Pollsters have made various tweaks. On Friday, we'll know which (if any) workedby Patrick Sturgis / June 7, 2017 / Leave a comment
As the person who chaired the British Polling Council/Market Research Society Inquiry into the failure of the 2015 election polls, I have been asked many times what the outcome of the snap election on Thursday will be. Yet my record of predicting election results makes me quite unsuited to answering this question— surprisingly for some, it seems. I went for a hung parliament in 2015, Remain in the EU Referendum, and Clinton in the US Presidential race, to name but a few of my more recent failures. So, your safest bet at this stage if you want to know the election result would be to add or subtract ~5 points from my prediction (which, for the record, is an 8-point lead for the Conservatives).
But I can perhaps comment more usefully on the matter of whether the pollsters have addressed what the Inquiry identified as the key problem with their methodology in 2015. This was that their samples were unrepresentative of the voting population: the procedures they used to recruit people into their samples simply resulted in too many Labour voters and too few Conservatives. The adjustment procedures they applied to the raw data didn’t fix this basic problem.
Now, to the uninitiated at least, this explanation may sound like stating the bleeding obvious, but there are many other things that can and do go wrong with polling: among them late swing, turnout weighting, deliberate misreporting, and herding. However, the evidence considered by the Inquiry suggested that, singly or collectively, these other factors contributed little or nothing to the polling miss. Unrepresentative samples was the problem.