Simulating Dutch polling error: where the chances come from
Abstract. Seat ranges, the chance of each party winning a seat and of coming first, and the chance of each combination reaching 76 come from 2,000 simulated elections, each party moved by its typical final-poll miss and the Kieswet count run on the result. There is no local error, because there is one national district. Because the PVV's and the VVD's errors have often run in opposite directions, drawing them independently makes combinations containing both more uncertain than the record suggests.
Summary
The seat ranges, each party's chance of winning a seat and of coming first, and the chance that each combination reaches 76 all come from 2,000 simulated elections. In each one, every party's national share is moved by a random error the size of its typical final-poll miss, and the Kieswet count is run on the result. There are no constituencies, so there is no local error: the whole of the uncertainty is in the national shares. The outputs are counts across the 2,000 runs.
Each party's error is drawn independently. For most Dutch parties that is a fair simplification. For the two parties missed most, the PVV and the VVD, it is not: their misses have often run in opposite directions, so their combined vote has been missed by less than the simulation assumes. We say below by how much.
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.
- Every outcome (3 October). The seats in every simulated election are shown as histograms at /nl/outcomes.
One run, step by step
- Move each party's national share. Add to each party's share a random error drawn from a normal distribution with the standard deviation below. A share cannot fall below 0.02%.
- Fill in the others. The parties not in the model take what the named parties leave, with a floor of 0.3%.
- Count the seats. The Kieswet count: full kiesdelers, then remainder seats by highest averages (From votes to seats: the Kieswet).
- Record each party's seats, whether it won at least one, which party came first (most seats, then most votes), whether each combination of parties reached 76, and which party led in each gemeente on the map.
The random number generator has a fixed seed, so the same polls give the same chances.
National error
The standard deviation for party \(p\) with share \(S_p\) is
where \(\text{RMS}_p\) is the root mean square of the party's final-poll errors at past elections (Measuring Dutch polling error, 2006–2025). In run \(r\),
Today's values:
| Party | Share | Elections measured | Typical final-poll miss (RMS) | Standard deviation used |
|---|---|---|---|---|
| PRO | 19.0 | 8 | 2.46 | 2.46 |
| PVV | 13.9 | 7 | 2.96 | 2.96 |
| JA21 | 11.8 | 3 | 0.57 | 0.57 |
| D66 | 10.4 | 8 | 1.22 | 1.22 |
| VVD | 9.2 | 8 | 2.15 | 2.15 |
| CDA | 8.7 | 8 | 2.38 | 2.38 |
| FvD | 7.6 | 3 | 0.92 | 0.92 |
| 50PLUS | 3.3 | 5 | 0.35 | 0.35 |
| PvdD | 3.0 | 7 | 0.40 | 0.40 |
| SP | 2.8 | 8 | 1.67 | 1.67 |
| SGP | 2.8 | 8 | 0.29 | 0.29 |
| Volt | 2.3 | 3 | 0.47 | 0.47 |
| CU | 2.1 | 8 | 0.63 | 0.63 |
| DENK | 2.0 | 3 | 0.37 | 0.37 |
| BBB | 1.2 | 3 | 0.53 | 0.53 |
The PVV's spread, about three points, comes from misses of 5.4, 4.1 and 2.4 points in one direction and up to 2.1 in the other. The VVD's and PRO's are a little over two. The small parties' spreads are a few tenths of a point, because the polls have measured them closely; at their size, a few tenths is the difference between one seat and none. The DNA, which has never stood, gets a spread that grows with its share.
The spread does not shrink with the correction. PRO's polls are corrected for half its average miss (Reading Dutch polls: the correction and its test), and then drawn with a spread equal to its whole typical miss, which includes the part the correction has already taken off. That errs on the wide side.
What the independent draw leaves out
At most Dutch elections the parties that were understated took their votes from parties that were overstated. The PVV and the VVD are the clearest case: their errors had opposite signs in 2010, 2012, 2017 and 2023. Their combined vote was missed by about 2.5 points in root-mean-square terms; drawn independently with spreads of 2.9 and 2.3, their sum has a standard deviation of about 3.7. So the simulation makes the right's total, and combinations containing both parties, more uncertain than the record of the two together. For the Jetten parties, D66, the VVD and the CDA, the independent draw and the record give about the same spread for the total, 2.9 points against 2.7.
