Netherlands General election15 May 2030in 1,271 days PRO largest party 91% D66 + VVD + CDA reach 76 0% Calendar Track record Research
Research · Netherlands
EN NL

Measuring Dutch polling error, 2006–2025

Abstract. We measure the error of the final polls at the Tweede Kamer elections from 2006 to 2025, all published the day before the vote: which polls count, how merged parties are measured, how seats-only polls are handled, the error party by party and on the gap between the two largest parties, each pollster's record and the long view. The centre-left (GroenLinks and the PvdA, now PRO) and the SP were overstated at six of seven elections; the largest misses belong to parties that rose late, the PVV in 2023 above all.

Polling Forecast · updated 9 November 2026 · 8 min read · Netherlands, polling error, final polls, PRO, GroenLinks, PvdA, PVV, VVD, late swing, pollsters

Summary

This paper sets out how we measure polling error in the Netherlands: which polls count as final polls, how parties that merged or split are measured, how the sources are graded, how the error is defined and what the record shows. At the seven Tweede Kamer elections from 2006 to 2025 the average of the final polls missed each party by between 0.7 and 1.3 points. That is less than in Italy or Poland, and the polls were taken the day before the vote. Two things stand out. GroenLinks and the PvdA, measured together as PRO, were overstated at six of the seven elections, and so was the SP. And the largest single misses belong to parties that moved late: the PVV was understated by 5.4 points in 2023, the VVD by 3.9 in 2017, D66 by 2.9 in 2021.

The record has three uses on the site: its direction sets the correction (Reading Dutch polls: the correction and its test); its size sets the width of the ranges (Simulating Dutch polling error: where the chances come from); and its pollster-by-pollster detail feeds the ratings (Rating Dutch pollsters: accuracy, the lead and openness).

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.

What is measured, and against what

The quantity that decides the seats is each party's share of the valid national vote, the votes cast for lists. Blank and invalid votes are not part of it, and the kiesdeler is calculated on it. We measure against that share, from the Kiesraad's results (The Dutch data: results, maps, statistics, members and polls).

Parties are measured by their line to the parties that will contest the next election:

  • PRO: GroenLinks plus the PvdA, added together, up to 2021; the joint GroenLinks-PvdA list in 2023 and 2025; Progressief Nederland, the merged party, from 2026.
  • CDA, VVD, D66, PVV, SP, CU, SGP, PvdD and 50PLUS as themselves.
  • FvD, JA21, BBB, Volt, DENK from 2021, and NSC in 2023 and 2025.

The DNA, formed by seven MPs who left the PVV in January 2026, has never stood and has no record. Adding GroenLinks and the PvdA together before 2023 is our choice: it measures whether the polls had the combined centre-left vote right, which is what matters for PRO's correction. It does not say how well each party was polled on its own.

Definitions

Let \(v_{p,t}\) be party \(p\)'s share at election \(t\) and \(y_{j,p,t}\) its share in final poll \(j\). The error of a poll is \(e_{j,p,t} = y_{j,p,t} - v_{p,t}\): poll minus result, so a positive error means the polls overstated the party. The industry error is the error of the simple average of the final polls,

\[ e_{p,t} = \bar{y}_{p,t} - v_{p,t}, \qquad \bar{y}_{p,t} = \frac{1}{n_{p,t}}\sum_{j} y_{j,p,t}, \tag{1} \]

with each pollster counting once. The average miss per party at an election is

\[ \text{MAE}_t = \frac{1}{P_t}\sum_{p} \left| e_{p,t} \right|, \tag{2} \]

and each party's record is summarised by its mean error (direction) and root mean square (size):

\[ \bar{e}_p = \frac{1}{T_p}\sum_{t} e_{p,t}, \qquad \text{RMS}_p = \sqrt{\frac{1}{T_p}\sum_{t} e_{p,t}^{2}}, \qquad \text{RMS}_p^2 = \bar{e}_p^{\,2} + s_p^2. \tag{3} \]

The gap error on the two parties that finished first and second, \(a\) and \(b\), is \(e^{\text{gap}}_t = e_{a,t} - e_{b,t}\); a positive value means the polls put the winner too far ahead.

The evidence: final polls the day before

What counts as a final poll

A pollster's final poll is its last poll before the election. Dutch law sets no ban on publishing polls, and the pollsters publish their last figures on the eve of the vote:

ElectionFinal pollsMedian days before the voteClosest to the voteFurthest
20022212
20063112
20101111
20123111
20177112
20214111
20233111
20253111
Live When the final polls were taken: days between the end of fieldwork and polling day Computed from the database, polls to 18 November 2026.

Fieldwork ends one or two days before polling day. A Dutch final poll is as late as a British one, and much later than a Spanish (about ten days) or Italian (about eighteen) one.

