ICM on why the polls overestimated the Lib Dems

Earlier in the week we also saw the publication of this article by Martin Boon and John Curtice, their take on why the polls overestimated the Lib Dems at the last election, based upon a call back survey of 1,200 of the respondents to their final survey.

The first reason Boon & Curtice suggest is a late swing, though they see this as explaining only part of the error. 95% of people who told ICM before the election they were going to vote Conservative reported actually having done so after the election, for Labour the figure was 93%, for the Lib Dems it was somewhat lower at 87%. That said, they say that the number of people switching towards the Lib Dems was almost as much as they lost, so this can only be a small factor.

Secondly, Boon & Curtice do not think differential turnout was a big factor. People who said they were going to vote Lib Dem before the election were not significantly more likely to tell ICM in the post election poll that they didn't vote.

Thirdly, ICM's don't knows disproportionately brokein favour of Labour, backing ICM's decision to use their "spiral of silence adjustment" (Boon & Curtice still refer to this as "shy tory syndrome" in their article, though in this case it was "shy" Labour voters). One of the conclusions they draw in the article is that it may be better to make the adjustment even stronger.

Finally, Boon & Curtice say ICM may have weighted the Lib Dems too highly in their past vote weighting, and that this will be high on the list of their investigations in coming months.

It is also worth reading Roger Mortimore of Ipsos MORI's reply in the comments to the article, where he makes a very valid point that underlines just how difficult it is to work out for certain what went wrong with the polls - many factors that could have skewed the final polls could also have skewed any call back survey. To take Roger's example, if the problem was people saying they would vote Lib Dem but not actually doing so on the day, those people might also have claimed to have voted when they didn't in a call back survey.

In that specific case, we will eventually have concrete evidence of whether differential turnout was a problem or not (the British Election Study recontact interviewees after the election and ask if they voted, but also check them on the marked electoral register to see if they really voted), but there are similar problems we will never be able to rule in or out for sure. If, for example, some people told pollsters they would support the Lib Dems but actually voted for someone else (for whatever reason - being the fashionable thing to say perhaps), they may have said Lib Dem in the ringback survey for that same reason.

UPDATE: And on more topical matters - tonight's YouGov voting intentions are CON 42%, LAB 34%, LDEM 17%.