When it comes to the AFL finals there are hot favourites … then there’s the data
When Essendon ran onto the field before the 2000 AFL grand final against Melbourne, few doubted the outcome.
The Bombers' all-star team had lost just one game all year with an average winning margin of 51 points and had utterly dismantled the Kangaroos in the first final, winning by more than 20 goals.
Of course, they won the grand final and no-one was surprised, especially not sports data scientist Darren O'Shaughnessy.
For years, he's been predicting the outcome of football games using computer modelling that he developed. The model this year has picked more winners with the highest level of accuracy in regards to the margins than most experts in the game.
The modelling has generated predictions for all potential 512 finals matches that could occur over the next few weeks and has turned up some intriguing results.
Richmond is the strongest premiership prospect he's seen coming into the finals since the Bombers' dominance in 2000.
The Tigers have a 47.7 per cent chance of winning the flag, the model predicts.
Not only is Richmond a well-rounded team, they've also been handed a dream draw.
"The teams that we rate second and third are Melbourne and Geelong and, of course, they play each other in an elimination final coming and Richmond potentially doesn't have to play the winner of that until the final," O'Shaughnessy said.
While it may not look like the modelling rates Geelong and Melbourne, with a 6.2 and 9.1 per cent chance respectively, that's only because they missed out on the top four and the double chance that comes with it.
But a Tigers flag is by no means a done deal. Last year, the Crows were strong favourites and suffered a surprise loss to Richmond in the final.
"There's always luck," O'Shaughnessy said. "Even though we've got Richmond as the outstanding team, they're still less than a 50 per cent chance of going all the way."
While some experts like to make their predictions based on teams peaking at the right time of the year, this modelling takes a longer-term view of a team's performance.
The work a team has put in across a year is much more important to the predicted outcomes than its last few games.
The same goes for injuries, one or two key players out won't change much about the way a team approaches a match. Sydney getting Lance Franklin back this week, for example, has not significantly improved their chances. In fact, they'll be lucky to see week two of the finals.
Then there's "finals footy", a seemingly unavoidable term if you watch or listen to more than a minute of AFL coverage over the next few weeks.
It's used to describe a tougher, more-intense brand of football that supposedly exists only in the finals.
O'Shaughnessy has tried to measure it. Among other things, he looked at the pressure on the ball carrier. There was an increase in sustained pressure, but a small one.
"But overall there are similar statistics, and tackling numbers are similar to what you would see between the two top teams throughout the year," he said.
Instead, clearances, high-quality shots on goal and the ability to force intercepts in any part of the ground are all key factors in this model.
Tip like an algorithm
If you're struggling with your tips, maybe try being less human.
"Most humans have those biases and they want to take more regard of those than the boring, this-is-what-the-computer-says idea," O'Shaughnessy said.
"And even little changes, like being over-confident in your assessment of the game, are enough to make you slightly less effective than the computer."
But all is not lost on us mere mortals. While the model has beaten out experts, former players and coaches and other computer predictions in the tips this year, there was one challenge O'Shaughnessy's predictions could not overcome: the extended family tipping competition.
His wife Amanda won that with a massive 146 correct tips out of 198 this year.
Reporting: Mark Doman
Development: Nathanael Scott
Design: Alex Palmer