Research · When the polls missed

Wilders' last-minute landslide: how the 2023 Dutch polls missed the PVV by nine seats

Abstract. The final polls showed a three-way tie, with the PVV on 27 to 29 seats; it won 37 seats and 23.5% as undecided and tactical voters on the right broke for Wilders in the last two days. All three pollsters missed in the same direction, the mark of a late swing rather than one firm's method.

Polling Forecast · 2 November 2026 · 6 min read · Netherlands, 2023 Tweede Kamer election, polling error, final polls, inquiry, pollster methods

Summary

On the eve of the Tweede Kamer election of 22 November 2023 the final polls showed a three-way tie: the PVV, GroenLinks-PvdA and the VVD each on 24 to 29 seats. Geert Wilders' PVV won 37 seats and 23.5% of the vote, eight to ten seats more than any final poll gave it, and finished twelve seats clear of GroenLinks-PvdA. The final polls put the PVV 0.9 points ahead of GroenLinks-PvdA; the real lead was 7.7 points, a miss of 6.9 points. The cause was a late swing: undecided and tactical voters on the right, many of them NSC and BBB supporters a few weeks earlier, broke for Wilders in the last two days, faster than any poll could record.

What the polls said

Three companies published regular national polls in the final week. All three released their last numbers on 21 November, the day before the vote.

Pollster	Fieldwork end	PVV	GL-PvdA	VVD	NSC	D66	BBB	CDA
Peil.nl (Maurice de Hond)	21 Nov	29	28	26	19	8	6	6
I&O Research	21 Nov	28	27	27	21	9	5	4
Ipsos/EenVandaag	21 Nov	27	24	29	19	11	6	4
Average (seats)		28.0	26.3	27.3	19.7	9.3	5.7	4.7
Result	22 Nov	37	25	24	20	9	7	5

Dutch polls are quoted in seats. I&O and Ipsos also published vote shares; Peil.nl publishes seats only, so its share below is seats divided by 1.5.

Vote share (%)	PVV	GL-PvdA	VVD	NSC
I&O Research	17.7	17.7	17.4	13.8
Ipsos/EenVandaag	17.1	15.1	18.4	12.5
Peil.nl (seat equivalent)	19.3	18.7	17.3	12.7
Average	18.0	17.2	17.7	13.0
Result	23.5	15.8	15.2	12.9

Tom Louwerse's Peilingwijzer, the poll of polls NOS publishes, closed on 21 November with the VVD on 26–30 seats, the PVV on 25–29, GroenLinks-PvdA on 23–27 and NSC on 19–23. The day before, the PVV range had been 22–28. The movement was visible; its size was not.

What happened

The Kiesraad's result: PVV 2,450,878 votes (23.5%, 37 seats), GroenLinks-PvdA 15.8% (25), VVD 15.2% (24), NSC 12.9% (20), D66 6.3% (9), BBB 4.7% (7), CDA 3.3% (5). Turnout was 77.7%.

Against the average final poll the PVV was 5.5 points and nine seats short. The VVD was overstated by 2.5 points and three seats, GroenLinks-PvdA by 1.4 points and one seat. NSC was almost exactly right at 13.0 against 12.9. Every other party was within about a seat. The whole miss sits in one party's share and the VVD's loss to it.

Even the Ipsos exit poll for NOS and RTL, taken among people leaving polling stations, had the PVV on 35, two seats short.

Why it happened: what the pollsters found

No Dutch body holds a formal inquiry into polling, and none was held after 2023. What the pollsters said, before and after, points one way.

The electorate was unsettled to the last day. On 21 November Sjoerd van Heck of Ipsos/EenVandaag told NOS that "just under half" of voters were certain how they would vote, a far lower share than at the previous election, and that I&O found fewer still. I&O's own final release said about two in three voters had not firmly decided. Both Van Heck and Peter Kanne of I&O named tactical voting on the left and the right as the factor that could decide the result.
Tactical intent on the right was measured but not converted. EenVandaag's panel of 30,000 (reported 20 November) found 43% of PVV supporters ready to vote tactically, the second-highest share after GroenLinks-PvdA. A Volkskrant survey found a quarter of BBB voters could still switch to Wilders and a fifth of NSC voters were considering a late switch.
The debates moved votes late. Maurice de Hond, writing on 18 November as the PVV jumped from fourth to joint first in his poll, credited Wilders' appearances in the televised debates and his focus on migration, an issue he said the other parties had left to him.
The academic election study confirmed the flows. First results from the Dutch Parliamentary Election Study (NKO), published on 21 December 2023 by Marcel Lubbers, Kristof Jacobs, Niels Spierings and Remko Voogd, found that 40% of people who in early November intended to vote BBB voted PVV, and about a quarter of those who had intended to vote NSC did the same. The PVV also drew people who had stayed at home in 2021, the group polls find hardest to reach.

The pollsters' reading, then, is late swing on top of a panel that could not see the final 48 hours, not a structural bias against PVV voters.

What our data shows

Our database holds all three final polls. Each pollster understated the PVV: Peil.nl by 4.2 points, I&O by 5.8 and Ipsos by 6.4. A spread of two points between houses, all missing in the same direction by far more than sampling error, is the signature of movement after fieldwork, not of one company's method.

The PVV's record in our Dutch series argues against a fixed anti-PVV bias. The final polls understated it in 2006 (2.4 points), 2010 (4.1) and 2023 (5.5), overstated it in 2012 (1.5), 2017 (1.4) and 2021 (2.1), and were close in 2025 (0.5). The error follows momentum: when Wilders is rising in the last week the polls lag him, when he is falling they lag that too. The 2023 PVV error is the second-largest single-party miss in our Dutch series since 2002, behind only the CDA's 5.9 points in 2002.

Two patterns from the wider series show up here as well. The GroenLinks and PvdA lineage was overstated again (1.4 points), as it has been at seven of the eight elections since 2002 in our data. And the VVD error swung sign: overstated by 2.5 in 2023 after being understated by 3.9 in 2017. In both years the party that became the right's vehicle for "largest party" voting gained in the last days.

What the pollsters changed

We found no published method change from any of the three firms after 2023. The changes were structural. In January 2024 Ipsos bought I&O Research, merging two of the country's three regular pollsters into Ipsos I&O. That matters for aggregation: Peilingwijzer had leaned on I&O since Kantar Public stopped regular polling in 2022, and it has excluded De Hond's polls since 2019, so the Dutch poll average now rests on fewer independent houses than at any recent election. At the 2025 election our data shows the PVV within half a point, while the late-rising party of that campaign, D66, was understated by 1.7.

What it means for our forecast

Our correction rule adds half of each party's average past miss to its polling, and keeps the correction only if doing so would have improved forecasts of past elections made without knowing their results. For the Netherlands it passes: with 2002 included, the average error falls from 1.12 to 1.09 points a party and the corrected forecast is closer at five of eight elections, so it is applied. It cannot catch a 2023-type miss on its own, because the PVV's error changes sign from one election to the next. What protects the forecast is the simulated polling error, which lets a single party's share move by several points after the last poll, as the PVV's did. The method and the backtests are in our paper on Dutch polling error, and each company's record is on our Dutch pollster ratings.

Sources

How to cite

Polling Forecast (2026). Wilders' last-minute landslide: how the 2023 Dutch polls missed the PVV by nine seats. Retrieved 2 November 2026 from https://www.pollingforecast.com/research/missed-nl-2023. Live figures change as new polls arrive; the date above is when the text was last revised.