Warren in my comments has highlighted a poll by Kindle Research in the Brighton Argus covering the three Brighton and Hove constituencies. The voting intention figures it gives are:
Brighton Pavilion: LAB 26%, CON 16%, GRN 12%, LDEM 5%, WNV 11%, DK 19%, Ref 7%
Brighton Kemptown: LAB 24%, CON 23%, GRN 8%, LDEM 7%, WNV 13%, DK 19%, Ref 4%
Hove: LAB 26%, CON 23%, GRN 7%, LDEM 7%, WNV 12%, DK 15%, Ref 5%
Once it’s repercentaged to exclude don’t knows, refusals and won’t votes, that works out as
Pavilion: LAB 41%(+5), CON 25%(+2), GRN 19%(-3), LDEM 8%(-8)
Kemptown: LAB 38%(-1), CON 36%(+2), GRN 13%(+7), LDEM 11%(-6)
Hove: LAB 38%(+1), CON 34%(-2), GRN 10%(+4), LDEM 10%(-8)
The poll was professionally conducted (ICM’s call centre did the fieldwork) and demographically weighted to each constituency. So why the rather counterintuitive results, especially when set against the ICM poll of Brighton Pavilion at the end of last year?
There are several reasons – firstly, once the don’t knows and won’t votes are taken away there were very small samples sizes of only 200 or so in each seat. Secondly it is implied that the poll didn’t prompt by party, when ICM included the Green party in their party prompt. As I wrote about the ICM poll, this probably helped the Green party and I wouldn’t normally do it, but it was probably justified in the particular circumstances in Brighton. There’s a fair argument against dealing with the unusual situation by not prompting for party at all, but this would have underestimated the level of support for the Lib Dems (and quite possibly the Greens).
Thirdly, unlike ICM the poll does not appear from their presentation and method statement to have factored in likelihood to vote, something which normally favours the Conservatives and penalises Labour. Finally, and most importantly, the poll did not have any political weighting – phone polls without political weighting of some sort will tend to grossly overestimate the Labour party.
With contrasting figures, I would put my faith in ICM rather than Kindle.