The British simulation avoids this by replaying whole past misses (Simulating UK polling error: where the chances come from). The replays on the what-if page do the same for the Netherlands (If we're wrong: replaying past Dutch polling misses on today's polls); the simulation does not.
What comes out
Seat ranges and winning a seat
For each party, the 80% range is the band within which its seats fell in the middle 80% of runs; the chance of winning a seat is the share of runs in which it won at least one.
| Party | Polls | Our bet | Seats | 80% range | 2025 | Largest party | Wins a seat |
|---|---|---|---|---|---|---|---|
| PRO | 20.0 | 19.0 | 30 | 25–34 | 20 | 91% | over 99% |
| PVV | 13.4 | 13.9 | 21 | 16–27 | 26 | 9% | over 99% |
| JA21 | 11.7 | 11.8 | 18 | 16–20 | 9 | 0% | over 99% |
| D66 | 10.0 | 10.4 | 16 | 13–18 | 26 | 0% | over 99% |
| VVD | 9.0 | 9.2 | 14 | 10–18 | 22 | under 1% | over 99% |
| CDA | 8.7 | 8.7 | 13 | 9–18 | 18 | under 1% | over 99% |
| FvD | 7.3 | 7.6 | 11 | 10–13 | 7 | 0% | over 99% |
| 50PLUS | 3.3 | 3.3 | 5 | 4–6 | 2 | 0% | over 99% |
| PvdD | 3.0 | 3.0 | 4 | 3–5 | 3 | 0% | over 99% |
| SP | 3.3 | 2.8 | 4 | 1–7 | 3 | 0% | 91% |
| SGP | 2.7 | 2.8 | 4 | 3–5 | 3 | 0% | over 99% |
| Volt | 2.4 | 2.3 | 3 | 2–4 | 1 | 0% | over 99% |
| CU | 2.4 | 2.1 | 3 | 1–4 | 3 | 0% | 99% |
| DENK | 2.0 | 2.0 | 3 | 2–3 | 3 | 0% | over 99% |
| BBB | 1.0 | 1.2 | 1 | 0–2 | 4 | 0% | 83% |
A party whose range starts at 0 has a real chance of falling below the kiesdeler. That chance depends as much on the spread as on the share: a party polled closely at past elections can sit near 0.67% with little risk, while one with a wider record cannot.
The largest party
In each run the largest party is the one with most seats, and on equal seats most votes. The chance of coming first is the share of runs in which each party was largest. Because the spreads differ, a party that trails narrowly but has been missed by more in the past can have a better chance than the gap in shares suggests.
Combinations and 76
For each combination on the coalitions page, the chance of 76 is the share of runs in which its parties' seats added up to 76 or more (Coalitions: the largest party, the 76 line and who ruled out whom). The chances of different combinations are not exclusive and do not add up to anything: in one run, several combinations can reach 76.
Gemeenten
For each gemeente, the simulation records how often each party leads there, with each gemeente's 2025 result moved by the national change of the run. This is for the map only; no seat depends on it.
What we know about the simulation
We can say how it was built, not yet how well it performs: seven elections are too few to check how often results fall inside the ranges. Three points matter.
The spreads come from seven elections, three for the newer parties. A miss larger than the PVV's in 2023 would fall outside what the spreads suggest.
Independent draws make combinations containing both the PVV and the VVD too uncertain, for the reason above.
The kiesdeler is exact. The only non-linear step, whether a small party wins its first seat, is computed by the count itself in every run, so the chances for the small parties need no further assumption.
Limitations
The simulation turns one projection into chances with the least machinery the evidence allows: a normal draw per party and the seat count. It does not model linked errors, late movement in the last days, turnout, or new parties. The polling-day version adds drift for the time left (From today to polling day in the Netherlands). What we don't know yet: limitations of the Dutch forecast collects the rest.
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). Simulating Dutch polling error: where the chances come from. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/netherlands-simulation. Live figures change as new polls arrive; the date above is when the text was last revised.


