More on Ipsos MORI's poll
The full tables for MORI's poll are now available here.
A notable finding there is that economic optimism has significantly dipped since the PBR, down from plus 10 last month to minus 4 now. I think this is the first poll on economic optimism since the PBR, but is entirely in line with the YouGov/Sunday Times poll that showed people becoming much more negative about the country's current economic situation after the PBR.
Turning to voting intention, there is a shift in propensity to vote, 68% of Tory voters now say they will definitely vote, compared to 61% last time. The proportion of Labour voters who say they will actually vote is down 2 points to 49%.
However, the biggest difference between this month and last is the make up of the sample itself. The past vote breakdown in this poll was CON 22%, LAB 28%, LDEM 9% (the equivalent, if you exclude did not votes and so on, of CON 33%, LAB 43%, LDEM 14%). Compare this to last month's past vote breakdown of CON 19%, LAB 30%, LDEM 10% (the equivalent of CON 29%, LAB 46%, LDEM 16%). It is the lack of political weighting that has produced such extreme switches in support in MORI's recent polls - ICM last weekend had a similarly pro-Labour sample, but their weighting brought it back to showing a much smaller swing.
MORI do not weight by past vote because of false recall. The question is not whether false recall exists or not (all the established pollsters accept that), but how variable it is. ICM and Populus believe it is generally steady and doesn't change wildly from month to month (the implication is that it is largely the result of a social desirability bias making people claim to have voted when they didn't, and the way people report tactical voting). MORI believe that it does genuinely vary slightly from month to month, and that there is the potential for it to change significantly (the implication there being that it changes with people's current political preferences).
Politically weighting polls makes them less volatile. However, MORI's contention would be that the public are genuinely very volatile, and that companies like ICM and Populus risk weighting out genuine volatility. There's no way of proving with certainty who is correct, though one thing to look at is the results of YouGov, who take a notably different approach. YouGov weight by party ID rather than past vote, but more importantly, they weight using information given by their panellists in the past (either on election day 2005 or when they joined the panel), rather than using data collected as part of that day's survey. So, while YouGov's weighting targets could be wrong, they can at least be confident that they are not changing from day to day. Their method would not dampen down genuine volatility, and their figures are one of the less volatile.
For those of you not bored to tears by discussion of methodology there is, incidentally, a one-day conference at the Royal British Academy next month (jointly hosted by the British Polling Council and the National Centre for Research Methods) discussing approaches to polling methodology, featuring Peter Kellner, Andrew Cooper, Martin Boon, Nick Moon and - most relevantly to this, John Curtice discussing whether it is safe to weight by past vote (Simon Atkinson of MORI is also there to put what I expect will be the counter argument!). More info here.
