Do the underlying numbers point to a different conclusion?

Just been looking at the detailed dataset from the the Telegraph’s Ipsos-MORI poll that came out late last night and in my view the underlying numbers are nothing like as good for Labour as the five point Tory lead might suggest.

After weighting for standard demographics we find that:

300 of those certain to vote in the sample said they had supported Labour at the last general election. Yet only 236 of everybody in the poll said they planned to vote Labour at the coming election.

229 of those certain to vote in the sample said they had supported the Tories at the last general election yet 274 of everybody in the poll said they planned to vote for the party at the coming election.

My simple calculation puts the 2010:2005 ratio for the Tories at 118.7% while with Labour it was 78.7%

So the MORI’s own numbers suggest that Labour is down more than a fifth on last time while the Tories are up by about a sixth. Given that the split in 2005 was L36.2-C33.2 then the latest poll, if it had had politically balanced sample, would have ended up with a lead a lot bigger than the reported 5%.

I know that this is me being mischievous and highly selective but it does show the massive challenge phone pollsters face – because of the systemic problem of the over-sampling of Labour past voters.

Mike Smithson