What we don't know yet: limitations of the Dutch forecast
Abstract. Every known weakness of the Dutch forecast in one place: a correction that passes narrowly and rests on PRO and the SP, late movers no average can foresee, parties that merge and split, seats-only polls, three pollsters, independent errors, coalition statements that may not last, the date and the pull-back, and the map. For each, how much it matters and what would change our mind.
Summary
The Dutch forecast has one part that is close to exact and several that are not. The seat count reproduces the last three elections to the seat, and with one national district there is no local projection to get wrong. The rest carries the usual weaknesses and some Dutch ones: a correction that passes narrowly and rests on two parties, a record dominated by late movers that no average can foresee, parties that merge and split between elections, polls that publish seats instead of shares, three pollsters, and coalition statements that may not survive the next campaign.
This paper sets out each weakness, how much we think it matters, and what evidence would change our mind. The British counterpart is What we don't know yet: limitations and open questions in our UK forecast; the method is in Reading Dutch polls: the correction and its test.
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.
- Only elections from 2000 shape the forecast (28 September). Earlier elections stay in the long view and on the history pages but no longer feed any test, correction or error size. In the Netherlands the gemeente swing is now tested on the eight pairs of elections from 2002; proportional swing was closer at all eight, so it stays.
- Pollster grades rescaled (13 October) (Rating Dutch pollsters: accuracy, the lead and openness).
A correction that passes narrowly
The correction takes off half each party's average past miss because, with each election predicted from the others, it was closer on average:
| Election | Polls' average miss | Corrected, leave one out | Verdict |
|---|---|---|---|
| 2002 | 1.96 | 2.10 | further out |
| 2006 | 0.75 | 0.53 | closer |
| 2010 | 1.21 | 1.18 | closer |
| 2012 | 1.11 | 1.18 | further out |
| 2017 | 1.26 | 1.16 | closer |
| 2021 | 1.08 | 1.08 | level |
| 2023 | 0.89 | 0.87 | closer |
| 2025 | 0.68 | 0.59 | closer |
| Average | 1.12 | 1.09 |
It passes by about six hundredths of a point per party, on seven elections, and it is further out at two of them. Most of its effect is two lineages: PRO, overstated at six of seven elections, loses −1.0 points today, and the SP about 0.6. For the PVV and the VVD, whose misses change sign, the correction does slightly more harm than good.
What would change our mind. The next election. The rule is checked every time the forecast is built, and if the correction stopped beating the polls on average it would switch off without any change to the method.
Late movers
The largest Dutch misses are parties that rose in the last days of the campaign and finished above every final poll: the VVD in 2017, D66 in 2021 and 2025, the PVV in 2023. Final polls are taken the day before the vote, so this is not movement hidden by a ban; it is voters deciding late, and polls unable to measure a move that is still under way. A historical average cannot say which party will move next, and in the record the late movers are different parties each time, so their misses mostly cancel in the correction and widen the spreads in the simulation.
What would change our mind. A measure of late movement that works across elections, such as the change in the polls over the last week, tested the same way as the correction. With seven elections we have not tried one.
Parties that merge and split
GroenLinks and the PvdA stood apart until 2021, together in 2023 and 2025, and merged into PRO in 2026. The forecast measures them together throughout, so PRO's correction is learned partly from two separate parties. The PVV lost seven MPs to the DNA in January 2026; the PVV's record belongs to the undivided party, and the DNA has none. NSC won 20 seats in 2023 and none in 2025; the BBB went from 7 to 4. A correction learned from parties' pasts assumes the parties are the same as they were, and in the Netherlands they often are not.
What would change our mind. Nothing but more elections with the same parties.
Seats instead of shares
Peil.nl publishes only seats, and so did most final polls before 2021. Their shares are seats ÷ 150, which rounds each party to the nearest 0.67 points and gives the parties with seats the votes of the lists without them. For polls that published both, the two differ by about a quarter of a point per party (Polls that publish only seats: deriving shares from seat headlines). That is small next to the polls' error, but it is in every seats-only figure.
