After the siren: 2018 Premiership Season
Congratulations to the West Coast Eagles on their thrilling victory over Collingwood to close out the 2018 Season! Starting this Elo rating project has provided me with a new insight into team performance, and I hope the same for the readers of this blog.
The rating system predicted 4 out of 9 finals, including correctly predicting the winner of the Grand Final. This is a lot lower than I would like, so there is room for improvement in the offseason. Like the BBL ratings I found that home ground advantage doesn't necessarily apply in the AFL, especially as West Coast were technically the home team (as they finished higher up the ladder) on Saturday but the game was played at Collingwood's home ground. Another change may be to move to an adjusted rating system similar to the BBL, where the rating is adjusted before the game to take into account the ground, rest days and other conditions. This would help I think in situations like the West Coast v Melbourne game where Melbourne had a 54% chance to win but obviously the home advantage for West Coast is significant.
This will be the last post about the AFL for a while as attention now moves to the A-League and BBL. Elo ratings work particularly well for soccer, and I'm interested to see how they go with predictions for this season. For the sake of clarity I keep two measures of accuracy, one counting a win or a draw as a correct prediction and one counting only a win as a correct prediction. Applying the predictions retroactively, the 2016-17 season had an accuracy of 76.4% (win+draw) and 50.7% (win only). The 2017-18 season had an accuracy of 68.6% (WD) and 49.3% (WO). The system is accurate 68.8% (WD) across all seasons.
For the BBL, it was much harder to get an accurate system. Using just ratings the overall accuracy was just 49.2%, so we'd actually be better just tossing a coin to predict a winner! Using an adjustment system that improves to 54.1%, which still isn't great. However I think a lot of this comes down to the lower number of matches in the first few seasons. The system (retroactively applied) correctly predicted 60.0% of BBL06 and 65.1% of BBL07, which included correctly predicting all three winners of the finals in both seasons. It will be interesting to see how this year goes.
Hopefully these predictions help you win your tipping competitions! Be sure to follow me on Twitter too!
The rating system predicted 4 out of 9 finals, including correctly predicting the winner of the Grand Final. This is a lot lower than I would like, so there is room for improvement in the offseason. Like the BBL ratings I found that home ground advantage doesn't necessarily apply in the AFL, especially as West Coast were technically the home team (as they finished higher up the ladder) on Saturday but the game was played at Collingwood's home ground. Another change may be to move to an adjusted rating system similar to the BBL, where the rating is adjusted before the game to take into account the ground, rest days and other conditions. This would help I think in situations like the West Coast v Melbourne game where Melbourne had a 54% chance to win but obviously the home advantage for West Coast is significant.
This will be the last post about the AFL for a while as attention now moves to the A-League and BBL. Elo ratings work particularly well for soccer, and I'm interested to see how they go with predictions for this season. For the sake of clarity I keep two measures of accuracy, one counting a win or a draw as a correct prediction and one counting only a win as a correct prediction. Applying the predictions retroactively, the 2016-17 season had an accuracy of 76.4% (win+draw) and 50.7% (win only). The 2017-18 season had an accuracy of 68.6% (WD) and 49.3% (WO). The system is accurate 68.8% (WD) across all seasons.
For the BBL, it was much harder to get an accurate system. Using just ratings the overall accuracy was just 49.2%, so we'd actually be better just tossing a coin to predict a winner! Using an adjustment system that improves to 54.1%, which still isn't great. However I think a lot of this comes down to the lower number of matches in the first few seasons. The system (retroactively applied) correctly predicted 60.0% of BBL06 and 65.1% of BBL07, which included correctly predicting all three winners of the finals in both seasons. It will be interesting to see how this year goes.
Hopefully these predictions help you win your tipping competitions! Be sure to follow me on Twitter too!
Comments
Post a Comment