The industry has a lot of work to do if it wants to be given another chanceby Martin Boon / June 9, 2017 / Leave a comment
What goes around, comes round.
Only yesterday, few people believed Survation’s Damian Lyons-Lowe, whose single point prediction for the Tories implying a hung parliament was uniformly derided. Similarly, YouGov’s seat projection model predicting a hung parliament was quietly downgraded almost off the agenda. It was supposed to be a Tory stroll.
Wrong, wrong, wrong—or should I say, right, right, wrong. The champagne corks will be popping in Survation Central, with that poll well in the frame for the most accurate. Equally, the dramatic successes of Yougov’s seat projection model might well have changed the way predictions are made moving forward, full stop.
It’s still uncertain what the final share of the vote will be, but somewhere around 43% for the Tories and 40% for Labour is likely. If that turns out to be correct, then (it might be somewhat surprising to hear) it’s not been a disastrous night for the pollsters in general—if you measure performance in traditional form of average error on the vote shares. The best poll will end up with an admirable error of around 0.5%, with the average across the final polls in the area of 2%.
But a hung parliament was not what the polls were saying for the most part, which means that for the second consecutive election the main storyline has been missed: a hung parliament instead of a usable Tory majority. This might be harsh—and some pollsters will have every right to exempt themselves—but delivering the right story is the main point of polls and most of us will be disappointed to have failed to do it, again.
But what’s remarkable about this polling story is the complete about-turn in the nature of error. Labour have been over-stated in the polls at just about every election in memory, but this time around the golden rule has been proved broken. With the exception of Survation, every polling company polled Labour too low, with one company under-shooting by 7-points.
Knee-jerk reactions are unhelpful, but the methods which didn’t work in 2015 did work on this occasion. New methods designed to confront the problems of last time made things worse. There’s a sampling inconsistency here that most of us pollsters have failed to deal with—there is much more work to be done, if, as an industry, we’re given another chance to do it.