AFL Probabilistic Tipping & Analytics
The CroyGlick Model is an AFL probabilistic match prediction system built on the foundation of the
Glicko-2 rating system,
originally developed by Mark Glickman for
chess and adapted here for Australian Football League matches. The model records dynamic ratings
for each team, adjusted after every match, incorporating the win record, the rating deviation
(rating uncertainty) and volatility of each team's performance over time.
Unlike the traditional Glicko-2 system, which focuses on individual "players" in zero-sum two-player
games, resulting in a binary win/loss outcome, the CroyGlick model adapts this framework to
accommodate the unique aspects of AFL, a team sport with variable scoring margins.
The model also incorporates a venue/home ground advantage factor, recognizing the significant impact
that playing at home can have on a team's performance. This factor is adjusted based on historic results,
some general calculated constants regarding home ground advantages, and travel factors when playing in a
different state.
The CroyGlick adaptation allows the model to account for the margin of victory or defeat, providing a more
nuanced understanding of team performance. The primary difference in this adaptation is the implementation
of a custom rating calculation, which takes into account many more factors than a simple win/loss outcome,
including the margin of victory, the expected margin based on team ratings, the expected influence of team
lineups and injuries, the venue advantage, and even momentum. This custom rating update has been extensively
tested and validated against years of historical AFL data to ensure its accuracy and reliability.
See below the disclaimers for more information on the key metrics used to evaluate the model's performance,
and historic seasonal performance metrics.
This is a passion project by a university student stats nerd who loves the footy, and is not affiliated with
the AFL, any teams, sports analytics organisations, or betting agencies.
The model is not intended to encourage, promote or facilitate gambling/betting activities. The model is for
educational and entertainment purposes only, providing insights into the dynamics of AFL matches through a
statistical lens.
You may view the
If you have gambling concerns, please seek help from:
Gambling Help Online: gamblinghelponline.org.au
National Problem Gambling Helpline: 1800 858 858
Mean Absolute Error is the average difference between the model's predicted and actual margins.
According to Squiggle, It's hard
to fluke a good MAE, because your tips have to be consistently close to the real margins.
Bits are a method of scoring that is similar to log loss. A correct tip is rewarded and an incorrect
tip punished, where confident correct tips are rewarded by less than a confident incorrect tip is punished.
From
Monash University Probabilistic Footy Tipping.