h1

The best test of a pollster is not how they’re currently doing against other firms but what happened last time they were tested

June 30th, 2019

I am afraid that I have to disagree with David Herdson on his latest Saturday thread about YouGov understating Labour. Firstly you cannot judge pollsters’ based on their current surveys when less than 5 weeks ago they were tested against a real election involving real voters.

In the two charts above I compare LAB and LD vote shares for the May Euros in their final published polls.  Just two of them can claim to have come out of the election well with the rest trailing some way back.

Just examine some of the exaggerated figures that some pollsters were record reporting for LAB where we had a range from 13% to 25%. The actual GB figures was 14%.

Now look at the second chart showing the final LD shares. These range from 12% to 20%. The actual GB share was 20,4%.

Apart from Ipsos MORI and YouGov the rest really did rather badly.

Because of the low turnout, the 37% that actually happened was broadly anticipated, this was always going to be a challenging election for polling because turnout was everything. If one party’s supporters were less likely to vote¬† then that presents the pollsters with serious challenges .

The other challenge, of course, was tactical voting generally by remain backing LAB voters to the parties they saw as being most likely to succeed in their region and so the vote could produce the maximum number of MEPs. This helped the LDs and, of course, the Greens to achieve the success that they did. Whatever mechanisms YouGov and Ipsos Mori use they were able to detect better what was the big characteristic of this election.

So when I look at the current polls I regard Survation and Opinium, of the recent ones, as LAB over-staters.

Mike Smithson