Research · When the polls missed

Thirteen per cent in the polls, second place on the night: Le Pen and the 21 April 2002 polling miss

Abstract. The final polls had Jospin on 17.8% and Le Pen on 13.0%; Le Pen took 16.9% and knocked Jospin (16.2%) out of the run-off. Every institute understated the far right while the outgoing left's smaller candidates were overstated, and the Commission des sondages pointed to telephone samples, cautious political weighting and institutes smoothing their published figures.

Polling Forecast · 2 November 2026 · 7 min read · France, 2002 presidential election, first round, polling error, final polls, inquiry, pollster methods

Every election the polls got badly wrong →

Summary

On 21 April 2002 Jean-Marie Le Pen took second place in the first round of the presidential election with 16.9% and Lionel Jospin, the Socialist prime minister, was knocked out with 16.2%. The final polls had Le Pen on 13.0% on average and Jospin on 17.8%, a lead of almost five points for Jospin that turned into a deficit of 194,600 votes. Le Pen was understated by 3.9 points and the left's smaller candidates were overstated. The polls caught the direction (Le Pen rising, Jospin slipping) but not the size, because their political weighting of a far-right vote that respondents under-declared by telephone was too cautious, and because many voters decided late in a field of sixteen candidates.

What the polls said

The law of 19 February 2002 had just shortened the publication ban to the eve and the day of each round, so polls could appear up to Friday 19 April. The last poll of each of the six main institutes:

Pollster	Fieldwork end	Chirac	Jospin	Le Pen	Laguiller	Bayrou	Chevènement
CSA	18 April	19.5	18.0	14.0	7.0	6.0	6.5
Sofres	18 April	19.5	18.0	12.5	6.5	6.0	6.5
Ipsos	18 April	20.0	18.0	14.0	7.0	6.0	6.5
BVA	18 April	19.0	18.0	14.0	8.0	6.0	6.0
Louis Harris	17 April	20.0	18.0	13.0	8.0	7.0	6.0
Ifop	13 April	20.0	16.5	10.5	7.0	6.5	8.5
Average		19.7	17.8	13.0	7.3	6.3	6.7

Five of the six had Jospin on exactly 18%. Ifop's last published poll was five days older than the rest, and it had Le Pen lowest. No poll had Le Pen ahead of Jospin, and no poll had them closer than four points.

What happened

The Conseil constitutionnel proclaimed Chirac 19.88% (5,665,855 votes), Le Pen 16.86% (4,804,713) and Jospin 16.18% (4,610,113). Bayrou had 6.84%, Laguiller 5.72%, Chevènement 5.33%, Mamère 5.25%, Besancenot 4.25% and Hue 3.37%. Abstention was 28.40%.

Against the average, Chirac was almost exact (0.2 points short). Le Pen was 3.9 points short and Jospin 1.6 points over. The error did not stop with them. Robert Hue was overstated by 1.9 points, Laguiller by 1.5, Chevènement by 1.3 and Mamère by 0.8. Taken together, Jospin, Hue, Mamère and Taubira, the candidates of the outgoing left coalition, polled 30.6% and won 27.1%. Le Pen and Bruno Mégret together polled 15.3% and won 19.2%. On the margin between the top two in the result, Chirac and Le Pen, the polls were 3.6 points out. On the margin that decided the run-off place, Jospin over Le Pen, they were 5.4 points out.

Why it happened: what the pollsters found

Commission des sondages, report on the 2002 elections. The regulator checked all 193 polls on the presidential election from April 2001 to 5 May 2002. It put the miss down to the method and the conditions rather than any single institute. Telephone polls could no longer reach people who used only mobiles; refusal rates were rising; and among those who did answer, the share giving no vote intention was rising too, so published figures rested on a few hundred people. It had warned in a communiqué of 18 December 2001 that scores under about 5% were not meaningful, and that "political" weighting was hardest for marginal or extremist candidates. Its most serious finding was what it called scores "hors fourchette": institutes publishing figures outside the range produced by their own weighting schemes, chosen on the basis of qualitative studies, the political context or other institutes' figures. It cited one second-round poll that published Jospin at 51% when its own weightings gave 54.3% to 56.7%. The commission said its pre-election warning that polls were not forecasts had been borne out on 21 April, and complained that France 2 had refused to broadcast it.

