Will the polls be right?
We approach a general election and we have the first sighting of the traditional "what if the polls are wrong" article. Thankfully, given that sometimes these beasts are extravaganzas of ill-informed and out-of-date tripe, this one is by Julian Glover and is generally sensible.
What Julian's article boils down to is that all the adjustments made to correct the polls after the debacle of 1992 have worked very well in an environment where Labour are on top and the Conservatives are the underdog, but will they still work in a political environment when the Conservatives are in the lead?
The purist answer is, of course, we don't know and we won't know for certain until after the coming General election. I am not, however, expecting any vast error.
Firstly, the polls agree with each other. As Julian correctly points out, this was also the case in 1992: everyone predicted the same result, and everyone was wrong. However, in 1992 the methods used by the polling industry were actually pretty homogenous. Everyone used quota sampling, everyone interviewed face-to-face. They were all wrong because they all made the same mistakes!
In 2010 there is no such uniformity - we have polls using quasi-random phone sampling, and polls using internet panels. We have polls weighted by party ID, past vote, or no political weighting at all. We have different approaches to don't knows and likelihood to vote - yet they all report much the same figures. It seems unlikely that all these companies are wrong in different ways that by co-incidence happen to produce the same answer.
The caveat to that is, of course, that pollsters tend to produce similar figures because it's better to be wrong together than risk going out on a limb and being wrong alone. However, for ICM, Populus and YouGov (and ComRes to some extent) there has been no significant shift in their methodology since before David Cameron became leader - yet their voting intention figures have charted the same reaction to the changing political scene.
Secondly, there is the
recent record of the polls. In 2005 the polls performed excellently. Neither are the polls entirely untested since the Conservatives moved ahead. YouGov, Populus and ICM all performed well in the European elections (ComRes less so). YouGov got the London mayoral election right (MORI didn't, but reviewed their methods afterwards). The shift in the balance of support does not seem to have upset their methods so far.
Thirdly, the changes since 1992 are not reliant upon assumptions of which party is in the lead. The political weighting that is applied to most phone polls these days does reduce Labour over-representation, but this doesn't seem to be a result of Labour's popularity (or lack of it). The proportion of people saying they voted Labour in the last election used to be more than actually had done... and still is now Labour are trailing behind. Labour over-representation in samples seems to be something to do with lifestyle or attitude, not a result of political popularity.
The other change that sprang from 1992 is the "spiral of silence" adjustment - the reallocation of don't knows by ICM and Populus to the party they say they supported at the last election. Originally this boosted the Conservatives, and I find a lot of people who assume it still does. In fact "shy Tories" are no more, and for more than five years it has tended to help Labour. Of course, it could be these "Bashful Brownites" behave differently to the "Shy Tories" of yesteryear... but they have already been tested in 2005 and the assumptions seemed to hold firm.
Of course, polls can get it wrong, as 1992 and 1970 will testify - and almost by definition it would likely come from somewhere unexpected. What if turnout rises, for example, or what if there is a high level of support for others - how would you factor in if they aren't standing everywhere? There is not, however, any particular reason to expect problems this time round.
