The Awful YouGov Poll

YouGov have produced a poll which puts the Conservatives on 44%, Labour on 26%, LibDems on 17% and others on 13% (others broke down as follows: SNP/PC 4%, Green 3%, BNP 3%, Respect 0, other 1%).

This gives the Conservatives a huge 18% lead, which comes as a surprise as the ICM poll just three days earlier put the Conservatives on 39%, Labour on 34%, Lib Dems on 19% and others on 9%, giving the Conservatives a 5% lead.

One or both of these poll is a rogue, (there is a huge difference in the reported Labour party figures).

The divergence is interesting because both YouGov and ICM weren't too far off about the 2005 election:

General Election result: Con 33.2%, Lab 36.2%, LibDems 22.6%, others 8%
YouGov May 3-4 2005: Con 32%, Lab 37%, LibDem 24%, others 7%
ICM May 1-3 2005: Con 32%, Lab 38%, LibDems 22%, others 8%

So both pollsters were pretty close to each other and to the final result. So why the current sharp divergence? The two pollsters use different methodology, ICM telephones people randomly and weights by certainty to vote, YouGov relies on a panel of voters on their database and doesn't weight by certainty. It's possible that as politics has moved out of the holding pattern of the last ten years, one of the methodologies has an underlying problem that is giving a skewed answer.

Both methods have vulnerabilities. ICM is vulnerable to the spiral of silence where people are embarassed to admit their affiliations because of the human desire not to be judged, even by a stranger on the phone - this could take the form of people who remember John Major's govt being shy to admit they are Tories, people who are shy to admit they like Gordon in the face of the bashing he's receiving and people who are shy to admit they are BNP.

Because YouGov's polls are on the internet, the spiral of silence problem is eliminated (as data is sent back anonymously). However they are vulnerable to people setting up multiple accounts (as they pay per poll done) and being swamped by activists joining en masse to try to move the polls.

Both pollsters are vulnerable to people mis-remembering who they voted for in the last election (eg people who voted LibDem and let in a Tory might want to conveniently "forget" this and people who are Labour to the core but voted LibDem to "keep the Tories out" might simply answer "I'm Labour" to the question). YouGov tried to get round this by polling their entire database on election day in 2005 to get a record of how they voted (and given how close their pre-election polls were, it's fair to say that most of their database would have answered truthfully). However, they are vulnerable to new panellists joining since 2005 giving the wrong information about how they voted in the last election.

Looking at the data for the recent YouGov poll, I'm struck by the BNP figures - can they really be polling roughly the same as the SNP and PC combined? Given how popular the SNP is? Or have a whole bunch of BNP supporters joined the YouGov panel to bump up their count (the BNP phenomenon is largely a post 2005 one)?

There is an easy way for YouGov to resolve this. They know that the panellists on their books at the 2005 election were sound and they also have the exact details on how they voted in that election. This group can therefore be used as a control. All they need to do is run three opinion polls, one sampling the pre-May2005 database, one sampling the post-May2005 database, and a third poll run based on the whole database. If there are sharp differences in the results, they can conclude that the post-May2005 panel has been corrupted and take action to sort this out. If there are no differences, they should conclude that their methodology and panel are sound, publicise the findings and stand by their polls.