The scare that mobilized

Did the first-round defeat set off a mobilization of Lula's electorate that changed turnout in the runoff? We test four predictions of this hypothesis with the vote in the same polling stations in 2022 and 2026, the profile of the electorate, 4.4 thousand street campaign events and the TSE report on the "Exército do Lula" channel. The mobilization happened, was larger where Lula was strong and began in the first 24 hours. But it was not "organic and then controlled": it was a mass adhesion to a structure organized from day one, and the effect of street rallies on the vote is small.

Daniel Pessini Sobreira · October 27, 2026 · TSE vote by polling station (2022 and 2026), electorate profile by polling station, Bora Lula events, TSE Technical Report No. 001/2026

1. +1.57 pp more turnout between rounds than in 2022, in the same polling stations. In 2014 and 2018 turnout fell.
2. +1.2 → +2.1 from the decile of polling stations most favorable to Flávio to the one most favorable to Lula. In 2022 the slope was the opposite.
3. 33 hours for Janones's channel to go from 55,000 to 1 million followers, between Oct 5 and 6, before Lula's video.

One hypothesis, four predictions

The analysis of what the polls and markets did not see raised the hypothesis that the Oct 4 result shocked Lula's electorate and set off an organic, autonomous mobilization, which was only partly controlled when André Janones's WhatsApp channel grew. This article tests that hypothesis with more data and with designs that allow it to fail.

It says four different things, and each has a prediction that the data can contradict. The predictions below were written before the tests were run.

Why a scare would mobilize

The calculus of voting. In classical theory, voting is worthwhile when the chance that one's own vote decides the outcome is high enough to cover the cost of going to the polls (DOWNS, 1957; RIKER; ORDESHOOK, 1968). That chance is minuscule for any voter, and so turnout depends less on the calculation and more on the sense of duty, social pressure and the perception that the election is close (BLAIS, 2000). A first round in which the poll favorite loses turns a "decided" election into an uncertain one: it is exactly the kind of shock that increases the perception that every vote counts.

Emotion. Affective intelligence theory says that anxiety leads people to pay more attention, and anger leads to action (MARCUS; NEUMAN; MACKUEN, 2000). Valentino et al. (2011) showed that anger, more than fear, increases high-cost participation, such as going to rallies and working in campaigns. Lula's unexpected defeat is a natural candidate to generate both emotions in his electorate. The literature on the "underdog" effect points in the same direction: whoever is behind may mobilize more (SIMON, 1954; MORTON et al., 2015).

Networks and social pressure. What moves turnout most is personal contact (GERBER; GREEN, 2000) and the pressure of people one knows (GERBER; GREEN; LARIMER, 2008). In an experiment with 61 million Facebook users, the message showing friends who had already voted raised turnout much more than the message without familiar faces (BOND et al., 2012). Mobilization, in this sense, is the activation of networks that already exist (ROSENSTONE; HANSEN, 1993).

Connective action and technopolitics. The part of the hypothesis that speaks of "organic and autonomous" mobilization and of later "control" echoes a debate in the sociology of movements. Bennett and Segerberg (2013) distinguish classical collective action, coordinated by organizations, from connective action, in which people come together through digital networks without a central organization, and from a hybrid type, "organizationally enabled" connective action, in which someone provides the platform and the tasks and adhesion is individual. Tufekci (2017) shows that purely networked mobilization grows fast but is fragile; the contentious politics tradition recalls that scale comes from mobilizing structures and political opportunities (McADAM; TARROW; TILLY, 2001; TARROW, 2011). Toret (2015) calls technopolitics the strategic use of networks to organize and expand collective action, starting from Spain's 15M. On the Brazilian case, Toret and Guerrero (2026) describe the "Exército do Lula" as a "coordinated, synchronized and centralized" activation, in which "lo centralizado era la orientación estratégica y lo distribuido eran las cuentas".

The Brazilian runoff. Turnout usually falls from the first round to the runoff: fewer offices are at stake, and part of the first-round electorate is left without a candidate (PONS; TRICAUD, 2018). The deciders are voters with no party and anti-partisans (SAMUELS; ZUCCO, 2018), and the geography of the Lula vote is concentrated in the poorer and less educated electorate, which is also the one that abstains most (NICOLAU, 2020).

