What's More Important: Who You Play or Where You Play Them?
/The benefits of playing at home have been extensively investigated both here on MAFL for Australian Rules football and more generally within the sports prediction community for this and other sports. Put simply, teams that play at home win more often and score more points than you'd otherwise expect them to after adjusting for the quality of the opponents they face.
Today I want to expand this style of analysis to look at the topic of game location on a venue-by-venue basis for each team, and to assess the importance of game venue relative to other characteristics of a contest such as the teams' Ratings and the short-term form of the participants.
THE DATA
My focus today will be on the games from the home-and-away portions of the 2000 to 2013 seasons and I'll be building 18 models, one for each team, seeking to explain a team's game margins (ie points scored less points conceded) as a function of:
- The Venue at which the game is played (included as a categorical variable with the reference level the "home" ground for the relevant team, defined as the ground at which it has played most games over the 14 year period).
- The team's Excess Venue Experience (the difference between the number of times it has played at the venue over the past 12 months less the number of times that its opponent has played there over the same period).
- The team's Excess Rating (the difference between its MARS Rating at the time and that of its opponent. For this blog I've used the same MARS Rating time series that I've been using for the MAFL Funds, which set all team Ratings to 1,000 at the start of the 1999 season, Rated Gold Coast at 1,000 in its initial season and Rated GWS at 900, and which uses the AFL designation of home team for Finals. This Ratings series will differ slightly from the All-Time MARS Ratings I've been favouring most recently, but probably not materially, except perhaps for GWS).
- The team's Excess Form (the difference between the team's average for-and-against differential over the past two rounds and the same differential for its opponent).
(In early runs of the models I also included Interstate Status and two other Excess Form variables spanning the most recent 6- and 16-week periods. All of these variables are highly correlated with other variables remaining in the models - Interstate Status, for example, is highly correlated with Excess Venue Experience - and were generally not statistically significant. I excluded these variables with Occam in mind.)
THE MODEL
I'm using a humble Ordinary Least Squares regression today to create the models for each team. As well, to estimate variable importance, I'm using the PMVD measure from the relaimpo package in R. This measure is more opinionated than the more common LMG variable importance measure, but has for me in the past provided more intuitive, if more data-sensitive, results. YMMV.
THE RESULTS - TEAM-BY-TEAM
The table below summarises the 18 individual models and can (nay must) be clicked to access a larger version.
Firstly, let's review these results on a team-by-team basis.
Adelaide
Over the 14 seasons, the Crows have played most of their games at Football Park (168), followed by Docklands (46), the MCG (26), and Subiaco (19). Looking at some of the other somewhat popular venues for Adelaide, we note that they've never played at Adelaide Oval, Marrara Oval, or Stadium Australia.
The coefficients in the Venue section of the table are the excess points that Adelaide can be expected to score at the venue relative to what they'd be expected to score if the game were played at their Home ground (Football Park) given the same Excess Venue Experience, Excess MARS Rating and Excess Form relative to their opponents. So, for example, Adelaide have scored 6.6 points per game more than you'd expect at Docklands, adjusting for Excess Venue Experience, Rating and Form, relative to what you'd have expected them to score if the same game were played at Football Park. In that sense, Adelaide have done "better than you'd expect" at Docklands (although I'd note that the difference is not statistically significant, as indicated by the absence of any symbols in the Sig column next to the coefficient for Docklands).
Subiaco is the only ground on which Adelaide have played more than a handful or games and have performed statistically significantly better or, as here, worse than expected. They've scored about a goal less than you'd expect when they've played there, which we can think about as Adelaide's home ground advantage relative to Subiaco.
Turning next to the Excess section of the table we see that Adelaide benefits from having Excess Venue Experience (after adjusting for all other variables in the model). For every additional game they've played at a Venue relative to their opponents they are expected to extend their game margin by 1.28 points.
As well, every additional Rating Point relative to their opponent is worth 0.54 points of game margin, and every additional point of Excess Form is worth 0.08 points of game margin. Neither of these two coefficients is statistically significantly different from zero however.
In total the model explains a little over 30% of the variability in Adelaide's game margins across the 14 seasons, which experience tells us is about the proportion we should expect of any acceptable model.
The final block records the results of a variable importance decomposition and shows that, for Adelaide, the Venue of a contest is about as important as the Rating of their opponent in terms of the variables' ability to explain Adelaide's performances.
Brisbane Lions
The Lions have a positive home ground advantage relative to every other venue on which they've played, a fact that's revealed by the full set of negative coefficients in the Venue section. For most grounds the deficit appears to be about 30 to 40 points, which we could interpret as indicating that the Lions are about a 6 to 7 goal better team at home than away given the same quality opposition.
Excess Venue Experience is largely of no consequence to the Lions, though it enjoys about a 0.6 point lift in margin for every additional MARS Rating Point and a 0.16 point lift for every additional point of Excess Form relative to its opponent, the latter indicating that recent form is a significant factor for the Lions. That said, Venue and Excess Rating explain over 90% of the tota' explained variability (36%) in the Lions' game margins.
Carlton
Until I performed this analysis I had no idea that the Blues hadn't faced the Cats at Kardinia Park in a home and away contest at any point in the previous 14 seasons. Nor did I realise quite how split had been their schedule, which has seen them play 106 games at the MCG and 98 at Docklands.
Relative to the G, Docklands hasn't been kind to the Blues. They've scored almost 3 goals per game fewer there than we'd expect here once we adjust for the other variables in the model. Amongst the other venues, Subiaco is probably the most interesting since the Blues have played there 22 times and conceded only a 1 goal home ground advantage to Freo or West Coast, on average.
Excess Venue Experience provides a small, positive but not statistically significant lift to the Blues. Excess Rating Points are worth about 0.65 points of game margin each, and Excess Form is worth about 0.13 points in margin per point in additional form.
Knowing the Blues' relative MARS Rating, however, provides almost all you need to know to best predict their eventual game margin. This difference, alone, explains almost 80% of the variability that is explained, in total, by the model (33%).
Collingwood
The Pies, as the Blues, have avoided playing the Cats at Kardinia and have also bounced between twin home grounds, though they've played a far larger proportion of their games at the MCG rather than at Docklands. Not that they would have minded playing more games at Docklands. Compared to the G, the same contest played at Dockands would be expected to finish with a 15.5 point better outcome for the Pies.
Actually, the Pies have better-than-expected records at a large range of venues compared to the G. The only venue at which they have played more than a handful of times and for which they have a negative coefficient is Subiaco where they have, on average, seen the Eagles or Freo enjoying an effective 3.5 point home ground advantage relative to the G. For the Pies over the past 14 seasons, almost every ground has been like a home ground - or even better than one.
The Pies' game margins also respond positively to Excess Venue Experience, Excess MARS Rating and Excess Form, all at statistically significant levels. Over 70% of the explained variability in those game margins can be attributed to the Pies' Excess Rating at the time, which, as we'll see is typical and defining for a Victorian team.
Essendon
The Dons are the third team to split their home games between Docklands and the G, though the first so far described with the majority of games played at the former. On average, adjusting for all other variables in the model, the Dons have performed about 4 points worse at the G than at Docklands. Like many of the teams, the Dons do better at home than at any other venue, in general by about 20 to 30 points. The Dons are also yet another team who've avoided the Cattery.
Subiaco is the frequently-visited ground that gives Essendon most trouble. They spot their opponents almost 23 points at that venue, which is a relatively large handicap for any team that's played at a venue more than just a few times (for whom any estimate of deficit will be more subject to sampling variation).
Essendon game margins respond positively to Excess Venue Experience, Ratings and Form, with most (79%) of the explained variability in these margins (29%) attributable to Excess Ratings.
Fremantle
Freo enjoy about a 15 to 25 point home ground advantage relative to the away venues at which they've played most often (Docklands and the MCG), with the G their most-dreaded, regular away venue.
Their results also respond positively to Excess Venue Experience, Ratings and Form. Venue is associated with almost one third of the explained variability in their game margins, a higher proportion than we've seen for the three Victorian clubs so far discussed, and a similar proportion to that for the non-Victorian teams of Adelaide and the Lions. There is, it seems, a clear delineation in the decomposition profile of Victorian and non-Victorian teams.
Geelong
When the Cats play a home game at Docklands they give up over 6 points relative to playing it at Kardinia Park, and when they play at the G they give up almost 19 points. That deficit makes the MCG about as friendly to the Cats in home-and-away contests as Football Park, and less friendly than Subiaco.
The Cats' game margin is positively related to their Excess Venue Experience, Rating and Form, though only in the case of Excess Rating is the coefficient statistically significantly different from zero.
Compared to other teams' regression models, the Cats' explains a relatively small proportion of the variability in their game margins (29%), about 60% of which is explained by Excess Ratings and 30% by the game Venue. That decomposition of explained variability, with a relatively significant proportion being explained by game Venue, is more similar to that of the non-Victorian teams than it is to the Victorian teams, which you might interpret as implying that playing at Kardinia Park is like playing interstate for most teams, Victorian or otherwise.
Gold Coast
With so few games to base inferences upon, some caution is required in interpreting the model for the Suns.
At this very early stage in their history it seems that they're about a 30 to 50 point better team at home than they are away, that Excess Venue Experience (of which they have little) is of no consequence, while Excess Rating and Form is important.
Amongst all the variable types in the model for the Suns, game Venue explains about 55% of explained variability, the highest proportion for any team. As well, the model for Gold Coast has the highest R-squared value of any team's.
GWS
Even more caution is essential in reviewing the Giants' model, which suggests that they are a 10 goals or more better team on their home ground than they are elsewhere.
Relative to the Suns, GWS game margins are explained by Ratings differences to a much larger extent and by game Venue to a slightly smaller extent. Recent relative Form makes almost no contribution to explaining the Suns' game margins, which might be interpreted as suggesting there's been little evidence of short-term spikes (or troughs) in their form leading to above- and below-average performances.
Hawthorn
The Hawks, surprisingly to me, have played 40 games at Aurora Stadium over the past 14 seasons and have, on average, performed almost 7 points better in games there than at home (the MCG) against similarly matched opponents.
Like most teams, at all other non-home venues the Hawks have performed relatively less well, generally by about 20 to 30 points at the more commonly-visited grounds. Football Park has troubled the Hawks most amongst those familiar venues, seeing them perform over 6 goals worse than at home against matched opponents.
Game margins of the Hawks are positively associated with Excess Ratings and Form, with about three-quarters of the explained variability in those margins being attributable to variability in Excess Ratings, giving them a decomposition profile similar to most of the other Victorian teams so far discussed.
Kangaroos
The Roos have played more than 10 games a season at Docklands over the past 14 years, and almost four a season at the MCG. The G is one of six venues at which the Roos have done better, relatively speaking, when they've played there relative to when they've played at their home ground for modelling purposes, Docklands. These venues include Football Park, where they've ventured 19 times and enjoyed 1.4 point better-than-average game margins, Kardinia Park (11 trips, 2.8 points performance lift), and Manuka Oval (17 trips, 1.7 point lift). The SCG is the venue at which they've performed worst, relatively speaking, amongst grounds they've visited more than just a few times.
Excess Venue Experience, Ratings and Form are all positively associated with Roos' game margins, with Excess Ratings explaining almost 70% of the explained variability in those margins.
