Germany Bundestag electionby 25 Mar 2029in 855 days CDU/CSU + SPD + Greens 99% Calendar Track record Research
Research · Germany
EN DE

The German data: results, boundaries, polls and licences

Abstract. Results, Wahlkreis boundaries, Strukturdaten and members come from the Bundeswahlleiterin under the Datenlizenz Deutschland and were checked 24 ways before use. Polls come from pollsters' own releases where they publish a readable series and otherwise from Wikipedia, graded C and marked; 405 of the 508 German polls are grade C. New polls are found by a daily check of three pollsters' pages and entered by hand or CSV, with validation before anything is stored; no aggregator's data is used.

Polling Forecast · updated 9 November 2026 · 17 min read · Germany, data sources, Bundeswahlleiterin, Datenlizenz Deutschland, Wahlkreise, hex map, Strukturdaten, polls, source grades, Wikipedia, regional polls, licences

Summary

This paper describes every data source behind the German pages: where each comes from, how it was changed, how it was checked and under what licence it is used. Results, boundaries, Strukturdaten and the list of members come from the Bundeswahlleiterin (the Federal Returning Officer) under the Datenlizenz Deutschland. Polls come from the pollsters' own releases (grade A) or, where a pollster publishes only through a client, the client's own article or chart (grade B); only where neither can be read do they come from Wikipedia's poll tables, graded C and marked "via Wikipedia" on the site. Of 508 German polls in the database, 171 are grade A, 114 grade B and 223 grade C. We use no poll aggregator's data, and nothing from wahlrecht.de except as a check.

Unlike the British pipeline (The UK data pipeline: collecting, reading and checking British pollsters' tables), which reads every figure out of the pollsters' own spreadsheets and PDFs, the German pipeline has no table readers: German pollsters publish headline figures, not tables. New German polls are read every morning from each pollster's own page, or its client's where the pollster publishes nothing itself, checked, and stored with their source link; a poll no reader gets can be entered by hand or by CSV import. Reading headline figures from pages is easier to get wrong than reading the British tables, and the checks described below exist to catch what they can.

What has changed since this paper was written

Updated 9 November 2026. The rest of the paper is as first published, except where a sentence below says it was corrected.

Principles

  1. Official data first. Every result, boundary and structural figure comes from the Bundeswahlleiterin or, for Landtag elections, the state returning officers. No result is taken from a secondary source where an official one could be reached.
  2. Every poll is cited. Each poll carries a link to where its figures were taken from, and a source grade: A for the pollster's own release (or an official notice), B for a press report, C for a secondary source. Each poll's page shows the link and what its grade means; a grade C poll is labelled as taken via Wikipedia and not yet checked against the release.
  3. No reuse of aggregators or other forecasters. DAWUM, wahlrecht.de's tables and other forecasters' data are not used as sources.
  4. Changes are stated. The Datenlizenz Deutschland requires modified data to be marked as such; every page footer says "data changed by Polling Forecast".
  5. Personal data is limited to what the official source still publishes. Where the Bundeswahlleiterin has withdrawn candidates' names, we do not fill them in from elsewhere.

Election results, 2005–2025

Results come from the Bundeswahlleiterin's open-data files: kerg2.csv for 2025 (the final result of 14 March 2025) and the btw_kerg archive for 2005 to 2021.

Level Elections What is stored
Germany and each Land 2005, 2009, 2013, 2017, 2021, 2025 Zweitstimmen and Erststimmen by party, eligible voters, voters, valid votes
Each of the 299 Wahlkreise 2025 Erststimmen and Zweitstimmen by party; the Erststimme winner; whether the winner was seated
Each of the 299 Wahlkreise 2021, recalculated onto the 2025 boundaries Erststimmen and Zweitstimmen by party

The CDU and CSU are stored separately and also together as CDU/CSU, since polls report them together; the combined rows are excluded wherever votes are summed. "Die Linke" covers the Linkspartei.PDS of 2005. The Freie Wähler are kept as a party from 2013. Everything else, including independents, is "others".

Two decisions about 2021 matter. The national 2021 figures are the original result of 26 September 2021, because that is the result the final polls were trying to predict (Measuring German polling error, 2005–2025). The currently valid official result includes the repeat election in parts of Berlin on 11 February 2024. The two differ only in Berlin and in the national totals, and no party's national share changes by as much as 0.06 points.

