The Problem with Swingometers
John Lanchester · 72 Hours to Go
So, 72 hours to go, and the polls are pointing with a quavering, accusatory finger at a hung Parliament. Or maybe it’s a confident, optimistic finger – anyway, that’s where they’re pointing. YouGov has the Tory/Lab/Lib Dem projection as 34/28/29, ICM has it as 33/28/28. UK Polling Report looked at all five of the polls published on Sunday, four of which showed some momentum for the Tories after the third debate, and all of which showed the Tories as falling short, with the lowest projection 264 seats and the highest 315 (the target, remember, is 326).
Those estimates assume something called a Uniform National Swing, i.e. that the vote will move by the same amount everywhere. This is probably an unsafe assumption, of the necessary-evil kind which always needs to be made in building a mathematical model. On UNS, the Tories need a 6.8 per cent swing away from Labour. But the Tories are, obviously, concentrating their efforts in marginal seats, which complicates things, as does the surge in support for the Lib Dems, which will hurt Labour in some places but will hurt the Tories in others; and in the last three elections the voters have shown a big propensity for anti-Tory tactical voting. If that changes then the electoral map will change too. For those of you who are far gone in poll nerdery, there is a fascinating, detailed and very technical analysis by Nate Silver here.
Silver is the US pollster whose site played a big role in analysing and explaining the process of polling during the presidential election. He says that ‘the science of UK electoral forecasting is not terribly advanced’, and does a good job of explaining just why it’s so complicated. He rejects UNS and has a different way of doing it which I’m not going to attempt to summarise. His model has the Tories on 299 and the Lib Dems on 120, ahead of Labour on votes but behind on seats.
The spread-betting markets, which have historically been more accurate than polls, are far more bullish about the Tories, and show them as clocking in at 323-328 seats: in other words, the middle of the spread is bang on the target for a majority. If Silver is right, you can make some money betting on the Lib Dems, who according the the spread are on 82-86 seats. Put this all together, and it fits the definition of ‘too close to call’.
Comments
He starts with votes achieved by each party in 2005, and then asks where, for each party, those votes will go in 2010; of everyone who voted Labour, for example, in 2005, how many will again vote Labour in 2010, how many Tory, how many LibDem.
The point of doing it like that is that you can reproduce the same UNS with many different combination of voter traffic, and it's the choreography of shifting allegiances, rather than the swing per se, which determines how the seats will fall.
The problem with this approach is there's not a huge amount of data on voter switching, and Silver's projections of how that might work are subjective. (There's a fairly amicable ding-dong going on around this issue between FiveThirtyEight and Robert Ford of Manchester University, who has a rather different methodology.) The responsible thing when there's a significant subjective element in your forecast model is to develop different scenarios, which is just what Silver does.
His baseline has the Tories on 299, but I've made my bets on the basis of the scenario Silver came up with assuming that Cameron did best in the third debate; 367 to the Tories, 154 to Labour, 97 to the LibDems. I've also got some money on a smaller overall majority for the Tories, and for the LibDems getting more than 100 seats. I'm not betting on there being a hung parliament; Intrade now has the probability of that outcome at less than 50% (down from the lows 60s before the third debate).