Since the election we’ve had three voting intion polls, from Survation, YouGov/Sun and ComRes/Mail. All of these pollsters have updated their weighting targets to take into account the election, which is standard practice and explains most of the change from the figures that polls were showing before the election. ComRes has also made fundamental changes to its methodology, with a new turnout filter.
The new filter essentially adjusts self-reported likelihood to vote for expected overestimation, which they model based on constituency turnout and demographics at the last two elections. As I remarked in the aftermath of the election, it is normal for respondents as a whole to exaggerate their LTV – the problem is that those who say they will vote but don’t, seem to be unlike voters as a whole. They’re typically younger and less affluent, both of which are more commonly associated with Labour voters than Tory voters, so this is likely to be a part of the structural bias in Labour’s favour that I mentioned before. The adjustment seems to be worth about four points on the Conservative lead.
As ComRes acknowledge, this isn’t likely to be the final word on methodology changes. But now that polls have at least been re-benchmarked to the election result, what do they show? The average of the three in May (excluding those before the election) gives figures of CON 40.7 LAB 30.0 LIB 7.0 UKIP 11.7 GRN 4.0. Compared with the election result, the polls average a three-point gain for the Conservatives and a one-point loss each for Labour, the Lib Dems and UKIP. This represents a 2% swing from LAB to CON since the election – roughly equivalent to a 50-seat working majority on current boundaries and a 100-seat landslide on the 2013 boundaries. The Tories haven’t had a month entirely in the 40s since March 2009.
Polling is likely to remain sporadic for the time being, but updates from pollsters on methodology are in the pipeline.