Melbourne
Melbourne are similar to the Roos in that they've performed better at a wide variety of grounds away from the ground at which they've played most often (for them, the MCG). Docklands, where the Dees have played 45 times, has been an 11 point better venue for them, while Subiaco, which they've visited 20 times, has been a 6 point better venue, and the Gabba, which they've graced on 13 occasions, has proven to be an 11 point better venue for games than those with equivalent opponents played on their home turf.
Football Park and Kardinia Park are the only venues at which they've played 10 or more games and performed less well than at home when playing equivalently capable opponents. Even at these two venues the relative deficit has been less than a goal.
Of course, it takes two components to define a relative performance, and I'd humbly suggest that the mostly positive and few, slightly negative coefficients we see against the different venues for Melbourne is due more to their inability to dominate at home than their exceptional abilities on the road.
The Dees' game margins are positively associated with their Excess Venue Experience, Rating and Form. They have the second-largest coefficient on Excess Venue Experience of all teams and the largest proportion of explained variability attributable to this factor (14%). Overall though, Excess Rating explains far more (63%) of their game margin variability, with Venue explaining most of the remainder (13%).
Port Adelaide
Reviewing Port Adelaide's results returns us to a more normal world where home ground performances are better than performances elsewhere. Port are about a 10 to 20 point better team at home than they are away. Across the venues at which they've appeared 10 or more times, they're a 5 point per game worse team at Subiaco and a 21 point per game worse team at the SCG.
As almost all teams, their game margins are positively associated with Excess Venue Experience, Rating and Form. The coefficient on Form in Port's model is the largest of any of team and, partly as a result, Excess Form explains a larger proportion of the explained portion of the variability in their game margins of any team except the Dogs.
Excess Rating, however, explains the absolutely largest share of explained variability (55%), with game Venue explaining much of the remainder (25%), giving Port Adelaide a decomposition profile more similar to a typical Victorian team's than to a non-Victorian team's.
Richmond
The Tigers have played most of their home games at the MCG over the 14 years under review, and have appeared at the Docklands a little less than half as often. The G's been kinder to them than has Docklands where they've performed about 4 points per game worse than at the G against equivalently matched opponents. Amongst their most-commonly visited away grounds they've been about 3 or 4 points a game worse at Football Park and Subiaco, and 18 points a game worse at the SCG.
Excess Venue Experience and Form have had little to say about the Tigers' performances, Excess Rating being a far more informative statistic. Over 85% of the explained variability in Richmond game margins - which, admittedly is small at only 23% - is explained by variability in Excess Rating. The Tigers' decomposition profile is, therefore, very typical of a Victorian team.
St Kilda
Of the models built for the teams that have contested all 14 of the seasons under review, the Saints' model has the highest proportion of explained variability (40.5%).
By far the majority of the Saints' games have taken place at Docklands, with the next most common ground, the G, seeing them only about one-quarter as often. At the MCG they've been about a 6 point better side than at Docklands against equivalent opponents. They've also performed better away from the G - at Subiaco where they've been a 1 point per game better side with 21 appearances, and at Football Park where they've been almost a 6 point per game better side with 17 appearances. At other grounds where they've played at least 10 times, they're about 3 points per game worse teams at the Gabba and the SCG against equivalent opponents.
Excess Venue Experience, Rating and Form are all positively and statistically significantly related to the Saints' game margins, with Excess Rating accounting for almost 65% of explained variability and Venue almost 20% more. The Saints' game margin decomposition profile is also that of a typical Victorian team, albeit with Excess Venue Experience contributing a bit more than the norm.
Sydney
The Swans have played the third fewest games at the same venue of any of the teams competing in all 14 seasons, having played just 128 games at the SCG. They've played 10 or more games at eight other venues, which is one venue more than any other team, the next most nomadic being the Roos', who've played at seven. At only one of those eight popular venues, the Gabba, do the Swans perform better than at the SCG against equivalent opponents; they're about a 9 point per game better side at the Gabba compared to the SCG.
