Pollster quality

Here I compare the final polls of the pollsters to the election outcome. The 2021 elections saw a possible reversal in pollster acduracy, with former top poll Ipsos (also called Barometer) scoring meagerly.

The average error averages all errors of a pollster, with the most recent error counting double.

Pollster errors

Lower numbers are better. The calculation is as follows:

  1. For each misassigned seat one point is given. In 2003 the Barometer gave the CDA two seats too few, the LPF one too few, the CU two too many, and the PvdA one too many. Thus it misassigned three seats and gets three points.
  2. If the pollster sees a party entering parliament but it wins zero seats, or if a party enters parliament but a pollster gives it zero seats, it gets an extra point.
    In 2010 Een Vandaag and TNS-NIPO both gave TON one seat, while it did not enter parliament. They both get a point because they were wrong about TON, in addition to the point for the misassigned seat.
  3. If the pollster incorrectly predicts the largest party it gets an extra point. In 2003 all pollsters showed the PvdA larger than the CDA, and they were all wrong and all get an extra point.

In 2010 I drew an average of the polls; I think I just averaged the numbers without weighting or other subtle tweaks. Since the pollsters were fairly close I also was fairly close. From 2012 on I use the results of my poll average.

Poll data over 2002-2006 come from Cijfers.net.

2021

The last thing the polls captured was a move from VVD to D66. What they didn’t capture was the move from the other left-wing parties to D66, which was clearly seen as the front-runner due to the VVD -> D66 move. Also, the polls missed a shift from PVV to FvD, though this story may be more complex. They did get the small parties right at the last possible moment (and the fact that they reported BIJ1 and BBB winning a seat might have helped those parties.)

2017

There was a genuine move from CDA and PVV to the VVD on election day, worth about 4 seats. Allowing for that move, it appears that once again the pollsters had the PVV at about the right amount of seats. It is now a normal party in the sense that its support can be predicted with a reasonable amount of accuracy.

2012

PvdA and especially VVD were underestimated by all pollsters — clearly a last-minute prime-minister race effect boosted them both, mostly to the detriment of the SP. Also, the PVV was overestimated for the first time, and by a lot less seats than it was underestimated in 2010. The theory that they lost last-minute seats to the VVD is defensible. So maybe the PVV is becoming predictable.

2010

In 2010 all pollsters underestimated Wilders’s PVV for the second time running, which is why it is a difficult party to predict. Some polls again missed the fact that the PVV would be the only extreme right party in parliament.

Also, the CU’s loss of one seat was not predicted (and Intomart even predicted a ridiculous 10 seats).

2006

All pollsters underestimated Wilders’s PVV, and only TNS-NIPO saw that it would be the only extreme right party — the others have one or two others win a few seats. Also, the SP polls diverged very widely.

2003

All pollsters but the PAM had the PvdA a tiny bit larger than the CDA, but in the elections the CDA became the biggest. This is a classic prime-minister-race effect.

2002

The big surprise was that the CDA grew a lot more than expected, and GL and CU lost one seat instead of winning a few. CU to CDA is a logical movement.

All pollsters got big newcomer LPF (List Pim Fortuyn) about right.

1998

Data as yet incomplete; Peil.nl is missing. And to be honest I’m not totally sure I got the very badly formatted data right, or whether it’s reliable. I mean, did they really predict 3 seats for Janmaat?