Associate Professor Michael Bailey has worked as a biostatistician for 20 years and has published over 400 manuscripts in peer reviewed medical journals that have been cited more than 14,000 times.
In 2005 he completed a PhD which outlined a statistical approach to predicting sporting outcomes and with a specific chapter dedicated to the Brownlow, he is the only known person to be awarded a PhD for the prediction of the Brownlow Medal winner.
Betfair has once again partnered with Associate Professor Michael Bailey to build a prediction model for the 2016 Brownlow Medal. The statistician known as ‘The Professor’ has this year’s Brownlow winner likely to be Patrick Dangerfield, with Dustin Martin and Dan Hannebery rounding out the top-3.
Dangerfield has an excellent track record on Brownlow night having polled at least 20 votes in each of the past four seasons. Prior to 2016, he had polled 97 Brownlow votes from 147 home and away matches for an average of 0.66 votes per game placing him in the top 10 current players. With 23 3-votes, 11 2-votes and only 6 1-votes, he has a remarkable ability to be rated by the umpires as the most influential player on the ground.
With a predicted total of 35 and his nearest rival 12 votes adrift, Dangerfield is given an 89% chance of winning, meaning his current price of $1.26 still represents value for punters that are prepared to take the short price.
Patrick Dangerfield from Geelong will go in to the 2016 Brownlow medal as the shortest price favourite in history and rightly so. Dangerfield has had one of the most outstanding individual seasons in the history of AFL football and is a chance to poll votes in all but two of the 22 home and away matches. Expect records to be broken for the most games polled in a season (currently 14 – Robert Harvey 1998, Dane Swan 2011 & Jobe Watson 2012) and total votes polled (currently 34 by Dane Swan 2011).
2016 Brownlow Medal Predicted Top 6
In 2016, Richmond midfielder Dustin Martin played six of the best 10 games of his 153 match AFL career and was duly awarded All-Australian honours for the first time. After polling 21 votes in the 2015 Brownlow to finish seventh, Martin is expected to perform better this year and is a narrow favourite to finish second to Dangerfield.
Both Sydney midfielders Dan Hannebery and Josh Kennedy are rated ahead of their more fancied Sydney counterpart, Luke Parker. Both Hannebery and Kennedy finished in the top 5 in 2015, and while both have had similar years to 2015, expect Hannebery to turn the tables on Kennedy and finish as the leading vote catcher for Sydney.
While the ultimate winner is clear-cut, the race for the minor placings is anything but clear, with upwards of a dozen players capable of finishing second or third.
With the 2016 all Australian captain, Geelong’s Joel Selwood in career best form, Dangerfield and Selwood may be the first players from the same team to finish first and second in the Brownlow since the infamous West Coast pairing of Ben Cousins and Daniel Kerr in 2005. With West Coast going on to lose the 2005 Grand final by 4 points to Sydney, Geelong supporters would be looking to avoid comparison.
With 150 Brownlow votes from 183 matches, Selwood has always been highly regarded by the umpires, and his career average of 0.82 votes a game places him 3rd amongst all current players.
In 2015, Sam Mitchell edged out fellow Hawthorn legend, Leigh Mathews as the player to poll the most Brownlow medal votes without winning. In 2016 he looks set to take that record one step further and become the most voted player under the current 3-2-1 system. While Mitchell has not had his best season in 2016, with 13 games in which he is rated in the top 3 on the ground, expect him to finish around the 20 vote mark.
Whilst Adelaide midfielder Rory Sloane is ineligible to win the 2016 Brownlow medal, expect him to figure prominently in the voting. In 2016, Sloane had a stellar year playing eight of the best 15 games of his 137 game career. With six potential best on ground performances and a further three vote catching games, he is also likely to finish around the 20 vote mark.
Predicted Top 12 Players for 2016
*Ineligible to win due to suspension
Predicting the Brownlow medal winner
By using past information, it is possible to identify specific match and player features that are related to who will be awarded votes. For example, 91% of all 3-votes, 78% of all 2-votes and 74% of all 1-votes are awarded to players from the winning team, with the probabilities further increasing with the margin of victory. Similarly, the number of possession that each player gets is strongly related to the number of votes polled, with the leading possession winner for each match receiving 3-votes approximately 39% of the time.
The natural ordering of the voting structure (3-2-1-0) lends itself to the use of a mathematical modelling process known as ordinal logistic regression. With a wealth of player information now readily available, by using information from over 2000 past matches, it is possible to identify and combine the effects of more than 20 contributing variables to produce a probability that each player will poll 3, 2 or 1 vote in each match. The main contributing variables in the model include features of the game such as kicks, marks, handballs, goals and margin of victory, however features relating to the specific player such as quality of possession and past voting performance are also important contributors. By combining these probabilities, each player is assigned a score from 0 to 3 for each match. By aggregating these scores over the entire season, it is then possible to assign each player with a predicted total of Brownlow votes for the season.
How well do the models perform?
Per game level
The model is remarkably good at identifying the leading players in each match, with the top 5 ranked players accounting for 96% of all 3-votes, 85% of all 2-votes and 73% of all 1-votes.
Furthermore, the leading ranked player in each match as determined by the modelling process will successfully poll 3 votes 58% of the time, 2 votes 22% of the time and 1 vote 9% of the time.
When matches are further categorised according to whether there was 1 clear standout player (as opposed to 2 or 3 standout players) the leading ranked player will successful poll 3 votes in 66% of all games.
Per season level
When individual match scores are aggregated for the season, the modelling process can accurately predict the season tally of votes for 84% of all 656 players to within 1 vote.
However, it is important to note that more than 70% of all footballers each season will not poll any votes at all.
When considering the top 10 leading players each season, the modelling process will typically identify about 7 or 8 of the top 10 players and 3 or 4 of the top 5.
When considering the winner, over the past 14 season, all winners have come from the top 4 ranked players each year, with the leading ranked player winning on 6 occasions, the second ranked player winning on 5 occasions and the third ranked player winning twice.
How the count is likely to eventuate
With the exception of losses in Round 2 (vs GWS) and Round 9 (vs Collingwood), in each match, Patrick Dangerfield has greater than 50% chance of polling votes. Accordingly, you can expect him to establish a clear lead by round 6 and continue to extend this lead throughout the course of the evening.
Brownlow voting per round
Historically, when considering individual teams, the leading vote getter for each team was ranked first by the model 84% of the time, with the 2nd ranked player by the model getting the most votes 11% of the time.
Leading players for each club for the 2016 season
Ranked Top 5 = Number of times each player was ranked in the top 5 players on the ground
Ranked Top 3 = Number of times each player was ranked in the top 3 players on the ground
Ranked 1st = Number of times each player was ranked as the best player on the ground
By aggregating all individual player scores to a team level it is possible to derive a predicted vote tally for each team. Using this approach, Sydney is the team likely to poll the most votes with a total of 98.
Team to poll the most votes for the 2016 season
Actual and predicted order and vote totals for the top 10 players for the past 5 years