The French data: results, maps, statistics, deputies and polls
Abstract. Official results, circonscription boundaries, INSEE statistics and the list of deputies come from French open data, almost all under the Licence Ouverte; polls come from pollsters' releases and Commission des sondages notices, with Wikipedia as a marked last resort. Every poll carries a source grade (A primary, B press report, C secondary): of the 172 French polls held in September 2026, 40 were grade A. The paper sets out each source, how results are grouped into candidate families and blocs, the checks, the daily check for new notices and the admin entry and CSV import, what we deliberately leave out, and the licences.
Summary
The French section is built from six kinds of data: official election results, circonscription boundaries, INSEE statistics, the Assemblée nationale's list of deputies, by-election results, and opinion polls. Results, boundaries, statistics and deputies come from official open data, almost all under the Licence Ouverte. Polls come from the pollsters' own releases and the notices they must file with the Commission des sondages; where no primary copy could be found we fall back on Wikipedia's poll tables, and mark every such poll. Each poll carries a source grade: A for the pollster's own release or official notice, B for a press report, C for a secondary source such as Wikipedia. On 26 September 2026 the database held 191 French polls: 91 grade A and 100 grade C, every one of the grade C polls a final poll from before 2017. Every morning the site reads each pollster's releases and the Commission des sondages' notices, checks any new poll and stores it with its source; a poll no reader gets can be entered by hand or by CSV import on an admin page, with the same checks. This paper describes each source, what we do to it, what we check, what we deliberately leave out, and the licences. The British counterpart is The UK data pipeline: collecting, reading and checking British pollsters' tables.
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.
- More frequent checks (13 October). After the morning run the poll readers check again every three hours until 11pm.
- Pollster grades rescaled (13 October) (Rating French pollsters: accuracy, the lead and openness).
Why the data deserve a paper
The French forecast rests on comparisons: final polls against results, today's polls against the 2024 round one in each circonscription, a family's share in 2002 against its share in 2022. Every one of those comparisons is only as good as the grouping and the sources behind it. A mislabelled candidate family or a transcription slip in a final poll moves the error estimates that decide whether we correct the polls at all (How we read French polls: what we correct, what we don't, and why). So the rule throughout is the one used for Britain: official data wherever it exists, a citation for every poll, a grade that says how close to the original the figure is, and no reuse of other forecasters' data.
Election results
Sources
All results come from the ministère de l'Intérieur through data.gouv.fr:
- Presidential elections, 2002–2022, national: every candidate in both rounds, for France as a whole (metropolitan France, overseas and French voters abroad). For each year and round the candidates' votes sum exactly to the valid votes (exprimés).
- Presidential election 2022 by circonscription: the ministry publishes results by legislative circonscription directly, so nothing had to be aggregated from polling stations. The 577 circonscriptions sum to the national totals exactly.
- Legislative elections of 2022 and 2024 by circonscription, both rounds: every candidate, with the ministry's label (nuance) and whether they were elected. In 2024, 76 seats were decided in round one and 501 in round two; in 2022, 5 and 572.
- Legislative elections of 2017, 2022 and 2024, national: round-one votes and final seats by nuance.
Presidential results for 1965 to 1995, used only for the long view (below), come from French Wikipedia's results tables, which reproduce the Conseil constitutionnel's proclamations. They are graded C.
What we do to them
Two transformations matter.
Candidate families. Candidates change between elections, so presidential results are grouped by family: Jean-Marie Le Pen, Marine Le Pen and the RN are one family; Jacques Chirac, Nicolas Sarkozy, François Fillon and Valérie Pécresse another; and so on. Some calls are judgements, and they are recorded in the source notes: Christiane Taubira, Jean-Pierre Chevènement and Corinne Lepage in 2002 and José Bové in 2007 go to "others"; Alain Madelin in 2002 goes with the Republicans, since his Démocratie libérale merged into the UMP that year; Mélenchon's 2012 Front de gauche run goes with France Insoumise.
