As regular readers and watchers of polling methodology will know, one of the questions facing pollsters is what to do with people who say don’t know or refuse to answer voting intention questions. A purist approach is to ignore them, to base reported figures only upon people who actually say who they would vote for or, for companies using a squeeze question, who they are most likely to vote for – this is the approach used by most pollsters in the UK. The alternative approach is to estimate how they would vote based on one of their other answers.
This re-allocation of don’t knows was pioneered by ICM after the 1992 debacle when the pollsters all vastly overestimated Labour. They noticed that there were a lot of former Conservatives now saying don’t know, and theorised that many of these were people who would still vote Tory, but were reluctant to admit to pollsters that they were supporting an unfashionable party. This was supported by recontact surveys after elections, when people who had said don’t know to pollsters prior to the election did tend to end up voting for the party they had done at the previous election.
As a result, ICM started reallocating 50% of don’t knows and refusals to the party they said they supported at the previous election. During the 1990s this invariably helped the Conservatives, and became known as the “Shy Tory adjustment”. The name stuck, so it was often still known as the shy Tories through the last two terms of the Labour government, when it actually tended to help Labour instead. Andrew Cooper of Populus called them “Bashful Blairites”.
In recent months it has changed again, polls are now showing a large proportion of people who voted Lib Dem in 2010 saying they don’t know how they would vote in an election tomorrow, and ICM’s reallocation of don’t knows is now favouring them. In ICM’s last three polls the re-allocation of don’t knows has bumped up the level of Liberal Democrat support by 2 points – yesterday’s topline figures of CON 36%, LAB 38%, LDEM 14% were CON 37%, LAB 40%, LDEM 12% without the don’t knows.
Naturally this leads us to the question of whether this is a sound thing to do. Certainly there is solid evidence to back up what ICM and Populus do. Re-contact polls after past elections always supported it, and ICM’s re-contact survey this time round found that about 50% of people who said don’t know in the pre-election polls did indeed end up voting for the same party they did in 2005. Equally, at past elections the adjustment has tended to make ICM’s figures more accurate. None of this guarantees it will still work in the future – the current political situation is rather unusual and these former Lib Dems may behave differently – but it’s a sound starting point.
The difference between re-allocating don’t knows and not doing so is more one of principle. Should polls report just what people say, or should we estimate what the people who refuse to answer think? There are good arguments for both, and in terms of voting intention I think it’s a positive that both are produced. We can see what people are actually saying, but also take account of the fact that there are lots of former Liberal Democrats out there saying don’t know, who may or may not filter back come election day. It’s just important to know which pollsters already factor in those don’t knows, and which ones don’t.