AngusReid - 34/40/11

There is a new Angus Reid poll that has topline figures of CON 34%(+2), LAB 40%(-3), LDEM 11%(nc). Changes are from their poll for the Sunday Express at the end of January which showed a 11 point Labour lead, the largest any company has shown since the election, so it might well just be a reversion to the mean. AngusReid's poll earlier the same week had shown CON 33%, LAB 41%, LDEM 12%, which is obviously within a single point of this one.

As an aside, I'm getting rather confused about Angus Reid's political weighting, which seems to vary an awful lot! The point of political weighting (and indeed other weighting) is to counteract bias in the sample and reduce volatility - whatever the random variation of the sample, after past vote weighting it should end up reflecting how the pollster thinks a perfectly representative sample would answer the question of how they voted at the last election.

Taking ICM as our comparison, their polls over the last three months have had past vote weighted to the equivalent of CON 36.7-8%, LAB 30.1-2%, LDEM 23.8-9%. It varies very slightly over time because of the way ICM calculate their targets - factoring the average of recalled vote over recent polls to try and account for changing false recall - but their targets for weighting over the last three months have changed by less than 0.1%.

Compare this to Angus Reid, their published tables have the weighted recalled vote varying to a much wider extent - over three months the preportion of 2010 voters saying they voted Conservative in their weighted sample has varied between 34.3% and 37.7%, Labour between 28.4% and 30.9%, Lib Dems between 23.8% and 26.6%. Whatever their past vote weighting is doing, it doesn't appear to be ensuring that the political make up of samples is the same from one poll to the next.