How we read Dutch polls
How we turn the Netherlands' polls into a forecast for the Tweede Kamer (150 seats, 76 for a majority): the 7 polls of the last 30 days, corrected for how the Netherlands' final polls have missed before. Today's average: PRO 20.0, PVV 13.4, JA21 11.7, D66 10.0.
Our rule everywhere: measure how far final polls missed at past elections, take off half the average miss, and keep the correction only if it would have made past forecasts better when each election is predicted without knowing its result. Then run the shares through the country's own electoral law.
Why we correct
Averaged over 2002–2025 the final polls missed by: PRO +2.0, VVD −0.4, PVV −1.1, D66 −0.7, CDA 0.0, SP +0.9 points (positive = too high). Predicting each election with the correction learned from the others cut the average miss per party from 1.12 to 1.09 points, closer at 5 of 8 elections, so we do it. The test →
The gain is small and does not hold at every election: 2002 and 2012 came out slightly worse. We keep the rule rather than pick and choose.
What decides the seats
- One national district. 150 seats shared in proportion to the national vote: seats follow shares almost exactly.
- The kiesdeler. A list needs 1/150 of the valid vote, about 0.67%, for a seat. Small parties near that line win or lose their only seat on a few thousand votes.
- The largest party. It usually leads the formation. In 2025 D66 and the PVV both won 26 seats; D66 was ahead by some 30,000 votes.
- Splits and mergers. GroenLinks and PvdA merged into PRO in 2026; seven PVV members left to form the Groep Markuszower (DNA). We measure past polls by lineage, and the DNA as polled.
- 26 final polls at 8 Tweede Kamer elections (2002–2025; 2002 has two pollsters). No final polls from before 2002 could be found: the pollsters' own archives start later, and none of 1986–1998 is online.
- Official results from the Kiesraad: national for every election since 1946, by gemeente since 2012.
- 546 poll averages at set distances before polling day, for the election-day test.
- No Dutch pollster publishes raw figures before its own weighting, so there is no raw-figures test here.
- Average: each pollster's latest poll within 21 days of the newest, equal weight per pollster, in vote shares. Where a poll published only seats (all of Peil.nl's), the share is seats ÷ 150.
- Correction: half the average past miss per party, rescaled to the same total. PRO takes the correction measured on GroenLinks and PvdA together.
- Seats: the Kieswet count (next tab).
- Chances: 5,000 simulations; each party moves by its typical final-poll miss (the root mean square of its past misses). From these come the 80% ranges, the chance of being the largest party and of each coalition reaching 76.
- Gemeenten: each party's 2025 share in every gemeente, scaled by its national change (proportional swing); the DNA gets its national share of the PVV's vote. This is for the map only: no seat depends on it.
- Polling day: the same, with extra drift for the days left: rmse(days) = 3.05 + 0.31 × ln(days / 30).
The Kieswet (chapter P) shares the 150 seats in one national count:
- The kiesdeler is the number of valid votes divided by 150.
- Each list gets one seat for every full kiesdeler it has.
- The seats left over go one at a time to the list with the highest average, its votes divided by its seats so far plus one. Only lists that won at least one full kiesdeler take part.
- List combinations (lijstencombinaties) were abolished for elections from 2021, so 2021, 2023 and 2025 follow these rules alone.
The check: the official results reproduced
Run on the Kiesraad's official votes, our count gives exactly the official seats for every party at all three elections.
