The polls did far better at the EU referendum than is widely perceived
High levels of postal voting mean we should look at the final 3 weeks not just the final polls
Just got back from holiday in the south of France and am focusing on the political session at a big betting conference that I am taking part in on Friday . Inevitably we will be looking back at what happened on June 23rd – the biggest political betting election ever.
There’s a widespread perception that the polls got the outcome wrong which is impacting on how their ongoing output is judged particularly by those who don’t like current leadership and Westminster voting intention numbers.
But, I’d argue, the pollsters did far better predicting the June 23rd outcome than they are being given credit for.
-
One simple fact. If you just take the polls from the start of June, when postal voting started, there were more published LEAVE lead surveys than REMAIN ones as can be seen in the table.
Even if you just take the final polls each firm published before June 23rd TNS, Opinium, Survation, ICM and YouGov all got it right to within the statistical margin of error.
YouGov’s average LEAVE share in the three weeks before the election was 51%.
There’s no doubt that for some firms their referendum polling was not good compared with what happened but this is by no means the complete picture.
My view is that the scale of early voting by post is impacting on final surveys. Many of those being sampled on June 22nd would have actually cast their postal ballots three weeks earlier which increases the possibility of them not giving an accurate response.
Interestingly looking at the table it was those polls about ten days out that tended to be the most accurate.
All the indications since the referendum was that the postal vote, about a fifth overall, went strongly for LEAVE which is not surprising given that older voters are the most likely to vote in this way. There are some who suggest that the REMAIN would have won if only votes cast on the day had been included.
Postal voting, I’d suggest, means we’ve got to rethink our addiction to judging polls on their final survey. We need to look at complete sets of survey from the time postal votes started being cast.
A final thought is that the polls that tended to be most highlighted were, for whatever reason, those with REMAIN leads. Maybe this impacts on our perceptions.