Churn changes
So many pollsters are carrying out general election surveys now - each with its own approaches and baselines - that the best way to get the
big picture is to focus not on specific party shares, but on the overall direction of travel. When a new poll comes out, the first reference point should be that firm's last survey, regardless of which media outlet commissioned it. And if various pollsters broadly agree that things are travelling in one direction or another, then you can conclude that they're probably correct.
Over the past couple of weeks, we have seen a shift. Until recently, the picture was, by and large, that Labour's gap in the polls was narrowing. But that now seems to have stalled. Several firms are reporting increases, if only small ones, in the Conservative lead.
So is this just a blip, or an important change? None of the firms is reporting Tory leads anything like those we saw last year. ICM's "churn data", which records how people switch between the parties, makes this particularly clear.
By asking its interviewees which way they voted in the previous election, the firm records how votes are moving. A key indicator is the retention factor: the proportion of 2005 voters who are sticking with the party they supported last time.
In October 2009, ICM's post-conference season poll indicated that the Conservatives were retaining 94 per cent of their 2005 voters (see graph, above). But this month's data suggests that this figure has fallen to 88 per cent. The numbers for both the Lib Dems and Labour have increased - by 7 and 8 per cent, respectively.
Split the difference
It may seem inexplicable when polling organisations carry out surveys at the same time, yet return vastly different numbers. Earlier this month, over the space of a few days, we saw Labour shares ranging from 26 per cent to 34 per cent - a result just 2 points down from their vote share at the last election (see chart, below).
These discrepancies can often be explained, at least in part, by looking at how each firm weights the raw data to ensure its samples are representative of the population. The most popular method is to ask how respondents voted in the last election, and then adjust the numbers in line with the actual result. Almost all firms that do this correct for the phenomenon of "misremembering" - that is, people who say they voted for
a party when they didn't. Firms that do not make an adjustment of this sort, such as Angus Reid, tend to return Labour shares that are on the low side.
YouGov, meanwhile, has a totally different approach to securing balanced samples. Among the data it stores about its panel members, on whom its surveys are conducted, is which party people say they most identify with. But voters quite often say they identify with Labour, and then vote for some other party - most often, the Liberal Democrats. So YouGov attempts to factor that in by identifying two categories of Labour voter - loyal and disloyal - while the Liberal Democrat quota includes only loyalists.
This explains, in part, why YouGov's recent polls have returned higher Labour shares than anyone else's.
Heavy betting
Among those who have ignored the polls that point to a hung parliament are political gamblers. Their collective view, as represented by the betting odds on the election's outcome, is that the Conservatives remain odds-on favourites to win an overall majority.
But there is no particular reason to trust the gamblers' predictions. In 2005, their money was on a three-figure Labour majority, rather than the eventual lead of 66 seats.
Mike Smithson is the editor of politicalbetting.com
