Research · When the polls missed

Right on the votes, wrong on the seats: the front républicain and France's 2024 projection miss

Abstract. The first-round polls were close (RN bloc 36.1 against 33.2, NFP 28.4 against 28.0), but every final seat projection made the RN bloc the largest group with 170 to 250 seats, and it came third with 142. More than 200 withdrawals between the rounds and unusually disciplined voters broke the transfer patterns the projections were built on.

Polling Forecast · 2 November 2026 · 8 min read · France, 2024 legislative election, polling error, final polls, inquiry, pollster methods

Summary

In the snap election of June and July 2024 the polls got the first-round vote broadly right and the seats badly wrong. The final first-round polls had the Rassemblement national and its allies on 36.1% and the Nouveau Front populaire on 28.4%; the result was 33.2% and 28.0%, a miss of 2.5 points on the lead. But every final seat projection had the RN bloc as the largest group in the Assembly, with 170 to 250 seats, and it finished third with 142, behind the NFP (178) and Ensemble (156 with Horizons). The cause was the "front républicain": more than 200 left and centre candidates withdrew between the rounds, voters followed their lead to an extent no past election had shown, and the seat models were built on past transfers.

What the polls said

Emmanuel Macron dissolved the Assembly on 9 June, the night of the European elections, and the first round was held on 30 June. The Commission des sondages counted 42 polls in the three-week campaign, 36 of them on the first round. The last poll of each institute:

Pollster	Fieldwork end	RN and allies	NFP	Ensemble	LR
Cluster17	27 June	35.0	29.5	20.0	7.5
Elabe	27 June	36.0	27.5	20.0	9.0
Harris Interactive	28 June	37.0	28.0	20.0	6.0
Ifop	28 June	36.5	29.0	20.5	7.0
Ipsos	28 June	36.0	29.0	20.0	8.0
Odoxa	27 June	35.0	27.5	21.0	7.0
OpinionWay	27 June	37.0	28.0	20.0	6.0
Average		36.1	28.4	20.2	7.2

(Elabe, Ifop and Ipsos included the divers droite in their LR figure.) Between the rounds almost no institute published second-round vote intentions. What they published were seat projections. The last of each, taken between 3 and 5 July:

Pollster	Fieldwork end	RN and allies	NFP	Ensemble	LR
Cluster17	5 July	170–210	165–195	130–160	30–50
Elabe	4 July	200–230	165–190	120–140	35–50
Harris Interactive	5 July	185–215	168–198	115–145	32–63
Ifop	5 July	170–210	155–185	120–150	50–65
Ipsos	4 July	175–205	145–175	118–148	57–67
Odoxa	4 July	210–250	140–180	115–155	40–60
OpinionWay	4 July	205–230	145–175	130–162	38–50
Midpoint average		205	170	136

What happened

In the first round (turnout 66.7%, the highest at a legislative election since 1997) the Ministry of the Interior's figures gave the RN 29.26% and Éric Ciotti's allies (UXD) 3.96%, together 33.2%; the Union de la gauche 28.0%; Ensemble 20.04% plus 0.72% for Horizons; and LR 6.57%. Against the average, the RN bloc was overstated by 2.9 points, the NFP by 0.4 and Ensemble by 0.2. All seven institutes had the RN bloc too high, by between 1.8 and 3.8 points.

After the second round on 7 July the RN held 125 seats and its allies 17, 142 in all. The Union de la gauche won 178, Ensemble 150 and Horizons 6, LR 39 and the divers droite 27. Against the midpoint of the final projections, the RN bloc was 63 seats short, the NFP 8 seats over and Ensemble 20 seats over. The RN bloc's result was below every institute's range, 28 seats under the lowest bound. The NFP's 178 fell inside five of the seven ranges, Ensemble's 156 inside two. Every institute had the order wrong: the bloc they put first came third.

So the vote miss was ordinary and the seat miss was not. The 2.9-point overstatement of the RN bloc is close to its average at French legislative elections (below). The seat error came from what happened after the first round, which no first-round poll was measuring.

Why it happened: what the pollsters found

Commission des sondages, activity report for 2024 (published February 2025). The commission called the legislative elections "a good vintage" for the polls on the strict polling exercise, the national first-round vote, and credited the short two-year parliament for giving respondents a sharper memory of their last vote to weight on. Performance was "much weaker" for seat projections, "even with fairly cautious seat ranges". It gave four reasons. Seat projections come from transfer matrices that infer future second-round behaviour from past behaviour, and that assumption failed: the front républicain was "as late as it was powerful", transfers from the right to the far right were far weaker than in 2022 and every other transfer far stronger than usual. The candidate line-up changed during the week: many withdrawals were decided late and were not always captured in the last surveys, fielded on the Tuesday and Wednesday. The movement itself grew on the Friday and Saturday and perhaps on polling day. And seat projections are not polls under the 1977 law, so the commission cannot check them; on 18 June 2024 it issued a communiqué asking the media to keep them apart from poll results. It also noted that for the first time all institutes polled on each respondent's actual local line-up, the "offre réelle", and that only 4,005 candidates stood against 6,290 in 2022.

Elabe, to the Assemblée nationale's commission of inquiry, 29 April 2025. Bernard Sananès and Vincent Thibault said their last survey ran on Wednesday 3 and Thursday 4 July, after nominations closed at 6pm on Tuesday 2 July; they had to collect the real second-round line-ups from the prefectures, and asked panellists about six configurations of contest. The seat projection built from those transfers was the only output given to the media.

Ipsos, same day. Brice Teinturier rejected the claim that 27 polls had all been wrong: seat simulations are not polls, but projections built from transfer matrices measured by polls, applied to 577 constituencies. Ipsos had chosen not to publish any seat projection before the first round. In his view the polling instrument "worked rather well" at the European and legislative elections.

Ifop, 7 May 2025. Frédéric Dabi said the first-round polls had been very good because turnout was massive, as in 1997.

What our data shows

The first-round miss in 2024 follows a pattern. In our database the RN and its allies were overstated in the first round of every legislative election since 2002: by 2.1 points in 2002, 1.0 in 2007, 1.8 in 2012, 4.3 in 2017, 0.5 in 2022 and 2.9 in 2024, an average of 2.1. 2024 was a little above that average, not an outlier.

The seat projections are a different story, and a worse one, with no stable direction. In 2017 the final projections gave Ensemble 430 to 470 seats; it won 350. In 2022 they gave the RN 20 to 60; it won 89, above every range, in an election where transfers from the right to the RN were far stronger than they would be in 2024. In 2024 they gave it 170 to 250 and it won 142. The projections assume voters will transfer as they did last time, and in each of the last three elections they did not. Before the first round of 2024 the gap was larger still: five institutes published projections that gave the RN bloc between 210 and 305 seats, and three of them had the absolute majority of 289 inside their range.

The vote and the seats also moved apart in the second round itself. The RN and its allies won 37.1% of the second-round vote, more than any other bloc, and still came third in seats, because so many of their contests had become two-way fights against a single opponent. Before the withdrawals the first round had produced more than 300 three-way runoffs; after them 89 were left.

What the pollsters changed

The changes are smaller than the miss, because the pollsters do not regard the seat projections as polls.

The actual line-up became standard practice. In 2024 every institute polled respondents on the candidates standing in their own constituency, at least late in the campaign. The commission welcomed it.
Ifop now weights on past legislative votes. Dabi told the inquiry in May 2025 that Ifop had begun, "in the last few months", to weight its political samples on the previous legislative vote as well as the presidential one.
Projections set apart. The commission asks institutes and media to publish seat projections separately from vote intentions, with a warning that they lack the same methodological safeguards. It has said that regulating them would need new law and would be hard to define, since academics and anonymous accounts publish projections too.
Ipsos's restraint. Already in 2024 Ipsos published no seat projection before the first round, because projections are built constituency by constituency and the line-ups were not yet known.

What it means for our forecast

Our legislative forecast treats the two errors separately. For the first-round vote, the RN bloc's overstatement has kept the same direction at all six elections since 2002, so our correction rule applies: we take off half the average past miss, and we keep the correction because it would have been closer to the result at six of six elections, each predicted without knowing its result. For seats, our model works constituency by constituency through qualification, withdrawal and the runoff itself, and it is fitted on 2024. Where the RN and two or more of its opponents qualify, every opponent except the strongest withdraws in the central projection; in the simulations each possible withdrawal happens with probability 0.71, the 2024 rate. That builds the lesson of 2024 into the seat count; it also means that if the next election saw transfers like 2022's, our seat figures would err the way the institutes' did in 2022. The method is in our paper on French polling error, and each institute's record is in our French pollster ratings.

Sources

How to cite

Polling Forecast (2026). Right on the votes, wrong on the seats: the front républicain and France's 2024 projection miss. Retrieved 2 November 2026 from https://www.pollingforecast.com/research/missed-fr-2024-leg. Live figures change as new polls arrive; the date above is when the text was last revised.