We almost have the full results for the European elections, so how well did the pollsters do? Here are all the polls taken over the final weekend, plus that final YouGov poll on the eve of polling.
Scotland is still to declare, but on the figures so far it looks as if YouGov and then Populus will have the laurels. ICM ended up severely underestimating UKIP support and overestimating Lib Dem support, while ComRes were out on Labour, the Conservatives and the Greens. I’ll update with a proper post, and what lessons we can learn, once the Scottish results have been declared.
UPDATE: I’ve now updated the table to include the Scottish figures so we can look at the final performance of ICM, Populus, YouGov and ComRes (MORI did not carry out any polling for the European election). Looking at the average errors of each company (that is, the average of the difference between each party’s actual share of the vote, and what each pollster had them in their final poll) YouGov performed the best – both their polls in the final week were closer than any competitor, with average errors of 0.83 and 1.23. Second were Populus, who slighly overestimated Conservative and UKIP support, but otherwise performed well with an average error of 1.57
ICM’s average error was 3.16 – this was down to them overestimating Lib Dem support and underestimating UKIP support. As regular readers will know, ICM do tend to produce higher levels of Lib Dem support than other pollsters do, however, in this case I don’t think the error in predicting the Lib Dem vote is part of a wider problem with ICM’s method – rather, I think it’s down to the way ICM and Populus asked the question of how people would vote. ICM prompted with just the main three party names, and then gave supporters of “other” parties a second list of prompts. Populus included minor parties in their main prompt. The results suggest Populus’s approach works better with phone polls (though YouGov’s increased accuracy compared with their 2004 performance suggests it’s the other way round online!)
ComRes’s final poll was furthest out, with an average error of 3.56. Their final poll before the election underestimated Conservative support, and severely overestimated support for Labour and the Green party (for whom the poll was carried out). My guess is that the skew towards Labour away from the Conservatives was due to the lack of any political weighting – exactly what went on with Green support I don’t know, it may well be a prompting issue – but at the moment I don’t know exactly how ComRes worded their question.