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

Measuring French polling error, 1965–2024

Abstract. We compare every pollster's final poll with the result at the presidential elections of 2002–2022 (by candidate family), the legislative elections of 2017–2024 (by bloc) and, more thinly, every presidential election since 1965. Round-one misses average about a point per family but change direction; round-two polls overstated the RN candidate at all three runoffs the RN reached; legislative polls overstated the RN and its allies three times out of three. Pollsters' seat projections between the legislative rounds missed by far more than the vote shares.

Polling Forecast · updated 9 November 2026 · 12 min read · France, polling error, final polls, candidate families, round two, legislative elections, seat projections, long view

Summary

This paper measures how far French opinion polls have missed, and sets out the evidence on which every other part of the France forecast rests. We compare each pollster's final poll before each round with the official result: 40 round-one and 37 round-two polls at the five presidential elections from 2002 to 2022, the smaller record of every presidential election back to 1965, 22 legislative round-one polls at the Assembly elections of 2017, 2022 and 2024, and the seat projections pollsters published between the legislative rounds.

Four findings carry the method. Presidential round-one polls have missed by about a point per candidate family on average, with single misses of up to five points, and in no steady direction. Round-two polls have overstated the Rassemblement national (RN) candidate at every runoff the RN has reached, by between 3.2 and 4.5 points. Legislative round-one polls overstated the RN and its allies at all three recent elections and understated the non-aligned right by about two points each time. And the seat projections published between the legislative rounds have missed by far more than the vote shares: the bloc that led round one came in below every pollster's range at each of the last three elections.

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.

Definitions

A final poll

A pollster's final poll is its last published voting-intention poll before the round in question, by fieldwork end date. Each pollster counts once per round. Where a pollster ran two series with the same final fieldwork date (Ifop in 2012 and Ipsos in 2022), only one counts. Polls of sub-populations, such as a survey of small-business owners or of a single community, are excluded. So are polls that were only reported, not published with figures. French law bans the publication of polls from the Friday midnight before polling day until the polls close, so most final polls finished their fieldwork within a week of the vote.

The error

For party or family \(p\) at election \(e\), let \(s_{j,p,e}\) be poll \(j\)'s share, \(\bar{s}_{p,e}\) the average over the \(n_e\) final polls, and \(r_{p,e}\) the official result as a share of valid votes (suffrages exprimés). The error is

\[ d_{p,e} = \bar{s}_{p,e} - r_{p,e}, \qquad \bar{s}_{p,e} = \frac{1}{n_e}\sum_{j=1}^{n_e} s_{j,p,e}. \tag{1} \]

The sign convention is the one used everywhere on the site: error = poll minus result. A positive error means the polls overstated the party. A party is measured at an election only if it appears in the result and in at least half the final polls. A candidate published as "under 0.5%" counts as zero.

In round two the shares are of the two-candidate vote, so the two errors are equal and opposite. When the RN candidate is overstated by 3.3 points, the opponent is understated by 3.3.

Summary measures

For each election we report the average miss per party: the mean absolute error across the families measured,

\[ \text{MAE}_e = \frac{1}{|P_e|}\sum_{p \in P_e} \left| d_{p,e} \right|. \tag{2} \]

This is the figure the leave-one-out tests compare. For each family we also keep the mean error \(\bar{d}_p\) across elections, which measures direction, and the root mean square error,

\[ \text{RMSE}_p = \sqrt{\frac{1}{n_p}\sum_{e} d_{p,e}^2}, \tag{3} \]

which measures size whatever the direction. The first decides whether a contest is corrected; the second sets the spread of the simulations (The 2027 presidential forecast: line-ups, runoffs and simulation).

Why candidate families

A British forecaster can track "the Conservatives" from 2015 to 2024. A French forecaster cannot track one label from 2002 to 2022. The Front national became the Rassemblement national in 2018; the RPR and UMP became Les Républicains; Emmanuel Macron's movement did not exist before 2016; Jean-Luc Mélenchon stood for the Front de gauche in 2012 and La France insoumise afterwards. Candidates also split families: in 2002 the mainstream right had both Jacques Chirac and Alain Madelin, and the far left had three candidates.

