Research · When the polls missed

Rutte pulls away, Wilders fades: the left-leaning panels behind the 2017 Dutch polling miss

Abstract. The final polls had the VVD only about five seats ahead of the PVV; Rutte won by 13 seats and 8.2 points. Every pollster understated the VVD, and the pollsters' own re-interviews found panels that leaned left plus a late move to Rutte during the row with Turkey.

Polling Forecast · 2 November 2026 · 6 min read · Netherlands, 2017 Tweede Kamer election, polling error, final polls, inquiry, pollster methods

Every election the polls got badly wrong →

Summary

For most of the winter before the Tweede Kamer election of 15 March 2017, Geert Wilders' PVV led the polls, and on average the final polls still had it only five seats behind Mark Rutte's VVD. The VVD won 33 seats and 21.3% of the vote; the PVV took 20 seats and 13.1%. The final polls put the VVD 3.6 points ahead of the PVV; the real lead was 8.2 points, a miss of 4.6 points. The pollsters' own checks after the vote found two causes: panels that leaned left and held too few VVD voters, and a late move to Rutte in the final days, during the diplomatic row with Turkey.

What the polls said

Dutch polls are quoted in seats out of 150. Five companies published final national polls on 13 or 14 March.

Pollster	Final release	VVD	PVV	CDA	D66	GL	SP	PvdA
Peil.nl (Maurice de Hond)	14 Mar	27	24	22	15	18	13	9
Ipsos (Politieke Barometer)	14 Mar	29	20	23	18	15	15	9
Kantar Public (TNS NIPO)	14 Mar	27	23	20	18	14	15	11
I&O Research	14 Mar	27	16	19	20	20	14	12
De Stemming (GfK/EenVandaag)	13 Mar	24	24	21	16	16	16	10
Average (seats)		26.8	21.4	21.0	17.4	16.6	14.6	10.2
Seat equivalent (%)		17.9	14.3	14.0	11.6	11.1	9.7	6.8
Result (seats)	15 Mar	33	20	19	19	14	14	9

The LISS panel also published a final estimate with the VVD on 24. Tom Louwerse's Peilingwijzer closed on 14 March with the VVD on 24–28 seats, the PVV on 19–22, CDA 19–21, D66 17–19, GroenLinks 16–18, SP 14–16 and PvdA 10–12. Three weeks earlier, on 22 February, Peilingwijzer had the PVV on 24–28 and the VVD on 23–27; on 30 November 2016 it had the PVV on 27–31. The story of the campaign was a PVV lead that faded, and the polls captured that. What they missed was how far the VVD would pull away.

What happened

The Kiesraad's result: VVD 2,238,351 votes (21.3%, 33 seats), PVV 13.1% (20), CDA 12.4% (19), D66 12.2% (19), GroenLinks 9.1% (14), SP 9.1% (14), PvdA 5.7% (9), ChristenUnie 3.4% (5), Partij voor de Dieren 3.2% (5), 50PLUS 3.1% (4), SGP 2.1% (3), DENK 2.1% (3), Forum voor Democratie 1.8% (2). Turnout was 81.9%.

Against the average final poll the VVD was 6.2 seats (3.4 points) short. The PVV was overstated by 1.4 seats (1.2 points), the CDA by 2.0 seats, GroenLinks by 2.6 seats (2.0 points) and the PvdA by 1.2. D66 was 1.6 seats short. Every pollster had the VVD as the largest party and every pollster had it too low: Ipsos was closest at 29, EenVandaag and LISS furthest at 24.

Why it happened: what the pollsters found

There was no formal inquiry. Most Dutch pollsters run their panels as standing samples, and after the election they asked the same people how they had actually voted.

NOS, 16 March 2017. The day after the vote NOS compared each final poll with the result: the VVD underestimated by at least four seats by everyone, by nine at EenVandaag and LISS; GroenLinks on an average of 17 against 14; the PVV on 21 against 20. Peilingwijzer had the order of the parties almost exactly right.
I&O Research, 1 May 2017. I&O re-interviewed its final sample (3,199 people: 2,618 from its own panel, 314 from PanelClix and 267 face-to-face interviews with Dutch people of migrant background; 86% response). The re-interview gave "more or less the same percentages" as the final poll. I&O concluded that late switching "hardly" explained its errors: its sample slightly over-represented GroenLinks and PvdA voters and under-represented VVD and PVV voters. Across all pollsters, it calculated, the final polls were on average 24 seats off the result in total; I&O's were 22.
NOS, 12 May 2017: "we had a lean to the left". Five of six pollsters (I&O, Kantar, Peil.nl, LISS and GfK/EenVandaag) shared their post-election re-measurements with NOS; Ipsos does not run them. Left-leaning voters were over-represented in most panels. The VVD and PVV together were about five seats too low; PvdA, GroenLinks and SP together about five seats too high; GroenLinks alone more than two seats too high on average. LISS and I&O showed the largest distortion, about five seats each. The pollsters judged that about half of the VVD's gap came from voters who decided in the last days to back the VVD, which they put down to Rutte's handling of the row with Turkey that weekend. Louwerse added that the VVD would still have been well ahead of the PVV without that late move, and that every pollster was low on the VVD, so it was not one house's chance error.

The two accounts differ on late swing: I&O found almost none in its own panel, the joint NOS review put half the VVD gap down to it. They agree on the sample: too many left-leaning respondents, too few VVD voters.

What our data shows

Our database holds the final polls of six firms (the five above and LISS). The VVD was understated by all six, by between 2.0 points (Ipsos) and 5.3 points (EenVandaag) in seat-equivalent terms. The PVV error split the field: Peil.nl and EenVandaag were 2.9 points too high, Kantar 2.3, LISS 0.9 and Ipsos 0.3, while I&O was 2.4 points too low. So the PVV overstatement in the average came from three houses; the VVD understatement came from all of them.

The left lineage, GroenLinks and PvdA added together, was 3.3 points too high in our average. Across the Dutch elections since 2002 in our data that lineage has been overstated at seven of eight, which fits the "lean to the left" the pollsters found. The VVD error is less stable: understated in 2012 (2.6 points), 2017 (3.9) and 2025 (2.3), overstated in 2023 (2.5). It moves with which party the right's tactical voters pick in the last days.

What the pollsters changed

I&O Research said it now knew, for almost every member of its panel, whether and how they had voted, and would use that when drawing samples and weighting results. It expected this to remove most of the over-representation of GroenLinks and PvdA voters and the under-representation of VVD and PVV voters.
The panel pollsters together told NOS on 12 May 2017 that they would investigate why their panels leaned left and would use post-election re-measurements after every election to adjust their panels, checking for example for too many older or highly educated members. Peter Kanne of I&O said knowing how every respondent voted would make later polls, municipal ones included, more accurate.

The polling field has since narrowed. Kantar Public stopped regular polling in 2022, and Ipsos bought I&O in January 2024.

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. The steady overstatement of the GroenLinks and PvdA lineage is the kind of lean it is built to catch. The VVD's swinging error is not, and for that the forecast relies on its simulated polling error, which allows a large party to move several points after the last poll, as the VVD did in 2017. The method and the backtests are in our paper on Dutch polling error, and each company's record is in our Dutch pollster ratings.

Sources

How to cite

Polling Forecast (2026). Rutte pulls away, Wilders fades: the left-leaning panels behind the 2017 Dutch polling miss. Retrieved 2 November 2026 from https://www.pollingforecast.com/research/missed-nl-2017. Live figures change as new polls arrive; the date above is when the text was last revised.