By Dr Will Jennings

In the Ipsos MORI Alamanac for 2012, Professor Roger Mortimore‘s ‘infallible method’ for predicting the outcome of British general elections on the basis of the previous season’s FA Cup final provides a bit of fun for watchers of British politics. It also makes a serious point about the dangers of ascribing “undue significance to patterns in past events when there is no reasonable excuse for assuming a causal link” (the old chestnut of correlation not equalling causation). Mortimore’s Sweet FA Prediction Model is as follows:

If the FA Cup holders at the time of the election are a team
who traditionally wear shirts in the Conservative colours of
blue or white, ignoring any black stripes, the Conservatives
win most seats at the election. On the other hand, if they play in Labour’s colours, red and/or yellow, Labour wins. (Ipsos-MORI Almanac 2012, p. 52).

Using this method (and with a bit of selective/liberal interpretation of the data), Mortimore’s Sweet FA model is able to correctly predict the outcome of 15 out of 17 post-war elections (and 4 out of 4 London Mayoral Elections).

Source: Ipsos-MORI Almanac 2012.

Election forecasting is not an exact science, and the Sweet FA model provides a nice reminder that successful predictions are not always based on a correct diagnosis of the underlying causal relationship. While forecasters like Nate Silver (and Paul the Octopus) can develop mythical status in offering highly accurate predictions of future events, these do not necessarily guarantee the prediction model will work in future, or that some outlier will lead to it being rejected further down the track.