Legislative blocs. Legislative results are grouped from the ministry's nuances into blocs: the united left (only candidates with the official NUPES or NFP label count), the presidential camp, Les Républicains and the UDI, the RN and its allies, Reconquête, and smaller groups (other left, other centre, other right, far left, regionalists, others). In 2017 there was no left alliance, so the Socialist, France Insoumise, Communist and Green nuances are combined. In 2024 the Ciotti wing of Les Républicains, labelled UXD, goes with the RN. Left candidates outside the NFP label in 2024 were dissidents or not endorsed, and go to "other left".
The groupings are ours. The ministry's own nuance is kept on every candidate and shown on the seat pages. Grouping by nuance does not reproduce the counts by parliamentary alliance that the press reports: by nuance the 2024 Assembly has 178 NFP seats and 156 for Ensemble, where counts by alliance give about 180 and 159, because a few winners backed by an alliance were labelled otherwise by their prefecture. The RN and allies' 142 seats match exactly.
Checks and known faults
The build scripts check that every circonscription has exactly one winner, that the seats won in round one plus those decided in round two make 577, that national totals equal the sum of circonscriptions, and that the same 577 codes appear in every file. Every check passed. The ministry's own 2024 files contain five circonscription-level inconsistencies of one or two votes between the candidates' votes and the stated valid votes, and the national round-one file sums to two votes below its own total. These are kept as published.
Circonscriptions are coded DDD-NN: the département code padded to three characters, then the seat number (075-18 is Paris 18th, 02A-01 Corse-du-Sud 1st). The ministry's overseas letter codes are converted to INSEE codes, and the 11 seats for French people living abroad are 099-01 to 099-11.
Boundaries and the two maps
The real map
The boundaries are the ministry's "contours géographiques des circonscriptions législatives" on data.gouv.fr, published in June 2024 and built from polling-station outlines; they are the 2012 boundaries, still in force. The file has 559 polygons covering metropolitan France, Corsica and the overseas départements. It has no shapes for New Caledonia (2 seats), French Polynesia (3), Wallis and Futuna (1), Saint-Martin and Saint-Barthélemy (1) or the 11 seats abroad. Those 18 are drawn as labelled squares.
The map is projected in Lambert-93 for the mainland and in each territory's own official projection for the overseas insets, simplified to about 1 MB, with Paris and its inner suburbs (54 seats) in a large inset. One fault in the source is kept as published: part of the interior of Guyane is not covered by the official outlines.
We did not use the Sciences Po and Toxicode map, which does cover the overseas collectivities and the seats abroad, because it is under the Open Database Licence, whose share-alike terms would extend to our own data.
The hex map
The default map is a hex cartogram of our own: one hexagon per circonscription, so Paris's 18 seats are as visible as the Creuse's one. It was generated from the official boundaries by pushing apart dense urban clusters, assigning each seat to a grid cell with the Hungarian algorithm, and then improving contiguity by simulated annealing. In the final layout every one of the 20 regions and 107 départements forms one connected block, with no holes. Corsica sits apart from the mainland; the overseas seats are in separate blocks below it; and the 11 seats abroad form a rough world map, from North America on the left to Asia and Oceania on the right.
Who lives here: INSEE statistics
Each seat page has a "Who lives here" table from INSEE's Portraits des circonscriptions législatives, which INSEE computes on the exact circonscription boundaries, plus INSEE's legal populations for the 2024 elections. The file has 124 indicators, mostly from the 2018 census, with income and poverty from the 2019 Filosofi data and population from 2021. The seat pages show sixteen of them (population, average age, unemployment, occupation, qualifications, retirement, poverty, living standard, housing and commuting), each against the median circonscription with a rank.
Two gaps are structural. INSEE does not cover the 11 seats abroad, so 566 circonscriptions have figures. Income and poverty are blank for 16 overseas seats that Filosofi does not cover. The unemployment rate shown is derived by us from INSEE's census counts of unemployed and employed people, not an INSEE published rate. The file also has no share of immigrants and no age-group shares, only the mean age and percentiles.
