Polls overstated the RN at all six Assembly votes since 2002

The polls have put the National Rally too high at every French Assembly election this century. Across the six legislative first rounds since 2002, the average of the final polls overstated the RN and its allies by 2.1 points, and it never once understated them. At the three presidential run-offs the party has reached, the final polls overstated its candidate by 3.7 points on average. With Marine Le Pen on 34.1 in our average for next April's first round, that record now bears directly on who is likely to win the Élysée.

The pattern does not hold everywhere. In presidential first rounds, the polls have missed the RN in both directions, and across Europe our data show no steady error for the radical right at all.

What we counted

We compared each pollster's final poll with the official result, using the polls and results shown on our France accuracy page. Shares are of valid votes. In Assembly elections the RN figure includes allied lists as the pollsters grouped them, so in 2024 it takes in Éric Ciotti's breakaway Republicans. A positive miss means the polls had the party too high.

The Assembly record runs one way. In 2002 the final polls had the Front National and its allies on 13.2; they took 11.1. In 2012 it was 15.4 in the polls and 13.6 on the day. The largest miss came in 2017, when all eight final polls had the party at 17 or 18 and it took 13.2. In the snap election of 2024, all seven final polls had the RN bloc between 35 and 37; it won 33.2. The two smallest misses, 1.0 points in 2007 and 0.5 in 2022, were within normal scatter, but they too were on the high side.

The run-offs tell the same story. In 2002 the polls gave Jean-Marie Le Pen 21.0 against Jacques Chirac, and he took 17.8. Marine Le Pen was polled at 38.4 against Emmanuel Macron in 2017 and won 33.9. In 2022 the polls had her on 44.8 and she won 41.5. In all three run-offs, every final poll had the Le Pen figure above the result.

The exception inside the exception

Presidential first rounds break the pattern. Since 2002 the RN candidate has been under-polled three times and over-polled twice, and the average miss is minus 0.4 points. The best-known miss was 2002 itself, when the polls had Jean-Marie Le Pen on 13.0 and he took 16.9, knocking Lionel Jospin out of the race. Marine Le Pen was polled at 15.9 in 2012 and took 17.9. In 2017 the polls were 0.9 points too high, and in 2022 they were 0.2 points too low. Going back to 1988, the RN was under-polled at five of seven first rounds.

So the same party, often with the same candidate and the same pollsters, has been polled about right or too low in the first round of a presidential race, and too high a fortnight later in the run-off and again a few weeks after that in the Assembly vote.

Our model handles the cases differently because a test tells it to. For each past election we worked out a correction from the other elections only and checked whether it would have helped. Correcting presidential first-round polls would have been further from the result at four of five elections since 2002, so our projection leaves them as published. Correcting the run-off polls was closer at four of five, and correcting Assembly polls was closer at six of six, so our projection uses both corrections.

The record is starker at lower-turnout votes. At the 2021 regional elections, when about a third of voters turned out, our data hold RN polls in eight mainland regions, and the RN list was polled too high in all eight, by between 3.9 and 9.7 points. In Île-de-France the polls had Jordan Bardella's list on 17.4; it took 13.1. In Auvergne-Rhône-Alpes the RN was polled at 22.0 and won 12.3. The 2024 European election was the exception: the RN was polled at 32.2 and took 31.4, a miss within normal scatter.

The March municipal elections offered one big-city check, and it had the same shape. In Marseille the final first-round polls put the RN's Franck Allisio between 29 and 34, and he took 35.0. Elabe's head-to-head run-off poll in February gave him 44 against Benoît Payan; he took 40.3 and lost, though the France Unbowed candidate withdrew after the poll was taken. One city proves little, but the shape matches the national record.

Europe has no shy far right

Outside France the picture changes. Since 2010 we hold 26 national elections in Germany, Spain, France, Italy, the Netherlands and Poland with polls for the main radical-right list. The final polls had that list too high 13 times and too low 12 times; once, for the AfD in 2025, they were exactly right at 20.8. The average miss is minus 0.1 points. Leave France out and the count is eight too high and ten too low.

The large misses outside France ran the other way. The PVV was polled at 18.0 in the Dutch election of 2023 and won 23.5. The Lega was polled at 13.6 in Italy in 2018 and took 17.4. Vox was polled at 12.5 in Spain in November 2019 and won 15.1. Lower-turnout Dutch votes do resemble France, though: the PVV was polled 3.8 points too high at the 2023 provincial elections and 5.2 points too high at the 2024 European election.

What the wider data do show is that new parties tend to be under-polled. Of 24 new parties and alliances at their first national contest in our records, 17 did better than their final polls. France again went its own way: Éric Zemmour's Reconquête was polled at 9.7 in 2022 and took 7.1.

What it means for 2027

Our average has Le Pen on 34.1, from the latest polls of seven pollsters, 16.4 points ahead of Édouard Philippe. A Toluna Harris Interactive poll for M6 and RTL published on 12 October put her on 35 to 37 depending on the line-up, and a Cluster17 poll for Politico in September had her on 30 to 31. On our record, those first-round figures carry no known lean.

The run-off is where the history bites. Six pollsters' latest Le Pen against Philippe polls average 53.7 for Le Pen. Our forecast takes that to 51.8 after its run-off correction, and gives her a 66 per cent chance of winning that pairing and an 80 per cent chance of the presidency overall. If the run-off polls missed by the full 3.7-point average of the last three, a 53.7 would become a dead heat at 50.0. A miss the size of 2017 would put Philippe ahead.

Whether Le Pen is on the ballot is now much less in doubt than in the spring. On 7 July the Paris appeal court upheld her conviction for embezzling European Parliament funds but cut her ban from office to 45 months, 30 of them suspended. The court said the ban she had been under since March 2025 had already remedied the breach of probity, Eunews reported. Public Sénat reported that she declared her candidacy on TF1 that evening and appealed to the Court of Cassation, which suspends the one-year sentence, to be served with an electronic tag, that the court imposed. She said Bardella would be her prime minister. Polls of a Bardella run are older: two pollsters' latest have him on 48.3 against Philippe.

The limits

These are small numbers: six Assembly elections and three run-offs. Final polls are taken days before the vote, and late movement counts as a miss here. Pollsters have changed their methods, and nothing guarantees an old pattern repeats. Seat projections do not share the direction either: between the rounds in 2022, pollsters gave the RN 26 to 48 seats on average and it won 89, while in 2024 they gave it 188 to 221 and it won 142.

L'Insoumission, the media platform of France Unbowed, argued in 2022 that polls routinely overstate Le Pen; our data support that for Assembly votes and run-offs, not for presidential first rounds. Ramona Bloj in Le Grand Continent found last month that polls 200 days out had been mixed for Le Pen. Our first-round finding agrees with her. The part that has gone unreported is the split: in France, the final polls have not understated the RN at a single Assembly first round this century or a presidential run-off.