The 2025 files were checked 24 ways before they were loaded, and all 24 checks passed. The Wahlkreis rows sum to the official national totals for every party on both votes; eligible voters and voters sum to the official 60,510,631 and 49,928,653; party rows sum to valid votes in every Land and year; the Land results for 2005–2017 and 2025 match the Bundeswahlleiterin's separate database of results since 1949 in all 595 party, Land and year cells compared; and the seats by party and Land match both that database and Heft 3 of the official report. The strongest check is the seat calculation itself: run on the official 2025 votes, it gives every party's seats, every Land list and the seated or unseated status of all 299 Wahlkreis winners exactly as the Bundeswahlleiterin did (From votes to seats: the Bundestag seat law).

2021 on the 2025 boundaries

The Wahlkreise were redrawn for 2025, so the 2021 Wahlkreis results cannot be compared with 2025 directly. The Bundeswahlleiterin publishes the 2021 result recalculated onto the 299 Wahlkreise of 2025 (btwkr25_umrechnung_btw21.csv), including the 2024 Berlin repeat election. We load it as published, mapping its "WGr/EB" and "sonstige" columns to others. The official file gives votes, not winners; the notional 2021 winners on the new boundaries are our own calculation, the party with the most recalculated Erststimmen: SPD 120, CDU 100, CSU 46, Greens 16, AfD 14 and Linke 3. The recalculated Wahlkreis party sums equal the official 2021 national totals, including the Berlin repeat.

The recalculated 2021 result is shown on each Wahlkreis page as the previous result. It is not used in the forecast, which swings from 2025.

Wahlkreis geometry and the hex map

The real map comes from the Bundeswahlleiterin's generalised boundaries for 2025 (299 polygons, UTM zone 32). We simplified them with a topology-preserving Douglas–Peucker algorithm at 100 metres on shared edges, so that neighbours keep identical borders, dropped fragments under 0.2 km², converted the coordinates to kilometres with north up, and rounded them to 0.1 km. The Land outlines are the union of their Wahlkreise. Berlin's twelve Wahlkreise, too small to read at national scale, are repeated in an inset at 2.75 times the scale. The result is about half a megabyte and keeps the islands and the Büsingen exclave.

The hex map gives every Wahlkreis one hexagon of the same size, so that a Wahlkreis in Munich counts visually as much as one in the Uckermark. We found two openly licensed tile layouts for German Wahlkreise, and neither could be used: one is a square grid that uses the 2017 and 2021 numbering (192 of 299 names differ at the same number from 2025), and the other is based on the 2017 map. We therefore built our own. The Wahlkreis centres are scaled and relaxed so that crowded cities spread out, assigned to a hexagonal grid, and then improved by simulated annealing that penalises long borders between Länder, long coastlines, Wahlkreise far from their real position and Länder split into pieces. The best of four runs was kept. Every Land is in one piece, there are no holes, Hamburg sits between Schleswig-Holstein and Lower Saxony, and Brandenburg wraps around Berlin as it does on the real map. The script is deterministic, so rerunning it gives the same layout.

Strukturdaten

Each Wahlkreis page has a section "Who lives here", built from the Bundeswahlleiterin's Strukturdaten for the 2025 Wahlkreise: 48 variables (population and foreign citizens, age groups, housing, cars, businesses, school leavers by certificate, childcare, household income and GDP, employment by sector, Bürgergeld recipients and unemployment) for the 299 Wahlkreise, the 16 Länder and Germany, 316 areas in all. We changed decimal commas to points and gave each column a code; the values are as published. Two cautions travel with the data. In 51 Wahlkreise in cities split between several Wahlkreise (Hamburg, the Hanover region, Berlin, Munich, Cologne and others) some columns show the whole city's value, or a value apportioned by population, not a figure for the Wahlkreis. And the reference years differ by indicator, from income in 2021 to unemployment in November 2024.

The Strukturdaten are not used in the forecast. They describe a Wahlkreis; they do not yet predict it. A seat model that uses them would have to beat proportional swing on a backtest first (What we don't know yet: limitations of the German forecast).

Members elected, and changes since

The 630 members elected in 2025 come from the Bundeswahlleiterin's final list of those elected: name, party, Land, Wahlkreis for directly elected members, and list place. Changes since the election come from the Bundeswahlleiterin's list of successors ("Veränderungen im 21. Deutschen Bundestag"); the version we hold, dated 23 June 2026, lists nine: one elected candidate who declined the seat, seven who resigned and one who died. Each successor is stored with the date they took the seat and the member they replaced. Two of the successors are unseated Wahlkreis winners of 2025, from Wahlkreise 282 and 290 in Baden-Württemberg.

The candidate list is a different matter. The Bundeswahlleiterin stopped publishing personal data on candidates from 15 September 2025, under § 86(3) of the Bundeswahlordnung. We therefore hold names only for candidates who were elected, for independents (whose surname appears in the results file) and for the two successors above. The other 1,988 Wahlkreis candidates of 2025, including 21 of the 23 unseated winners, are stored without a name. We did not take their names from other sites, because the official source withdrew them for data-protection reasons.

State election results

Six Landtag elections have been held since the 2025 Bundestag election, and the site uses them as a check on the pollsters (Rating German pollsters: accuracy, the lead and openness). Results come from the state returning officers:

Election Result source Grade
Hamburg, 2 March 2025 Official final result (Landeslisten) A
Baden-Württemberg, 8 March 2026 Official final result, Zweitstimmen A
Rheinland-Pfalz, 22 March 2026 Landeswahlleiter's notice of the final result A
Sachsen-Anhalt, 6 September 2026 Statistical office results page A
Berlin, 20 September 2026 Landeswahlleiter's provisional result A
Mecklenburg-Vorpommern, 20 September 2026 Landeswahlleiter's provisional result A

Table: State election results used.

The Berlin and Mecklenburg-Vorpommern results are provisional until the final results are declared. For each election the polls are each pollster's last poll that finished within 31 days of polling day. Of the 24 such polls, 12 come from the pollster's own release (A: FGW's Politbarometer-Extra and infratest dimap's own pages), 4 from the commissioning outlet's chart (B: BILD for INSA, NIUS for INSA in Sachsen-Anhalt, n-tv for Forsa) and 8 from Wikipedia's state election pages (C), where the only release is behind a paywall or on a site that forbids automated access, or no release was found. Every A and B figure matched the Wikipedia row except one sample size.

Polls

What is held

Set Polls Used for
Final polls, Bundestag 2005–2025 53 (52 usable) Polling error, backtest, lenses, pollster ratings
National polls since the 2025 election 387 Today's average and the forecast
Regional polls since the 2025 election 68 Shown on the Länder page; not in the forecast
Final polls, 1953–2002 15 pollster-elections The long view only
Polls through each term, 2005–2025 3,985 Averages at set distances before polling day, for the election-day test (From today to polling day in Germany)
Final polls at six Landtag elections 24 pollster-elections State-election check

Table: German poll data. The first three sets make the 508 polls counted on the sources page.

Where the figures come from

Four pollsters publish their own series in a form we can read, and their polls are grade A:

  • Forschungsgruppe Wahlen: the Politbarometer archive from 2004, about 570 releases, parsed from the release text for both the Projektion and the raw "Politische Stimmung". Releases the parser could not read completely were dropped or keyed by hand from the text.
  • infratest dimap: its own Sonntagsfrage series back to 2000, with fieldwork and sample size from the ARD-DeutschlandTREND release pages where those are still online.
  • Allensbach: its own Sonntagsfrage table from July 2010. The table gives no sample sizes; those come from Wikipedia where the dates match.
  • Ipsos: the monthly series in its public chart. Fieldwork dates and sample sizes come from Wikipedia.

The other pollsters, since the 2025 election:

  • YouGov: every monthly poll from YouGov's own result-table PDFs (A), including its East and West columns.
  • GMS: its own "Politikbus" press releases (A), with the chart where the text gives only changes; two 2026 polls never reached its press page and stay C.
  • pollytix: the table on its own Wahltrend page, where each poll stays for about 20 days, and one client report (A); six older polls stay C.
  • Forsa: forsa publishes no results itself, so every poll is read from the chart in n-tv's own RTL/ntv-Trendbarometer article (B).
  • Verian: verian.de shows only a holding page, so polls are read from FOCUS's own FOCUS-Sonntagsfrage articles where FOCUS prints every figure (B); eleven stay C.
  • INSA: INSA's site shows only its latest Meinungstrend (older ones are in its paid archive) and most BILD reports are behind BILDplus, so 24 polls come from INSA's page or free BILD articles (A or B) and the other 138 stay C.

In all, 117 of the 387 national polls since the election are grade A, 111 grade B and 159 grade C. Every A and B figure was compared with the Wikipedia row it replaced: the party figures agreed except in a handful of polls (a GMS chart and text that disagree, a Forsa "others" figure that made the Wikipedia row add up to 101), and several dates and samples were corrected (a pollytix poll Wikipedia dated a month early among them).

How they are checked

Before loading, every poll's shares were summed (every current poll sums to between 98 and 101) and every primary poll was matched against the Wikipedia row for the same pollster and fieldwork end. Sixty-six matched exactly on every party. The differences were resolved in favour of the pollster: Allensbach's own table has the CDU/CSU at 29.5 where Wikipedia has 29 for March 2025, and Ipsos's chart has others at 4 where Wikipedia has 5 for the same month. The check also caught a parsing error of our own: six FGW releases from October 2025 to January 2026 say that "all other parties, including the BSW and FDP" had 12%, which the parser had first read as FDP 12. These are now stored as others 12, with no BSW or FDP figure, because FGW did not publish one.

The final polls were checked against wahlrecht.de's final-poll tables, as a check only. 251 of 252 party figures agreed; the exception, Emnid's 2013 FDP figure, is left blank. Emnid's 2009 final poll could not be found in any usable source and is missing. Forsa's 2005 final poll was published only as ranges and has no shares.

Conventions worth knowing

  • CDU/CSU. Polls report the Union together. INSA also publishes the CDU and CSU separately; we store both and use the combined figure. Land polls keep the CDU and CSU separate, as the parties stand.
  • Undecided. Forsa publishes a combined share of non-voters and undecided, stored separately and not part of the 100. FGW and Infratest dimap state an undecided share only in some releases. The definitions differ and are not compared across pollsters.
  • FGW's raw figures are stored beside its Projektion for every FGW poll since the 2025 election. They cover the named parties only (FGW does not publish an "others" figure in the Stimmung), so they sum to between 90 and 98.
  • Dates. Wikipedia's tables give fieldwork dates; FGW's and Infratest dimap's series give release dates. Where only one date is known it is stored as the fieldwork end. The 2013 table on Wikipedia gives one date per poll, probably the publication date.
  • The long view. No primary archive of final polls exists online for 1953 to 2002. Those figures were keyed from Wikipedia (C), and there are no final polls at all for 1957, 1961, 1965, 1969, 1980, 1983, 1987, 1990 or 1994.

Regional polls

A few pollsters publish Bundestag voting intention for East and West Germany or for a single Land. Since the 2025 election we hold 68 such polls: 30 for the East and 29 for the West (YouGov monthly, INSA, and one Forsa), five for Mecklenburg-Vorpommern (infratest dimap and Forsa), three for North Rhine-Westphalia (Forsa) and one for Bavaria (Forsa). YouGov's 38 East and West figures come from its own result tables (A), where they corrected two Wikipedia rows (Greens and BSW swapped in the East in June 2025; the FDP in April 2026); three Forsa and infratest dimap polls come from the commissioning paper's or broadcaster's chart (B); the other 27, mostly INSA's, stay C. YouGov's East column appears to include Berlin: its weighted share matches the East plus Berlin, though YouGov does not say so.

The Länder page shows them against the 2025 result in the same area. They do not enter the seat projection: with so few, and most of them from two pollsters, the national average moved by each Land's 2025 result is the better estimate (What we don't know yet: limitations of the German forecast). They are there so that a reader can see whether the East and West are moving differently from what proportional swing assumes.

The daily check and entering new polls

The daily check

Every morning at seven, the same background service that collects British pollsters' tables runs a reader for each German pollster. Each reader fetches the pollster's own results page, or the client's where the pollster publishes nothing itself, and reads the newest poll or two from it:

Pollster Page read Grade
Forschungsgruppe Wahlen Politbarometer archive A
infratest dimap Sonntagsfrage page A
Allensbach Sonntagsfrage table A
Ipsos Sonntagsfrage page A
YouGov monthly article and its result-table PDF A
INSA Meinungstrend on its news page A
GMS press page A
pollytix Wahltrend table A
Forsa n-tv's Trendbarometer chart B
Verian FOCUS's Sonntagsfrage articles B

A poll we already hold (same pollster, same fieldwork end within a day) is skipped. A new one must pass the same checks as a poll typed in by hand, and more: every main party must be present, the shares must add up to between 85 and 101.5, and the fieldwork must have ended in the last 45 days. A poll that passes is stored with its source link and grade, and the pages are rebuilt. A poll that fails is not stored; the reason is logged and shown on the admin page, where it can be entered by hand after a look. Every attempt is logged: stored, rejected, blocked, failed, nothing found or up to date. Every request goes through the same polite fetcher as the British check: a named user agent, robots.txt obeyed, no logins and no workarounds.

The readers have limits. Each is written against a page's current layout, so a redesign shows up as "failed" or "nothing found" until the reader is changed. INSA's weekend poll is not on INSA's page. GMS does not post every release, and FOCUS keeps no list of its Sonntagsfrage articles, so a Verian poll the magazine prints only in print, or publishes more than about three days before a check, is missed. pollytix's polls stay on its page for about 20 days, so they are caught only if the check runs within that time.

