Thursday, May 9, 2013

The AFL Power Rankings and the Effect Of The Season Break


In last week’s ‘AFL Power Rankings’, I noted that the Bombers’ fall in ranking from 6th to 9th may be a bit controversial given that they had just won their sixth straight game to start the 2013 season. Essendon also fell down the rankings over at AFL Footy Maths, moving into 12th spot.

On Twitter, Greg Jericho suggested that perhaps both ranking systems needed to change their weightings for matches in the previous season. It is certainly something I have thought about before. Currently, in my rankings system, a team’s result from four matches ago carries more weight than its result from five matches ago, but it makes no difference whether those matches were one week apart or six months apart. There are some reasons to think that a team’s performance may change by more over a season break than within a season – players age over a season break by as much as they do within a season, players leave and new players are recruited, and so on. What I need to find out is if the relationship between a team’s performance from the end of one season to the start of another season is very different from the variation in a team’s performance within a season. There are two possibilities, each with its own implication:
 
1)      They are very different – and that means a match should, all other things equal, carry less weight if it was played in the previous season than if it was played in the current season.

2)      They are not very different – and therefore the weight of a match should only depend on how many matches ago it was for the team.
To test this, I summed up the net margin for each team over its first five matches of a season, and then compared that to the net margin for each team over its next five matches of that season, and the net margin for each team over its last five (home and away) matches of the previous season. Obviously net margins will depend on which teams that team plays, but I’m assuming that for the purpose of this comparison, those effects will even out across teams. I only did this for each of the seasons from 2008 to 2012, because strange as it may seem, I don’t have a spreadsheet of the result of every match in AFL history lying around, and so I had to undergo the tedious exercise of entering the results in for seasons prior to the advent of the Power Rankings.  That does not give a large sample size, so all that is stake here for the moment is how far I am willing to pursue this question further.

Correlations of AFL teams’ net margins over first five matches of season with next five matches of season and last five (home and away) matches of previous season

Season
Correlation with next five matches of season
Correlation with last five matches of previous season
2012
0.46
0.51
2011
0.61
0.44
2010
0.36
0.38
2009
0.74
0.58
2008
0.65
0.27

As you can see, over the past five seasons there is, on average, a greater correlation between a team’s net margin over its first five matches of the season and its next five matches of the season than with its last five (home and away) matches of the previous season. Again, it’s only five seasons of observations – it might be that the previous fifty seasons show the reverse pattern! – but perhaps the question of the effect of the AFL season break on performance is something I should give further thought. If only I had a spreadsheet of every VFL/AFL result ...

No comments: