Revealed – How the pollsters produce their numbers
People who like to probe behind the headline figures that are issued by the polling firms are going to have a field day following the introduction of the British Polling Council rules requiring pollsters to publish the background data that was used.
For the first time the main pollsters are having to publish a specified range of information that went into producing their headline figures so that interested observers can subject them to proper scrutiny. This is a very welcome move and will allow us to see in detail if there is something in the pollsters’ methodology that has caused the inherent over-statement of Labour for nearly two decades.
The internet poll by YouGov in the Telegraph poll yesterday had Lab 35: CON 32: LD 23. But according to the information that YouGov has provided the following weightings based on the surveyed individuals’ recall of how they voted in 2001 were applied – CON 28: LAB 56: LD 13.4 So from this we can calculate:-
When we did a similar calculation on the November ICM figures we observed that “you cannot easily dismiss the fact that such a large proportion of those saying they were Labour last time have apparently deserted the party.”
We are sure Lib Dem supporters will be delighted that applying the 172% to what actually happened in 2001 produces a projected share of 32.3%. The Tory figure moves to 37.4% and Labour to 26.25%. What’s extraordinary about this calculation is that it very nearly mirrors what actually happened in the local elections on June 10 this year when the votes split – CON 38: LD 29: LAB 26.
We acknowledge that this is a bit mischievous and we are certainly not saying that this will happen. But the more information you get the less clear things look.
The polling commentator, Anthony Wells, has got a good analysis and concludes that it is “bizarre” that with the published weightings YouGov has tended to give Labour lower ratings than the other pollsters.
The only betting conclusion is BE VERY CAUTIOUS. Full odds round-up.