Given the considerable divergence between Scotland and the rest of Great Britain in recent months, a number of pollsters now publish England sub-breaks. This makes intuitive sense, in an "English polls for English votes" kind of way. Any weighting bias of the sort that can arise in Scotland is likely to be tiny as England is such a large component of Great Britain.
Some have remarked that the swing from CON to LAB in England is noticeably larger than it is across Great Britain as a whole. Because the Scottish Conservative vote seems to be holding up pretty well from 2010, while Labour appear to be down around 15 points, this has to be offset in England and Wales to give the voting intention figures we see for GB. Since that’s where the marginals are, this appears significant.
But if you actually look at English marginals, they don’t seem to be behaving like the rest of England – they actually seem to be behaving more like Great Britain as a a whole. That might be a coincidence, but the swings have consistently been smaller:
In fact in the second half of 2014 the swing was 1.5 points smaller in the marginals. Why are battleground seats behaving differently? Firstly it’s worth remembering that these England numbers are before the “spiral of silence” adjustment is applied, whereas the constituency polls include it. But that doesn’t account for a net difference of 3 points.
Next, as I wrote previously when I looked at the skew in detail, there could be a ‘hidden’ incumbency effect, or regional patterns (since marginals aren’t distributed evenly across the country). And we can’t rule out an “in house” house effect, between national and constituency polls, (although cross checks suggest this isn’t a significant factor – more on that in a moment).
But what seems to be a bigger factor concerns the Lib-Lab switchers – a significant part of Labour’s support nationally has come from former Lib Dems. A look at the 56 English seats that the Tories won by less than 10 points from Labour in 2010 shows an average Lib Dem vote share of 17.6%, compared with 24.2% in all English seats. Through decades of tactical squeezing (and probably fewer resources being put to work), there were considerably fewer Lib Dem voters in the Lab-Con marginals than nationally. Has the drop therefore been smaller? It would appear so:
In other words, it seems that Lib-Lab switchers are (as a proportion of 2010 voters) disproportionately few in most of the places where Labour needs them (where the Lib Dem vote is falling 4 or 5 points less), and more plentiful where there were more of them to begin with, seats whose outcome they are less likely to alter. To paraphrase Neal Lawson, the wrong Lib Dems are voting Labour.
All of this has two consequences. Firstly, applying separate uniform swings to England (or England and Wales) and Scotland helps address the Scottish problem (to some extent), but it still doesn’t get England right. Secondly, aggregating back up to Great Britain as a whole, this marginals picture (especialy when combined with Scotland) has a significant impact on the skew. In 2010, the skew was about 4 points – Labour and the Conservatives would have had equal seats if Labour were 4 points behind in the popular vote. This time that is very unlikely to be the case.
But you don’t have to take my word for it. The electionforecast model, which is based on constituency-level YouGov microdata, paints an extremely similar picture. At the time of writing, their forecast has the Tories ahead by a few seats, but a few days ago, when they had the two largest parties level, it was on the basis of vote shares just one point apart:
— NumbrCrunchrPolitics (@NCPoliticsUK) February 6, 2015
So unless it’s actually that close, I still expect that the party with the most votes across Great Britain will also have the most seats. But the message doesn’t seem to have got through to everyone – statements such as “Labour can be the largest party on substantially fewer votes” continue to be made – contrary to the evidence. Sometimes this reflects a lingering uniform swing mindset, sometimes it suits someone’s narrative.
Bottom line – the evidence suggests the skew has pretty much gone.