France Presidential election, round one18 Apr 2027in 148 days Le Pen wins 99% Philippe wins 1% Calendar Track record Research
Research · France
EN FR

From votes to seats in the National Assembly

Abstract. Each of the 577 circonscriptions starts from its 2024 round one, moves by uniform swing, and goes through the legal thresholds for round two, a withdrawal rule and a model of two-candidate duels fitted on the 403 duels of 2024. The duel model picks 370 of those 403 winners in sample but, fitted on one election and tested on the other, does little better than naming the round-one leader, and pollsters' own seat projections have missed by even more. The paper describes each step, its tests, the poor fit for overseas and abroad seats, and why the 80% ranges are too narrow.

Polling Forecast · updated 9 November 2026 · 13 min read · France, National Assembly, seat model, uniform swing, two-round system, withdrawals, duel model, overseas seats, simulation

Summary

This paper describes how the site turns legislative voting-intention polls into seats in the National Assembly. Each of the 577 circonscriptions starts from its 2024 round-one result, grouped into seven blocs, and moves by the national change in each bloc. The legal thresholds decide who wins outright and who reaches round two. Where the RN and two or more of its opponents qualify, the weaker opponents withdraw. Each remaining duel is decided by a model of the round-two share, fitted on the 403 duels of 2024, that uses every bloc's round-one share.

The model reproduces 2024 well, as it should, since it was fitted on it: given the actual round one, the duel model picks 370 of 403 winners. It predicts other elections poorly. Fitted on 2022 and applied to 2024, it picks 64.5% of duel winners, against 61.0% for simply naming the round-one leader; fitted on 2024 and applied to 2022, it picks 74.8%, against 78.0% for the leader. The pollsters' own seat projections have missed by even more. The seat ranges should be read with that record in mind.

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.

  • Only elections from 2000 shape the forecast (28 September). Earlier elections stay in the long view and on the history pages but no longer feed any test, correction or error size. In France the presidential tests now run on the five elections from 2002 to 2022: the round-one correction stays off and the round-two correction stays on.

Why seats are hard in France

In a British seat the winner is whoever comes first. In a French circonscription, round one only sorts the candidates. A candidate wins outright with over half the valid votes and at least a quarter of registered voters. Otherwise the top two go through, together with anyone who has the votes of at least 12.5% of registered voters. Qualified candidates may withdraw, and in round two whoever has most votes wins.

So the seat depends on four things: each bloc's round-one share, turnout (which sets how many clear the 12.5% bar), who withdraws, and where the voters of eliminated and withdrawn candidates go. A party with a third of the vote can win a majority if its opponents are divided and their voters stay at home in round two, or a quarter of the seats if its opponents stand down for each other and their voters follow. The RN's 2024 result is the plainest example: it led round one with 33.2% for itself and its allies and won 142 seats, third among the blocs.

The model below takes each of those steps in turn. It plays the part for France that From votes to Commons seats: UK swing models and their test plays for Britain.

The inputs

Legislative polls

No Assembly election is due, and pollsters ask about one only now and then. The last legislative polls were taken on 8 October 2025. The projection averages the latest poll from each pollster within 60 days of the newest, with equal weight per pollster. Pollsters have kept polling the presidential first round, so when the newest presidential poll is more than 30 days newer than the newest legislative one, we move each bloc's legislative average by that bloc's change in the presidential polls over the same period: from 7 October 2025 to 19 November 2026 at present. The legislative polls keep what is particular to an Assembly vote, such as the right's local candidates and the left's single list; the presidential polls say how far each bloc has moved since. Most of these polls test several scenarios for how the left and the far right would stand. We take the scenario with a single list for the left (La France insoumise, the Greens, the Communists and the Socialists together) and the RN standing with Éric Ciotti's UDR, the arrangement of 2024, and fall back on the poll's first scenario if that one was not tested.

The average is corrected by half the mean final-poll miss of 2002–2024, because that correction was closer than the polls at 6 of 6 past elections (Measuring French polling error, 1965–2024). The largest adjustments are to the RN and its allies (down about 1.3 points before rescaling) and the non-aligned right (up about 1.1).