We therefore measure by candidate family. A family's share in a poll, or in the result, is the sum of its candidates' shares. The families we use are:

Family Candidates it includes (examples)
National Rally Jean-Marie Le Pen (2002, 2007), Marine Le Pen (2012–2022)
Republicans Chirac and Madelin (2002), Sarkozy (2007, 2012), Fillon (2017), Pécresse (2022)
Socialists Jospin (2002), Royal (2007), Hollande (2012), Hamon (2017), Hidalgo (2022)
Ensemble (Macron camp) Macron (2017, 2022)
Centre (UDF, MoDem) Bayrou (2002–2012)
France Unbowed Mélenchon (2012–2022)
Far left Laguiller, Besancenot, Poutou, Arthaud and others
Sovereigntists de Villiers (2007), Dupont-Aignan (2012–2022), Asselineau (2017)
Reconquête Zemmour (2022)

The Greens, the Communist Party, the hunting and rural party (CPNT), Bruno Mégret's MNR and a residual "Others" family make up the rest. The grouping is a judgement, and some cases are arguable: Bayrou's centre and Macron's camp are kept apart, because Bayrou stood against the mainstream right and left from a different position in 2007 and 2012 and backed Macron in 2017. Families matter most for the RN, whose line from father to daughter is what lets three runoffs be read as one series.

The evidence

Pollsters per election run from six in 2002 to eleven in 2022 for round one, and from three in 2002 to eleven in 2022 for round two. The 2002 runoff is the thinnest record: only CSA, Ipsos and Sofres are listed, and Sofres' "final" poll is one taken on election night after round one. No runoff polls from BVA, Ifop or Louis Harris for 2002 could be found.

Sources are graded on the site. From 2016 onwards most polls rest on the pollster's own release or its notice to the Commission des sondages, which every published electoral poll must file under the law of 19 July 1977. The notices from before 2016 are not online, so every poll from 2002 to 2012 rests on secondary sources, mainly the poll tables on Wikipedia, checked against the pollsters' published records where those survive. The checks are in The French data: results, maps, statistics, deputies and polls.

Presidential round one, 2002–2022

Party19651969197419811988199520022007201220172022Average
Republicans−1.6−4.0+0.9+1.5+3.6+2.4+0.2−2.8−0.7−0.7+3.1−0.2
Socialists−4.7+3.4+1.2−4.5+0.9−3.3+1.6−1.6−0.7+1.3+0.3+0.2
Others+1.9−0.3−0.8+0.3−0.1−0.10.00.0–−0.4−0.6−0.3
Far left–−0.1–+0.2−0.9−0.1+0.6+1.3+0.1+0.3+0.3+0.5
Centre (UDF, MoDem)+4.4+4.7−2.1−0.8−0.1–−0.6−0.7+1.1––−0.1
Communist Party–−3.8–+3.6+0.2+1.0+1.9+0.6––+0.3+0.9
Greens–––−0.4−1.3+0.2+0.8−0.2+0.1–+0.1+0.2
National Rally––––−2.4−1.3−3.9+3.3−2.0+0.9−0.2−0.4
Sovereigntists–––––+1.3–−0.2−0.1−0.7+0.2−0.2
France Unbowed––––––––+2.5−0.4−4.8−0.9
Hunting and rural party––––––−0.6+0.3–––−0.1
Ensemble (Macron camp)–––––––––−0.1−1.5−0.8
MNR––––––−0.1––––−0.1
Reconquête––––––––––+2.7+2.7
Live Presidential round one: final-poll average minus result, by family (points) Computed from the database, polls to 19 November 2026.

The typical miss is about a point per family. The average miss per party was 1.02 in 2002, 1.10 in 2007, 0.91 in 2012, 0.60 in 2017 and 1.26 in 2022. But the averages hide large single misses, and those misses do not keep one direction:

  • The far right was understated by 3.9 points in 2002, when Jean-Marie Le Pen reached round two against every poll's expectation (a miss examined by Durand, Blais and Larochelle, 2004), overstated by 3.3 in 2007, understated by 2.0 in 2012, and within a point in 2017 and 2022.
  • The mainstream right was understated by 2.8 points in 2007 and overstated by 3.1 in 2022.
  • Jean-Luc Mélenchon was overstated by 2.5 points in 2012 and understated by 4.8 in 2022.
  • Éric Zemmour was overstated by 2.7 points in 2022, his only election.

