Are Labour really ahead in Glasgow East?

Last weekend the Sunday Telegraph published an ICM poll conducted in Glasgow East showing Labour ahead by 14 points. It's a good lead, but a lot of people seem unconvinced by the poll. Let's look at the reasons.

Firstly, the weighting of the poll. Some people have flagged up that on the unweighted figures Labour are only 4 points ahead of the SNP. This is irrelevant - the unweighted figures are skewed, even before we get to the politics they contained far too many people in social classes AB, and far too few people in social class D.

A few people have raised the question of the political weighting. ICM base their target weights on the last general election result, from back in 2005. Some people have worried whether respondents might have actually answered with how they voted in the Scottish elections in 2007, when the SNP did better. If that were the case, ICM should have weighted past SNP voters to a higher point. Confusion between the two past elections is a possible cause of a higher level of false recall than is usual, but ICM do specify in their question that they want to know how people voted at "the last General Election in May 2005", they don't just say last election, and they do assume some level of false recall anyway. There's no way to rule out this factor, but neither is there any evidence to suggest it is affecting the result.

What the weighting does flag up is how unrepresentative the raw sample was. The proportion of people who owned their homes outright was twice what it should have been, the proportion of ABs in the sample was well over twice the proportion there really is in Glasgow East. At the most extreme extent, only 1% of the original sample voted Lib Dem in 2005 and they needed to be weighted up by a factor of 6 to get to the correct proportion.

None of this is a great surprise - the levels of severe social deprivation is some parts of Glasgow East are going to make them very difficult indeed to accurately poll. The more a sample relies upon weighting to make it representative though, the less robust the results are, and they weren't actually very robust to start with: the sample size of the poll was 516, but once non-voters and some of the don't knows were taken out, the final figures were based on 373 respondents, so we have a margin of error of 5% or so anyway.

There is also the issue of ICM's reallocation of don't knows, which boosted Labour's lead from 11 points to 14 points. Normally this is a sound adjustment based on how people have behaved at past general elections. We cannot, however, have the same confidence that they behave that way in by-elections, and at Crewe and Nantwich it made ICM's figures less accurate, not more so.

So, should we trust ICM's figures? Well, there isn't anything wrong with what ICM have done, it is just a very difficult contest to poll, as evidenced by the small sample size and the extreme weighting ICM were forced to resort to. While the large margin of error and sample problems should make us slightly wary, a 14 point lead is very substantial. The biggest caveat is probably the least technical - a poll is only a snapshot, and it was conducted with two weeks of the campaign still to go. Even if Labour were 14 points ahead a week ago, it doesn't follow that they'll still be there next week.