Is this the way to work out how the firm does its weightings?
As regular visitors will know Anthony Wells of UKPollingReport and I have been in discussion for some time with the ComRes polling organisation trying to work out how it does its past vote weighting – a measure that all but one of the pollsters adopt in one form or another to ensure that their samples are politically representative.
The process is quite simple. Respondees are asked how they voted last time and a weighting is applied so that the sample is broadly in line with what actually happened on May 5th 2005. I say broadly because it’s recognised that a proportion mis-remember or have false recall so the numbers you get are not going to be precise.
Each of the firms has its own formula for dealing with this which generally assumes, amongst other things, that a significant proportion of those polled will say “Labour” when they didn’t. The operation of this from the latest ICM and Populus polls is featured in the panel above and as can be seen there was in each case a sharp scaling back of the Labour numbers. In both polls the LD weightings were increased though ICM had to reduce its Tory numbers while Populus had to boost its figures for the party.
This has been fairly standard stuff for ICM since the mid-90s and for Populus since its foundation six years ago.
ComRes also says it applies a similar adjustment but there is nothing in the data-sets it issues to indicate how the weighting process actually work. What would really provide clarity would be something similar to the tables reproduced above.
What’s concerned Anthony and I is that what are said to be the weighted figures – after the adjustments have been made – seem to fluctuate by big amounts from poll to poll. This really shouldn’t happen – the Populus and ICM weightings only move by minuscule amounts from poll to poll.
I think this is probably down to something in ComRes’s presentation though it will be hard to comment with confidence on their next survey, expected tomorrow night, until we can see data like that featured above.
UPDATE 1945. I have now received a pile of data from ComRes which I will analyse with Anthony Wells.
Helping to keep PB going. A number of people have said that they’d like to help with PB’s running costs by making a donation. We would be most grateful. To do so please click on the button below.