Only the far left leans consistently, overstated at all five elections, by between 0.1 and 1.3 points. The Communist Party was overstated at all three elections it was measured. Neither matters to who reaches round two.

The individual polls vary more than the average. In 2017 Elabe's final poll missed by 0.26 points per family and Le Terrain's by 1.31. In 2022 the best final poll (Odoxa) missed by 1.51 and the worst (Kantar) by 2.37; every pollster underestimated Mélenchon, and in the final week that miss was large enough to matter, since he finished 1.2 points behind Marine Le Pen for second place.

The test in Forecasting French elections from imperfect polls: an overview asks whether taking off half each family's average past miss would have helped. At four of the five elections it would have been further out, because the average of misses that change sign is close to zero for most families, and where it is not, the next election tends to go the other way. The round-one polls are therefore taken as published. Their errors still matter: each family's RMSE sets the spread of the round-one simulation.

Presidential round two, 2002–2022

Party1965 round two1969 round two1974 round two1981 round two1988 round two1995 round two2002 round two2007 round two2012 round two2017 round two2022 round twoAverage
Republicans−0.2−0.7––−0.5+0.9−3.2+0.7−1.1––−1.2
Socialists+0.2–+0.8+0.2+0.5−0.9–−0.7+1.1––+0.2
Centre (UDF, MoDem)–+0.7−0.8−0.2––––––––
National Rally––––––+3.2––+4.5+3.3+3.7
Ensemble (Macron camp)–––––––––−4.5−3.3−3.9
Live Presidential round two: final-poll average minus result (points) Computed from the database, polls to 19 November 2026.

Round two is where the pattern is. The RN candidate reached round two in 2002, 2017 and 2022, and each time the final polls put that candidate too high: by 3.2 points in 2002, 4.5 in 2017 and 3.3 in 2022. Every single final poll at those three runoffs overstated the RN. In 2017 the closest polls (Ifop and Ipsos) had Macron at 63 against a result of 66.1; the furthest (Kantar) had 59. In 2022 the closest (OpinionWay) had 57 against 58.6, and the furthest (AtlasIntel and Odoxa) had 53.

The two runoffs without an RN candidate missed by about a point, in opposite directions: the Republican candidate was overstated by 0.7 in 2007 and understated by 1.1 in 2012.

Why the RN has been overstated at runoffs is not something these data can settle. The possibilities include late movement against the RN once the runoff is set, differential turnout, and samples or weights that give too much room to RN voters. What matters for a forecast is that the direction has been the same three times out of three, and that the leave-one-out test agrees: correcting would have been closer at four of five runoffs, and further out only in 2007, the smallest miss and one with no RN candidate.

The gap between the top two

Round two is decided by the gap between two candidates, and qualifying for it by the gap between second and third. The error in a gap is the difference of two errors, so it can be larger than either. For each election we take each final poll's gap between the two leading families minus the real gap, and average:

Election Top two Real gap Polls' gap miss Worst poll
2002, round one Republicans / National Rally 6.9 +4.1 +6.1
2007, round one Republicans / Socialists 5.3 −1.1 −4.3
2012, round one Socialists / Republicans 1.5 +0.1 +2.1
2017, round one Ensemble / National Rally 2.7 −0.9 −2.7
2022, round one Ensemble / National Rally 4.7 −1.3 −3.7
2002, round two Republicans / National Rally 64.4 −6.4 −8.4
2007, round two Republicans / Socialists 6.1 +1.5 +3.9
2012, round two Socialists / Republicans 3.3 +2.2 +3.7
2017, round two Ensemble / National Rally 32.2 −8.9 −14.2
2022, round two Ensemble / National Rally 17.1 −6.6 −11.1

A positive miss means the polls had the leader too far ahead. At the three RN runoffs the polls understated the winner's margin by between 6.4 and 8.9 points on average, and the worst single poll by up to 14.2. At the two closer runoffs of 2007 and 2012 they overstated it, by 1.5 and 2.2 points. The 2012 runoff shows why this matters: the real gap was 3.3 points, and a miss of that size in the other direction would have changed the winner.

