Polling post-mortem

Final vote shares are:

CON: 42.4% LAB: 40.2% LIB: 7.1% SNP: 3.1% UKIP: 1.9% Green: 1.6%

Remember this plot? So, while both the two major party shares look plausible on their own, when you take them together we're a little bit away from the most probable area. This tells us (perhaps not that surprisingly, given the spread and the fact that not everyone could be right) that lots of pollsters did a pretty crumby job!

Top marks go to Survation, who in their final poll predicted:

Con Lab LD SNP UKIP Green 41% 40% 8% 4% 2% 2%

(this is as bang-on as you can get with a poll of 3000 people, realistically!)

There was a real tendency among most of the other pollsters to underestimate the Labour vote share this time, a fact I want to highlight, as it does go against the received wisdom about polling. Good in some ways to know that the bias from 2015 was over-corrected, as it suggests that the sampling issue has been addressed (major attempts were made to fix this but many pollsters added some additional corrections which turn out to not have been necessary). Worst for this were the pollsters who were cautious about youth turnout and assumed it wouldn't change - I suspect we'll see some analysis in the next few days saying youth votes were way up (in fact I've seen various unverified claims say turnout for under-25s was above 70%). Notable, too, was the fact that UKIP were seriously overrated by most pollsters (Survation being the sole exception). A couple of percent might not seem like much, but in practice they got half as many votes as most pollsters were predicting.

All in all, not a perfect outing for the pollsters, but proof it's not just voodoo. A pretty solid performance in many seats for YouGov's fancy model and the exit poll vindicated on the scale of the big picture, which is what you're trying to measure, after all! So, that's that, no need for the pollsters to pack up shop just yet.