A few days ago, I doodled this:
The idea was to use a three way swingometer to show the changing fortunes of the three parties with respect to each other in terms of votes cast today compared to 2005, or as input to a visualisation that would allow folk to play around with swings from one party to another (based on election 2005 results) and see the effect on likely allocation of seats this time round. But then I went and did something else…
Anyway, via John Naughton just now, I picked up on this post – Labour Danger: Uniform Swing Calculations May Understate Risk to Incumbents – which comments about the dangers of the uniform swing model whose only attraction, as far as I can tell, is that it is “fairly easy to calculate”. Instead, “an alternative approach [is proposed] which, while also based on fairly simple assumptions, is potentially more robust. The approach works by assigning shares of one party’s 2005 vote to another. For instance, what happens if 10 percent of people who voted for Labour in 2005 defect to the Conservatives, 15 percent of Labour’s voters defect to LibDems, and 10 percent of the Conservatives’ voters defect to LibDems?”
I don’t have time today to hack together my three way swingometer thingy, though something similar already exists here:
…but maybe if we get another election after the summer, I’ll hack it together for then…?;-)
PS see also UK General Election 2010 – Interactive Maps and Swingometers.
One thought on “Debunking Uniform Swing… Maybe Next Time…?”
I was very strongly struck by two things as the results came in.
First was how totally bogus the two-party ‘Swing’ reported by the BBC seemed to be for pretty much every seat. You’d get a result like Con +10% LD -4% Lab -4% reported as something like a 6.7% LibDem-to-Con Swing. Most constituencies had very complex results, and the overall impression I got was that the between-seats standard deviation in this election was very high indeed. It seemed surreal and pointless to reduce this complexity to a single figure.
Second was how good it was overall in guesstimating final seat results anyway. The exit poll prediction from BBC/Sky turned out spectacularly close: the prediction was Con 307, Lab 255, LD 59, Others 29. We got Con 306, Lab 258, LD 57, Others 28.
The moral I’m taking home is the power of a simplifying assumption for estimating the big picture, even when it’s manifestly wrong on the details.
Comments are closed.