Peter Kellner admits “error” in the pollster’s weighting system?
Last week while I was on holiday in France Adam Boulton’s blog carried a piece suggesting that YouGov’s London samples included too many older voters, who traditionally are more inclined to vote Conservative. He suggested that if the weightings had been in line with official population statistics then Boris would have had a lead of 6% – not the 10% found in the latest survey.
So is there something in the Boulton suggestion and should not that be affecting the betting?
Below are the age weightings as they have appeared on YouGov’s data pages. The first is from the end December 2007 survey which had Ken one point ahead. The second is from the end February survey which had Boris 5% clear. Then there are the two March surveys where the Johnson leads were 12% and 10% respectively.
Just look at the numbers ringed in red on the graphic and see how the weighted proportion of over-55s has changed. The figures to notice are the weighted ones – those in bold and the overall samples.
I make it that in December the over 55s represented just over 22%. In the later surveys this has been at 38%
I put this to YouGov’s Peter Kellner. This was his response:-
An error crept into our weighting system. We have gone back to first principles and recalibrated our London demographic weights according to the latest official data. The over 55s in our remaining polls will be weighted to 28%.
We have also reviewed our weighting policy regarding BME electors. These will be weighted to 26% – their proportion in the 2001 census. (Incidentally, on this point, I have seen the figure for BME Londoners stated to be 29%. This is true of the total population, but because of the different age distribution, analysis of ONS data shows that the figure is around 40% of under 18s, and 26% among those aged 18+.)
We re-ran our last poll using the new weights for age and BME electors; it made no material difference to the overall results. This is because the party ID weights remain unchanged; it would take quite large differences in the loyalty and defection rates among different demographic groups to make a significant difference to the overall weighted results.
To my mind, the key fact in this campaign so far is that around one-in-five people who “generally speaking” think of themselves as Labour say they would vote for Boris Johnson. If Livingstone can get most of this group to return to the fold (plus do better on second preferences), he might still win; if he can’t, he loses.
Peter’s final point relates to YouGov’s standard practice of weighting samples in line with “party ID” to ensure that they are politically representative. ICM, Populus and ComRes do this by asking respondents if and how they voted at the last general election. The samples for YouGov surveys come from their “polling panel” on whom they have a lot of data already including responses to the question of which party members most identify with.
I have highlighted what those identifying themselves as Labour say they will do on May 1st. Only two thirds plan to support Ken; between a fifth and a quarter said Boris with 7% saying Paddick. The nearest comparative figures from last week’s ICM poll based on how people voted at the last general election broadly support the YouGov trend.
It’s clear that Ken is having trouble retaining London Labour voters from the general election which is why he is involved the biggest fight of his life.
In the betting the Ken price is 1.72/1. Boris is the 0.51/1 favourite.