These statistics are descriptive. The seat model does not use them: every circonscription moves by the same national change from its 2024 result (From votes to seats in the National Assembly). We show them because readers want to know what a seat is like, not because they improve the projection.
Deputies
The list of deputies comes from the Assemblée nationale's open data for the 17th legislature, under the Licence Ouverte, cross-checked against the Assemblée's separate list of sitting deputies on the same day. It holds 678 mandates since July 2024: the 577 deputies elected then, plus substitutes who replaced deputies appointed to the government or who resigned, former ministers returning to their seats, and by-election winners. Each seat page shows the current deputy, their parliamentary group today, and every change since 2024 with the Assemblée's stated reason.
The Assemblée's official portraits are not reused. Its site terms say photographs may not be reproduced without prior agreement, and even its "free" photographs may not be cropped. Instead, deputies' photos come from Wikimedia Commons where Wikidata links one to the deputy (566 of the 678), cropped square to the face and credited to the photographer under the file's own licence; the others show an outline. Many Commons photos were taken at events rather than in a studio, so they vary more than the Assemblée's portraits would. Portraits of presidential candidates also come from Wikimedia Commons.
By-elections
Ten seats have held by-elections since July 2024, after a resignation, an annulment by the Conseil constitutionnel, or a deputy joining the government without a substitute able to sit.
| Circonscription | Date | Elected | Bloc | Share |
|---|---|---|---|---|
| Ardennes 1re | 8 Dec 2024 | Lionel Vuibert | Others | 50.9 |
| Isère 1re | 19 Jan 2025 | Camille Galliard-Minier | Ensemble (Macron camp) | 64.3 |
| Hauts-de-Seine 9e | 9 Feb 2025 | Élisabeth de Maistre | Republicans and UDI | 59.7 |
| Jura 2e | 6 Apr 2025 | Marie-Christine Dalloz | Republicans and UDI | 73.8 |
| Saône-et-Loire 5e | 25 May 2025 | Sébastien Martin | Other right | 58.6 |
| Paris 2e | 28 Sep 2025 | Michel Barnier | Republicans and UDI | 62.6 |
| Tarn-et-Garonne 1re | 12 Oct 2025 | Pierre-Henri Carbonnel | National Rally and allies | 52.1 |
| Français établis hors de France 5e | 12 Oct 2025 | Nathalie Coggia | Ensemble (Macron camp) | 65.0 |
| Loiret 1re | 25 Jan 2026 | Stéphanie Rist | Ensemble (Macron camp) | 62.1 |
| Haute-Savoie 3e | 1 Feb 2026 | Antoine Valentin | National Rally and allies | 59.1 |
Two of the ten are graded A: Paris 2nd (the returning officers' official record for round one and the Ville de Paris's polling-station open data for round two) and the 5th seat for French people abroad (the foreign ministry's result sheets). The other eight are graded C, transcribed from French Wikipedia, which cites the prefecture documents. We tried the primary sources: each prefecture website closed the connection on the first request, and the ministry's by-election results pages returned a block page. We did not try to get round those blocks. data.gouv.fr has no ministry dataset for by-elections after 2016. The ministry's nuance is known for only three of the ten contests, so the others are assigned to a bloc from the party label, with the rule recorded against each candidate.
By-elections have low turnout and strong local candidates, so the site shows them against the same seat's 2024 round one and does not use them in the forecast.
Polls
Where the figures come from
Every poll figure has a source link. In order of preference:
- The Commission des sondages notice. Under the law of 19 July 1977 every published electoral poll must file a notice, which states the sample, fieldwork dates, questions and weighting criteria. Some notices also print the results and the raw figures. The Commission's terms restrict reproduction of its documents, so we extract figures only and cite the notice's address; no notice is stored.
