The tables for MORI’s monthly poll are now up on their website so we can dig around inside them and look at the maths. Obviously with such a surprising shift in support, the thing I looked at first was Liberal Democrat support. What actually caused that jump in their figure?
The sample itself wasn’t massive more Liberal Democrat – last month 9% of the sample said they had voted Lib Dem in 2005, this month 10% said they had. The raw numbers of people saying they were voting Lib Dem were up from 17% to 20%, but again, that’s a lot less than 8 points! A major factor seems to be the filtering by likelihood to vote.
I have written a long article on the site here looking in detail at how pollsters deal with likelihood to vote. The simplified version though is as follows…
YouGov ignore it,
Populus – weight by it, so someone who says they are 9/10 likely to vote is worth 90% of someone who says they are 10/10 likely to vote (and so on),
ComRes – do similar, but entirely exclude those who are less than 5/10 likely,
Ipsos MORI – filter by it, so someone who says they are 10/10 likely to vote is counted, and someone who says they are 9/10 likely to vote (or lower) is excluded,
ICM – also filter by it, but less strictly, taking those who rate their chances at 7/10 or higher.
In last month’s MORI poll, of all the people who said they would vote Liberal Democrat, only 47% of people said they were 10/10 certain to vote. In this month’s MORI poll 69% of Liberal Democrats said they were 10/10 certain to vote, so a much larger proportion were included in the topline voting intention, contributing to the massive increase in Lib Dem support.
Interestingly though it wasn’t a massive shift in the likelihood of Lib Dem supporters to vote. Last month 81% of Lib Dem supporters said they were 7/10 likely to vote or above. This month 85% of Lib Dem supporters said they were 7/10 likely to vote or above. What actually happened is that lots of Lib Dem supporters who had said they were very likely to vote, rating their chances at 7 to 9 out of 10, moved to saying they were certain to vote, but because it tipped them over the 10/10 point it moved them from being excluded from the poll to being included. It’s the result of having a straight cut off, rather than a scale like ComRes & Populus do.
Filed under: Methodology