As party political attention focuses on the General Election – albeit with the likelihood of a by-election in Cambridgeshire South when Andrew Lansley is packed off to Brussels to be the UK’s contribution to the Commissioner Corps -the opinion polls will become increasingly important in telling us the possible outcome for May 2015.
The Political Betting blog reports that the “Polling Observatory” at Manchester University have released their latest report in which they seek to estimate current electoral sentiment by pooling all the currently available polling data, while taking into account the estimated biases of the individual pollsters (“house effects”). This allows them to assess whether the reported vote intention for a given pollster is above or below the industry average.
This is interesting as party supporters have their favourite pollsters, who they rate more highly than the others. For example, UKIP supporters love talking up Survation polls and many reject You Gov polls because Peter Kellner is married to the EU’s High Representative for Foreign Affairs and Security Policy, Baroness Catherine Ashton. Tories on the other hand prefer the results that Populus publish, while contending that Survation always understates the true level of Tory electoral support; and so on.
While the results are on the Manchester Uni page linked above, Political Betting’s love of Datawrapper has seen them plot the results on a party by party basis. Makes interesting reading….
This becomes even more interesting when you look at these ‘house effects’ with the European Elections in mind, where we have already covered the accuracy of pollsters in light of the election results.
The issues seem to have their basis in both polling method (phone or internet) and adjustment methodology (how they deal with undecideds and those who won’t say who they plan to vote for).
Autonomous Mind 
If you average these effects – even Survation’s huge bias towards UKIP – they pretty much cancel each other out, to within half a percentage point. So we can expect a “poll of polls”, even uncorrected for the various “house effects”, to be pretty accurate.
In fact, without Survation’s “rogue”, the other pollsters would have been less accurate, on average, about the UKIP result. So while its own polling was wildly wrong, its fans in the party did have a point. The same could be said about ICM and the LibDems: wrong, but pulling the average error closer to zero. It’s more obvious there from the chart, because Opinium also pulls the UKIP average up. But the effect is roughly the same, reducing the error by about 0.3 percentage points. (Without Opinium and Survation, the “average house effect” of the other pollsters would have been to underestimate the UKIP result by more than 4 points.)
What’s also interesting – and I’m only going visually by the charts here; I haven’t worked it out – Lord Ashcroft seems to be the least biased of all, over all four parties. It’s close with ComRes Online, I think.
Ashcroft’s polls tend to have much bigger samples.
Ashcroft’s poll about the Tories beating UKIP in the Euros was so wrong that it makes you wonder if there was a bit of social engineering at work.
In looking at the accuracy of polls against out-turn, you would probably need to factor in 1.5-2% votes lost by UKIP to the soundalike ‘An Independence for Europe’ party. On a complex ballot paper, those not too skilled at voting would probably thought that they were voting for UKIP. Most of those will be potential UKIP voters in 2015.
Not trying to make any partisan points about UKIP, rather trying to observe that there are several factors.
I think this automatic assumption that votes which went to AIFE were in error is misguided. There are many ex UKIP supporters who will not vote for the party but still wish to vote for a withdrawalist option. I know two who deliberately voted for Nattrass’ group for that reason.