- The pollster's own release: Ifop, Toluna Harris Interactive, Elabe, Ipsos, Odoxa and Cluster17 publish results documents or articles.
- Wikipedia's poll tables, under CC BY-SA 4.0, as a last resort. No poll of the current cycle now depends on them; they remain the source of final polls before 2017. Every such poll is marked "via Wikipedia" on the polls page, and its source link keeps the pollster or newspaper link that Wikipedia cites.
We checked Wikipedia against primary documents wherever both existed. Where they differ, the primary document wins. Reading every 2027-cycle poll from its release found more errors in the Wikipedia rows we had used: ten line-ups in five polls named Sarah Knafo where the release has Éric Zemmour; five line-ups lacked a candidate the release includes (Dominique de Villepin, David Lisnard, Bruno Le Maire); one showed an ecologist in a line-up that had none; eight figures were out by one to two points; and one "poll" dated 22 April 2026 was the Marine Le Pen line-ups of Harris Interactive's 28–30 April poll, now stored as part of that poll. The earlier check had found three Wikipedia errors in 2026 OpinionWay polls (figures that differed from the notice with line-ups missing, a missing candidate, and one figure out by a point), and one error in a notice itself, where OpinionWay's notice swapped the labels of Marine Le Pen and Éric Zemmour; the newspaper's and Wikipedia's figures settle which is which.
Source grades
Every poll carries a grade, and the admin page will not store a poll without one:
| Grade | Meaning |
|---|---|
| A | Every figure found in the pollster's own release or official notice |
| B | Taken from a press report of the poll |
| C | Taken from a secondary source, in practice Wikipedia |
A poll is graded A only if all its figures were located in the primary document; a line-up with even one figure found only on Wikipedia stays C. On 26 September 2026 the database held 191 French polls: 91 grade A, 100 grade C, none grade B. Every poll of the current cycle is grade A: 61 presidential polls (46 for round one, 15 with round-two match-ups) and 10 legislative polls. So are 20 of the final polls of 2017 and 2022. Notices before 2016 are not online, so every final poll from 2002 to 2012 and every legislative final poll is grade C.
The grade describes the source, not the pollster. The pollster ratings are a separate matter (Rating French pollsters: accuracy, the lead and openness).
The sets of polls
- Final polls, 2002–2022: each pollster's last poll before each presidential round, and before round one of the legislative elections of 2017, 2022 and 2024. Round-two polls must end on or after round one. Where one pollster published two series with the same final date, both are stored and the second is excluded from averages.
- Legislative seat projections: each pollster's last seat range between the two rounds, 17 in all across the three elections. Where a pollster published ranges by party rather than bloc, the bloc's range is the sum of the party ranges, which overstates its width.
- 2027 presidential polls: every national poll since September 2024 whose results could be read in a primary source, with every line-up (hypothèse) it tested. Each line-up is stored as a separate scenario, with the candidate, the family and, where printed, the raw figure.
- Current legislative polls: voting intention for a snap Assembly election, February to October 2025, by list, with the left and RN configurations tested. No legislative voting-intention poll with fieldwork in 2026 was found.
Some polls were left out on purpose: a poll of a sub-population (a sample of LGBT respondents in February 2026, a poll of small-business owners in October 2025), and five 2027 polls and two legislative polls whose notices contain only the method and whose results could not be found in a primary source. One poll commissioned by a minor candidate and published on his own page is included, with a note.
The build scripts check each line-up's total, and flag any under 98.5% or over 101.5% for checking rather than correcting it by guesswork. Since every 2027 line-up was read from its release, the only one outside that band is an Ifop line-up that its release prints adding up to 102. One 2025 Harris Interactive poll for Regards was left out because it tested party slots ("a left-union candidate") rather than named candidates.
The long view and the horizon series
Two further sets support the track record and the election-day model.
