Method

The polling figures come from the open database of dawum.de, published under the Open Database License (ODbL). It contains every representative election poll for the Bundestag, the state parliaments and the European Parliament since 2017, with pollster, client, fieldwork period, sample size and method. We import the database every hour.

For each parliament we take only the most recent poll from each pollster within a time window: 45 days at federal level, 120 to 180 days in the states depending on how often they are polled. These polls are averaged without weighting and rounded to one decimal place. A party is only included in the average if at least half of the pollsters report it. The average is not a forecast; it is the current state of the polls.

Seats are allocated using the Sainte-Laguë/Schepers method among the parties that clear the electoral threshold set by the relevant electoral law. The basis is the parliament's statutory size without overhang and compensatory seats. Direct mandates, basic-mandate clauses and regional peculiarities such as Bremen's separate threshold are not taken into account. The seat distribution is therefore an approximation.

We show every alliance of two or three parties that would hold more than half the seats under the calculated seat distribution. Alliances that merely add another party to a smaller coalition which already has a majority are hidden. Whether an alliance is politically feasible or desired is not part of the calculation.

Pollsters differ in method, weighting and sampling, and therefore measure the same party systematically slightly higher or lower. For each poll by a pollster we calculate the rolling average of all other pollsters on the same day and average the difference per party over the past twelve months (federal) or 24 months (states). That is the house effect. The adjusted average subtracts these effects from the current polls and shows how much the current picture depends on which pollsters have published most recently. A house effect does not say which pollster is right.

For every election since 2017 we compare each pollster's final poll within 30 days of the election with the official result. The yardstick is the mean absolute error across all parties the poll reports. The results are taken from the official determinations of the Federal Returning Officer and the state returning officers, compiled via the results tables of wahlrecht.de and spot-checked against the primary sources. Preliminary results are marked. Parties that polls do not report separately count as Others in the comparison.

Constituency results for the 2025 federal election come from the Federal Returning Officer's kerg2.csv file, district results from the district file and municipal results from the polling-district statistics: we sum all polling districts of a municipality according to the Federal Returning Officer's key file. In joint municipal associations (Ämter, Samtgemeinden, Verbandsgemeinden) postal votes are often counted together and cannot be attributed to a single municipality; these municipalities show polling-station votes only and are marked accordingly, with the postal vote listed as a separate row under the association. All three sources are published under the Data licence Germany, attribution 2.0, and are linked on every page. State elections by municipality will follow as soon as the respective state returning officer provides machine-readable data; each source is documented individually.

For each parliament we draw random poll values per party 3,000 times from a normal distribution around the current average and recalculate the seat distribution each time. The spread per party is the square root of two components: the sampling error of the average (from the mean sample size and the number of polls) and a systematic component derived from accuracy, namely the mean absolute error of the final polls before all elections since 2017 per party, converted to a standard deviation (factor 1.25, at least 0.8 points). Parties are drawn independently and not normalised to 100%; correlations between parties are ignored, which tends to understate rather than overstate the uncertainty. The random generator is seeded with the poll values, so the same data always produce the same figures. The probabilities describe the uncertainty of today's polling, not the result of the next election.

Strongholds are the areas with a party's highest second-vote share; for municipalities we only count places with at least 1,000 voters, for cities at least 40,000, so that tiny municipalities do not dominate the lists. For the urban-rural divide we group all municipalities into five classes by number of eligible voters and sum the votes before calculating shares. Municipalities whose postal votes were counted at association level are included with their polling-station votes only.

Administrative boundaries come from the Verwaltungsgebiete 1:250 000 (VG250) of the Federal Agency for Cartography and Geodesy, © GeoBasis-DE / BKG 2025, Data licence Germany, attribution 2.0. We have simplified them for display in the browser and converted them to geographic coordinates; the shapes are therefore suitable for maps, not for surveying. Each area is coloured by the party with the highest share, and the labels show that share.

Representative polls of 1,000 to 2,000 respondents have a statistical margin of error of around two to three percentage points for large parties and about one point for small ones. Differences of this size, including between pollsters, are no evidence of a change.

Every election is available as JSON at /api/umfragen/{wahl} and as CSV at /api/umfragen/{wahl}.csv, with the overview at /api/umfragen. For AI systems, /llms.txt describes all entry points. The only condition is that this site and dawum.de are credited as the data source.

• 7 September 2026: Individual poll pages, chronicle, state profiles, Bundesrat, election glossary, constituency results of the 2025 federal election. Seat allocation per state using the method in force there (Hare/Niemeyer in six states, d'Hondt in Saarland). Governments of Baden-Württemberg (Greens, CDU) and Rhineland-Palatinate (CDU, SPD) entered after the 2026 elections. Saxony-Anhalt corrected to 83 statutory seats.
• 6 September 2026: House effects and accuracy added, official results of 33 elections since 2017 stored. Hamburg 2025 and Rhineland-Palatinate 2026 marked as preliminary.
• 5 September 2026: 2025 federal election: share of Others corrected from 3.1% to 4.6%, because the Free Voters are counted under Others in the polls.
• 5 September 2026: Site launched with polling average, seat distribution, coalitions and trend.

The sentences about gaps, movements and majorities on the poll pages are generated from the figures. They contain no assessments that do not follow from the data.