ComRes have a poll of marginal seats out tonight covering the forty most marginal seats with Labour and Conservative in first and second place (so 25 with Tory incumbents, 15 with Labour incumbents). Collectively the vote in these seats was CON 37%, LAB 37%, LDEM 18%, UKIP 3% at the last general election. In today’s poll ComRes found current support of CON 33%(-4), LAB 35%(-2), LDEM 8%(-10), UKIP 17%(+14). That’s a swing from Conservative to Labour of just one point, far lower than the swing shown in ComRes’s GB polls (it would be the equivalent of a national poll showing a Conservative lead of five points) suggesting Labour are doing worse in key marginals than in the country as a whole.
ComRes also break down the figures for the Conservative held and Labour held seats (though given the sample was only 1000 to begin with, caveats about sample size obviously kick in here). In the Tory held seats the Conservatives have a lead of 2 points (no change since the election), in the Labour held seats Labour have an 8 point lead (up 6 since the election). What that means is if these figures were repeated at the general election none of these Con/Lab seats would change hands at all – the Tories would hold theirs, Labour would hold theirs. In practice it wouldn’t work like that of course, there isn’t a uniform swing and some seats would probably switch in both directions, but it’s a suggestion that there isn’t really a swing to either Lab or Con in the key marginals.
A word about polls of marginal seats. I’m always a bit wary of reading too much into them. In theory they *should* be far more useful in predicting elections, in practice… perhaps not, probably because of how uncommon they are. They tend to be rare one-offs and show contradictory things: the last Ashcroft poll of marginals showed Labour doing better in marginals, the previous Ashcroft marginal poll has show the Tories doing a tiny bit better, this one shows the Tories doing much better. That might be different methods, or change over time (there are years between those polls!) or just normal margin of error. With so few such polls it’s impossible to tell – yet because they are so rare there’s a temptation to read a lot into them. You shouldn’t. Rarity of a poll doesn’t decrease its confidence interval. The good news is that this ComRes poll is apparently the start of a series, so assuming they are relatively frequent we will have a better chance to at least be able to look at trends and averages over time and how they relate to ComRes’s national figures.