Regional crossbreaks again

There was a time when I used to spend an inordinate amount of time telling people not to spend too much time looking at regional cross breaks in voting intention polls - they have small samples, large margins of error, and are not internally weighted. Needless to say, it goes double or triple for London boroughs! There is a post on LabourList casting doubt on the ComRes poll because it shows wacky figures in some of the Borough crossbreaks and I've seen similar comments on Twitter.

There are only around 20-40 people in the London sample in each borough and often less than 20 who have given a voting intention, so the margin of error is about plus or minus 22 points, even if you did have a representative group, which really isn't plausible with that few people. In other words, if a sample that small showed a borough as 50/50 Ken and Boris, all you could say would be that Ken was between 28% and 72%, and the same for Boris.

Polls are supposed to be representative at the level they are conducted at - a London poll will be representative of London, a GB poll will be representative of GB. They don't claim that they are necessarily representative of sub-samples of that whole, and if you go down to small enough sub-samples, you will find absurd results. It doesn't invalidate the total results.