One good thing: a better electoral forecasting model

So it looks like the local elections didn’t go so badly. While we’re in the intermediate phase between Corbyn’s Labour doing something reasonably well, and them throwing it all away through some sort of terrible cake-and-arse juggle, I’d like to take note of something.

John Curtice reckoned we should expect a net-loss of about 170 seats. Jeremy Corbyn said very publicly that he didn’t believe Labour would lose any seats at all. I’m not sure whether he intended this as a prediction, or just felt he should put a brave face on things and cheer up the troops, but it amounts to predicting a net-loss of zero seats.

One thing pretty much everyone agrees on is that the bad polling had a profound, if subtle, effect on the 2015 election campaign. Right up until the last minute, it was possible to believe – even necessary, on the grounds that you should base beliefs on the best available evidence – that we were going to win.

People who worried about a coalition with the SNP were worried because the polls looked like that might happen. People who worried that Ed Miliband wasn’t keen enough on coalition with the SNP and decided to go Green or not bother voting did so because the polls looked like a Labour/SNP alliance would be necessary. I know at least some of these people existed because they used to shout at me on Twitter. Tories who thought the public was scared of the SNP acted as they did because the polls looked that way. Sizable chunks of the Tories’ policy agenda that just look weird in the post-2015 context only existed because the polls looked like another coalition was the only way they had a hope of getting back in. So much of the ambiguity and chaos of the 2015 election is down to the fact that politics is largely perceived through polling, and the polling was crap.

Blaming the pollsters is futile. For a remarkably policy- and data-oriented Labour team, Ed Miliband’s staff seem to have comprehensively failed to be intelligent consumers of polling. As a result, it wasn’t until final canvass returns from the West Midlands were analysed with two days to go that anyone realised the polls might be wrong, and the result was the Edstone. For those of us who weren’t privy to those data sets, we just had to wait for the epic punch in the guts that was the 10pm exit poll on the night.

If Corbyn meant it about not losing any seats, that forecast was off by 23 net seats out of 866 contested or 2.6%. That’s pretty good. I recently reviewed a market share forecast I prepared in 2012, and was more than pleased – in fact, ecstatic – to be within 4%. The nature of forecasting is that it’s very hard to tell quality (the technical term is skill) from good luck. The nature of forecasting is also that when I started drafting this post, the score was -2 out of 866 or 0.23%, and my point was stronger by a factor of 10 or thereabouts. Either way, it’s not 170.

But Labour would have been vastly better off in 2012-2015 had it been able to derive quality forecasts from polling or canvass data. If party HQ has a much better analytical capability, or the quality and coverage of canvassing is much better, this is an important fact about the practicalities of politics.

It’s just a pity we keep going on about Hitler. Also, the e-mail hasn’t been quite as bad as the leadership campaign, but as of yesterday 21 out of the first 50 messages in my inbox came from different bits of Labour. I’m having words with Tom Watson about it. Seriously:

@yorksranter can I talk to you about this after the election? It's a helpful case study of how not to do email.

— tom_watson (@tom_watson) May 4, 2016