Blocs

The polls, the results and the model use slightly different groupings, so the model works with seven blocs:

Model bloc Ministry labels (nuances) and poll groups it includes
Left (NFP and other left) Nouveau Front populaire, other left
Ensemble and centre Ensemble, other centre
LR and other right Les Républicains and UDI, other right
RN and allies RN, UDR and other far-right allies
Reconquête Reconquête
Far left far left
Others and regionalists others, regionalists

The 2024 baseline

Each circonscription \(i\) starts from its 2024 round-one result: the share \(b_{i,k}\) of valid votes for bloc \(k\), its registered voters \(R_i\) and its valid votes \(V_i\). With \(B_k\) the national 2024 share of bloc \(k\) and \(S_k\) the corrected national share today, the projected round-one share is

\[ x_{i,k} = \frac{\max\!\left(0,\; b_{i,k} + S_k - B_k\right)}{\sum_{m} \max\!\left(0,\; b_{i,m} + S_m - B_m\right)} \times 100. \tag{1} \]

This is uniform swing (Butler and Stokes, 1969): every seat moves by the same number of points. It is the simplest projection there is, and we chose it for that reason, not because it has been tested against alternatives in France. It takes no account of incumbency, of local candidates, or of the fact that a bloc gaining ten points nationally is unlikely to gain ten points in a seat where it had five.

Round one: outright wins and qualification

Turnout is held at its 2024 level in each seat. Let \(k_1\) be the leading bloc. The seat is won outright if

\[ x_{i,k_1} > 50 \quad\text{and}\quad \frac{x_{i,k_1}}{100}\,V_i \;\ge\; 0.25\,R_i. \tag{2} \]

Otherwise bloc \(k\) qualifies if it is in the top two or if

\[ \frac{x_{i,k}}{100}\,V_i \;\ge\; 0.125\,R_i. \tag{3} \]

The thresholds are those of the electoral code. Because they are set on registered voters, not votes cast, they bite harder where turnout is low. In 2024 turnout was high (68.8% of registered voters in metropolitan France), and three or more candidates cleared the bar in 312 of the 501 circonscriptions that went to round two, by our count from the ministry's files. At an election with lower turnout, the number of three-way contests would fall.

Withdrawals

In 2024 most three-way contests became duels. By our count from the ministry's files, 91 of those 312 circonscriptions still had three or more candidates in round two, and 225 qualified candidates withdrew. (224 of them were in those 312 seats; the other was in a seat with only two qualifiers. Figures reported at the time, 216 withdrawals from 306 possible three-way contests, leaving 89, differ from this recount; the model uses the recount.) Of those 225 withdrawals, 131 were by candidates of the Nouveau Front populaire and 78 by Ensemble candidates, nearly all of them standing down so that a single opponent faced the RN.

The model follows that pattern. Where the RN and two or more of its opponents qualify, every opponent except the strongest withdraws. The opponents counted are the left, the centre, the mainstream right, the far left and others; Reconquête never withdraws. Where the RN is not among the qualifiers, nobody withdraws, and a three-way round two goes ahead. In the central projection the rule is applied every time. In the simulation each possible withdrawal happens with probability \(224/312 \approx 0.72\), the 2024 rate.

This is a deliberately plain rule. It ignores the deals made in particular seats, the candidates who refused to stand down, and the possibility that the parties will behave differently next time. In 2024 the presidential camp and the left stood down for each other; whether they would do so again, and whether the mainstream right would join in, is a political question the model cannot answer.

The duel model

Form

For each pair of blocs \((A, B)\) that met in round two in 2024, we fit a linear model of A's share of the two-candidate vote:

\[ y_{A} - \tfrac{1}{2} = \beta_0 + \beta_1\,(x_A - x_B) + \sum_{k \ne A,B} \gamma_k\, x_k, \tag{4} \]

where the \(x\) are round-one shares as fractions and \(k\) runs over the other five blocs. The coefficient \(\beta_1\) says how much of a round-one lead carries into round two. Each \(\gamma_k\) says how the voters of bloc \(k\), whose candidate is out of the race, split between A and B: positive when they lean to A. The model is fitted separately for each of the nine pairs that occurred, by least squares with a small ridge penalty (0.02 for pairs with at least 20 duels, 0.2 for the thinner ones) to keep the coefficients of rare pairs from running wild. The predicted share is held between 2% and 98%. A pair with no fitted model is decided by the round-one leader.

What it learned

Three pairs account for most of the duels, and their coefficients show how 2024's round two worked:

Duel (A against B) Duels in 2024 \(\beta_1\) Left voters Ensemble voters LR and right voters Others' voters Reconquête voters
Left against RN 157 0.56 – 0.25 0.17 0.15 −0.09
Ensemble against RN 129 0.48 0.37 – 0.07 0.21 0.06
LR and right against RN 61 0.50 0.48 0.25 – 0.09 0.01

Every bloc's eliminated voters except Reconquête's favoured the RN's opponent, but by very different amounts. Left voters backed a mainstream-right candidate against the RN strongly (0.48) and a centrist fairly strongly (0.37). Centrist voters backed the left against the RN (0.25) as much as they backed the right (0.25). Right-wing voters barely helped a centrist against the RN (0.07), and helped the left a little more (0.17).

