Statistical model tackles World Cup 2002 predictions. Do they work?

 

Here is an article about statistical predictions for the Japan 2002 World Cup. Have a look how close they were. Do you think statistical models work for these kind of events?

A new statistical model of the World Cup football tournament could provide a more accurate set of predictions than either bookmakers and television pundits, according to Henry Stott, a mathematician at the University of Warwick, UK.

Stott has used a modelling technique more commonly applied to financial risk assessment to predict the outcome of every game in the 2002 competition, as well as the ultimate winners

The approach combines human intuition, in the form of bookmakers' odds, statistical analysis and brute force computer simulation. Stott first developed a similar system to pick the winners of the 1998 World Cup after entering an office sweepstake.

"I did a much cruder analysis and did quite well," he says. "Then I started to realise that it's quite an interesting synthesis of human and computer thinking."

Peter Haigh, a statistician at the University of Sussex, says the approach is much more comprehensive than regular punditry. "I strongly believe in simulations as a reliable way to generate estimations," he told New Scientist. "It's based on the sound idea that, if you want to come up with an estimate you have to take into account the path to the final."

Twice the chance

Although many bookmakers' favourites Argentina were still given the greatest chance of winning the tournament (13.2%), the model provides a number of surprising results. Brazil (9.3%) were given better odds than France (8.7%), whereas bookmakers rate France's chances almost twice as high. England (6.9%) were also given a better chance of triumphing than many people's hot tip Italy (6.7%).

In the case of Brazil, Stott believes recent poor results have overly affected people's assessment of the team's chances. He cautions that his probabilities and bookmaker's odds are only comparable in relative terms, because bookmakers reduce their prices in order to give them a profit.

The biggest surprise came in individual matches. Stott's system gave a much higher chance that underdogs would upset strong teams than bookmakers. For example, South Africa are given only a 10% chance of defeating Spain by most bookmakers, while Stott calculated this probability to be 28%.

Game of two halves

Stott's method, developed with BBC 5Live, involves calculating the strength of each team based on performances during qualification matches, as well as official FIFA rankings and bookmakers' odds. But unlike some other ranking systems, Stott also included a measure of each team's unpredictability, or "patchiness", based on previous results.

For individual matches, one team's strength and unpredictability was then compared to that of their opponents to determine the probability that each side would win, lose or draw.

To determine the overall probability that a team would win the whole tournament, Stott used a computer to run hundreds of thousands of virtual matches.

Portugal (7.9%) turned out to be the fourth most likely team to become world champions, after Argentina, Brazil and France. The team with the lowest probability of ultimate victory chance is China with a tiny 0.3 % chance, though this means they are still overpriced at bookmakers at 750-1.