Seats, not shares

Dutch pollsters lead with seats. Where a final poll published vote shares, we use them. Where it published only seats, as Peil.nl always does and most pollsters did before 2021, we take seats ÷ 150 as the share. That rounds each party to the nearest 0.67 points, and it gives the parties with seats the share that small lists without seats really won, about 1.4 points in 2025. Polls that publish only seats: deriving shares from seat headlines measures the cost: for polls that published both, the two figures differ by about a quarter of a point per party.

The elections

The database holds 26 final polls at the 8 elections from 2006 to 2025, and two more for 2002, which appear only in the long view. We found no final poll for 2003, and only one for 2010. The pollsters are Peil.nl (Maurice de Hond), Ipsos and I&O Research (merged in 2025 as Ipsos I&O), Verian (NIPO, TNS NIPO and Kantar Public under earlier names), and, in 2017 only, De Stemming, the EenVandaag panel and the LISS panel.

Grading the source

Every figure carries a source grade. Grade A is the pollster's own report; grade B the commissioning outlet's article; grade C a secondary transcription, here Wikipedia's poll tables. Old releases have mostly gone from the pollsters' sites, so only the 2025 polls of Ipsos I&O and Verian and the 2023 Ipsos seats are grade A; the other final polls are grade C. Every primary figure we could read matched the Wikipedia row.

A worked example: 2025

The 2025 election, held on 29 October after the PVV left the Schoof cabinet, had three final polls, all ending on 28 October:

Pollster Published D66 PVV VVD PRO CDA JA21 FvD
Peil.nl seats only 14.67 15.33 14.67 15.33 13.33 6.00 4.67
Verian shares 15.8 18.5 10.3 15.9 12.4 5.7 3.9
Ipsos I&O shares 15.4 14.8 10.9 15.0 12.7 7.4 4.0
Average 15.29 16.21 11.96 15.41 12.81 6.37 4.19
Result 16.94 16.66 14.24 12.79 11.79 5.95 4.54
Error −1.6 −0.4 −2.3 +2.6 +1.0 +0.4 −0.4

Table: Final polls and result, Tweede Kamer 2025, national vote shares, points. Peil.nl's figures are seats ÷ 150. PRO is the GroenLinks-PvdA list.

In seats: the result was D66 26, PVV 26, VVD 22, GroenLinks-PvdA 20, CDA 18. Every final poll had GroenLinks-PvdA on 23 to 25 seats, three to five too many. Verian had the VVD on 16 and Ipsos I&O on 17; it won 22. Peil.nl had the VVD exactly, from seats only. D66's late rise was caught in direction by all three, but not in size. The average missed by 0.68 points per party, the smallest miss in the record.

The record, 2006–2025

Party20022006201020122017202120232025Average
CDA−5.9+0.4+2.4−0.3+1.2+1.3−0.2+1.00.0
CU+1.1+0.6+0.8+0.4+0.5+0.6+0.4−0.1+0.5
D66+0.6−0.4−0.3−0.5−0.6−2.90.0−1.6−0.7
PRO (GL + PvdA)+3.9+1.8+1.0−0.9+3.3+2.9+1.4+2.6+2.0
SGP−0.40.0+0.3−0.5−0.20.0−0.3−0.1−0.2
SP−0.6+0.5−0.5+4.1+1.1+1.4+0.4+1.1+0.9
VVD+1.2−0.2+1.5−2.6−3.9+0.5+2.5−2.3−0.4
PvdD–−0.50.0−0.2−0.10.0+0.90.00.0
PVV–−2.4−4.1+1.5+1.4+2.1−5.4−0.4−1.1
50PLUS–––−0.1+0.3−0.6−0.2+0.3−0.1
BBB–––––−0.3−0.9+0.1−0.4
DENK–––––−0.6+0.3+0.1−0.1
FvD–––––−1.5+0.4−0.4−0.5
JA21–––––−0.9+0.2+0.4−0.1
Volt–––––−0.4+0.6+0.4+0.2
NSC––––––+0.10.00.0
Live Tweede Kamer: final-poll average minus result, by party (points) Computed from the database, polls to 18 November 2026.

Four points stand out.

The centre-left has been overstated. PRO's predecessors were overstated at every election but 2012: by 3.3 points in 2017, 2.9 in 2021 and 2.6 in 2025. The SP was overstated at every election but 2010, by 4.1 points in 2012.

Late movers are understated. The PVV in 2023 (5.4 points, after the final debates), the VVD in 2017 (3.9) and D66 in 2021 (2.9) all rose in the last days and finished above every final poll. The direction of these misses depends on which party moved, so they cancel across elections.

The PVV and the VVD change sign. The PVV was understated by 2.4 to 5.4 points in 2006, 2010 and 2023 and overstated by 1.4 to 2.1 in 2012, 2017 and 2021. The VVD was understated by 2.3 to 3.9 points in 2012, 2017 and 2025 and overstated by 1.5 and 2.5 in 2010 and 2023. A correction learns little from them.