2021 · kiesdeler 69,485.68 · all seats match
| List | Votes | Full seats | Remainder seats | Our count | Official |
|---|---|---|---|---|---|
| VVD | 2,279,130 | 32 | 2 | 34 | 34 |
| D66 | 1,565,861 | 22 | 2 | 24 | 24 |
| PVV | 1,124,482 | 16 | 1 | 17 | 17 |
| CDA | 990,601 | 14 | 1 | 15 | 15 |
| SP | 623,371 | 8 | 1 | 9 | 9 |
| PvdA | 597,192 | 8 | 1 | 9 | 9 |
| GL | 537,308 | 7 | 1 | 8 | 8 |
| FvD | 523,083 | 7 | 1 | 8 | 8 |
| PvdD | 399,750 | 5 | 1 | 6 | 6 |
| CU | 351,275 | 5 | 0 | 5 | 5 |
| Volt | 252,480 | 3 | 0 | 3 | 3 |
| JA21 | 246,620 | 3 | 0 | 3 | 3 |
| SGP | 215,249 | 3 | 0 | 3 | 3 |
| DENK | 211,237 | 3 | 0 | 3 | 3 |
| 50PLUS | 106,702 | 1 | 0 | 1 | 1 |
| BBB | 104,319 | 1 | 0 | 1 | 1 |
| BIJ1 | 87,238 | 1 | 0 | 1 | 1 |
| Other lists | 193,756 | 0 | 0 |
2023 · kiesdeler 69,551.51 · all seats match
| List | Votes | Full seats | Remainder seats | Our count | Official |
|---|---|---|---|---|---|
| PVV | 2,450,878 | 35 | 2 | 37 | 37 |
| GL-PvdA | 1,643,073 | 23 | 2 | 25 | 25 |
| VVD | 1,589,519 | 22 | 2 | 24 | 24 |
| NSC | 1,343,287 | 19 | 1 | 20 | 20 |
| D66 | 656,292 | 9 | 0 | 9 | 9 |
| BBB | 485,551 | 6 | 1 | 7 | 7 |
| CDA | 345,822 | 4 | 1 | 5 | 5 |
| SP | 328,225 | 4 | 1 | 5 | 5 |
| DENK | 246,765 | 3 | 0 | 3 | 3 |
| PvdD | 235,148 | 3 | 0 | 3 | 3 |
| FvD | 232,963 | 3 | 0 | 3 | 3 |
| SGP | 217,270 | 3 | 0 | 3 | 3 |
| CU | 212,532 | 3 | 0 | 3 | 3 |
| Volt | 178,802 | 2 | 0 | 2 | 2 |
| JA21 | 71,345 | 1 | 0 | 1 | 1 |
| BVNL | 52,913 | 0 | 0 | 0 | 0 |
| 50PLUS | 51,043 | 0 | 0 | 0 | 0 |
| BIJ1 | 44,253 | 0 | 0 | 0 | 0 |
| Other lists | 47,045 | 0 | 0 |
2025 · kiesdeler 70,479.93 · all seats match
| List | Votes | Full seats | Remainder seats | Our count | Official |
|---|---|---|---|---|---|
| D66 | 1,790,634 | 25 | 1 | 26 | 26 |
| PVV | 1,760,966 | 24 | 2 | 26 | 26 |
| VVD | 1,505,829 | 21 | 1 | 22 | 22 |
| GL-PvdA | 1,352,163 | 19 | 1 | 20 | 20 |
| CDA | 1,246,874 | 17 | 1 | 18 | 18 |
| JA21 | 628,517 | 8 | 1 | 9 | 9 |
| FvD | 480,393 | 6 | 1 | 7 | 7 |
| BBB | 279,916 | 3 | 1 | 4 | 4 |
| DENK | 250,368 | 3 | 0 | 3 | 3 |
| SGP | 238,093 | 3 | 0 | 3 | 3 |
| PvdD | 219,371 | 3 | 0 | 3 | 3 |
| CU | 201,361 | 2 | 1 | 3 | 3 |
| SP | 199,585 | 2 | 1 | 3 | 3 |
| 50PLUS | 151,053 | 2 | 0 | 2 | 2 |
| Volt | 116,468 | 1 | 0 | 1 | 1 |
| BIJ1 | 40,360 | 0 | 0 | 0 | 0 |
| NSC | 39,408 | 0 | 0 | 0 | 0 |
| Other lists | 52,154 | 0 | 0 |
Votes: Kiesraad, final results. Lists below half a kiesdeler without seats are grouped under other lists.
Correcting final polls
Each election is predicted with the correction worked out from the other elections only, so it never sees the result it is tested on. Correcting Dutch polls was closer at 5 of 8 and further out at 2 (average miss per party 1.09 against 1.12 for the polls as published). So our projection uses the correction.
| Election | Polls' average miss | Corrected | Verdict |
|---|---|---|---|
| 2002 | 1.96 | 2.10 | Correction further out |
| 2006 | 0.75 | 0.53 | Correction closer |
| 2010 | 1.21 | 1.18 | Correction closer |
| 2012 | 1.11 | 1.18 | Correction further out |
| 2017 | 1.26 | 1.16 | Correction closer |
| 2021 | 1.08 | 1.08 | Level |
| 2023 | 0.89 | 0.87 | Correction closer |
| 2025 | 0.68 | 0.59 | Correction closer |
Pulling back towards the last result
Closer than the polls as they stand, so used on the election-day page. Details →
The gemeente map: proportional or uniform?
The map moves each party's last gemeente share by its national change in proportion. Replayed on every pair of Tweede Kamer elections since 1946 (Kiesraad results by gemeente, parties by their 2026 lineage) and counting only the 8 pairs from 2000 on (the model uses no earlier election; the older pairs are shown but not used), proportional swing named the largest party in 2842 of 3158 gemeenten, uniform swing in 2707; its share error was smaller at 8 of 8 pairs. So the map stays proportional. No seat depends on it.
| Pair | Gemeenten | Largest party right, proportional | … uniform | Share error, proportional | … uniform |
|---|---|---|---|---|---|
| 2023 → 2025 | 342 | 289 | 247 | 0.7 | 1.2 |
| 2021 → 2023 | 338 | 282 | 273 | 1.0 | 1.6 |
| 2017 → 2021 | 339 | 325 | 311 | 0.8 | 1.0 |
| 2012 → 2017 | 380 | 330 | 320 | 1.0 | 1.2 |
| 2010 → 2012 | 409 | 378 | 367 | 0.8 | 1.1 |
| 2006 → 2010 | 420 | 358 | 318 | 1.2 | 1.7 |
| 2003 → 2006 | 447 | 421 | 424 | 1.2 | 1.4 |
| 2002 → 2003 | 483 | 459 | 447 | 0.8 | 1.1 |
| 1998 → 2002 not used by the model | 481 | 434 | 426 | 1.6 | 2.0 |
| 1994 → 1998 not used by the model | 525 | 459 | 469 | 1.6 | 1.4 |
| 1989 → 1994 not used by the model | 619 | 551 | 553 | 1.7 | 2.2 |
| 1986 → 1989 not used by the model | 698 | 671 | 667 | 0.9 | 1.0 |
| 1982 → 1986 not used by the model | 693 | 634 | 640 | 1.7 | 1.7 |
| 1981 → 1982 not used by the model | 771 | 718 | 734 | 1.2 | 1.2 |
| 1977 → 1981 not used by the model | 799 | 764 | 758 | 1.6 | 1.4 |
| 1972 → 1977 not used by the model | 838 | 788 | 785 | 1.9 | 1.8 |
| 1971 → 1972 not used by the model | 860 | 797 | 809 | 2.4 | 2.2 |
| 1967 → 1971 not used by the model | 862 | 824 | 824 | 2.2 | 2.1 |
| 1963 → 1967 not used by the model | 936 | 906 | 902 | 2.3 | 2.7 |
| 1959 → 1963 not used by the model | 976 | 962 | 961 | 1.6 | 1.9 |
| 1956 → 1959 not used by the model | 996 | 967 | 974 | 1.5 | 1.7 |
| 1952 → 1956 not used by the model | 1002 | 975 | 972 | 1.8 | 1.9 |
| 1948 → 1952 not used by the model | 1009 | 975 | 977 | 2.1 | 2.0 |
| 1946 → 1948 not used by the model | 1016 | 981 | 979 | 2.2 | 2.6 |
Gemeenten with the same code at both elections; a gemeente that absorbed others keeps its code. Before 2012 the Kiesraad Databank gives list votes only, so shares are of the valid vote. The 1946 file is 1% short of the national valid vote.
- Late deciders. Dutch voters make up their minds late, and final-poll misses of five seats or more for one party happened in 2021, 2023 and 2025. A historical average corrects only the part that repeats.
- Three pollsters. With so few, one pollster's lean moves the average.
- Seats only. Peil.nl publishes seats, not shares, and most of its posts are for subscribers.
- Statements change. The coalition table uses what parties said during the 2025–26 formation.
- The Netherlands has no ban on publishing polls before an election.
Figures as of . Published by Polling Forecast.




