Tag Archives: Stats


Toward a model of adjusting player stats for the “personality” of the team and the league

As I said in my first article on this topic I think that Ted Knutson from StatsBomb has a good idea. Individual player’s stats are deceptive and need to be grounded in the context of the “personality” of the team he plays for and the League he plays in. The idea of adjusting individual’s stats for the personality of the team, in his formula he used possession, is solid. The problem is that possession doesn’t map to defensive stats because possession is simply a ratio of passes.

The reality of football is that teams vary wildly in terms of their defensive aggression. How often a team attempts tackles or gathers interceptions has a very weak correlation to possession stats. I want to be clear that I’m not trying to hammer anyone here. Rather, I’m trying to build on Ted’s idea and create a new model which will help us to understand individual player’s stats better.

To that end, take a look at the chart of Premier League teams below (click to embiggen):


In this chart I have sorted the 20 League teams by possession. I have broken down each team’s tackles and interceptions. Then I have taken the League averages for the tackles and interceptions (added together) and the averages for tackles and interceptions. Then I have done a +/- over the average for total, tackles, and interceptions.

What you can see very clearly is that the top five teams who dominate possession also are among the most aggressive defensive teams in the Premier League, with the exception of Arsenal who are essentially an average defensive team. Those top five teams are making more passes than average (that’s what possession measures) and more tackles+interceptions than average. The next three teams are known as defensive teams and yet they have fewer tackles+interceptions numbers than average.

I think that the Chelsea numbers are a good clue as to why teams who have a reputation for boring and defensive football also are less active defensively. If you look at the whole chart, teams with a negative overall defensive number are all well known to be disciplined defensive sides. They aren’t diving into tackles or playing the passing lanes. They sit back, in two banks of four, and play positional defense. That’s my theory anyway.

So, perhaps we could adjust player’s stats based on an overall “defensive aggression” number? Even if we did that, however, we might be wrong adjusting certain player’s stats, like say tackles, on a team like Swansea who (as you can see from the chart) aren’t a very tackle oriented team.

Looking at the next two columns we see a further breakdown of each team’s defensive personality. For example, the most aggressive tackling teams are Stoke, Liverpool, Crystal Palace, Southampton, and Manchester City. I think that’s a fair assessment of the Premier League, actually. Stoke and Liverpool play smash-mouth football, which may offend some Liverpool fans but for anyone who has seen them play you know that they are scrappy and a bit dirty on the defensive end. Crystal Palace is another interesting team. I bet they avoided relegation last year simply with their amazing defensive work rate, that +299 with + in both tackles and interceptions is really crazy.

Regardless of whether you agree about my analysis of Liverpool and Crystal Palace, I think I’ve made my point here with regards to using possession as a means of adjusting player’s stats. Instead we should probably be looking at the personality of the team and using that to adjust a player’s stats. How, exactly, we adjust for that I haven’t figured out yet. I don’t have a math genius here to bounce ideas off of, like Ted does. If you’re a math wiz and you want to help, just leave me a comment below. I’ll get back to you.

In the mean-time, one more thing to consider: a team’s stats also vary from league to league. For example, the Bundesliga is a far more aggressive league in terms of tackles than the Premier League.

BundesligaRemember that Bundesliga teams only play 34 games a season and yet they are averaging 339 more tackles per season than their Premier League rivals! That’s 10 more tackles a game. Per team. For my money that seems like a much more important thing to adjust for than possession.

How do we do that? I don’t know, yet!





Why we shouldn’t adjust defensive stats for possession

“Have you thought about adjusting your stats for possession?” – Everyone since June

Last month the venerable Ted Knutson published an article strongly suggesting that us stats folks should adjust basic defensive stats like tackles and interceptions for possession. The theory being that the more possession (offense) a team owns the fewer opportunities that same team has to make defensive actions. If true, then we should adjust player’s individual stats for their team’s possession. It’s a good theory with only two minor flaws: 1) it’s wrong and 2) it’s wrong. I know technically that is just one flaw but it seems like such a big flaw that I felt it deserved to be repeated.

