London poll - the post mortem

Fun over, here's the sensible grown-up look at how the final polls in the London race compared to the result. The final result and the final polls from each of the three pollsters who carried out regular polling in the race (ICM's single poll was very early in the campaign) are below:

The accuracy of the polls was thrown into unusual prominence for a couple of reasons - firstly because, rather than the usual politican's response to unfortunate polls ("I never look at polls" or "the only poll that counts is on polling day"), Ken Livingstone spent much of the campaign viciously attacking YouGov's methodology, criticism that was picked up in some of the papers and which has resulted in a narrative in some quarters of the race not just being Johnson vs Livingstone, but also being YouGov vs MORI.

A second reason was the closeness of the race. In 1997 and 2001 some pollsters actually performed very poorly, but it didn't receive much coverage because they all got the right winner. If the polls predict a comfy Labour win, and Labour do indeed win comfortably, no one is too fussed that the polls showed their lead to be rather comfyier than it actually was. Here we had one pollster showing Johnson comfortably ahead, while rival pollsters showed Livingstone very marginally ahead. When pollsters show different winners the difference suddenly becomes very apparent.

The result, of course, is that YouGov's final poll was almost spot on with Johnson and Livingstone's shares of the vote, though they overestimated the level of support for Brian Paddick and slightly underestimated the level of support for "others". The reaction will be that YouGov have been vindicated.

Rather less positive will probably be the reaction to Ipsos MORI, who have been painted into the role of YouGov's opponent. Their final poll was outside the margin of error for Livingstone and Johnson, but in fairness was conducted over a week before polling day. It is possible that there was a swing towards Boris Johnson in that final week and that MORI were not wrong at all - a poll is but a snapshot in time and there can be no guarantee that people won't change their mind before they actually cast a vote. That is one possible explanation, but of course, it's also possible that there are other reasons, dealing with turnout, a sample that is too Labour, a shy-Boris supporter effect or something else entirely.

I expect MORI will also point out that even they didn't expect turnout as high as it was in their poll and that they published a paper a day or two before the election making predictions about how, with a lower level of turnout they would expect Boris to be marginally ahead. The question will be whether these perfectly good points will be enough to defend their reputation in the court of public (or in this case, commentariat) opinion.

UPDATE: Ben Page of Ipsos MORI has left the comment below: "At Ipsos MORI we will be reviewing our political polling methodology to look at lessons learned and publishing our conclusions and any changes we make in a few months’ time" (more below). In my opinion that's the best thing to do. There is nothing to be gained from just blaming a result that looks wrong on a late swing, the right thing to do is to look in detail at the figures, try and work out what if anything did go wrong, and see how it can be put right.