YouGov's Peter Kellner published a piece last week, reading the last rights to uniform national swing. The article’s seat illustration, showing the Conservatives and Labour on equal seats, despite the Tories getting fewer votes, provoked lots of debate, much of it centred around the Conservatives’ incumbency bonus (or lack thereof) among the class of 2010. In this post I conduct a similar analysis with a few modifications to estimate what I refer to as the skew, sometimes called the “electoral bias” – the difference in popular vote share between Labour and the Conservatives at which the two largest parties have equal seats.
Peter Kellner’s first step was to separate Scotland from England and Wales, which is the non-controversial part. As anyone reading my regular piece will know, Labour have slumped in Scotland since the referendum, with a substantial swing to the SNP (currently running at about 20% since 2010). I pointed out back in October that this has huge implications for the skew, because Labour would lose a huge chunks of Scotland’s seats, but a far smaller proportion of Scottish votes. In technical terms, the distributional efficiency of the Labour vote would be far lower.
Using his vote share figures, a GB-wide UNS would give seat totals of CON 275 LAB 325, but building in the SNP surge changes this to 263-302.
The interesting thing in this scenario is that the Conservatives lose almost half as many seats compared with a UNS projection as Labour do. How is this possible when they only have one seat in Scotland? The answer is that Labour’s drop in Scotland is so large that it (combined with the Scots Tories’ relative stability) has caused the overall GB swing to diverge noticeably from that in English and Welsh seats, among which the LAB-CON marginals lie. So Labour loses seats to the SNP but gains a chunk of them back from the Conservatives. My numbers differ slightly from Peter Kellner’s, seemingly due to rounding somewhere.
I like this two-step approach a lot, but it isn’t foolproof. The first question might be picking up some tactical voting, while the second might not be picking up all of it. I’ve been comparing constituency polling with national polling by the same pollster and, contrary to what is often reported, recent swings have been 0-1% smaller in the marginals than nationally. Tim Smith’s research suggests a difference of around 1% historically – a bit bigger, but clearly less than 2%.