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

If we're wrong: replaying past Dutch polling misses on today's polls

Abstract. Nine readings of today's Dutch polls: as they stand, our corrected projection, and each Tweede Kamer election's final-poll miss since 2006 replayed in full, all through the same seat count. The misses of 2017, 2021 and 2023, each with a late mover, change the largest party; none gives the parties of the Jetten cabinet a majority. We explain the count of gemeenten whose leading party changes.

Polling Forecast · updated 9 November 2026 · 5 min read · Netherlands, lenses, scenarios, past polling misses, what-if, largest party, late swing

Summary

A lens is one stated assumption about how wrong today's polls are. For the Netherlands the site shows nine readings of the same polls: the polls as they stand; our projection, which takes off half the average past miss because that passed its test (Reading Dutch polls: the correction and its test); and seven replays that take each Tweede Kamer election's final-poll miss since 2006 and apply it in full to today's average. Every reading goes through the same Kieswet count, and we show them side by side without averaging them.

Two things follow from the Dutch record. The replays change the largest party: under the misses of 2017, 2021 and 2023 a different party comes first, because those were the elections with a large late mover. And no replay brings the parties of the Jetten cabinet near 76: their seats vary by about a dozen across the readings, well short of a majority in every one.

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 a lens is

Let \(M\) be the parties the forecast models, \(v_p\) today's average for party \(p\) (each pollster's latest poll within 21 days of the newest, equally weighted) and \(S = \sum_{p \in M} v_p\). For a past election \(t\) the error on party \(p\) is \(e_{p,t}\), final-poll average minus result. The replay of election \(t\) subtracts that error from every party it measured, with a floor of 0.1, and rescales the modelled parties to their original combined share:

\[ v^{t}_{p} = \tilde{v}^{t}_{p} \times \frac{S}{\sum_{q \in M} \tilde{v}^{t}_{q}}, \qquad \tilde{v}^{t}_{p} = \begin{cases} \max\left(0.1,\; v_p - e_{p,t}\right) & p \in P_t,\\ v_p & \text{otherwise.} \end{cases} \tag{1} \]

The parties measured change from one election to the next. PRO (GroenLinks and the PvdA together before 2023), the PVV, the VVD, D66, the CDA, the SP, the CU, the SGP and the PvdD are measured at all seven; 50PLUS from 2012; JA21, FvD, the BBB, Volt and DENK from 2021. The DNA is measured at none, and moves only through the rescale.

Each replay applies a whole past miss to the polls as they stand, not half of it to our projection. The projection is the rule's reading; the replays are what the Kamer would look like if this election's polls missed exactly as a past one's did.

After the replay, every reading goes into the same count (From votes to seats: the Kieswet).

The readings

The polls as they stand. The average with no adjustment.

Our projection. Half the average past miss taken off each party measured at two or more elections, then rescaled.

Seven past misses. One replay for each of 2006, 2010, 2012, 2017, 2021, 2023 and 2025, built from 26 final polls.

The reader's own. The what-if tool lets a reader set an error for each party and runs the same count. Its presets are the readings above. The prediction tool does the same from shares the reader types in.

Every reading through the same count

ReadingPROPVVJA21D66VVDCDAFvDLargestD66 + VVD + CDA
The polls as they stand31201815141311PRO42
Our projection30211816141311PRO43
If the polls miss like 200225211814122211PRO48
If the polls miss like 200628241816141211PRO42
If the polls miss like 201030271816111011PRO37
If the polls miss like 201232181816181311PRO47
If the polls miss like 201727191917201211PRO49
If the polls miss like 202127171920131114PRO44
If the polls miss like 202329291815101410PVV39
If the polls miss like 202527221718171212PRO47
Live Seats under each reading of today's polls Computed from the database, polls to 18 November 2026.

The table is recomputed whenever a poll is added. Four features follow from the shape of the record and should survive most changes in the polls.

The projection and the polls differ by a few seats. The correction takes about a point off PRO and a little off the SP and the CDA, and gives it to the PVV, D66 and the VVD. PRO loses two or three seats; the others gain one or two each.

2017, 2021 and 2023 change the largest party. In 2017 the polls understated the VVD by 3.9 points; replayed today, the VVD comes first. In 2021 they understated D66 by 2.9; D66 comes first. In 2023 they understated the PVV by 5.4; the PVV comes first. These are the three elections at which one party rose late and finished well above every final poll.

2012 is the centre-left's best replay. In 2012 GroenLinks and the PvdA were understated and the SP overstated by 4.1 points; replayed today, PRO gains and the SP loses.

No replay gives the Jetten parties a majority. D66, the VVD and the CDA together range from the mid-forties to the upper fifties. The combinations that reach 76 all need PRO or the PVV (Coalitions: the largest party, the 76 line and who ruled out whom).

Why the replays move seats smoothly

In a system with constituencies a small shift in votes can move many seats at once. Here it cannot: one seat is 0.67 points of the vote, so a replay moves each party's seats roughly in proportion to the points it adds or takes away. The only jumps are at the kiesdeler, where a small party wins its first seat or loses its last, and in the order of the largest parties, which can change on a few tenths of a point.

Gemeenten at risk

The map shows the leading party in each gemeente, with its 2025 result moved by the national change. For each gemeente the site counts how many of the seven replays give it a different leading party from the projection:

\[ F_g = \sum_{t} \mathbf{1}\!\left[\, w^{t}_{g} \neq w^{B}_{g} \,\right]. \tag{2} \]

Today 226 of the 345 gemeenten and islands on the map change leading party under at least one replay. That is a large share, because four or five parties are within a few points of each other nationally, and it matters less than it would elsewhere: no seat is won in a gemeente. The count is shown on the map for interest, not as a forecast.

Limitations

Seven past misses are not a distribution. The replays show what known misses would do, not how likely each is, and they are not independent: 2017, 2021 and 2023 all have a late mover, but a different one each time.

Late movers cannot be replayed into the right party. The next late mover may be any party. A replay can only move the party that moved last time.

Lineage. PRO's replays before 2023 apply the combined miss of GroenLinks and the PvdA. The PVV's apply to a party that has since split; the DNA is untouched.

Parties without a record move only through the rescale.

References

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). If we're wrong: replaying past Dutch polling misses on today's polls. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/netherlands-lenses. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like