Introducing the PBC Average of Weighted Polls

With David Cameron set to celebrate his first year as Tory leader on Wednesday we thought it would be a good time to launch PBC’s new measure of polling performance – the average of weighted polls. The numbers featured in the chart are from those firms which employ some system to ensure that their samples are not skewed by their methodology and are also members of the British Polling Council.

The latest Average, based on surveys that took place in November, has with changes on October, CON 36.7% (-1.3): LAB 32.3% (+0.3): LD 19.3% (+0.6).. So the Tories would be up 3.5% on what they got at the General Election, Labour would be down 3.9% and the Lib Dems down 3.4%.

Given that the Tories need a margin of about 5% just to win most seats then Labour would have come on top in all but three months of David Cameron’s first year.

It has been said many times before but it needs to be emphasised that to win a Commons majority the Tories need a margin of 10% on votes. Labour can retain a majority even if the party is level-pegging with the Tories.

So only in May, August and October does the PBC Average have the Tory margin big enough for them to win most seats.

Punters seem to be more optimistic about the prospects for the Tories than the PBC Average would suggest. The latest betting on which party will win most seats has the Tories on 0.8/1 with Labour on 1.3/1.

A note on the Average. Included in the figures for each month are the results from Populus, ICM and YouGov. Excluded are the polls from BPIX for the Mail on Sunday because the pollster is not a member of the BPC and surveys from Ipsos-Mori and Communicate Research which do not employ past vote weighting or other device to deal with possible sample error. If NOP, the most accurate pollster at the 2005 General Election, resumes doing surveys then its numbers would be part. Also if any pollster changes its methodology then that would lead to them being dropped or included. Where a pollster carries out more than one survey in a month then the last one is featured.

I think that this is a much better way of showing underlying trends than just focussing on one pollster or, as with other “polls of polls” embracing surveys where no remedial measure has been taken to deal with potential sample bias. For one of the huge challenges that pollsters face is that when you ask people how they voted last time then many more will tell you Labour than actually voted for Blair’s party on May 5th 2005. A consistent average of an 8% overstatement has featured since the General Election.

What the pollsters featured in the Average do is to take some remedial measure to ensure that their numbers are representative. At least some of the 8% can be put down to sample bias and the firms have their own ways of making adjustments.

Mike Smithson