So the votes have (finally) been counted everywhere and the results are known, so the unpicking can begin. Each of the elections deserves proper analysis in its own right, which I'll do in due course, but here I'm going to give an overview of Super Thursday. Yes, I said Super Thursday. As Matt Chorley from the Times put it:
FFS. Now people complaining about using Super Thursday. 45m voters, 7 diff elections: 3 govs, London, 1st Corbyn test. Get over yourselves
— Matt Chorley (@MattChorley) May 5, 2016
There were some general themes across all of the contests:
- Polls, on the whole, were accurate
- The pro-Labour bias seems to have been fixed in these lower-turnout elections (we can’t yet say whether it has among a general electorate)
- In these contests, phone and online fieldwork got similar results in most cases, once timing effects are taken into account
- However online polls, almost without exception, overstated support for UKIP
The last of these has significance for the EU referendum because UKIP voters are overwhelmingly pro-Brexit, while other parties’ supporters are to some extent split. In contests where UKIP’s vote share was polled (in other words, London mayoral first preferences, assembly constituency and list votes in London and Wales, and lists for the Scottish Parliament), 12 online polls were conducted in the final two weeks. Not one of them underestimated UKIP support, while 10 of them overestimated it.
On average, online polls had UKIP 3 to 4 points too high on both London Assembly ballots and both Welsh Assembly ballots, while in the London Mayoral race and Scottish regional lists the errors were smaller, but still sizeable relative to the party’s vote share. By contrast, there were only three phone polls across these contests, of which one had UKIP a point too low, one a point too high and the third spot on.
While support for leaving the EU and votes for a pro-Brexit party are clearly not the same thing, all of this is at least consistent with the analysis I did with James Kanagasooriam at Populus in Polls Apart – online panels seem to be too Leave-y. You may also recall some of our critics attacking our analysis, and the British Election Study and British Social Attitudes Survey that it cited, on the grounds that those surveys reported UKIP vote shares that were too low. In the light of these results, they might wish to reconsider their conclusions.
To take each contest in turn:
NCP’s expectation was that Khan would win with a vote share towards the lower end of the 56-60% range that the polls had been showing – he got 56.8%. The slightly surprising thing was the less-bad-than-usual turnout of 45.3%. Despite the impression that Tory voters weren’t enthused by their candidate and that Labour voters weren’t enthused by their party, Londoners of all stripes showed up to vote.
On the thing that mattered most, the polls were spot on. All of the polls conducted late on had Khan on 56 or 57%. The Survation phone poll had the winning candidate of 60%, but the fieldwork was done before the Ken Livingstone row. The penultimate YouGov poll, which was done at a similar point in time, also had Khan on 60%, so it does appear that there was a subsequent swing towards Goldsmith. Additionally, across the two ballots for the Assembly, YouGov (whose poll was the only one) nailed the Labour-Tory gap.
But beneath the surface, a few things stood out. UKIP, as elsewhere, was overestimated by everyone besides Opinium and ComRes, while the Greens were understated, as were the Womens Equality Party on the Assembly list (though the one poll of Sophie Walker in the Mayoral contest was exactly right).
Wales did see Labour fall further short a of a majority as predicted, but it could have been worse – they are now 2 seats short of the winning line. Only YouGov polled here, and their polls were again pretty accurate. This is important because of all the main races last Thursday (London and Scotland being the others), Wales is the most representative of Britain as a whole in terms of attitudes to the EU.
The average absolute errors were slight, and there was no sign of the traditional pro-Labour bias (in fact, Labour did slightly better than the polls). As elsewhere, UKIP did far worse – 3.5 points on the final and recall constituency polls and 3 points on each list vote poll. To be clear, this is not to single out YouGov – it’s likely that if other online pollsters had polled Wales they would have had similar results.
That the SNP won a third term with a huge lead was no surprise, but the Scottish election was pretty eventful. I said it was more likely than not that the Conservatives would come second ahead of Labour. In fact they not only came second, they did so well in constituency seats that (with a bit of help from the Lib Dems) they were able to deprive Nicola Sturgeon of an overall majority, and in doing so, chalked up their biggest gain in vote share in Scotland since 1931.
The polls were slightly off with all of them underestimating the Tories and nearly all overestimating the SNP. The pollsters will undoubtedly look carefully at what happened, but on the face it, there is a tell-tale pattern – the later a poll was, the more accurate it was. This looks a bit like the 2014 independence referendum and the 1970 Westminster election, where the polls were about right when they were done, but people actually changed their minds after being polled.
But even the actual popular vote shares would have returned Nicola Sturgeon with an overall majority on a uniform swing, so it will be interesting to see how much of the non-uniformity was down to pro-union tactical voting. At first glance, there are certainly signs of it. Additionally, rural Eastern areas that have tended to be strong for the SNP but which voted heavily against independence (think “tartan Tory” country) showed big direct swings away from the nationalists.
The local elections call for detailed analysis, but the big picture is that relatively little changed in terms of seats or council control. As expected, Labour made a net loss of councillors (-18), although the Conservatives did too (-48), with the Lib Dems (+45) and UKIP (+15) gaining. The BBC PNS and Plymouth NEV measures, estimates of the vote share had every ward in Great Britain been fully contested, were as follows:
BBC projected national shares (changes vs 2012) CON 30 (-1) LAB 31 (-7) LIB 15(-1) UKIP 12 (not calculated for 2012) 3% Lab to Con swing — NumbrCrunchrPolitics (@NCPoliticsUK) May 6, 2016
Rallings & Thrasher/Times (Nat’l Equivalent Vote, chgs vs 2012):
I don’t think either main party should be over the moon with these results. Labour avoided the bloodbath that many had predicted, however the only conclusion that can sensibly drawn from that is that expectations were too low. And while it’s true that Labour is at about the same sort of level (relative to the Conservatives) as at the equivalent point in the last cycle (2011), history suggests that Labour got off surprisingly lightly in 2015 considering its local election performance. In 2020 it might not.
But the Conservatives shouldn’t be too happy either. Given Labour’s recent problems and poor fundamentals (particularly its polling on leadership and the economy), the Tories might have been expected to do rather better than five years ago. As such, these numbers are consistent with them winning in 2020, but with a majority, not a landslide.
Nothing to see here, both should have been easy Labour holds, and they were. But congratulations the two new MPs, Chris Elmore (Ogmore) and Gill Furniss (Sheffield Brightside and Hillsborough).