'Possession adjusting': An essay.

Hello everyone, and I hope you’ve had a great Easter weekend if you celebrate it (religiously or culturally (yum, chocolate)), or just a great week if you don’t. I hear it’s Passover too! In England, the weather’s nice and M&S pre-mixed Mojitos are flying off the shelves!

Um, anyway.

The following is on ‘possession adjusting’. A basic definition from mine and Ashwin Raman’s football analytics glossary:

Often used with defensive statistics, these are metrics that have been adjusted to reflect the amount of possession a team has had.

Now we begin…

1.

In the beginning, there were no stats.

Then there were shot stats, and it turned out that they roughly matched who the good strikers were.

And then… there was the wilderness.

Around 2013-2014, WhoScored and Squawka sprang up and stats gradually hit the mainstream. People did not know what the stats meant. Even the smart people. And most of us, and I include myself in this, are not born smart.

From shots, sprung expected goals. As the tips of the tree of football, the attacking goalmouth, was fed the sun of statistical enlightenment, its rays gradually worked their way further down the ecosystem. Shot assists, and then expected goals assisted, and then ball progression.

But defensive stats stayed difficult. After resisting the idea for too long, believing that Opta’s figures contained some magic bullet, I accepted that the best way to work with (the currently available) defensive stats was to use them to build an idea of a player’s style or role within their team’s system.[1]

There are numerous problems with using defensive stats as a sign of good defending. The fact that some players are instructed to be more defensively proactive than others, for one. Are large numbers a sign of good execution, or just tactical role?

Another is that some players just don’t get the opportunity to do much active defending. When I was starting out blogging, it was around the 2014/2015 season. John Terry and Gary Cahill were in a soon-to-be-title-winning Jose Mourinho team and Burnley were getting their first taste of the Premier League. Some centre-backs – the defensive position I was focussed on – were just told to back off and sit behind their midfield.

An idea that I toyed with back then, and which Ted Knutson has continued[2], was about adjusting the defensive statistics in order to level the playing field. If Terry and Cahill were bunkered behind a midfield, surely it’s not fair to judge their (to come up with random figures) 1 tackle per game with the 3 tackles per game that other centre-backs were averaging.

‘Possession adjusting’ was born.[3] The concept is simple. Players on teams who spent most of the game defending (well, without the ball) get their defensive numbers dampened a little, while players on teams who have a lot of the ball get their numbers boosted.

2.

Recently, I’ve been thinking a lot about this.

Somebody (and I can’t remember who, to my annoyance) pointed out to me several years ago that both teams in a match are naturally going to have a similar number of turnovers. It’s just the maths behind one object, the ball, being tossed between two parties. Quirks of refereeing and data collection mean the figures are unlikely to match exactly, but they should be pretty damn close.

This came back to me recently, and I started thinking about how this concept applied to adjusting defensive stats. Let’s call it ‘turnover adjusting’, compared to ‘possession adjusting’.

Here are the theoretical arguments underpinning those two approaches:

‘Possession adjustment’: Making defensive actions, like tackles or interceptions, are partly due to role and partly due to opportunity. If a team spends most of the match without the ball, their players have more opportunity/more time in which they can win the ball back.

‘Turnover adjustment’: There are only a finite amount of times that the ball can be turned over in a match, and the number of turnovers will broadly match up between the two teams. The team who dominates possession, therefore, will naturally benefit from as many turnovers as the team they’re dominating. The turnovers that they do benefit from, then, are just condensed into a smaller period of time. Some matches are more bitty, therefore have more turnovers, therefore have more chances to ‘earn’ defensive action stats than others.

3.

I’ll be honest, I find both theories fairly convincing. They’ve been circling around in my head and, just when I think I’ve settled on one being correct, the other muscles back in.

I should briefly talk about why adjusting defensive stats might be worthwhile.

