Monday, August 23, 2021

AFL Rankings: Round 23 2021

It looks a very close race in this year’s finals series, but I’ve a few bucks on the Brisbane Lions to win the 2021 flag.

It was an exciting finish to the 2021 home-and-away season, with positions changing significantly by the week – and by the minute on the weekend in the final round.

The Brisbane Lions, who were outside the top four over the past month, snuck into fourth in the final minute of their match against the Eagles, giving them a double chance in the finals.

The Western Bulldogs, who were on top of the ladder after Round 20, lost their last three matches to finish fifth, meaning they are now in an elimination final and must win all of their finals.

Port Adelaide finished in second with their two-point win against the Dogs, which is their highest position since Round 2, and earned themselves a home final.

And Melbourne won their first minor premiership in 57 years, which usually means they should not get a home ground disadvantage unless they lose in the first week, but who knows this year?

Premiership favouritism has changed often in recent weeks, and the finals series looks like it will be no different. In the current TAB betting, the top four teams are all at odds of less than $6.00, and Stats Insider has all of the top four teams at a probability of between 20 and 25 per cent. Adding more complexity is that – with state restrictions due to COVID – we don’t even know where teams will play each week if they do win; for example, whether Port Adelaide gets a home preliminary final if they win this week, or which state the Victorian sides would host finals in, or even where the freaking Grand Final is going to be played!

Nevertheless I make picks for the finals every year, and this year will be no exception. Here are my predictions for every match of the 2021 AFL finals series:

Courtesy of AFL.com.au Predictor

So I’m picking the Brisbane Lions, who currently top my rankings, for the premiership, and reflecting my confidence I’ve put a whole three bucks on it. I’ll enjoy my sixteen dollars return – which I will have already spent on another six-pack to get me through lockdown – in the ‘quarter of a chance’ event it comes off.

Sunday, August 15, 2021

AFL Statistics Series #5: What Happens In a ‘Possession Chain’?

This is the big one

Of my AFL statistics posts, this one is perhaps going to be the most important, or at least the most detailed. In this post, I am going to break down ‘possession chains’. I am going to see how they are most often started, maintained, and ended.

As most of the common actions recorded in AFL statistics take place as part of these possession chains – either as being part of them in attack, or in stopping them in defence – the aim is to get more of a ‘proper sense’ of the value of these actions. For example, one trendy claim over the past few years has been to say that hit outs by rucks do not mean much, since often a team wins the hit outs but loses the match. That is true, but it does not mean that winning hit outs isn’t valuable in its own right: their value just needs to be put in the context of all the other actions that contribute to a team scoring (or preventing the team from being scored on).

The game we are going to examine

To examine possession chains, I am going to use a log of possessions from the 2018 Grand Final between the West Coast Eagles and Collingwood. In part I chose that game because I meant to do this post two years ago (!), and at that time it was the most recently played match. I also chose it in part because it is a game I don’t mind re-watching. (Admittedly I do love re-watching Richmond’s premiership wins more, but I decided not to choose one of those in case analyzing one of those in detail did end up ruining my enjoyment of that match.)

To help with the analysis, I copied and pasted the actions of every player from AFL Stats Pro for this match – back when AFL Stats Pro still had useful, detailed information like this. This means that how the actions have been categorized in this post have been largely based on how that website categorized them, with some minor changes where I needed to fit things into my categories for analysis.

Returning to what a possession chain is

The concept of a ‘possession chain’ I believe originally started with Champion Data, though it has been used by others since, including HPN Footy. Possession chains can be started in one of three ways:

Clearances – getting the ball out of a stoppage;

Intercepts – getting the ball off the opposition; or

Kick-Ins – getting the ball after the opposition has kicked a behind.

Chains are ended by:

Scoring – kicking a goal, which is obviously the ultimate aim of ending a possession chain, or a behind;

Turnovers – giving the ball to the opposition; or

Stoppages – mostly ball-ups and throw-ins, but which I’ll also stretch here to mean events that stop play such as the end of the quarter.

There are some actions – including possessions even – that do take place outside of these chains. For example at the start of the 2018 Grand Final, before Luke Shuey gets the first clearance from a free kick 45 seconds in, there are two hitouts, two tackles, a contested possession, and a ‘one percenter’ (a smother) that are all captured within AFL statistics, but have occurred before and outside the start of the first possession chain. Nevertheless, most actions take places within possession chains, so looking at these will still tell us a lot about what happens during a football match.

For ease of analysis, I will count every time after a clearance that there is a change in the team which has ‘control’ of the ball as an ‘intercept’ to one team, and a ‘turnover’ to the other team. The data I took from AFL Stats Pro, which recorded player ‘possession gains’ rather than ‘intercepts’, did not attribute these to events such as a free kick from an opponent kicking the ball out of bounds on the full (possibly as the player taking the resultant free kick did not ‘earn’ possession), or if a player kicked the ball off the ground without picking it up (in this case a player still gets a disposal). Alternatively, Footywire appears to have counted these as intercepts and turnovers, but there are still a few occasions where an intercept was recorded for one team without a turnover being recorded for the other. I will add these in so intercepts for one team = turnovers for the other. (If these were not counted as turnovers they would be counted as part of ‘possession chains stopped’.)