First, you have to understand what possession is and how it is measured. Opta measures possession simply as a ratio of passes between teams. Ergo if Arsenal make 600 passes and West Ham make 400 passes the possession numbers at the end of the game are reported as 60-40. Measuring possession as pass ratio has caused a good bit of consternation among stats nerds for a few years now. As you know, many teams, like West Ham, intentionally waste time when they have the ball in order to prevent teams like Arsenal playing. At the end of a match like that, however, a team with faster passing might end up dominating possession as something like 70-30 when in actuality the possession numbers were more like 50-40-30 with 30% of the ball being out of play.

Don’t believe me that a ball could be out of play for 30% of the match? Well, if we use the alternate method of tracking possession, the “chess method” where a clock is actually deployed and time measured for when a team has control of the ball we see that Bayern, for example, led the Champions League in possession last season at 65% and yet only had, on average, 38 minutes of play. 65% of 90 is 58.5. Where did the other 20 minutes go? Time wasting and other tactics to slow the game down.

But even if we were to accept the idea that possession is a “good enough” measure of offensive dominance and thus defensive opportunity it still doesn’t explain why a team like Barcelona who routinely have 68%+ of possession would wrack up fewer tackles than a team like West Ham who routinely have ~40% of possession. If the idea that less possession = more opportunity and therefore inflated individual and team defensive stats we should see West Ham, Fulham, and Crystal Palace blowing away teams like Barcelona in terms of total tackles and interceptions.

Or even if we were to stay in the Premier League and take Southampton, who led the Premier League in possession, and compare them to the bottom teams we should see the bottom teams have more tackles and interceptions. After all, they have more opportunity.

But rather than seeing a direct correlation between possession and defensive stats, we actually get a very poor correlation as you can see below. Southampton and Man City led the League in possession while West Ham, Crystal Palace, and Fulham were the bottom three in possession.


Courtesy http://www.squawka.com/comparison-matrix

How is it possible that Southampton made 1802 attempted tackles and interceptions while West Ham, a team known as a tough tackling side who Stuart Robson loves to commend for “earning the right to play”, makes just 1456 of the same actions? That’s 346 fewer actions, almost ten a game! And Southampton has 58% of the possession compared to 42% of the possession for West Ham? The answer is simple: possession, or lack thereof, has little to nothing to do with tackles and interceptions.

If we were to apply Knutson’s formula to a comparison matrix between Mark Noble and Morgan Schneiderlin we would artificially inflate Schneiderlin’s tackles and interceptions numbers on the inverse basis of their possession % and artificially deflate Noble’s numbers on the assumption that he has more chances to tackle and intercept. The only problem is that here, between these two teams, that would be exactly wrong.

The problem is in the assumption. The assumption is that every time a team makes a pass it presents an opportunity to intercept. Similarly, that every time a player makes a dribble it presents an opportunity for a team to tackle. But not every team is trying to win the ball back with every opposition possession. Noble tackles less because his team tackles less. If you’ve seen them play you know that they are content with the opponents having the ball in their own half. Meanwhile, if you watch Southampton, they press. They try to force turnovers.

That’s just one example, there are others. In fact, Ted’s own regression analysis showed an R2 value of just .40. Only 40% of the variation can be explained by the model. That leaves a whopping 60% unexplained. I’ve done a fair number of regression analyses and I would probably never publish a .40 much less make some of the sweeping statements that Ted makes.

The thing about stats and especially football stats is that you have to keep them grounded in context: in the context of a team’s playing style, in the context of a player’s abilities and history, in the context of a league, the context of that league’s overall competition, and in the context of a league’s refereeing style.

I do think Ted is on to something. The idea that playing style matters to the end stats is bang on and you do see me taking into account a player’s contributions to the team’s whole. That said, possession only measures pass dominance. So, if I were to adjust anything for possession it would be passes. If I wanted to adjust defensive stats between two players, I might do something similar to what I have done above and compare those two teams’ tackles+interceptions ratios. Or even better, find out what the team’s tackle-interception ratios are in the League and then adjust for that.

But possession? I’m not even sure I like the possession stat as a “possession” stat much less as something I would use to adjust defensive output.




Wenger after Khedira but where would he fit at Arsenal?

