Thursday 29 May 2014

Statistical models and the 2015 General Election


So now we've got lots and lots of polls. We've got a massive dataset from the local and European elections. And we've got about eleven months left until the General Election. The result should be resolving in front of us, based on statistical modelling of past experience, polling trends and electoral data. Right? Well, hold your horses. We're getting a better idea. We have the range of likely results. We're able to exclude, historically, what we definitely think won't happen. And experts' prognostications are beginning to converge, as they usually do. But we're not quite there yet.

So let's take a look at the range of easily-available statistical models - and what they tell us about the general election. We're going to leave the Liberal Democrats out of things for a moment, because we're focused on which of the 'big two' is going to be the largest party. And, let's be honest, it's best to leave them to their private grief for now.

Leo Barasi's blog, using poll data one year out from General Elections: Conservatives 283 seats; Labour 323 seats.
Rod Crosby's simple regression calculation, based on last week's council polls (implied): Conservatives 326; Labour 262.
Stephen Fisher's model, based on the polls' behaviour in the run-up to previous elections: Conservative 307; Labour 284.
Chris Prosser's approach, looking at the last local election before the big one: Conservatives 304; Labour 279.

Oh, you want an average? Well, don't we all. The Barasi and Crosby approaches are the crudest, and probably best to rate lower down the ladder of any overall look at these forecasts. And averaging a load of shots in the dark doesn't mean all that much. So this is basically statistical rubbish. But a mean score of all those seat counts? That'd give the Conservatives 305 members of the House of Commons, and Labour 287. That's a pretty tight, but still a clear, judgement in anyone's language. What happens next will depend on what the Liberal Democrats think about going on in a Conservative-domianted government.

So here's the rub: Labour are unlikely to win the 2015 General Election, at least outright. Their leader's ratings are pretty poor, while David Cameron's (above) are mediocre to okay-ish, and modelling past elections (opens as PDF) on the basis of 'satisfaction with the Prime Minister' gives us pretty strong results.

But it could be close in terms of seats. And Labour still has at least a stab at being the largest party. Less than 20 seats may divide the parties in total. We've been saying this for months, of course - perhaps years. But it's good to have it all parcelled up and quantified. It's a comfort blanket of (apparent) mathematical certainty. And it shows that, on the basis of past elections, Mr Cameron's probably going to be with us for some time yet. But that 'probably' hides a wealth of uncertainty.