The long view, 1965–2022

French polling began with the 1965 presidential election, when Ifop's final poll caught Charles de Gaulle being forced into a runoff. We have sourced the final polls of every presidential election since, though coverage before 2002 is thin: one pollster in 1965 and 1981, two or three otherwise. They come from the pollsters' published records and contemporary press reports, each cited in the database, and are measured against the Conseil constitutionnel's results.

ElectionFinal pollsAverage miss per partyBiggest miss
196513.19Socialists −4.7
196922.69Centre (UDF, MoDem) +4.7
197421.25Centre (UDF, MoDem) −2.1
198111.62Socialists −4.5
198811.17Republicans +3.6
199531.22Socialists −3.3
200261.02National Rally −3.9
200761.10National Rally +3.3
201280.91France Unbowed +2.5
201790.60Socialists +1.3
2022111.26France Unbowed −4.8
Live Round one: final polls against the result, every election since 1965 Computed from the database, polls to 19 November 2026.
ElectionFinal pollsAverage miss per partyBiggest miss
196510.20Republicans −0.2
196920.71Centre (UDF, MoDem) +0.7
197420.81Centre (UDF, MoDem) −0.8
198110.24Centre (UDF, MoDem) −0.2
198820.48Republicans −0.5
199520.86Republicans +0.9
200233.21National Rally +3.2
200770.73Republicans +0.7
201281.11Republicans −1.1
201784.48National Rally +4.5
2022113.32Ensemble (Macron camp) −3.3
Live Round two: final polls against the result, every election since 1965 Computed from the database, polls to 19 November 2026.

Two things stand out. In round one, the size of the typical miss has not shrunk much in sixty years. Apart from the first two elections, with one and two polls, it has stayed between 0.6 and 1.7 points per family, and the 2022 miss (1.26) was no smaller than those of 1974 (1.25) or 1995 (1.22). This fits the cross-national finding that polling error has not become smaller over time (Jennings and Wlezien, 2018).

In round two, every runoff before 2002 missed by less than a point per candidate. The three large runoff misses in the whole record are the three RN runoffs. The long view is not used in the correction, which rests on 2002–2022 only, because the older record has too few polls per election and different families. It does put the round-two pattern in context: large runoff misses are not a general feature of French polling but, so far, a feature of runoffs against the RN.

Legislative round one, 2017–2024

Legislative polls are grouped by bloc: the united left (the NUPES in 2022, the Nouveau Front populaire in 2024, and the equivalent parties in 2017), the presidential camp, Les Républicains and the UDI, the RN and its allies (including Éric Ciotti's UDR in 2024), Reconquête, the far left, and the candidates the ministry labels as other left, other centre, other right and others. The grouping follows the ministry's own candidate labels (nuances).

Party200220072012201720222024Average
Others+1.5+0.30.0+1.4+2.7+1.6+1.3
Other right−0.7−1.3−0.6−2.1−2.2−2.1−1.5
Other left0.0−0.7−1.3−1.1−0.5−0.5−0.7
Far left+0.70.0+0.4+0.10.0−0.1+0.2
Republicans and UDI−3.7−0.5+0.2+1.4−0.9+0.1−0.6
New Popular Front (united left)+0.6+0.4−0.5−0.4+0.8+0.4+0.2
National Rally and allies+2.1+1.0+1.8+4.3+0.5+2.9+2.1
Other centre−0.1+1.2+0.7––−0.3+0.4
Ensemble (Macron camp)–––−1.8+1.0−0.6−0.5
Reconquête––––+1.6+0.7+1.1
Live Legislative round one: final-poll average minus result, by bloc (points) Computed from the database, polls to 19 November 2026.