“Please do a stat comparison of Khedira and Schneiderlin!” – everyone on Twitter

If absence makes the heart grow fonder then ignorance makes the heart grow clamorous and there has never been racket so clamorous as the rumored transfer of Sami Khedira to Arsenal. It is a noise like a a chest full of kids toys thrown down the stairs. Squeaking, chirping, clanging, and ultimately some crying.

In terms of stats when I’m looking at defensive midfielders for Arsenal I’m looking for high numbers of tackles, interceptions, and long passes per game. I also want to see a low turnover rate and low dispossessed numbers along with high pass completion rate and a large number of passes per game. A high completion rate of long passes is a bonus, a high number of blocks per game is a bonus, and a high number of key passes is a bonus.

Shots per game are not something I look for in a DM. If I see a player with high numbers of tackles, interceptions, passes, and shots I’m looking at someone who is more likely a box-to-box midfielder. For example, Aaron Ramsey. 2nd at Arsenal in tackles per game, 1 interception per game, 3rd among outfield players for long balls per game, second in passes per game. Looks like a DM until you see that he is second in all those stats to Mikel Arteta and he averages almost +1 key pass per game over Arteta, almost +2 shots per game over Arteta, and almost +1 dribble per game more than Arteta. If Arteta is the prototypical Arsene Wenger DM, Ramsey is the prototypical Arsene Wenger box-to-box midfielder.

But is there an actual “prototypical” Arsene Wenger anything? Two years ago Arteta and Song played next to each other in midfield. They both had about the same number of tackles per game, interceptions, and passes. Song had more turnovers and dispossessed numbers, more assists (11!) and far fewer shots and fewer key passes. They both had about the same number of attempted long passes but Song was no where near as good at it as Arteta (88% v. 68%) and nor were his overall passing percents as good as Arteta. I joke but it did look a lot like Arsenal were playing with a holding attacking mid next to a defensive box-to-box player. I’ll let you sort out which was which.

All of which is to say that Wenger doesn’t have “prototypical” anything. I think he even may intentionally eschew such things. So, where does Khedira fit in all of this and what are Khedira’s stats?

Well, Khedira’s stats suck but they aren’t really his fault. First, he’s been injured and I don’t need to tell you that a player has to stay healthy to compete for places on a team like Real Madrid. Second, when he does play he plays next to one of the best midfielders of his generation in Xabi Alonso. What’s Khedira’s role in that team? He does a little of this, does a little of that. but takes a back seat to Xabi.

The result is that his stats look like crap. Let’s take the main three DM stats from last season, tackles, Interceptions, and long passes:

Khedira comp

Khedira’s not the main tackler, nor the main man to intercept the ball and is certainly not the deep distributor. Because Xabi, like Elvis, is everywhere on the pitch already doing all those things. But weirdly, even when he was at Stuttgart he never shone too brightly. His numbers there were pretty terrible to be honest and yet Real Madrid paid €14m to get him from Stuttgart and then paid him €100k a week in salary! So, we can’t look at his stats from his club football and draw any conclusions.

In the end, Khedira is an absolutely fascinating player. He’s not been given a free role at Real Madrid to showcase what he can do and yet despite his back seat role at Real Madrid, and injury nightmare, and despite his kind of mediocre performances for Stuttgart he’s been selected to play for Germany 51 times. (record scratch).

Yep, Khedira is a mainstay in the German national team. Has been since he was a kid. He’s captained the German U21 side and been selected in every senior team match he was available for since 2010. This is a German team with an embarrassment of players available in midfield to play the defensive midfield role and Jurgie Löw chooses Khedira over all of them.

We’ve also seen that he’s not “just a defensive midfielder” for Germany. He can go forward, he dribble, and he can set his teammates up for goals. As you might expect from a player who has had to play multiple roles over his career to get time at Real Madrid he is versatile and classy.


So, for once I have to put down the stats and ask “what kind of midfielder is Sami Khedira?” He’s a German international midfielder. A German international midfielder who is about to help his country win the World Cup.

And where would he fit at Arsenal? My guess is “in the midfield, somewhere, doing all kinds of cool things to help Arsenal win”.