Adjusting defensive stats isn’t about seeing which are the ‘good’ defenders, but if you’re using the defensive stats to determine a player’s role, then you need to isolate that role as much as possible. In other words, you have three things that can affect a player’s stats. Their execution, their tactical role, and things outside their control like the quality of their or their opponents team and, from that, how much of the ball either side sees.[4]

On the most basic level, it’s kinda useful/interesting to know who the ‘high-volume’ defensive players are in football. Who the players who are just really damn active at breaking up their opponents’ play are.

You want to know this because, as a team, you might be looking for a player to play that kind of role, and you want to know who’s doing it at the moment. You don’t want your list of potential candidates skewed in a way that means the player’s role on the team they’re currently on might not translate to the team that you’re on.

(Or you’re in the media and you want to remove some easy-to-remove caveats around using statistics if, indeed, this particular caveat is easy to remove).

4.

There’s an argument, though, that there’s no point in adjusting defensive stats at all. We don’t adjust attacking stats like shots or expected goals just because a striker plays for a good team, for example.[5]

But why should you use possession or the amount of turnovers in a game to adjust defensive stats? What are you looking for in the numbers? Do you get closer to this with that adjustment?

Why am I asking all these questions?

The last one is one I can actually answer. I’m asking because I don’t have a firm opinion.

I think I come down on the side against possession adjusting. If it’s supposed to be a way to even things out between similar roles on different teams, I’m not sure it does it. The theory behind ‘turnover adjustment’ seems to me to hold true. Players on high-possession teams will be getting rewarded just because their team keeps the ball (I think).

Possession adjustment also assumes that possession is the determining factor in opportunity for proactive defending across all positions on the pitch. I think this is flawed.

It also depends on the defensive action itself. Even with my doubts about possession adjusting, it makes more sense for tackles and interceptions than clearances, the majority of which come from clearing crosses or long balls.

Statsbomb have a stat called ‘pressures’. It does what it says on the tin – it counts the amount of times a player pressures the person on the ball. I mention this because, crucially, it isn’t a ‘turnover’ stat like tackles or interceptions. If there’s any defensive stat that makes sense to possession adjust – at least as I theorise it – it’s these pressures.

so yeah. I am probably ‘against possession adjusting’, but I’m not sure I’ve decided what, if anything, I am ‘for’. Thank you for listening.

If you want to get in touch on this, you can find me on Twitter @EveryTeam_Mark or, if you’re a subscriber, reply to this email!

Footnotes:

[1]The same, really, is true of shots. There are forwards, like Roberto Firmino, whose job is only partly to be a direct goal threat. They’re rare though, which is why the primitive measures like shot volume tended to match up with good strikers. Tactical variety among defensive players is far more common.

[2]I believe that I came to the idea independently of him, but I can’t exactly guarantee that I hadn’t seen him or someone else talking about it and just can’t remember it.

[3]Although I did something slightly different, combining possession with the share of shots on target and blocked shots that a team conceded. The idea for me was that those shot numbers would give a sense of what the midfield was letting through.

[4]You could also argue that a player’s role should be taken across all facets of the game. A ‘search-and-destroy’ type of defensive role on a high-possession team won’t be the same as a ‘search-and-destroy’ type of role on a low-possession team, I think. This complicates things slightly, but is another argument against possession-adjusting defensive stats (I think?). i.e, you can’t just transplant one abstract notion of a particular player role from one type of team to another, the act of having more or less possession will change the role enough that they are no longer the same role.

[5]Maybe we should! Who knows! There is an argument that, instead of splitting stats up on a ‘per 90 minute’ basis, we should do it on a ‘per X possessions’ (usually per 100 possessions) basis. Thinking about it, this incorporates both of the ‘possession adjustment’ and ‘turnover adjustment’ theories. However, it’s far less intuitive than normalising stats on a ‘per game’ basis, and you also don’t necessarily solve the problem of knowing to what extent normalising the stats is even a worthwhile thing. (For example, the point raised in the previous footnote still applies).