Turnout rose as it had not risen

Turnout fell from the first round to the runoff in 2014 and 2018 and rose +0.4 point in 2022. In 2026, it rose +1.9.

The national comparison mixes changes in the electorate and in polling stations. The clean comparison is station by station: 442,458 polling stations in 5,547 municipalities existed with the same code and at the same voting location in 2022 and in 2026. In them, turnout rose +1.92 points between the 2026 rounds and +0.35 between the 2022 rounds: +1.57 points above the pattern, which, applied to the roughly 159 million eligible voters, comes to something like 2.5 million more voters than a "normal" runoff would have brought to the polls. Prediction H1a is confirmed.

The mobilization came from where Lula was strong

If the mobilization came from Lula's electorate, the extra turnout should appear in the polling stations where he was already strong, and not in the others. The chart divides the polling stations into ten groups of equal size, from the most favorable to Flávio to the most favorable to Lula in the 2026 first round, always relative to the mean of the municipality itself, and shows how much turnout rose between rounds in 2022 (gray) and in 2026.

In 2022, turnout rose slightly less in the most Lula-leaning polling stations: one standard deviation more of Lula margin corresponded to −0.08 point. In 2026 the sign reverses: +0.23 point per standard deviation, and the difference between the two years grows from +1.23 in the decile most favorable to Flávio to +2.13 in the one most favorable to Lula. The effect holds up under controls for blank and null votes in the first round, age and schooling (coefficient of +0.22 point per standard deviation, with a standard error of 0.01). Prediction H1b is confirmed.

Who the extra voter was

A different question is who the votes of those who started voting went to. Within each municipality, the polling stations where turnout rose most gave Lula 53 extra votes and Flávio 35 for every 100 extra voters. In 2022, the proportion was almost the same: 52 for Lula and 38 for Bolsonaro.

The voter who returned to the polls was no different from those of other elections: voting for Lula at a ratio of three to two. What changed was how many returned and where. This is consistent with the estimate in the analysis of the victory, in which those absent in the first round who voted in the runoff gave 3.0M votes to Lula and 0.6M to Flávio.

Organic, but organized from the first day

The most specific part of the hypothesis is the order of events: first a spontaneous mobilization, then organization around Janones's channel. The TSE technical report on the channel (TRIBUNAL SUPERIOR ELEITORAL, 2026), built from public messages, makes it possible to date the growth hour by hour: about 55,000 followers on the morning of Oct 5; half a million at 7:19 a.m. on Oct 6; 1 million at 3:55 p.m.; 1.5 million at midnight on Oct 7; 3.3 million on Oct 11. The invitation text ("Acabei de entrar. Gente, entra!"), published on the night of Oct 5, was forwarded 344,000 times. In audio from the same day, Janones set the goal: "se eu conseguir colocar meio milhão de pessoas até segunda-feira a gente bate essa eleição".

The sequence has three stages. The channel explodes between Oct 5 and 7, before the video in which Lula asked people to join (Oct 7). Street events grow from Oct 8, with peaks on weekends (the Paulista rally on Oct 11, and the eve of the election, with more than 600 events); between Oct 9 and 14 there were events in 243 municipalities. The polls move only afterward: Lula's average in valid votes stays between 47.5% and 48% until Oct 12 and rises to about 49% between Oct 13 and 17.

The adhesion was spontaneous in the sense that millions of people joined on other people's invitation within two days, with no paid announcement and without the official campaign. But it did not happen before the structure: it happened inside it. The same TSE report describes the channel's operation as "coordinated, centralized and synchronized", with "missions", deadlines and requests for confirmation from Oct 5. This is the hybrid type that Bennett and Segerberg (2013) call organizationally enabled connective action, and what Toret and Guerrero (2026) sum up as central orientation and distributed accounts: "Janones no esperaba a que un contenido se hiciera viral: organizaba la viralidad". Prediction H1c is confirmed only in part: there was immediate mass adhesion, but not an autonomous phase that was later "controlled".

