Original URL: http://www.theregister.co.uk/2010/07/13/world_cup_stats/
World Cup Stats Prof: I was right all along
Holland had 88% probability of not winning World Cup
Spain's extra time goal against Holland in the World Cup final denied our resident stats expert victory in predicting the tournament winner. Or did it?
Before a ball had been kicked Dr Ian McHale, senior lecturer in statistics at Salford University, had modelled the teams' performances in the run-up and had Holland as favourites to win.
But as an academic in the field of statistics, what he'd have preferred to have said on the matter is that at the start of the tournament Holland had an 11.86 per cent (lets round it to 12 per cent) probability of winning it. This made them favourite, as Spain had an 11.47 per cent probability, with Brazil at 8.17 per cent.
“Actually there was an 88 per cent probability they wouldn't win. So I was right. By far the most likely outcome was that the wouldn't win,” said McHale. (I assure you he was laughing while he was saying this.)
McHale is actually pretty pleased with his model's performance. Before the start of the tournament Holland were ranked 4th in FIFA's tables, and the bookies had them at considerably longer odds than Spain, Brazil, Argentina and England (you wot?).
But perhaps the real result was predicting Uruguay as third favourites after the group stage.
“It's still pretty pleasing that Holland were in the finals and Uruguay got to the semi-finals – I don't think many people would have predicted that,” said McHale.
Spain's loss to Switzerland in their opening game was what affected their probability of winning the tournament in McHale's model.
“If Spain had beaten Switzerland, then Spain would have been very slight favourites,” he says. The model had Holland with a 55 per cent probability of winning the final, following the semi-final results. If Spain had beaten Switzerland they'd have had a 53 per cent probability of beating Holland.
The model uses the results of recent games, and FIFA team rankings. Losing to a far lower ranked team has far more weight than losing to a similar or higher ranked team. “If Spain had lost to Brazil it wouldn't have mattered as much as Spain losing to Switzerland, and again it wouldn't have mattered as much as Spain losing to say Trinidad and Tobago,” said McHale.
How should the result of the Switzerland-Spain game have been handled? McHale is marvelling that the result of an opening game could affect predictive probabilities in the final.
“Mathematically it's correct,” he said. “The Switzerland result was information that Spain are vulnerable, but then as a football fan you could say its a bit of an outlier and should probably ignore it.”
Statistical models can handle outliers, but in football it's difficult to identify what they are. Was Algeria holding England to a 0-0 draw an outlier?
“One of the attractions of football is that the results have a high degree of randomness,” said McHale. “It does make predicting difficult but doesn't make it impossible.
Dr McHale will use his results to illustrate modelling to second-year business and economics undergraduates who dismiss some modelling examples as giving obvious results. Holland and Uruguay's performances weren't obvious. “It's going to be a good teaching tool,” he said.
Next stop for the model is a research paper on whether Champions League games have an adverse effect on domestic results. It uses data on how far the teams have to travel to play, and Dr McHale has already got a research monkey collating this and the results.
As for the game itself, “It was a disappointing match,” he said. “No matter who they play against Spain reduced the team to a chasing ball team. Holland looked pretty good against other teams - they looked like the passing team. When they came up against Spain they couldn't do it.” ®