Sunday, August 5, 2012

Swinburne Footy Tips

Footy tipping is a popular pastime in Australia amongst those who follow - to whatever degree - Australian rules football.  Almost all workplaces have a tipping competition, and online tipping is a a popular marketing tool.

Swinburne University statistics department has been producing footy tips since 1980, and these tips have achieved both reasonable predictive accuracy and publicity for the university.

I was interested to see if Swinburne's tips are more accurate than consensus tips - as represented by betting odds.

And not surprisingly, the answer is NO.  Of the 196 matches, Swinburne tipped the outcome of 145 matches accurately, whereas consensus picked the outcome of 151 matches correctly. 

Swinburne's correct tips were:-

- tipped by Swinburne and consensus          139
- tipped correctly by Swinburne only               6

Both Swinburne and Consensus tipped 32 games incorrectly.

The fact that Consensus had a better tipping record than Swinburne is not surprising. Researchers at University of Melbourne studied football tipping performance in 2003 and concluded that a consensus forecast outperformed individual tipsters.

The performance measure is absolute number of games correctly tipped. What these figures don't show is if the individual experts (and Swinburne algorithm ) are better at forecasting lower ranked wins. Although given the number of wins forecast by both Swinburne and Consensus, this  seems unlikely.

The superiority of the consensus probably explains the success of the predictive analytics website .  Kaggle uses crowdsourcing to obtain the best solution to a predictive analytic problem.