The long view, 1965–1995. French polling began with the 1965 presidential election. The final polls of the six elections from 1965 to 1995 come from French Wikipedia's election articles, and for 1995 round one also English Wikipedia: 105 figures from 22 pollster-rounds, all grade C. Coverage is thin. In 1965 Ifop was the only pollster; for 1981 only Sofres is listed; several final polls from 1981 to 1995 were not found. From 1977 to 2002 the law banned publishing polls in the week before each round, so some "final" polls in these years were never published in France before the vote; they are flagged. These polls feed the long-view tables on the track-record page and nothing else. The round-one table is below.
| Election | Final polls | Average miss per party | Biggest miss |
|---|---|---|---|
| 1965 | 1 | 3.19 | Socialists −4.7 |
| 1969 | 2 | 2.69 | Centre (UDF, MoDem) +4.7 |
| 1974 | 2 | 1.25 | Centre (UDF, MoDem) −2.1 |
| 1981 | 1 | 1.62 | Socialists −4.5 |
| 1988 | 1 | 1.17 | Republicans +3.6 |
| 1995 | 3 | 1.22 | Socialists −3.3 |
| 2002 | 6 | 1.02 | National Rally −3.9 |
| 2007 | 6 | 1.10 | National Rally +3.3 |
| 2012 | 8 | 0.91 | France Unbowed +2.5 |
| 2017 | 9 | 0.60 | Socialists +1.3 |
| 2022 | 11 | 1.26 | France Unbowed −4.8 |
The horizon series, 2002–2022. For the election-day test we need polls at set distances before polling day, not only final polls: every national round-one poll in the year before each presidential election (997 polls) and every poll of the eventual runoff pairing (923). These come from English Wikipedia's poll lists, grade C, and are reduced to 198 poll averages at six distances with the previous result alongside (From today to polling day: drift in French presidential polls). Where a poll tested several line-ups, the one kept is the one testing the most eventual candidates and the fewest others.
Keeping it current
The daily check
Every morning (7 a.m. by default) the site runs its poll readers. For France there is one reader for each pollster that publishes results on its own site (Ifop, Elabe, OpinionWay, Odoxa, Ipsos, Toluna Harris Interactive, Cluster17, CSA, YouGov, AtlasIntel and Verian) and one for the Commission des sondages' list of notices, which catches polls whose figures appear only in the notice. Each reads the newest releases, keeps those for the 2027 presidential election or the next Assembly election, and reads every line-up.
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: the main candidates or blocs must be present, each line-up must add up to between 85 and 101.5 (a round-two match-up to about 100), and the fieldwork must have ended in the last 45 days. A poll that passes is stored with its source link and grade A, and the pages are rebuilt; one that fails is logged with the reason and shown on the admin page. Every request goes through a polite fetcher: it obeys robots.txt, identifies itself, and never logs in or works round a block. Aggregators whose data we may not reuse are not read at all. Notice PDFs are read and not kept, as the Commission's terms require.
French law bans publishing, broadcasting or commenting on polls on the eve and day of each round. The admin page has a switch, off by default, that freezes the France pages for that window, showing the figures as they stood at the start of the eve; polls are collected either way.
The readers depend on each page's layout, so a redesign shows up as "failed" or "nothing found" until the reader is changed, and the Commission's site sometimes refuses connections; the reader then tries again the next morning.
Entering a poll
New polls are entered at an admin page, by form or by CSV import. The form takes the pollster, client, fieldwork dates, publication date, sample, mode, source grade and source link, and the figures one per line: RN 33 or RN 34 36 for a published and raw figure, Marine Le Pen | RN 33 for a named candidate, with a line starting # for each new line-up. The CSV import takes one row per figure with the same fields as named columns.
Before storing, the page checks that the election and every party or family code exist, that a new pollster has a name, that there is a fieldwork end or publication date and it is not in the future, that the grade is A, B or C, that there is a source link, that shares lie between 0 and 100 and no line-up adds up to more than 101.5%, and that the pollster does not already have a poll for that election ending that day. Any problem stops the poll and is shown.
Polls added this way take identifiers from 1,000,000 upwards, so the build scripts, which reload their own polls from the source files, never touch them, and only these polls can be deleted from the admin page. A new poll changes the averages, chances and seat projections on the next rebuild; the site caches the French forecast for 30 minutes.
What we don't use
- Other forecasters and aggregators. France-Vote, sondages2027.fr, Civiqo, Vote-Scope, PolitPro and Politico's poll of polls were used, at most, as indexes to find polls. None offers an open licence, and we take no figures from them.
- Wikipedia, except as a last resort. It is the source of every grade-C poll and of the 1965–1995 long view, and every such figure is marked. Archived links that Wikipedia gives as its sources are kept as citations but were not fetched.
- Maps under share-alike licences, for the reason given above.
- The Assemblée's official portraits, whose terms do not allow reuse.
- Anything behind a login, a block or a paywall. Where a site refused automated requests we recorded the fact and used the next source down, graded accordingly.
Licences
| Data | Source | Licence |
|---|---|---|
| Presidential results 2002–2017, legislative 2017 | Ministère de l'Intérieur, data.gouv.fr | Licence Ouverte 1.0 |
| Legislative results 2024, circonscription boundaries | Ministère de l'Intérieur, data.gouv.fr | Licence Ouverte 2.0 |
| Presidential and legislative results 2022 (final) | Ministère de l'Intérieur, data.gouv.fr | No licence tag on the dataset; the ministry publishes results as open data |
| Presidential results 1965–1995 | Wikipedia, reproducing the Conseil constitutionnel | CC BY-SA 4.0 |
| Circonscription statistics | INSEE | Licence Ouverte 2.0 |
| Deputies | Assemblée nationale open data | Licence Ouverte |
| By-elections | Prefecture and Ville de Paris records, foreign ministry; otherwise Wikipedia | Official records; Licence Ouverte; CC BY-SA 4.0 |
| Polls | Commission des sondages notices, pollsters' releases; otherwise Wikipedia | Figures cited to source; CC BY-SA 4.0 |
| Candidate and deputy portraits | Wikimedia Commons | Each file's own licence |
The 2022 row deserves a note. The four definitive 2022 result files on data.gouv.fr carry the licence field "not specified", unlike the ministry's other election files. The ministry's results are published as open data, but anyone republishing these particular files should check before relying on that. Attribution on the site reads "Ministère de l'Intérieur – data.gouv.fr".
Limitations of the data
Most historical poll figures are secondary. Every final poll from 2002 to 2012, every legislative final poll, and the whole horizon series are grade C, because pre-2016 notices are not online and no pollster archive was found for those years. Our checks of Wikipedia against notices for 2017 and 2022 found it accurate for final polls, but errors turned up in 2026 figures, and there is no way to check the older ones.
The candidate families and legislative blocs are our own groupings. They are stated and applied consistently, but a different reasonable grouping (Bayrou's centrists with the presidential camp, say, or the 2017 left split by party) would change some past errors.
The by-election figures are mostly secondary, and the long view is thin before 2002.
The INSEE statistics describe the population in 2018 and 2019, not today, and do not cover the seats abroad.
References
Loi n° 77-808 du 19 juillet 1977 relative à la publication et à la diffusion de certains sondages d'opinion.
Ministère de l'Intérieur. Résultats électoraux; Contours géographiques des circonscriptions législatives. https://www.data.gouv.fr.
INSEE. Portraits des circonscriptions législatives – données (2022). https://www.insee.fr/fr/statistiques/6436476.
Assemblée nationale. Données ouvertes de la XVIIe législature. https://data.assemblee-nationale.fr.
Commission des sondages. Notices des sondages. https://www.commission-des-sondages.fr/notices/.
How to cite
Polling Forecast (2026). The French data: results, maps, statistics, deputies and polls. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/france-data. Live figures change as new polls arrive; the date above is when the text was last revised.



