Last night I wrote that I wouldn't be staying up because:
In the event, I did stay up and we got all three. First of all, in Clacton, we can see that the polls, though taken six weeks or so before the date, were broadly accurate:
In the case of UKIP, the polls averaged 60% and they got 59.7%. The Tories did 2 or 3 points better and Labour about 3 points worse, but under the circumstances, no error of note.
Up north it was a very different story. The Labour share was 8 points lower and UKIP 8 points higher:
That is huge. In the 1992 general election, the overall error in opinion polling was something in the order of 9 points – this is roughly double that. Why were the pollsters fairly accurate in one contest between UKIP and a ‘big two’ party, but the same two pollsters so wrong in another? It is possible that there might be a Carswell effect, a turnout effect, or maybe a difference in sampling bias between these two equal and opposite seats. More on polling to follow.
Having a Labour vs Ukip and a Conservative vs UKIP contest on the same day gives us a rare chance to look at combined results, or more interestingly, combined swings:
The so-called “LibLabCon” parties are all down on this measure, with a swing from LAB to CON of 7.6%. Applied nationally, that would put Labour 8 points ahead of the Conservatives. But governing parties normally take a hit in by elections. Incidentally, the changes would also show UKIP on 51% and heading for a landslide next May. Anthony Wells did a great writeup, including the reasons why by elections are a strange animal.
One final point… The swing Douglas Carswell achieved from himself (in different colours) was 44.1%. This is, of course, astonishing, but Simon Hughes, who was a Sky News Pundit last night, holds on to his 1983 record swing of 44.2% – Carswell would have needed just 112 more votes to beat it. He should have demanded a recount!
I’m currently writing a detailed piece on polling, to be published soon. To get it first, be sure follow this blog on Twitter: