Wednesday, 31 March 2010

Politics is always improved by a little injection of Science.

We're into the last few weeks before the General Election. At least, I assume we are, our Glorious Leader having apparently neglected to officially call one, and seeming to want to hang on until the last possible minute. And who can blame him, frankly? The poll numbers have been showing the Conservative lead evaporating recently, going down from a lead of around 18% last July to a mere 7% or so. The Budget seems to have pushed it back in their favour (not unusually, given how people don't tend to like being told they're going to pay more), but whether that's the start of a new trend back towards the blue end of the scale or a mere blip on the way to the great British public being completely undecided is something we just won't know for a little while longer.

Either way, it's an exciting time in politics, and the pollsters are having a lovely time. But how do we interpret the poll results? Well, I've been letting my geeky side (which is, I must admit, pretty enormous compared to my non-geeky side) run riot a little more than usual, with the result that I'm now in a position to put forward a tentative projection.

There's two main ways of predicting Parliamentary results based on poll figures. The first is the Uniform Swing Projection. In a nutshell, this method compares the vote share predicted by a poll with the vote share at the previous election to calculate the swing towards or away from each party, then applies that swing to the vote share in each separate constituency.

It's not the most sophisticated way of doing predictions, and suffers from a few limitations. For example, a strong negative swing can result in a prediction of zero votes for a party, which is fairly unlikely. It also fails to take into account boundary changes and the size of a constituency, and is likely to hit particular problems in an election like this one where a number of MPs have resigned or have had their reputations tarnished by the expenses scandal. It's a good place to start, though, and gives you a reasonable idea of the general trend of the election.

So, what's the prediction? Well, based on the most recent YouGov poll listed on the UK Polling Report, which gives figures of 38% Conservative, 31% Labour, 19% Lib Dem, 12% other, I'm projecting a Hung Parliament with the Conservative Party short of a majority by 51 seats. That's not far off UK Polling Report's current projection of a Conservative shortfall of 19, so I think I'm in the right neck of the woods.

Of course, that's only one poll, and I've heard bad things about YouGov (it's an online outfit, which does make them susceptible to outside influence). So let's see some more recent results, with my projections.

  • 29th March (YouGov), Con 39% Lab 32% LD 18%. Projection: Hung Parliament, Con down by 35
  • 29th March (Opinium), Con 38%, Lab 28%, LD 18%. Projection: Hung Parliament, Con down by 3
  • 28th March (ComRes), Con 37%, Lab 30%, LD 20%. Projection: Hung Parliament, Con down by 59
  • 27th March (BPIX), same results as ComRes above
All of which seems to suggest that there may be quite a bit of variability, but the Conservatives have a lot of work to do if they want to form a majority government.

I've got some more work to do on my projection model, and I've had some ideas for getting it a bit more accurate, so watch this space for more details (I'm also going to open-source my code shortly so you can play around with it for yourself). Until then, I'm going to leave the parties to frantically scrabble for those extra few votes...

No comments: