AFL Statistics Series #3: Building a System of Player Value

Almost two years ago, I started a series of posts about statistics in the Australian Football League. The aim of the posts was to set out how much I thought the various statistics in the AFL mattered, and how they connected, in telling you how a team was able to score and stop its opponents from scoring.

Part of my interest was also to come up with a system for rating players’ contribution to winning. The AFL Player Ratings system is too complex for an amateur analyst like me to replicate. Other systems are simpler, but are less intuitive in how they measure value. If possible, I would like my system to be in terms of the player’s contribution to the team’s and opposition’s actual scores. While that would be the most elegant way of quantifying value, I’ll go into further below the potential challenges that presents.

After months of false starts and ends, here is where my thinking has led to. This isn’t the system, but it’s how after much consideration I now plan on building it.

Get the ball, move the ball, aim for the goals

As I said in my first post, the aim of Australian football is to get the ball towards and through your goals, and to prevent your opponents from doing the same. All AFL statistics can be classified in terms of these objectives.

The simplest part to capture is the result of a player’s shots on goal. Goals and behinds are easily available, and therefore we can easily see a player’s points per scoring shot compared with the league average. I went through in my second post in this series how this could be measured, and I don’t expect to talk about it too much more in the rest of these posts. The measure can be improved by using expected score, and accounting for kicks out of bounds, but since I’m aiming for a system that can mostly work with simple, historical statistics I’m probably going to leave this part alone and focus most of my exploration elsewhere.

Most of the work will be in the part that is often used to measure the impact of players – how often a player gets the ball and gets it down the field to help their team score (see table above). In other words, how much does a player do to generate shots on goal for their team? Relevant to this will be measures of:

- how often a player gets the ball – i.e. possessions;

- how the player got the ball – did they win it themselves in a contest, or were they part of a chain of possessions for their team; and

- how far they get the ball down the field – which can be measured by metres gained, but also by inside and rebound 50s, and also even just by kicks and handballs. 

The other part of ‘attack’ is how many points a player costs their team by not retaining possession. Turnovers are the most commonly cited statistic in relation to ‘wasting’ possession nowadays. However, many turnovers happen not through a player kicking the ball directly to an opponent, but by kicking to a contest which their team loses, so perhaps we will need to be a bit more nuanced in distributing the ‘blame’ for losing the ball.

Measuring a player’s contribution to defending, at least conceptually, should be the reverse – what does a player do to prevent or not prevent the opposition generating scoring shots? The common defensive measures – tackles, spoils, and intercepts – capture the former. Harder to capture is the latter, i.e. what does a player not do to prevent the opposition from scoring? Potentially this is part of why midfielders tend to occupy many of the top positions on rating systems – they are in more contests, and we can relatively easily measure the contests they win, but it is harder to measure the contests they lose.

Scoring is more variable than possession

As I said above, I would like my system to be in terms of the player’s contribution to the team’s and opposition’s actual scores. Sounds appropriate, right? However, many player rating systems do not have those relativities.

Consider the 2019 Grand Final (see table below). Richmond’s score was over four times greater than Greater Western Sydney’s. On the SuperCoach rating system though, Richmond’s collective rating points were less than 20 per cent higher. That is more similar to the difference in metres gained, and the difference in disposals was even closer – the Tigers had only one more disposal.   

Getting the ball towards goal for a score requires not only having possession, but also having the right sequence of possessions. Usually, it will not be just three consecutive kicks towards goal – it may be two kicks forward, one kick the other way, another kick back, a handball back the other way … until eventually a team gets the ball far enough forward that it scores.

What is the value of what the team’s players did then? Is it how often they got the right sequence of possessions to score? Or is it just how often they got the ball, and got it forward? What is the ‘true’ value of high possession-getters in ‘bad’ teams? That will be one of the questions I wrestle with in future posts.