Friday, May 7, 2010

UK Election Results

With the results declared in 649 constituencies, here's a summary of the UK election results. First, let's look at the likely number of seats for each party (the remaining one will probably go Tory in a delayed election), with their likely final GB vote share in parentheses:

CON - 307 (37%)
LAB - 258 (29.7%)
LD - 57 (23.5%)
Other - 28 (9.8%)

Of the four polling averages that I calculated just before Election Day, the most restrictive one, counting only last-minute polls by established firms, performed best. Compared to that average, the Tories got +1.2%, Labour got +1.7%, while the Lib Dems got -3.5%. That 5-point Lib Dem loss relative to the other parties explains their woefully disappointing result - they actually lost 5 seats relative to 2005! This election can also be considered a failure for pollsters, especially Angus Reid, which completely screwed up their entry into the UK election polling scene by lowballing Labour's share at 23-24% throughout the campaign.

As it turns out, the uniform national swing (UNS) model would have indeed overstated Labour's seat count and understated the Tory one, though not quite by as much as I thought, and certainly be less than than FiveThirtyEight thought. However, because the Lib Dem vote share barely changed in the last election, UNS, like just about any sensible model, would have pinned down their seat number quite well.

The outcome of this election is right in the middle of the chaos range, where neither Labour+Lib Dem (315) nor Conservatives+Unionists (317) have a majority out of the 645 non-Sinn Fein seats. If the Tories don't get Lib Dem support, they will need some combination of SNP (6), Plaid Cymru (3), SDLP (3) and Green (1) backing. This will be very interesting, as all four of these parties are left-leaning. The excitement isn't over yet!

Of the 14 projections I noted, 10 accurately predicted a lack of working majority for any party, with the Tories as largest party. LSE was inaccurate in having Labour as the largest party, while The Times, Angus Reid and BC Iconoclast were wrong in projecting a Conservative working majority (with unionist support in the case of the Times).

Here are the 3 closest projections for each of the major parties:

CON: 1 (tie). YouGov and UK Polling Report (305); 3. Canadian Election Watch (310)
LAB: 1. Election Prediction Project (245); 2 (tie). YouGov and Electoral Calculus (235)
LD: 1. YouGov (80); 2. LSE (81); 3. The Times (82).

Adding up the absolute deviations for the 3 big parties gives the following ranking of projections:

1. YouGov - 48
2. Electoral Calculus - 62
3. ComRes - 63
3. UK Polling Report - 63
5. Riding By Riding - 67
6. Election Prediction Project - 68
7. Trendlines - 74
8. Canadian Election Watch - 84
9. The Times - 85
10. FiveThirtyEight - 105
11. LSE - 108
12. TNS - 126
13. Angus Reid - 161.5
14. BC Iconoclast - 212

Essentially, due to the Lib Dems' unexpected weakness and Labour's unexpected strength, projections relying more on UNS did better than those taking into account geometric vote transfers. With a middle-of-the-pack ranking, I can't say that I'm very happy, but at least I wasn't laughably bad...

The silver lining - and I think a rather important one - is that the most important number in these projections is the Tory seat count, since that dictates how easy/hard of a time David Cameron will have in forming a government. On that account, I ranked 3rd out of 14, just 1 off YouGov and UK Polling Report!