The other six pairs had between 1 and 30 duels each; the left against Ensemble (30) is the only one of them with more than a dozen. Their coefficients rest on very little.

A worked example

Take a seat where round one gives the RN 38%, the left 35%, Ensemble 15%, the mainstream right 8%, Reconquête 2% and the far left and others 1% each, and the left and the RN go through. With the rounded coefficients for the left against the RN,

\[ y_{\text{Left}} - 0.5 \approx -0.011 + 0.56(0.35 - 0.38) + 0.25(0.15) + 0.17(0.08) - 0.09(0.02) + 0.06(0.01) + 0.15(0.01) \approx 0.023, \]

so the left takes about 52% and wins a seat in which it trailed by three points in round one. This is the mechanism behind the RN's 2024 seat total, and it is why a projection that looked only at round one would overstate the RN.

Three-way contests

Where three or more candidates remain, the model looks for the candidate who would beat each of the others in a duel, using the duel model for each pair. If there is none, the round-one leader wins.

How well it works

In sample

Given the actual 2024 round one and the actual finalists, the duel model picks 370 of the 403 two-candidate round twos correctly. Naming the round-one leader would pick 61.0%. Running the whole chain (the 2024 round one in every seat, our qualification and withdrawal rules, and the duel model) gives the right bloc in 521 of 577 seats, with these totals:

Bloc Model on the 2024 round one Actual 2024
Left (NFP and other left) 205 192
Ensemble and centre 163 162
LR and other right 65 69
RN and allies 130 142
Others and regionalists 14 12

Even on the election it was fitted to, the chain gives the left 13 seats too many and the RN 12 too few. Part of that is probably the withdrawal rule, which removes every weaker opponent of the RN, whereas in reality 91 contests stayed three-way. These are in-sample figures and flatter the model.

Out of sample

The honest test is to fit on one election and predict another. We have only two elections with circonscription results in the database, 2022 and 2024, and we used each to predict the other:

Fitted on Tested on Duels Duel model Round-one leader
2024 2024 (in sample) 403 91.8% 61.0%
2022 2024 403 64.5% 61.0%
2024 2022 551 74.8% 78.0%

Out of sample the duel model is about as good as naming the round-one leader, and in one direction worse. The problem is not the form of the model, which fits 2024 closely, but that round-two behaviour changed between the two elections: rates learned from one did not carry over to the other.

The projection assumes that round two works as it did in 2024. That is the natural assumption for an election soon after 2024, and the worst one if the parties and their voters stop standing down for each other. It is the weakest link in the chain, and it is the main reason the 80% ranges are wide.

How pollsters' seat projections have missed

Pollsters face the same problem, and their record is a useful warning. Between the rounds they publish seat ranges for each bloc, with round one already known.

ElectionBlocProjected (average range)WonRanges that held it
2017Ensemble (Macron camp)437–4673500/3
2017Republicans and UDI67–881300/3
2017Left31–52581/3
2017National Rally and allies1–680/3
2017Others3–843/3
2022Ensemble (Macron camp)258–2982451/7
2022Left (NUPES)161–2061310/7
2022National Rally and allies26–48890/7
2022Republicans and UDI54–72645/7
2022Other left6–12221/6
2022Other right3–7110/5
2022Other centre2–443/4
2022Others5–1010/7
2024New Popular Front (united left)155–1851785/7
2024Ensemble (Macron camp)121–1511562/7
2024National Rally and allies188–2211420/7
2024Republicans and UDI40–58425/7
2024Other left12–17142/2
2024Other centre6–861/1
2024Others10–1930/7
Live Seat projections published between the rounds against the seats won Computed from the database, polls to 19 November 2026.

The bloc that led round one won fewer seats than every pollster's range at each of the last three elections: the presidential camp in 2017, the NUPES left in 2022 and the RN in 2024. The RN in 2022, projected at 26–48 seats, won 89. These projections had the benefit of the real round one; ours does not. We have no reason to think our model would have done better, and the out-of-sample tests above suggest it would not.

Overseas and abroad

Of the 577 seats, 27 are overseas and 11 represent French people living abroad. They are projected like the rest, by uniform swing from 2024, and that fits them badly.

Their politics differ. In the 2024 round one, the RN and its allies took 34.0% in metropolitan France but 15.4% overseas and 11.0% abroad. Overseas, most votes went to candidates labelled other right (21.7%), other left (21.4%) and regionalist (21.2%). Abroad, the left (34.5%) and Ensemble (32.7%) dominated. The overseas seats were won mostly by candidates outside the national blocs: in 2024, 8 by other-left candidates, 6 by regionalists and 5 by other-right candidates. Abroad, Ensemble won 9 of 11.

Two consequences follow. A national swing measured mainly in metropolitan France is applied, point for point, to seats where the bloc barely stands; if the RN gains three points nationally, it gains three in Guadeloupe too. And because regionalists are grouped with "others", most of the seats in the "Others and regionalists" row of the projection are overseas. Turnout is also far lower outside metropolitan France: 41.9% overseas and 36.6% abroad in the 2024 round one, against 68.8% in the metropole, so the 12.5%-of-registered bar is much harder for a third candidate to clear there. Nothing in the model is fitted to these seats specifically.

The simulation and the 80% ranges

The central projection is one run of the chain with our projection and every withdrawal applied. The ranges come from 1,000 simulated elections, each built as follows:

  1. Each bloc's national share is moved by a random error with a standard deviation of 3 points for the RN and the left, 2.5 for Ensemble, 2 for the mainstream right and 1 for the others.
  2. In each circonscription, each bloc's share receives independent local noise with a standard deviation of 2.5 points where the bloc had more than 3% in 2024, and 0.75 where it had less.
  3. Each possible withdrawal happens with probability 0.72.
  4. Every duel is decided by the duel model, as in the central projection.

The 80% range for a bloc runs from the 10th to the 90th percentile of its seats across the runs. A bloc has a majority in a run if it wins 289 seats or more.

Bloc2024PollsOur betSeats80% rangeMajority
Left (NFP and other left)29.733.033.6291204–31022%
Ensemble and centre22.015.015.27262–1420%
LR and other right11.015.016.29082–1120%
RN and allies33.231.831.110869–1470%
Reconquête0.82.82.300–00%
Far left1.11.00.900–00%
Others and regionalists2.24.33.71614–170%
Live National Assembly: round one by bloc and projected seats Computed from the database, polls to 19 November 2026.

The national standard deviations are set by judgement, not fitted. They are of the order of the final-poll misses by bloc (the RN's legislative misses of 4.3, 0.5 and 2.9 points have a root mean square of 3.0), with no extra allowance for the age of the polls. The simulation does not vary the duel model's coefficients, and it does not vary turnout. Given the out-of-sample record, the duel step is the largest source of uncertainty, and it is left out of the ranges. So the ranges are too narrow, probably by a good deal. No bloc has a majority in 78% of the runs. The direction of that finding is clear from the central projection and every lens; the exact figure should not be taken literally.

If the polls miss as they did before

The ranges show how far chance could move the result. The lenses ask something different: what happens if today's polls are wrong in the way they were wrong at a past election. Each replays one past election's final-poll miss, bloc by bloc, on today's legislative polls and runs every seat again.

ReadingLeftEnsembleLRRNReconquête
The polls as they stand28674831180
Our projection29172901080
If the polls miss like 200227881104990
If the polls miss like 200729168941080
If the polls miss like 20123195887970
If the polls miss like 20173188388720
If the polls miss like 202228767951130
If the polls miss like 20243007995880
Live Seats under each reading of today's legislative polls Computed from the database, polls to 19 November 2026.

91 seats change hands under at least one past miss. The Assembly page marks them, and the what-if tool lets a reader set their own error for each bloc. If we're wrong: replaying past polling misses on today's French polls treats the lenses in full.

By-elections

Ten by-elections have been held since the 2024 election. They are shown on the by-elections page and are not used in the model: each is one seat, usually with low turnout and a local story, and none was decided in round one.

Limitations

The duel model is the weak link. It is fitted on one election and does little better than the round-one leader out of sample. If withdrawals and transfers change, the projection will be wrong in the way pollsters' projections have been wrong.

The polls are old. The projection rests on polls taken in autumn 2025. If an election is called, the first new polls will matter more than anything here.

Uniform swing. It ignores incumbency and local candidates, both of which matter a great deal in French legislative elections, and it moves every seat by the same number of points regardless of the bloc's local strength.

Turnout is fixed. Holding 2024 turnout fixes how many candidates clear the 12.5% bar. Lower turnout would mean fewer three-way contests and fewer withdrawals.

Alliances are fixed. The model assumes the left stands as one list and the RN with the UDR, as in 2024. A split left, or a broader alliance on the right, would change round one in most seats.

Overseas and abroad. These 38 seats are projected by a method that fits them badly.

References

Butler, D. and Stokes, D. (1969). Political Change in Britain. London: Macmillan.

Code électoral, articles L126 (election in round one) and L162 (conditions for standing in round two).

Ministère de l'Intérieur, results of the legislative elections of 2022 and 2024 by circonscription, via data.gouv.fr (Licence Ouverte).

How to cite

Polling Forecast (2026). From votes to seats in the National Assembly. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/france-assembly. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like