At the 2010 election, the Conservatives were underestimated by 1.1 points and Labour underestimated by 1.8 points, but the Lib Dems were overestimated by 3.3 points. As a result, the LAB-CON spread was polled very accurately indeed, 2010 being the only one of the 12 elections I examined where the polls were within a point of the spread outcome, while it was the Lib Dems that found their hopes of a breakthrough dashed. Two principal theories have been put forward to explain what happened:
- Loud Liberal theory: The Lib Dems rode a wave of popular approval (remember Cleggmania?) and response bias led pollsters to overstate their support at the roughly equal expense of the other two main parties. These suggestions sometimes include talk of a late swing away from the Lib Dems as habitual Labour and Conservative supporters took fright at the prospect of “letting [the other] in” (though ex-post analysis by Martin Boon and John Curtice suggests that the impact of this particular factor was small). This theory would suggest that the accuracy of the LAB-CON spread in the polls was genuine, having simply become more and more accurately measured over time, whereas overestimation of Lib Dem support was specific to that party, not the other two.
- Shy Labour theory: the Conservative vote was underestimated as per its historical tendency. But this time, the Labour vote was underestimated as well. Labour had been in power for 13 years, the recession was fresh in the mind, making Labour voters now less likely to admit their support. This had happened to some extent in 2005 following the invasion of Iraq, when Labour’s overestimation was less marked than its historical average. But 2010 saw the emergence of Nick Clegg’s Lib Dems as an obvious, progressive alternative to Labour that some even thought could win the election. If that sounds preposterous, remember that there were no fewer than seven polls that had the Lib Dems in the lead, all in the final month of the campaign. The effect of this was enough for Labour’s vote share actually to be underestimated, and this shy Labour factor offset the shy Tory factor, consistent with the inaccuracy of the Lib Dem polling being a symptom of the shy Labour factor, not the cause. The unusual accuracy of the CON-LAB spread was therefore a fluke.
Which of these theories we believe is of great importance, because as the most recent ‘field test’, our understanding of it has an outsize bearing on how many will look at the current election: If we believe the first of these theories, then with Cleggmaina long gone, we have no reason to think that the polls will be systematically biased, and it is safe to take them at face value. If we believe the second, then the shy Labour factor was a blip and with Labour now in opposition, shy Labour voters should be gone too, meaning that we can expect the polling to return to its historical tendency, overstating Labour support.
So which is it? Sometimes, polling internals can tell you things ex-post that aren’t obvious ex-ante. One innovation after 1992 was ICM’s spiral of silence adjustment, which reallocates 50% of likely voters who are “don’t knows” or refuse to give a voting intention to the party they voted for at the prior election. It’s possible that it isn’t a big enough adjustment but it can be an indication of silence that might be hidden elsewhere, or that a party’s voters are being less forthcoming than usual. It has certainly improved the accuracy of the polls that use it, though remember at all times that individual polls have a margin of error.
We know from this John Curtice paper that the adjustment boosted the Conservatives by three points in 1997. NCP has not been able to obtain data for 2001, but in late 2002 (from which point onwards the full tables are publicly available online) it was close to neutral. During New Labour’s latter years, particularly from the time of the 2003 Iraq war onwards, it typically pointed to a shy Labour effect. But as the following charts show, it exhibited some interesting behaviour around the time of elections.
(Continued, please navigate using numbered tabs…)