The Eagles famously won the 2018 Grand Final over Collingwood by five points, or more specifically by five behinds; that is they had five more scoring shots. They did this despite having five less clearances and five less kick ins than the Magpies (see table below). What fueled their greater number of scoring shots was getting 13 more intercepts.


As Richmond showed over their years of recent success, winning the ‘intercept battle’ is usually more important that winning the ‘clearance battle’, just because these happen more often and start more possession chains. Each individual clearance and intercept counts equally for starting a possession chain though.

Every possession chain started equals a possession chain ended. The Eagles having 13 more intercepts also means they had 13 less turnovers, yet only five more of their possession chains ended in scores. Where the Pies made up a lot of that difference was in ‘stoppages’ – 11 more of the Eagles’ chains ended in a stoppage in play compared with those of the Magpies.

Why bother with this concept?

Teams get themselves in a position to score by starting and maintaining possession chains, and they are prevented from scoring by turnovers or stoppages. Note this is true of many team ball sports, but I think there are reasons why it is more useful conceptually for Australian football than many other sports. Taking a sport like soccer for example, while you can only score from an ‘unbroken’ chain of possession, given how few scores there are it is clear that almost every chain is ‘broken’, probably making the concept less useful there. Taking another example of basketball, in that sport analysts have focused a fair bit on the number of possessions a team has (where ‘possessions’ means there something more like ‘possessions chains’ rather than the number of individual times a player touches the ball), but many possessions there are often not ‘won’ – they simply come from alternating possessions after scores. It is also relatively easy in basketball to progress the ball down the court into scoring position, even if scoring itself is a 50-50 prospect.  

Other football codes like rugby and American football may be more similar in that a team must progress the ball down the field into scoring position through a ‘chain’ of possession, and a team more often is doing something – at least successfully defending – to win possession back. For those codes too you could look at how teams win or lose games in part by the number of ‘chains’ they have and how they win or lose possession.

What makes Australian football relatively unique though is the commonality in which the ball is in dispute. In rugby and American football the ball is usually clearly in one team’s hands. We can often tell reasonably easily what the outcome of disposing the ball was in those sports, for example by whether a pass was deemed ‘complete’ or ‘incomplete’. In Australian football though a kick may have maintained possession of the ball, turned it straight over to the opposition, or quite often put the ball into a disputed situation where each team has a decent chance of getting possession. As we will see, there are turnovers from ‘clangers’, and then there are disposals that result in turnovers arguably as much through the result of a contest than through the result of the player who put the ball into that contest.

That is why I wanted to break down possession chains in AFL in more detail to get a sense of how often various actions were leading to a team winning control of the ball or losing it, and through this how they contribute to points ‘for’ and ‘against’. Hence we’ll now look in more detail at the role of various actions – kicks, handballs, tackles, etc. – in starting, maintaining, and ending possession chains.

Australian football is ‘aerial ping-pong’

Let’s quickly look at how long a possession chain usually lasts. In the 2018 Grand Final, the average number of possessions per possession chain was 2.6. (Note again some possessions do not occur in possession chains as defined above.) Over 60 per cent of chains lasted only one or two possessions (see table below). Apart from a monster 19-possession chain by West Coast, no possession chain lasted more than nine possessions.

This is largely due to the ‘contested’ nature of Australian football, even (or some may argue especially) at the top level. Forty-six per cent of possessions in this match were contested. As contested possessions are, by definition, possessions won when the ball is in dispute, then that shows around half of possessions were won when the other team had about an equal chance of winning the ball, hence making long possession chains difficult. The difficulty of the attacking team maintaining possession when the ball was contested may have been even harder than that, as we’ll see further below.

Contested possessions win the ball…

We’ll look first at how possession chains start. In the 2018 Grand Final, over three-quarters of possession chains started from a contested possession (see table below). Only around 10 per cent of possession chains started from an uncontested possession – indeed, about as many chains for Collingwood started from kick ins than from uncontested possessions. This suggests that for a team to gain (rather than retain) possession of the ball, the majority of the time they’ll need to win it through a contest.

Clearances of course start a significant proportion of chains, accounting for around a third of all chains started. However, intercepts – not clearances – still accounted for the majority of chains started through contested possessions. When one hears the word ‘intercept’ the image that might come to mind is that of a tall defender sailing across the backline to take an uncontested mark. More intercepts occur though as a result of contests. More also occur at ground-level rather than up in the air; there were 45 intercept marks in the match, which is less than a third of all the intercepts that occurred.

Following on from that point, just under half of possession chains were started through ‘ground ball gets', which are contested possessions won at ground level excluding free kicks. Chains started through free kicks, hit outs to advantage, and contested marks were all about as common as each other, accounting for less than 10 per cent each of all chains started. (A player on the receiving end of a hit out to advantage can record a contested possession, but not a ground ball get.)

This I think helps to show why hit outs have only a minor influence overall on the result. However that should not be considered as a knock against the value of rucks, as – with 22 players per team – usually any individual player is going to have a minor influence overall on the result. Through his hit outs to advantage in the 2018 Grand Final Brodie Grundy started 15 possession chains for his team, which even if you give some of the credit to the player who received the hit out, is a relatively major contribution to winning the ball from one player.

… uncontested possessions retain the ball

Next we’ll look first at how possession is retained. For the most part possession chains in the match were maintained through uncontested possessions, with around 85 per cent of possessions after the start of a chain being uncontested (see table below).

Almost all of the possessions that followed handballs, which made up close to half of all the possessions which maintained a chain, were uncontested. For kicks, around two-thirds of kicks that maintained a possession chain resulted in an uncontested possession for the team, and around half of them resulted in uncontested marks.

One thing I noticed is that the number of contested possessions that maintained a possession chain is much less than the number of contested possessions that were intercepts. This could suggest that the defending team has a better chance of winning a contest. I don’t know how specific that finding is to this match though.

So if a handball is safer, should a team try not to kick the ball then? Of course not - kicks usually gain a team far more territory than a handball does. The ideal possession chain for a team would be something like a series of forward kicks to uncontested marks; the catch is that if you’re kicking long you’re quite often kicking to a contest.

Note that for kicks that resulted in a non-mark contested possession, some of these have been classified as ‘effective kicks’ and some as ‘ineffective kicks’. Both ‘effective’ and ‘ineffective’ kicks can lead to contests, as a long kick of greater than 40 metres that goes to a contest is classified by Champion Data as an effective kick. That makes it hard just from looking at kicking efficiency statistics to tell how many of these effective kicks went to teammates, and how many actually resulted in a player’s team losing possession through a turnover.

Contests end possession

Finally, let’s look at how possession chains most often end. In keeping with the theme of football as ‘aerial ping-pong’ the majority of possession chains finish up in turnovers (see table below), rather than stoppages or scoring.

‘Turnovers’ in the AFL though are often quite different from the turnovers that we generally think of in basketball or other codes of football, where one team loses the ball straight to the other. As implied above by the data on intercepts, three-quarters of turnovers in the 2018 Grand Final actually occurred through the opposition winning a contest. For around 40 per cent of those turnovers from contested possessions, the final disposal in the chain was classified as an ‘effective kick’.

This means that a long kick to a contest seems to be a more common way for a possession chain to end than a clanger that puts the ball straight down the opponent’s throat. In this match there were only 23 clanger kicks that ended a team’s possession through an uncontested possession for the other team. That doesn’t mean clangers don’t matter, as they are still the worst way to dispose of the ball, just that they are not the most common source of turnovers.

For slightly less than half of these turnovers there was an opposition tackle or ‘one percenter’ (spoil or smother) involved in the change of possession. This indicates that both tackles and spoils play an important part in the ball changing hands, but also that many changes in possession can happen without one of these events. Put another way, sometimes a tackle or spoil creates the contest that leads to a change in possession, but often it comes about just through the attacking team’s kick to a contest – either because the defending team is pressuring the ball, or otherwise.

Tackles and spoils were more important in this match to chains that ended in stoppages, with around three-quarters of chains stopped having a tackle or one percenter involved. This makes sense as often stoppages in play are the result of the defending team laying a tackle, or punching the ball over the boundary line. Disposal-wise, as with turnovers, the last disposal in chains prior to a stoppage was more often a kick than a handball.

In summary

Australian football is a game of contests. It starts with a bounce in the middle in which eight players try to win possession of the ball. It has many similar re-starts all over the ground from stoppages. If you get the ball your aim is to progress it uncontested down the field towards a scoring position, or at least put it where you have a decent chance of winning possession again. If you don’t have the ball, ideally you force the opposition into giving the ball straight back to you, but more often you try to at least get the ball into a contest again.

An important thought I want to leave this post with is that I think the value of contested and uncontested possessions depends on how often they occur. If most possessions are contested possessions – something that might for example happen in lower-level junior games – then the value of each individual contested possession is probably worth less, as winning the ball is expected to lead to less progress. On the other hand, if most possessions are uncontested possessions then winning the ball in a contest in order to start a chain of possession becomes more important.

Along with my previous posts on gaining territory and scoring accuracy, I should have most of what I need now to evaluate teams and players – or at least most of what I need from a conceptual view given the unavailability of some data. Hopefully when I next get back to this I’ll have much of my final framework for evaluating a player’s contribution to winning.