Methodology

Every presidential candidate since 1989 is assigned to one of two blocs: left (the PT camp and allies) or right (the camp now organized around PL). There is no center bloc: in a system decided by 50% + 1 every candidacy ends up adding to one of the two sides, and two blocs keep every election comparable on the same scale.

Strength (1–3) says how clearly the candidate or party belongs to the bloc in that era: 3 = core of the bloc; 2 = usual ally or intermediate position; 1 = classified by elimination, lowest confidence.

Parties are classified by era (valid_from–valid_to) because they change sides over time. The classification is editorial, versioned and open to review: every row carries its rationale and source.

Party taxonomy 1 · candidates 1 · updated 10/19/2026 · 27 rows flagged “under review”

Parties

Candidates

A state’s lean index is the weighted average, over the 2026, 2022, 2018, 2014 runoffs, of the gap between the state margin (right − left) and the national margin, with weights 0.4 / 0.3 / 0.2 / 0.1 from most recent to oldest. Positive = the state votes further right than Brazil; negative = further left.

lean(UF) = Σ_i w_i · (margem_UF,i − margem_BR,i)

Rating by the absolute value of the index: ≥ 15 pts solid; 8–15 likely; 3–8 lean; < 3 toss-up.

A state is a swing state (shown in yellow across the site) when it meets at least one of three criteria: (1) it changed bloc at least 3 times across the 10 elections from 1989 to 2026, using round 2 where there was a runoff and round 1 in the years without one (1994 and 1998); (2) it changed bloc at least 4 times over every round — counting round 1 by the bloc of the state’s most-voted candidate and round 2 by the winner, so a turn between the two rounds of the same year counts (Minas Gerais in 2026: right ahead in round 1, left in round 2); summing each bloc’s votes in round 1 would manufacture turns in three-way races, hence the most-voted rule; or (3) its lean index is below 8 pts in absolute value (“lean” or “toss-up” rating). Yellow is reserved for this classification and used for nothing else. The Swing states page shows the full series, each state’s index and how much every state moved between round 1 and round 2 in runoff years.

See swing states →

The headline comes solely from the transfer model: a state-by-state emulation of how round-1 votes redistribute between the two finalists in the runoff, weighted by electorate and expected abstention. Uncertainty is propagated by Monte Carlo simulation. Polls and the incumbency prior are computed and shown for comparison only.

Editorial decision: the headline number does not use polls; they appear only as a comparison.

Statistical forecast based on official TSE results and on polls registered with the TSE; it is not an opinion poll (TSE Resolution 23.600/2019).

National uncertainty: 4.77 pts standard deviation, estimated from the historical swing of the national result between round 1 and round 2, floored at 2.50 pts. (sd of year effects α_y (between-round national swing) floored at 2.5 pp)

Simulation: 20,000 scenarios with a national error of 4.77 pts standard deviation, a 3.00 pts error per state and 0.60 regional correlation. The win probability is the share of simulations in which each bloc exceeds 50% of valid votes.

Scenarios

• Base — the model as described above.
• High abstention in the Northeast — lower turnout in the Northeast region, reducing those states’ weight in the national total.
• Strong transfer from eliminated candidates — votes of eliminated candidates migrate to the finalist of the same bloc above the historical pattern.

Comparison: polls and prior (not part of the number)

Average of the selected polls: 48.89% for the left (raw), 47.18% after the house-effect correction. Incumbency prior: 51.27% for the left. Both appear next to the projection for the reader to compare, without changing the headline.

Institute house effects

For each institute we compare the margin (left − right) of its final polls before the 2022 rounds and the 2026 round 1 with the official result. The mean signed error is subtracted from the 2026 polls; reliability (0–1) weights each institute in the average.

In the elections used for calibration, polls on average overstated the left-bloc candidate by 2.81 pts. The correction therefore moves the poll average from 48.89% to 47.18% of valid votes for the left — toward the right-bloc finalist.

Leave-one-year-out validation using transfer + prior only — there are no historical polls in the repository — so the error is wider than the full ensemble’s.

Official TSE results by municipality and zone (1989 by state only, since the TSE did not publish that year by municipality). Each downloaded file is identified by its SHA-256 hash.

• TSE — Repositório de Dados Eleitorais — votacao_candidato_munzona and detalhe_votacao_munzona files per year.
• TSE — Divulgação de resultados — official per-state JSON used by the Live page.
• IBGE — Malhas — state and municipality boundaries, simplified.
• Polls: only institutes registered with the TSE; the registration number and release URL accompany every row of polls.json.

The site uses only public, aggregated data from the TSE (totals by municipality, zone and state) and from IBGE. There is no personal voter data at any stage. The site sets no cookies and uses no personal identifiers; audience metrics are aggregate and anonymous.

Exported data and taxonomy: CC-BY 4.0, with attribution to the TSE and IBGE. Site code: closed.

Download the data →