Look what happens to Labour when pollsters don’t adjust
The June 4th election for the EU parliament was the last occasion when we could test the output of a range of pollsters against real votes in a national election – and the outcome is featured in the above UKPollingReport table. The numbers to focus on are the Labour and Tory shares.
Of the four pollsters that did EU election polls ICM and Populus weighted their sample by what respondents said they did at the last general election – YouGov weighted by party ID. Look at how close these firms were with the eventual Labour share.
Only ComRes did not apply any measure to ensure a politically balanced sample and look at its Labour and Tory shares compared with the outcome and the other pollsters
I never understood why in this specific poll ComRes did not apply the standard past vote weighting calculation that it uses for Westminster voting intention surveys but the outcome is helpful in demonstrating what happens when you don’t have such a mechanism.
For the key challenge facing the phone pollsters is that, almost without exception, whenever they make a mass of unsolicited randomised calls they find a much higher proportion of respondents who said they voted Labour at the 2005 general election than actually did do.
This is being highlighted now as a follow-up to yesterday morning’s article on the December ICM survey and to help put in context further polling as we get nearer the general election.
When Anthony Wells at UKPollingReport works out his regular polling averages he has a formula that places much less emphasis on the findings of pollsters that don’t embrace such an approach.
Meanwhile I commend the commentary on current polling by Martin Kettle in his latest Guardian column.