New Harris/Metro poll

There is a new Harris poll in this morning's Metro. The topline figures are CON 35%(-1), LAB 28%(nc), LDEM 17%(-1). Others are up to 20%. Changes are from Harris's previous poll and are well within the margin of error.

The Metro, for some unknown reason, have chosen to draw comparisons from the poll before that - perhaps because it makes it look more dramatic. They claim this poll shows support for the main parties being hit by the fuss over the fake lobbyist sting, which is rather tenuous given that the fieldwork was conducted between last Wednesday and this Monday (so two thirds was before the story broke) and that it doesn't actually show any notable movement from the previous poll.

Anyway, I have also had a look at the tables for the poll, so have some more details of Harris' method. Unlike most other UK pollsters, they do start out by asking people whether they are registered to vote and excluding the 4% who say no and 3% who are not sure. They also use a squeeze question on people who say they are undecided, and while they ask likelihood to vote using a verbal scale, I don't think they actually filter or weight by it.

On weighting, Harris use age, gender, education, region and internet usage (obviously all users are online, but some people are more online than others - this is their way of controlling differences between "fast responders" and "slow responders"). Education is unusual, as is the absence of social class as a weighting variable. Finally Harris weight by their own "propensity scoring" - a figure they have calculated themselves to deal with the differences between people who join online panels and take surveys, and the majority of people who do not. It is based on attitudinal, behavoural and demographic characteristics and calculated by comparing the online sample to a representative face-to-face sample, and weighting as appropriate.