What would change our mind. Pollsters publishing vote shares; or finding that seats-derived shares lean one way for a party, which would show up in the record.
Three pollsters
Three pollsters publish Dutch voting intention: Peil.nl, Ipsos I&O and Verian. Each counts for a third of the average, so one pollster's lean moves it noticeably. In recent polls Peil.nl has had FvD several seats above the other two. With three pollsters there is little to average out.
What would change our mind. More pollsters. Nothing else can fix it.
Independent errors
The simulation draws each party's error independently. The PVV and the VVD, the two parties missed most, have been missed in opposite directions at four of seven elections, so their combined vote is more certain than the simulation makes it (Simulating Dutch polling error: where the chances come from). Combinations containing both have ranges that are too wide.
What would change our mind. Drawing linked errors from past elections' patterns, as the British simulation does, once there are enough elections to make the patterns more than anecdotes.
Coalition statements
The coalitions page marks combinations whose parties ruled each other out during the 2025–26 formation (Coalitions: the largest party, the 76 line and who ruled out whom). Those statements were made about the seats of 2025, some by leaders who may not lead at the next election, and one about GroenLinks-PvdA before it became PRO. Parties say new things in every campaign. The page records statements; it does not forecast them.
What would change our mind. New statements, which will be added with their dates and sources as they are made.
The date and the pull-back
The site uses the next regular date, 15 May 2030, and the election-day forecast widens the ranges for more than three years of drift. An early election is quite possible with a minority cabinet. The pull-back towards 2025 moves the PVV towards the undivided PVV's result, which probably overstates it now that the DNA has split off (From today to polling day in the Netherlands).
What would change our mind. A dissolution, which would bring the date forward and narrow the ranges; a DNA result at a provincial or municipal level, which would show how much of the PVV's vote it holds.
The map
The map moves each gemeente's 2025 result by each party's national change. It is for illustration only, because no seat depends on it, and it will be wrong wherever a party's change is uneven, which after a merger or a split it usually is.
The law
The forecast applies the Kieswet as it stands: one national district, the kiesdeler and no other threshold. Reforms have been discussed, a regional district system under the Schoof cabinet and a threshold under the Jetten coalition, and neither has been tabled as a bill. A reform before the next election would change the count.
Summary
| Limitation | How much it matters | What would address it |
|---|---|---|
| A narrow correction | Most of its effect is PRO and the SP | The next election; the test reruns itself |
| Late movers | The largest misses, not foreseeable | A tested late-movement measure |
| Merged and split parties | Corrections learned from parties that changed | More elections with the same parties |
| Seats instead of shares | About a quarter of a point per party | Pollsters publishing shares |
| Three pollsters | One pollster's lean moves the average | More pollsters |
| Independent errors | Ranges too wide for combinations with the PVV and the VVD | Linked errors from past patterns |
| Coalition statements | May not survive the campaign | New statements, added as made |
| The date and the pull-back | Wide ranges; the PVV probably pulled too far | A dissolution; a DNA result |
| The map | Illustrative only | – |
| The law | A reform would change the count | – |
Table: Known limitations of the Dutch forecast.
The forecast's claim is narrow: Dutch polls, corrected by a rule that has earned its place so far, carried through a seat count that reproduces the last three elections exactly, with the coalition arithmetic and the statements that constrain it shown side by side. The next Tweede Kamer election will be the first real test, and we will publish how it went.
References
Kieswet, hoofdstuk P (Zetelverdeling).
Jennings, W. and Wlezien, C. (2018). Election polling errors across time and space. Nature Human Behaviour, 2(4), 276–283.
How to cite
Polling Forecast (2026). What we don't know yet: limitations of the Dutch forecast. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/netherlands-limitations. Live figures change as new polls arrive; the date above is when the text was last revised.