Entering a poll

New polls are entered on an admin page, one at a time or by CSV. A poll needs its country and election, pollster (with a name if the pollster is new), fieldwork end or publication date, source grade, source link and figures; client, fieldwork start, sample size, mode, region and notes are optional. Figures are typed one per line as a party code and share, optionally followed by a raw share (for FGW's Stimmung). The CSV import takes one row per figure, with the columns in any order; rows with the same poll details make one poll. A template is on the page, and a dry run reports what would be stored.

Before anything is stored, every poll is checked: the election must exist, the fieldwork end cannot be in the future, the grade must be A, B or C, a source link is required, every party code must be known, every share must be between 0 and 100, a scenario's shares cannot add up to more than 101.5, and the pollster must not already have a poll for the same election, date and region. A poll that fails any check is reported with its problems and not stored. Polls added on the page take identifiers from 1,000,000 up, so the loading scripts, which reload their own polls, never touch them; only polls added on the page can be deleted there. Storing or deleting a poll rebuilds the forecast at once.

What we don't use

  • Poll aggregators. DAWUM's data and wahlrecht.de's tables are not used as sources. wahlrecht.de was consulted only to check our final polls after they had been built. The daily check does not visit aggregators.
  • Other forecasters' outputs, including academic forecasts and pollsters' seat projections. YouGov's 2025 MRP model is excluded.
  • Wikipedia, except as a last resort. Where a pollster does not publish its own series in a readable form, Wikipedia's tables are the fallback, graded C and marked "via Wikipedia" on the poll pages. One caveat: almost every Wikipedia poll row cites wahlrecht.de rather than the pollster, so our grade C figures are, indirectly, largely wahlrecht.de's compilation. Where Wikipedia links the pollster's or client's own page, we add that link to the source.
  • The Internet Archive. No figure comes from archived copies of pages.
  • Candidate names from secondary sites, for the reason given above.
  • Bundestag photographs, except as a fallback. Members' photos come from Wikimedia Commons where Wikidata links one (403 of 639), cropped to the face and credited. For the others the site shows the photo published with the member's biography on bundestag.de, credited to the photographer named there and shown whole, uncropped, because the Bundestag's picture terms allow use with credit but not editing.

Licences

Data Rights holder Licence Credit
Results, Wahlkreis boundaries, Strukturdaten, members elected Die Bundeswahlleiterin, Wiesbaden Datenlizenz Deutschland – Namensnennung 2.0 "© Die Bundeswahlleiterin, Wiesbaden 2025"; data changed
Geodata in the boundaries GeoBasis-DE / BKG as above "Grundlage der Geoinformationen © GeoBasis-DE / BKG 2024"
Strukturdaten columns on employment, Bürgergeld and unemployment Bundesagentur für Arbeit via the Bundeswahlleiterin as published
Landtag results State returning officers Official results cited per election
Poll figures from pollsters Each pollster Published figures, cited link on each poll
Poll figures from clients (Forsa, Verian, some INSA) n-tv, FOCUS, BILD Published figures, cited link on each poll
Poll figures from Wikipedia Wikipedia contributors CC BY-SA 4.0 "via Wikipedia" on each poll
Party leaders' and members' photographs Photographers on Wikimedia Commons Each file's licence per photograph
Members' photographs without a Commons image bundestag.de biographies, photographer as named there Bundestag picture terms: credit, no editing per photograph, shown uncropped

Table: Licences.

The Datenlizenz Deutschland – Namensnennung 2.0 allows reuse, including commercial reuse, with attribution and a note that the data has been changed. Ours has been: parties re-coded, figures aggregated, percentages computed, geometry simplified. Poll figures carry no open licence; we treat them as published facts and cite the release for each.

Known gaps

Gap Effect What would fix it
Most INSA polls grade C 138 of INSA's 162 polls rest on Wikipedia, and indirectly on wahlrecht.de INSA publishing its past Meinungstrends openly, or BILD's reports being free to read
INSA's second weekly poll, Verian and GMS not always online Some new polls are missed by the daily readers INSA publishing its weekend poll, Verian's own site returning, GMS posting every release
Emnid 2009 final poll missing; Forsa 2005 has no shares Two fewer final polls A primary source for either
Berlin and Mecklenburg-Vorpommern results provisional Two state tests on provisional figures Load the official final results when declared
No names for most 2025 candidates Wahlkreis pages name only the elected Nothing, by design
No final polls for nine elections between 1957 and 1994 The long view has gaps An open academic compilation, if one exists

Table: Known gaps in the German data.

How to cite

Polling Forecast (2026). The German data: results, boundaries, polls and licences. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/germany-data. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like