Durand, Blais and Larochelle, presented at Université Laval in February 2003 (published in 2004). They found Le Pen understated by three to six points depending on the institute, well outside sampling error, and named several causes at once: fieldwork of about 2.2 days on average, against six or seven in Canada, which favours the easily reached; under-coverage of unlisted numbers and of older people without their own phone; heavy use of quotas, past-vote weighting and the allocation of undecided voters; and a herding effect, the "effet Saint-Guillaume", among pollsters trained at the same school who smoothed toward each other. They asked the institutes to publish their weighting methods and their unweighted results.

Jean-Paul Grémy (2011) went back over the twelve polls published in the last two weeks. Le Pen rose steadily from 9.5% (Ifop, 11–12 April) to 14% (Ipsos, 17–18 April); a straight-line extrapolation from the fieldwork dates gave him about 17% on the day. In Ipsos's last poll 41% of respondents said they could still change their minds.

The pollsters, looking back. Before the Assemblée nationale's commission of inquiry in 2025, Frédéric Dabi of Ifop said that "in the telephone era there were sometimes wild gaps between a raw result and a weighted one, for Jean-Marie Le Pen for example", and that in 2002 all the polls had shown Chirac stalling, Jospin falling and Le Pen rising. Brice Teinturier of Ipsos said that in the 1980s and 1990s the past vote recalled in telephone samples was so far from reality that weighting had to do a great deal of work.

What our data shows

Every one of the six institutes understated Le Pen, by between 2.9 points (CSA, Ipsos, BVA) and 6.4 points (Ifop, whose poll was the oldest). The four polls that finished on 18 April averaged 13.6% for Le Pen, still 3.2 points short, so the miss was not only a matter of stale fieldwork. It was also not a late surge that came from nowhere: the trend was in the data, and the institutes stopped short of where it was heading.

The pattern across French presidential first rounds in our database fits the commission's account. The National Front candidate was understated in 1988 (by 2.4 points, from one surviving poll), in 1995 (1.3) and in 2002 (3.9), the largest miss on that family we hold. In 2007, after the shock, the polls overstated Le Pen by 3.3 points, which looks like a correction that went too far. Since then the misses have been smaller and have changed sign: understated by 2.0 in 2012, overstated by 0.9 in 2017 and almost exact in 2022 (0.2 short) for Marine Le Pen.

The other half of 2002 is the left. Our average had the outgoing coalition's four candidates 3.5 points too high in total. Jospin's own error was 1.6 points; most of the rest came from Hue (1.9) and Mamère (0.8), partly offset by Taubira, who was 0.7 points short. Chevènement, outside the coalition, was 1.3 points too high as well. A small overstatement for the left and a large understatement for the far right, in a race decided by 0.7 points, was enough.

What the pollsters changed

The commission tightened its rules. In its 2002 report it asked institutes to supply every element from outside the survey used in their adjustments, and asked the press to publish the range of weighted results beside the chosen figure. Most institutes, at a meeting on 17 May 2002, found the idea sound but said newspapers would not accept it. By 2007 the commission required institutes to keep the same method for their indicators throughout a campaign and to justify any figure outside the range. It now requires a reference column so that each candidate's figure comes from the same estimator.
New methods arrived quickly. In 2007 Ipsos ran a daily rolling barometer from early March, and OpinionWay presented its polls of online-only samples to the commission.
The move online. Over the following decade the institutes moved from telephone to online panels. In its 2024 report the commission noted that predictive accuracy had "progressed considerably" with that change; Teinturier told the 2025 inquiry that past-vote recall in online panels is close enough to reality that weighting is much less brutal than it was.
The law changed. The law of 25 April 2016 requires every published poll to state its margins of error, and the institutes' notices are now published on the commission's website.

What it means for our forecast

The 2002 miss is in the record our French presidential forecast learns from. In round one, the far-right family's past misses set how widely its share can move in our simulations. In round two we apply a correction learned from the three runoffs with an RN candidate, under our usual rule: half the average past miss, kept only because it would have been closer to the result at four of five runoffs when each was predicted without knowing its result. Small candidates are the second lesson of 2002: a few points of overstatement spread across several of them can decide who reaches the run-off, which is why we read a close race for second place as close even when every poll has the same candidate ahead. 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). Thirteen per cent in the polls, second place on the night: Le Pen and the 21 April 2002 polling miss. Retrieved 2 November 2026 from https://www.pollingforecast.com/research/missed-fr-2002-pres. Live figures change as new polls arrive; the date above is when the text was last revised.