In AFL, playing at home is a distinct advantage, albeit perhaps a little less of an advantage than it once was. So, around this time of year, I usually spend a few days agonising over the allocation of home team status for each game in the upcoming season.
For those of you more familiar with team sports other than Australian Rules the notion of indeterminacy in home team status might strike you as odd. To my knowledge, no-one's ever thought to themselves "Hmmm, Manly v St George at Brookvale Oval ... Sea Eagles or Saints home game? I'd better check Big League", but one of the (many) quirks of AFL is that teams share home grounds. For example, Adelaide and Port Adelaide share Footy Park; the Eagles and Freo share Subiaco; Collingwood, Essendon and a cavalcade of other Melbourne-based teams share the G; and a few more teams share Docklands.
Another wrinkle is that the AFL's governing body occasionally switches a team's notional home game to its opponents home ground. So, for example, in 2009 we had Collingwood playing Geelong at the G in what was, according to the AFL, a Geelong home game and, more comically, in 2007 we had Melbourne playing a home fixture against the Lions at the Gabba.
During the off-season I've revisited the issue of home team status with a view to finding a less arbitrary way of determining home team status for those games where it's contentious. There are, I'd suggest, two components to home ground advantage (HGA):
- the predominance of a team's supporters in the crowd
- the team's familiarity with the venue
The first component is best proxied by what I've previously called "notional" home team status - basically, the team with this component of HGA is the team named as the home team in the Record. If the AFL deems team X to be at home for a particular game then extra tickets are made available for that team's supporters and so, ceteris paribus, that team should enjoy a preponderance of its fans at the game.
For the second component I've created a new variable, called Venue Experience, which is the number of games that the team has played on the relevant ground - in home-and-away games or finals - in the previous 12 months. For any give game during the season I can calculate the Venue Experience for each of the participating teams, and with these two numbers I can calculate what I call the Excess Venue Experience of one team relative to the other.
Does the notion of Venue Experience have any predictive power?
Over the past 11 seasons, notional home teams - the teams named as home teams by the AFL - have won 58.7% of games, but the rate at which they have won has declined with diminishing Excess Venue Experience. Put another way, the more familiar the notional home team's opponent is with the venue at which a game is played, the less likely the notional home team is to win.
Indeed, if the notional away team has played 6 or more times more often than the notional home team at the current venue - say it's played there 8 times in the past 12 months and the home team has played there only twice - the away team is, empirically, more likely to win than to lose. There have been 86 such clashes over the past 11 seasons of which the away team has won 48 or about 56% of them. Statistically speaking, the 44% victory rate for notional home teams in these games is significantly lower than the victory rate of home teams in all other games.
To give you a sense of how home team victory rates increase as Excess Venue Experience grows, here's a summary table of winning statistics for notional home teams for various Venue Experience excess ranges:
The winning rate of notional home teams with an excess Venue Experience of 10 or more games (67.3%) is also statistically significantly different from notional home teams winning rate in all other games.
So, empirically, it appears that Venue Experience matters over-and-above notional home team status. In other words, both of the HGA components described earlier seem to matter.
To test this more formally I built a statistical (binary logit) model to explain the win/loss result for the notional home team with Venue Experience for the two teams as the only regressors. Drawn games and finals were excluded from the fit and, for each game, a predicted probability of victory for the notional home team was generated. Using a simple log probability scoring rule, the predicted probabilities from this model scored better than - that is, were better calibrated than - a simple model in which the notional home team is predicted to win every game with probability 58.7%.
(The model, by the way, was:
logit(Pr Notional Home Team Wins) = 0.107 + 0.0446 x Home Team Venue Experience - 0.0500 x Away Team Venue Experience
The two non-intercept coefficients are significant at less than 0.1%.)
The effects of Venue Experience are, it appears, so small in magnitude that they're difficult to detect in the results of any single season. Fitting a model to the data for each of the previous 11 seasons taken individually, the coefficient on Home Team Venue Experience is statistically significant at the 10% level or higher in only 4 of those seasons, though it does carry the expected positive sign in 8 of them. The coefficient on Away Team Venue Experience is also statistically significant at the 10% level or higher in only 4 of those seasons, but it carries the expected negative sign in all 11 seasons.
In 2010, Venue Experience was almost irrelevant. The predictions from the binary logit model built for that season in the same way as described earlier were better calibrated than those of the equivalent naive model, but only barely so. Notional home teams facing a 7 game or greater Venue Experience deficit, of which there were only 5 during the entire season, won just 40% of the time. Those games aside, notional home teams won at pretty much the same rate - around 62% - regardless of the level of their excess Venue Experience.
Nonetheless, I think Venue Experience has shown itself to be sufficiently predictive across most of recent football history that I'm going to include it in all (both?) the MAFL Fund models for 2011. I won't, however, as I have done in previous years, be switching home team designation. The home team for every game will therefore be the home team as decreed by the AFL - it'll just be the case that some home teams will have a Venue Experience excess and some a deficit.
Having demonstrated the small but persistent importance of Venue Experience over the period 2000 to 2010, it's interesting to take a look at how often each team has played a home game during the home-and-away portion of the season at a venue where it had had the same or lesser Venue Experience than its opponent on the day. Here's the list:
- Carlton 47
- Essendon 46
- Kangaroos 39
- Bulldogs 38
- Hawthorn 37
- Richmond 37
- Melbourne 34
- Geelong 29
- Collingwood 22
- St Kilda 12
- Fremantle 9
- West Coast 7
- Port Adelaide 6
- Adelaide 5
- Sydney 1
- Brisbane 0
One way of thinking about this data is to note that Carlton, for example, has on average played 4 of its 11 home games each season at a venue where its opponent has had the same or more recent Venue Experience. Brisbane, in contrast, has played no such game in 11 seasons. That's gotta be worth something.