The rallies brought people to the polls, and barely changed votes

The last prediction is the most demanding: street mobilization would have its own effect on the vote. The pro-Lula events recorded in Bora Lula, a platform that gathered public agendas of movements, the campaign and local groups, add up to 4,421 occurrences between Oct 5 and 24 in 979 municipalities, home to 65% of the electorate.

Comparing cities with and without events is misleading, because events happen in big, Lula-leaning cities. So we compare each of the 978 municipalities with events with the five most similar municipalities without events in the same state (electorate, margin and turnout in the first round, turnout and swing between the 2022 rounds, age and schooling), and repeat the calculation with a regression with state fixed effects.

In cities with events, turnout rose +0.33 point more than in comparable ones (95% interval from +0.07 to +0.71), and +0.27 above the 2022 pattern (from +0.09 to +0.49). The swing, however, was not larger: −0.07 point (from −0.39 to +0.25). The regression points to a larger swing (+0.72), but it does not survive the comparison with similar cities, which is the more demanding test.

The test with rain on rally days (the reduced form of Madestam et al., 2013) depends on daily rainfall from Oct 10 to 24 in each municipality, which is still being downloaded; the result will enter this section when the series is complete.

The pattern is consistent with the literature: street mobilization tends to bring to the polls those who already agree, and persuades little (GERBER; GREEN, 2000; ROSENSTONE; HANSEN, 1993). Prediction H1d is confirmed for turnout, on a small scale, and not for the swing.

What remains of the hypothesis

The scare of Oct 4 mobilized, and mobilized the right electorate: turnout rose as it had not risen, and rose where Lula was strong. That mobilization, however, was not a spontaneous wave that someone later tamed. From the first hours it was the adhesion of millions of people to a command structure that offered simple tasks and visible goals, exactly the arrangement the literature on connective action describes. And its effect on the result came more through turnout, which yielded something on the order of 2 to 2.5 million net votes for Lula, than through the conversion of Flávio voters, which the street data do not show.

How the numbers were calculated

• Polling station panel. 2022 ballot-box bulletins (first round and runoff) and 2026 vote by polling station (TSE), matched by code (state, municipality, zone, station) and kept only when the voting location is the same and there are at least 50 voters in each round: 442,458 polling stations. Turnout = voters ÷ eligible voters in each round.
• Slopes. Least squares weighted by electorate, with each variable measured as a deviation from the municipality mean (municipality fixed effects) and standard errors clustered by municipality. Controls: blank and null votes in the 2026 first round, voters aged 16 to 24, aged 60 or older and with higher education (electorate profile by polling station, TSE, July 2026).
• Marginal voter. Within-municipality slope of the change in each candidate's votes between rounds on the change in turnout, per 100 eligible voters. It is an ecological estimate (KING, 1997): it describes polling stations, not people.
• Events. Public views of Bora Lula (acao_publica and turno_publico), downloaded on Oct 27, with no personal data: type, date, municipality (by name and coordinates). The platform began recording on Oct 8 and gathered public agendas of movements, the campaign and local groups, plus events found on social media; it measures the density of announced mobilization, not attendance. Matching: for each municipality with events, the five closest without events in the same state on seven standardized characteristics; interval by bootstrap (2,000 resamples).
• Channel. Public messages of the channel recorded in Technical Report No. 001/2026 of the AEED/TSE and press figures after Oct 11.
• Limits. None of this data identifies who, individually, was mobilized. The events are not random, and the matching controls only for what is observable. The 3-week window and the polls with few waves do not allow a precise estimate of how long the mobilization took to become vote intention. Other explanations for the higher turnout, such as free transit in some cities and the weather on the two Sundays, were not measured here.
• Files. hipotese1.json and hipotese1-municipios.json; sources and data in analise-hipotese1/; code in hipotese1_analise.py.

References

BENNETT, W. Lance; SEGERBERG, Alexandra. The logic of connective action: digital media and the personalization of contentious politics. Cambridge: Cambridge University Press, 2013.