The RN and its allies were overstated at all three elections: by 4.3 points in 2017, 0.5 in 2022 and 2.9 in 2024. In 2024 every one of the seven final polls put the RN bloc between 35 and 37 per cent; it took 33.2. The candidates labelled "other right" were understated by about two points each time (2.1, 2.2 and 2.1), and "others" overstated by between 1.4 and 2.7. The last two patterns are probably partly a matter of labels: pollsters ask about national parties, and many local right-wing candidates who stand without a party label are hard to prompt for. Whatever the cause, it has been steady.

With three elections, the leave-one-out test predicts each from the other two. Correcting was closer all three times, though only just in 2022 (1.13 against 1.14). The Assembly projection uses the correction (From votes to seats in the National Assembly).

In round one the gap between the leading two blocs was missed by −1.4 points in 2017, +0.1 in 2022 and +2.5 in 2024, when the polls had the RN too far ahead of the left.

Seat projections between the rounds

Between the two legislative rounds, pollsters publish seat ranges for each bloc. These are not voting-intention polls; they are the pollsters' own models of round two, applied to the round-one result. They have missed by far more than the vote shares.

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.

At each of the last three elections the bloc that led round one won fewer seats than every pollster's range. In 2017 the presidential camp was projected at about 437–467 seats and won 350. In 2022 the NUPES left was projected at about 161–206 and won 131, while the RN was projected at 26–48 and won 89. In 2024 the RN and its allies were projected at about 188–221 and won 142. In each case round one was already known, so the misses come from round two: who withdrew, and how the voters of eliminated candidates split. The 2024 miss is the clearest case. After round one, more than 200 qualified candidates withdrew, most of them to leave a single opponent to the RN, and the voters of the withdrawn candidates followed more often than the projections assumed.

A caveat on the table: where a pollster published a bloc as several party ranges, as Ipsos did in 2024, the bloc range is the sum of the components, which overstates its width.

This record is the main reason the Assembly projection on this site carries wide ranges, and the main reason we do not trust our own duel model more than its tests allow (From votes to seats in the National Assembly).

What the record supports

The record supports three different treatments, one per contest, and they are what the leave-one-out tests produce when the same rule is applied to each:

Contest Elections Direction of misses Average miss, polls Average miss, corrected Verdict
Presidential round one 5 changes sign 0.98 1.06 take as published
Presidential round two 5 RN overstated 3 of 3 2.57 1.86 correct
Legislative round one 3 RN overstated 3 of 3 1.21 0.97 correct

The correction is the same half of the average past miss in each case (How we read French polls: what we correct, what we don't, and why). What differs is whether the misses have leaned the same way often enough for the average to carry information. Where they have not, the size of past misses still enters the forecast as uncertainty.

Limits

Few elections. Five presidential elections and three legislative ones are little evidence. One new election could reverse any of these verdicts. The round-two correction rests on three runoffs with an RN candidate.

Secondary sources before 2016. The polls of 2002–2012 come from secondary tables, checked where possible. A transcription error in one of them would move a family's average by a fraction of a point.

Families are a choice. Measuring Marine Le Pen against her father's record, or Pécresse against Chirac's, assumes that polls treat a family's voters similarly across candidates. That is plausible for the RN and less so for the mainstream right, whose candidates have ranged from 31 per cent to under 5.

Final polls are not the whole story. These are errors in the last days before a vote. Polls taken months out miss by much more, for reasons that have little to do with how polls are made; From today to polling day: drift in French presidential polls treats that separately.

Raw figures. We cannot yet measure error at each stage of a French poll, as we do in Britain (Why we start from British pollsters' raw answers rather than the published headline), because only a handful of historical final polls have raw figures on file: OpinionWay in 2017 and 2022, Ifop in 2017 and Kantar in 2022.

References

Durand, C., Blais, A. and Larochelle, M. (2004). The polls in the 2002 French presidential election: an autopsy. Public Opinion Quarterly, 68(4), 602–622.

Jennings, W. and Wlezien, C. (2018). Election polling errors across time and space. Nature Human Behaviour, 2(4), 276–283.

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, results of the presidential elections 2002–2022 and the legislative elections 2017–2024, via data.gouv.fr (Licence Ouverte).

How to cite

Polling Forecast (2026). Measuring French polling error, 1965–2024. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/france-polling-error. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like