Are the ComRes numbers down to weightings changes?

Are the ComRes numbers down to weightings changes?

CON 30(-10) LAB 22(+1) LD 18 (nc)

Tonight’s second shock poll is from ComRes for the Independent which shows a dramatic collapse in the Tory vote to just 30% while Labour is up one point at 22%. Fieldwork took place from Friday until Sunday at exactly the same time as this evening’s MORI poll which had C40-L18-LD18.

So how come that all the other firms which have polled in recent days have the Conservatives in the range of 39 -41% while ComRes is showing a figure which is even down a tenth on what Michael Howard’s party got at the 2005 general election?

Part of the answer lies in the way it handles past vote weightings which is the device used by the firm alongside ICM and Populus to try to ensure that the sample is politically balanced. Respondents are asked how they voted last time and weightings are given which take account of what happened four year ago.

So in the same way that you weight for gender and age based on known data the past vote calculation uses the results last time as its base-line with an adjustment to take into account so called “mis-remembering”

With Populus and ICM these figures almost never vary and when they do it is by small fractions. For some reason the ComRes numbers move all over the place (see below) and as I’ve recorded here before I’ve been trying since last year to find out why.


Just contrast that with ICM’s weightings in the same period.

The difference between ComRes’s weightings tonight and the standard approach of ICM is to inflate the Labour headline figures and to deflate the Tory ones. My guess is that if ICM’s weightings had been applied then the CON-LAB margin would have been fifteen percent or more.

The ComRes methodology dates back to just March 2007 so, as yet, it is untested in a general election. This evening I put these points to Andrew Hawkins, the ComRes CEO. This was his response:

ComRes’s past vote weight is calculated by taking into account the past vote recall figures from the current poll and an average of the past twelve polls, this is used with the actual outcome in 2005 to calculate the weight.Some firms use target marginal weights; ComRes uses factor weighting (taking a figure for past vote recall weighting centered around 1). This score will up- or down-weight respondents based on their past vote recall in correspondence to the figure assigned for factor weighting.
These weights take into account recall bias and sample bias in comparison to the actual outcome in 2005. Of course the specific proportions of the weight given to the target of the actual outcome and the recall figure (taking account of bias) do vary between all polling firms.
I appreciate that the technical detail of target marginal versus factor weighting don’t always make for the most clarity and for that reason (if for no other) we are looking at how we present and manage data so as not to risk misleading readers. I am happy to commit to making public our conclusions on this before the end of June.

Mike Smithson

Comments are closed.