1900: 83.5% tipping accuracy. The best season for model 9 tipping accuracy was very early in the VFL’s history and with a lot less games played per season than there is now. However, the predictability of the season is highlighted by the lack of predictions out by more than 7 goals and the superiority of home teams. The predicted and actual scores were also a lot lower than in today’s modern era (predicted home team scores shown below for 1900).
1907: 53.5% tipping accuracy. The worst season was also early in the history of the AFL and highlights a large proportion of home team losses that were not expected – and if there had been about another 20 goals to some of the home teams that lost then the average accuracy would have skyrocketed as a lot of games were won by a couple of kicks.
Predicted score for home team versus actual (season 1907)
All predictions for 2000 onwards. Moving forward one hundred years there are both a lot more games to predict and a lot more home team failures in the eyes of model 9.
2012 and 2013 tipping accuracy. Home teams not performing as well (or away teams performing better than expected). Home ground advantage diminishing in the modern era – Melbourne based teams no longer have a home ground advantage amongst themselves but do with interstate teams.
Games involving Hawthorn from 2000 onwards
West Coast by comparison
West Coast v Hawthorn (Hawthorn away team)
Hawthorn v West Coast (including home ground in Tassie)
Why all of the above graphs? Well I am looking for visual trends and clues to which factors I can legitimately focus on (other than downloading BOM weather data for over 100 years of weekends to add weather as a factor in the model - the shear quantity of weather data is a bit daunting).
I am working on changing the home ground advantage over time -where in the old days it was a combination of local ground knowledge and a parochial supporter base. Over time this has migrated to grounds being essentially the same but there is a difference in local crowd intimidation (lessened in finals?? due to more ambivalent supporters/ corporate ticket holders going to the peoples game?). I think this thinking can sit well with the analysis of MAFL (I am not too humble to stand on the shoulders of Statisticians in the search to make my model more legitimate/ quasi legitimate)
Probably need to go to Dan Murphys to stock up on research aids... and settle in for some number crunching...