Elsewhere, they're about a 1 point per game worse team at Football Park and at their alternative home ground, Stadium Australia, and a 23 point per game worse team at Kardinia Park.
The Swans' game margins are positively associated with Excess Venue Experience, Rating and Form, though Excess Venue Experience and Form make only a small contribution to explaining these margins. Variability in Excess Rating can be tied to about 60% of the explained variability in game margins, while game Venue explains a little under one-third more, making Sydney's decomposition profile similar to that of a typical non-Victorian team.
West Coast
West Coast have played 10 or more games at only five venues including 165 at Subiaco. Of the four other venues, the Gabba's been best for the Eagles as they've been almost 19 points per game better there than at home against equivalent opposition. They've also been about 1 point per game worse at the MCG than at home, and about 7 points per game worse at Docklands and Football Park.
Excess Venue Experience, Ratings and Form are all positively associated with West Coast game margins, with Excess Rating variability accounting for just over one-half of the explained variability in those margins and Venue accounting for another 37%. The Eagles too, therefore, have the decomposition profile of a non-Victorian team.
Western Bulldogs
The model for the Dogs explains only 27% of the variability in their game margins, the second lowest proportion of all teams. The Venue-related coefficients in that model show that the Dogs have been about 7 points per game better at the MCG than at Docklands, and about 5 to 10 points worse than at the MCG at other venues where they've played 10 or more games (viz Football Park, the Gabba, and Subiaco).
Excess Ratings and Form are most associated with the Dogs' game margins, with Excess Form accounting for almost 30% of the explained variability in those margins, representing the largest contribution of this variable for any team. Venue accounts for only 10% of the explained variability, the smallest contribution made by this variable for any team. Excess Rating accounts for just over 60%. The Dogs' decomposition profile is thus more like that of a typical Victorian team than of a non-Victorian team, but with a larger contribution of Excess Form.
RELATIVE CONTRIBUTIONS TO EXPLAINED VARIABILITY IN GAME MARGINS
In discussing the model results for each team I've made frequent reference to the typical game margin decomposition profiles of Victorian and non-Victorian teams, by which I mean the proportions of the explained variability in a team's game margin that can be attributed to the game Venue, Excess Venue Experience, Excess Rating and Excess Form.
The following chart might help to make this distinction clearer.
The defining difference between Victorian and non-Victorian teams is the proportion of explained variability in game margin attributable to game Venue compared with that attributable to Excess Ratings.
The variability in game margins for Victorian teams, who inevitably play the bulk of their games in Victoria on grounds where they enjoy relatively little Excess Venue Experience (see this previous post for details), tends to be better explained by differences in team Ratings than in game Venue. In contrast, the variability in game margins for non-Victorian teams, who can partially or fully enjoy and exploit the benefits of a far-away, rarely visited (by opponents) home ground, can proportionately be better explaned by game Venue, though differences in team Ratings are also important.
SUMMARY AND CONCLUSION
Overall, we can explain about 30 to 35% of the variability in team game margins mostly on the basis of where the game is played and the difference in Ratings between the competing teams.
For games involving a non-Victorian team, generally about 30 to 40% of the variability in the final game margin can be attributed to the Venue at which the game is played and 50 to 60% can be attributed to the difference in the teams' ratings. Alternatively, for games involving Victorian teams, only about 10 to 20% of the variability in game margins can be attributed to the game Venue, and 60 to 80% can be attributed to the teams' Ratings.
On average, for those teams for whom it matters:
- Excess Venue Experience is worth about 1 to 2 points of game margin for every game of additional Experience
- Excess Ratings are worth about 0.5 to 0.8 points of game margin for every additional Rating Point difference
- Excess Form is worth about 0.1 to 0.2 points of game margin for every additional point of superior Form
So, to answer the question posed in the title of this blog, game venue (and Excess Venue Experience) is somewhat important - moreso for non-Victorian teams - but Excess Rating is much more important for all but a handful of teams.