More on Ipsos MORI's 6 point lead

The tables for Ipsos MORI's poll are now up on their website, so we can dig about and see what's actually happened.

Regular readers will know that the big difference between MORI and other pollsters is that MORI do not politically weight their sample. All the pollsters including MORI weight their samples by known demographic figures like age, gender, social class and region. All except MORI also use political weighting, normally weighting by how people claim they voted at the last election.

There's a much longer explanation of why most pollsters do this and MORI do not here, but the very short version is that people aren't very good at recalling their 2005 vote. ICM, Populus and ComRes take the view that past recall is pretty stable over time and can be estimated well enough to weight by, MORI take the view that it's too unstable and should not be used for weighting. The result is that MORI's samples run the risk of varying politically from month to month more than those of other companies (though MORI would claim the opposite - that other companies risk weighting out genuine public volatility). In a case like this, it raises the question of to what extent the shift is down to people changing their minds, and what extent it is just a more Labour sample.

In MORI's poll last month which showed a 17 point Tory lead, amongst those who voted in 2005 32% said they voted Conservative, 43% Labour and 16% Liberal Democrat. In this month's poll which shows a 6 point Tory lead the figures of recalled 2005 vote break down as Conservative 29%, Labour 46% and 16% Liberal Democrat - so a 6 point change in the recalled lead from 2005. (For reference, ICM weighted their sample so recalled 2005 vote was Conservative 33%, Labour 38% and Lib Dem 22% - even they don't weight to the actual figures because of false recall).

So part of the reason for the shift in MORI's voting intentions since last month is that their sample appears to have had significantly fewer people who voted Tory in 2005 and significantly more people who voted Labour in 2005 (again, from MORI's point of view at least some of that would be changes in the level of false recall).

However, this does not explain the whole difference. If you look at the rest of the survey's innards, there is real movement in Labour's favour too. Likelihood of Labour supporters voting has increased, that of Conservatives deceased - but in both cases the change is too small to be significant. More interestingly there was an increase in the likelihood of people who voted Labour in 2005 to vote Labour now - in October 65% of former Labour voters said they would back the party this time, this month 72% of Labour's 2005 voters said they would back the party.

In conclusion, while a lot of the massive shift in voting intentions in this poll appeares to be down to sample variation, there's a modest firming of the Labour vote too. We should still wait for some more polls to see whether that itself is real, and whether other companies also pick up a firming of the Labour vote. I wouldn't, however, expect any company using political weighting to show quite such a narrowing of the lead.