Last week, Lord Aschroft created a stir when, in his latest round of constituency polling, he showed Ukip within 12 points of Ed Miliband in Doncaster North. He put the party on 28 per cent, up from 4 per cent in 2010.
But the eagle-eyed Anthony Wells – YouGov’s chief political analyst, editor of UK Polling Report, and provider of May2015‘s historical polling data – spotted an error. The poll was weighted to include too many Tories and too few Labour voters. Ashcroft has always been unerringly impartial in his reporting of his polls, this was a simple data entry error.
The new weightings have put Miliband ahead by 29 per cent, not 12, and give him a higher share of the vote than 2010, rather than a lower one. Our recent, pre-poll suggestion that Miliband could face a battle for his seat now has little support.
The graph below shows how the new poll differs from Ashcroft’s previous poll, and the results of the 2010 election.
This wasn’t an inconsequential poll. Ashcroft’s pre-release tweet was shared on Twitter nearly 600 times.
Released tomorrow polling in Doncaster North showing UKIP 2nd so if Tories tactically vote for UKIP Miliband loses #votetorygetlabour!
— Lord Ashcroft (@LordAshcroft) November 26, 2014
But how exactly did Ashcroft wrongly weight the poll? There are two steps to creating a representative poll: an initial sampling of the population and then a weighting of them to make sure they are demographically representative (e.g. the correct proportion of age groups, gender, party ID, etc.).
The error in Ashcroft’s poll was at the weighting stage. His weightings are meant to begin with the 2010 result in Doncaster and make some allowance for the past recall of voters (he asks those he has sampled who they voted for in 2010, which will differ slightly from the actual result). The problem with his first poll was that the past recall numbers for the Tories and Labour were mixed up.
The new weightings are shown below (here are the tables for the first and second poll). As the graphs show, the old weightings implied the Tories won the seat in 2010. The new weightings are close to the 2010 result (past recall accounts for the slight difference), and mean more Labour voters are in the poll.