The small parties are polled closely. The CU, the SGP, the PvdD, 50PLUS, DENK and Volt have rarely been missed by more than a point.

Using equation (3):

Party Elections Mean error RMS error Share of RMS² that is bias
PVV 7 −1.04 2.94 13%
VVD 7 −0.64 2.27 8%
PRO 7 +1.73 2.17 63%
SP 7 +1.16 1.77 43%
D66 7 −0.90 1.30 48%
CDA 7 +0.83 1.21 47%
CU 7 +0.46 0.53 75%

Table: Summary of industry errors, 2006–2025, points. The last column is \(\bar{e}_p^2 / \text{RMS}_p^2\).

For PRO, the CU and to a lesser degree D66, the CDA and the SP, a large part of the error is steady. For the PVV and the VVD almost none of it is. A correction that takes off half the mean helps where the steady part is large, and does little harm where it is small, because there the mean is close to zero. That is what the test finds (Reading Dutch polls: the correction and its test).

The gap between the two largest parties

The largest party usually leads the formation, so who comes first matters.

ElectionTop twoReal gapPolls' gap missWorst poll
2002CDA / PRO (GL + PvdA)5.9−9.9−11.2
2006CDA / PRO (GL + PvdA)0.7−1.4−3.4
2010PRO (GL + PvdA) / VVD5.8−0.5−0.5
2012PRO (GL + PvdA) / VVD0.6+1.6+2.1
2017VVD / PRO (GL + PvdA)6.5−7.1−9.8
2021VVD / D666.8+3.4+5.8
2023PVV / PRO (GL + PvdA)7.7−6.9−7.7
2025D66 / PVV0.3−1.2−3.0
Live The gap between the two largest parties: final polls' average error and the worst poll (points) Computed from the database, polls to 18 November 2026.

Before 2023 PRO is GroenLinks and the PvdA added together, so a "top two" containing it is our lineage, not a party that stood. Two elections stand out. In 2017 the polls had the VVD and the combined centre-left close; the VVD won by 6.5 points. In 2023 the polls had the PVV and GroenLinks-PvdA level; the PVV won by 7.7. In 2025 the gap between D66 and the PVV was 0.3 points, and the average poll had the PVV slightly ahead.

Pollster by pollster

Pollster Elections Average miss per party
Ipsos I&O 1 0.98
Peil.nl 6 1.06
Ipsos 6 1.08
I&O Research 3 1.21
Verian 5 1.22
De Stemming 1 1.72
EenVandaag Opiniepanel 1 1.72
LISS panel 1 1.83

Table: Average miss per party of each pollster's final polls, 2006–2025, points.

The three long-running pollsters are close to each other, between 1.06 and 1.22, next to a spread between elections of 0.68 to 1.26. Peil.nl, which publishes only seats, is not worse for it. The ratings (Rating Dutch pollsters: accuracy, the lead and openness) use these records, pulled towards a poor Dutch final poll.

How the record is used

Our rule takes off half the average miss and keeps the correction only if it would have improved past forecasts, each election predicted without knowing its result. In the Netherlands it does, by a small margin: 1.09 points per party against 1.12 (Reading Dutch polls: the correction and its test). The record also sets the size of the simulation's error for each party (Simulating Dutch polling error: where the chances come from): the PVV's typical miss of about three points gives it the widest range.

The long view

ElectionFinal pollsAverage miss per partyBiggest miss
200221.96CDA −5.9
200630.75PVV −2.4
201011.21PVV −4.1
201231.11SP +4.1
201771.26VVD −3.9
202141.08PRO (GL + PvdA) +2.9
202330.89PVV −5.4
202530.68PRO (GL + PvdA) +2.6
Live Final polls against the result, every Tweede Kamer election we could source Computed from the database, polls to 18 November 2026.

2002, the election after Pim Fortuyn's murder, had the largest miss in the record: the CDA was understated by 5.9 points. It is shown for context and not used in the correction, because the parties of 2002 (the LPF above all) have little to do with today's. For 2003 we found no final poll.

Limits

Seven elections, three pollsters. Every figure here rests on seven numbers per party, fewer for the newer parties, and on three pollsters at most elections.

Lineage. PRO before 2023 is two parties added together. The PVV's record belongs to a party that has since split.

Seats-derived shares. Most final polls before 2021 published only seats. Their shares are seats ÷ 150, which rounds and gives the parties with seats the votes of lists without them.

Secondary sources. Most final polls are Wikipedia transcriptions; the primary figures we could read all matched.

Final polls only. These errors say nothing directly about a poll taken a year out; that is From today to polling day in the Netherlands.

References

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

Shirani-Mehr, H., Rothschild, D., Goel, S. and Gelman, A. (2018). Disentangling bias and variance in election polls. Journal of the American Statistical Association, 113(522), 607–614.

How to cite

Polling Forecast (2026). Measuring Dutch polling error, 2006–2025. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/netherlands-polling-error. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like