BLAIS, André. To vote or not to vote? The merits and limits of rational choice theory. Pittsburgh: University of Pittsburgh Press, 2000.

BOND, Robert M. et al. A 61-million-person experiment in social influence and political mobilization. Nature, v. 489, n. 7415, p. 295-298, 2012.

DOWNS, Anthony. An economic theory of democracy. New York: Harper & Row, 1957.

GERBER, Alan S.; GREEN, Donald P. The effects of canvassing, telephone calls, and direct mail on voter turnout: a field experiment. American Political Science Review, v. 94, n. 3, p. 653-663, 2000.

GERBER, Alan S.; GREEN, Donald P.; LARIMER, Christopher W. Social pressure and voter turnout: evidence from a large-scale field experiment. American Political Science Review, v. 102, n. 1, p. 33-48, 2008.

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

MADESTAM, Andreas; SHOAG, Daniel; VEUGER, Stan; YANAGIZAWA-DROTT, David. Do political protests matter? Evidence from the Tea Party movement. The Quarterly Journal of Economics, v. 128, n. 4, p. 1633-1685, 2013.

MARCUS, George E.; NEUMAN, W. Russell; MACKUEN, Michael. Affective intelligence and political judgment. Chicago: University of Chicago Press, 2000.

McADAM, Doug; TARROW, Sidney; TILLY, Charles. Dynamics of contention. Cambridge: Cambridge University Press, 2001.

MORTON, Rebecca B.; MULLER, Daniel; PAGE, Lionel; TORGLER, Benno. Exit polls, turnout, and bandwagon voting: evidence from a natural experiment. European Economic Review, v. 77, p. 65-81, 2015.

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

PONS, Vincent; TRICAUD, Clémence. Expressive voting and its cost: evidence from runoffs with two or three candidates. Econometrica, v. 86, n. 5, p. 1621-1649, 2018.

RIKER, William H.; ORDESHOOK, Peter C. A theory of the calculus of voting. American Political Science Review, v. 62, n. 1, p. 25-42, 1968.

ROSENSTONE, Steven J.; HANSEN, John Mark. Mobilization, participation, and democracy in America. New York: Macmillan, 1993.

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

SIMON, Herbert A. Bandwagon and underdog effects and the possibility of election predictions. Public Opinion Quarterly, v. 18, n. 3, p. 245-253, 1954.

TARROW, Sidney. Power in movement: social movements and contentious politics. 3. ed. Cambridge: Cambridge University Press, 2011.

TORET, Javier (coord.). Tecnopolítica y 15M: la potencia de las multitudes conectadas. Barcelona: Editorial UOC, 2015.

TORET, Javier; GUERRERO, Rocío. El Ejército de Lula. Tecnopolítica a la altura de los tiempos. El Salto, Madrid, Oct. 27, 2026. Blog Atenea Cyborg. Available at: https://www.elsaltodiario.com/atenea_cyborg/ejercito-lula-tecnopolitica. Accessed on: Oct 27, 2026.

TRIBUNAL SUPERIOR ELEITORAL (Brazil). Assessoria Especial de Enfrentamento à Desinformação. Informação Técnica n. 001/2026 – AEED/TSE. Petição Cível n. 0602791-73.2026.6.00.0000. Brasília, Oct. 2026. Available at: https://static.poder360.com.br/uploads/2026/10/area-tecnica-relatorio-janones-exercito-lula.pdf. Accessed on: Oct 27, 2026.

TRIBUNAL SUPERIOR ELEITORAL (Brazil). Open data: votação por seção eleitoral 2022 e 2026; perfil do eleitorado por seção eleitoral 2026. Brasília: TSE, 2026. Available at: https://dadosabertos.tse.jus.br. Accessed on: Oct 27, 2026.

TUFEKCI, Zeynep. Twitter and tear gas: the power and fragility of networked protest. New Haven: Yale University Press, 2017.

VALENTINO, Nicholas A. et al. Election night's alright for fighting: the role of emotions in political participation. The Journal of Politics, v. 73, n. 1, p. 156-170, 2011.

What the polls and markets did not see · What gave Lula the victory · Placar home page