the separation. All the measures are of places, not of people, and do not allow inferences about individual behavior (ROBINSON, 1950; KING, 1997). The TSE profile has no income, and race is reported for only a minority of voters, so it was left out.
• Files. bolhas.json, bolhas-municipios.json and the polling places by state in bolhas-locais/; code, data and README in analise-bolhas/.

References

ALKMIM, Antonio Carlos; TERRON, Sonia Luiza. O Brasil é realmente um país polarizado? Análise das eleições presidenciais de 1989 a 2018. Estudos Avançados, v. 36, n. 106, p. 6-32, 2022.

ANSELIN, Luc. Local indicators of spatial association: LISA. Geographical Analysis, v. 27, n. 2, p. 93-115, 1995.

ALLPORT, Gordon W. The nature of prejudice. Cambridge, MA: Addison-Wesley, 1954.

BISHOP, Bill; CUSHING, Robert G. The big sort: why the clustering of like-minded America is tearing us apart. Boston: Houghton Mifflin, 2008.

BROWN, Jacob R.; ENOS, Ryan D. The measurement of partisan sorting for 180 million voters. Nature Human Behaviour, v. 5, n. 8, p. 998-1008, 2021.

CALDEIRA, Teresa Pires do Rio. Cidade de muros: crime, segregação e cidadania em São Paulo. São Paulo: Editora 34; Edusp, 2000.

CHEN, Jowei; RODDEN, Jonathan. Unintentional gerrymandering: political geography and electoral bias in legislatures. Quarterly Journal of Political Science, v. 8, n. 3, p. 239-269, 2013.

DUNCAN, Otis Dudley; DUNCAN, Beverly. A methodological analysis of segregation indexes. American Sociological Review, v. 20, n. 2, p. 210-217, 1955.

ENOS, Ryan D. The space between us: social geography and politics. Cambridge: Cambridge University Press, 2017.

FUKS, Mario; MARQUES, Pedro Henrique. Polarização e contexto: medindo e explicando a polarização política no Brasil. Opinião Pública, v. 28, n. 3, p. 560-593, 2022.

FUKS, Mario; RIBEIRO, Ednaldo; BORBA, Julian. From antipetismo to generalized antipartisanship: the impact of rejection of political parties on the 2018 vote for Bolsonaro. Brazilian Political Science Review, v. 15, n. 1, 2021.

HUCKFELDT, Robert; SPRAGUE, John. Citizens, politics, and social communication: information and influence in an election campaign. Cambridge: Cambridge University Press, 1995.

INSTITUTO BRASILEIRO DE GEOGRAFIA E ESTATÍSTICA. Estimativas da população residente: tabela 6579. Rio de Janeiro: IBGE, 2026. Available at: https://sidra.ibge.gov.br/tabela/6579. Accessed on: Oct 29, 2026.

JOHNSTON, Ron; MANLEY, David; JONES, Kelvyn. Spatial polarization of presidential voting in the United States, 1992-2012: the "big sort" revisited. Annals of the American Association of Geographers, v. 106, n. 5, p. 1047-1062, 2016.

KAPLAN, Ethan; SPENKUCH, Jörg L.; SULLIVAN, Rebecca. Partisan spatial sorting in the United States: a theoretical and empirical overview. Journal of Public Economics, v. 211, 104668, 2022.

KING, Gary. A solution to the ecological inference problem: reconstructing individual behavior from aggregate data. Princeton: Princeton University Press, 1997.

LAYTON, Matthew L.; SMITH, Amy Erica; MOSELEY, Mason W.; COHEN, Mollie J. Demographic polarization and the rise of the far right: Brazil's 2018 presidential election. Research & Politics, v. 8, n. 1, 2021.

MARQUES, Eduardo; TORRES, Haroldo (ed.). São Paulo: segregação, pobreza e desigualdades sociais. São Paulo: Senac, 2005.

MARTIN, Gregory J.; WEBSTER, Steven W. Does residential sorting explain geographic polarization? Political Science Research and Methods, v. 8, n. 2, p. 215-231, 2020.

MASSEY, Douglas S.; DENTON, Nancy A. The dimensions of residential segregation. Social Forces, v. 67, n. 2, p. 281-315, 1988.

MORAN, P. A. P. Notes on continuous stochastic phenomena. Biometrika, v. 37, n. 1-2, p. 17-23, 1950.

MUMMOLO, Jonathan; NALL, Clayton. Why partisans do not sort: the constraints on political segregation. The Journal of Politics, v. 79, n. 1, p. 45-59, 2017.

MUTZ, Diana C. Hearing the other side: deliberative versus participatory democracy. Cambridge: Cambridge University Press, 2006.

NICOLAU, Jairo. O Brasil dobrou à direita: uma radiografia da eleição de Bolsonaro em 2018. Rio de Janeiro: Zahar, 2020.

OPENSHAW, Stan. The modifiable areal unit problem. Norwich: Geo Books, 1984. (Concepts and Techniques in Modern Geography, 38).

PETTIGREW, Thomas F.; TROPP, Linda R. A meta-analytic test of intergroup contact theory. Journal of Personality and Social Psychology, v. 90, n. 5, p. 751-783, 2006.

REARDON, Sean F.; O'SULLIVAN, David. Measures of spatial segregation. Sociological Methodology, v. 34, n. 1, p. 121-162, 2004.

ROBINSON, W. S. Ecological correlations and the behavior of individuals. American Sociological Review, v. 15, n. 3, p. 351-357, 1950.

RODDEN, Jonathan A. Why cities lose: the deep roots of the urban-rural political divide. New York: Basic Books, 2019.

SAMUELS, David J.; ZUCCO, Cesar. Partisans, antipartisans, and nonpartisans: voting behavior in Brazil. Cambridge: Cambridge University Press, 2018.

SOARES, Gláucio Ary Dillon; TERRON, Sonia Luiza. Dois Lulas: a geografia eleitoral da reeleição (explorando conceitos, métodos e técnicas de análise geoespacial). Opinião Pública, v. 14, n. 2, p. 269-301, 2008.

TERRON, Sonia Luiza; SOARES, Gláucio Ary Dillon. As bases eleitorais de Lula e do PT: do distanciamento ao divórcio. Opinião Pública, v. 16, n. 2, p. 310-337, 2010.

TOBLER, Waldo R. A computer movie simulating urban growth in the Detroit region. Economic Geography, v. 46, sup. 1, p. 234-240, 1970.

TRIBUNAL SUPERIOR ELEITORAL (Brazil). Open data: votação por seção eleitoral 2018, 2022 e 2026; eleitorado por local de votação 2018, 2022 e 2026; perfil do eleitorado por seção eleitoral 2026. Brasília: TSE, 2026. Available at: https://dadosabertos.tse.jus.br. Accessed on: Oct 29, 2026.

VILLAÇA, Flávio. Espaço intra-urbano no Brasil. São Paulo: Studio Nobel; FAPESP; Lincoln Institute, 1998.

ZUCCO, Cesar. The president's "new" constituency: Lula and the pragmatic vote in Brazil's 2006 presidential elections. Journal of Latin American Studies, v. 40, n. 1, p. 29-49, 2008.

What gave Lula the victory · The comeback neighborhood by neighborhood in the state capitals · Placar home page