Possession is weird

by imothyt

I was thinking about something this weekend.

Possessions in a football match are usually measured by the number of passes or touches a team makes* but the reality is that the actual number of times one team has the ball over the other team should be extremely close to 50%-50%.

This is just how football works. You could have the ball from kickoff to the 45th minute – pass around, never let the opponent have the ball, never take a shot and at the 2nd half, the opponent could do the same.

If you shoot the ball, unless the ball is blocked out, the opponent gets it back. If you kick it out of play the opponent gets it. If the opponent tackles and wins the ball, the opponent gets the ball. This basic fact means that every possession either ends with a shot or a turnover.

I tried to make the stats I have available to me (whoscored.com and StatsZone) square this equation and they cannot. The problem with these public data sets is that they don’t tell me when a tackle won possession (which is rarer than you think), when a clearance lost possession, when a shot was blocked and thus didn’t lose possession, etc. They do have a few direct change of possession stats: interceptions, offsides, failed pass, and dispossessed. But if both teams have nearly the exact same number of possessions (+-1), what does interceptions really tell us?

It also begs the question about “possession adjusted” stats. A few years back I wrote a piece on here explaining why possession adjustments make me uneasy (that article was lost in the move to this server). Basically, the rate at which a team tackles, intercepts, clears the ball, etc, has almost zero correlation on shots taken, shots prevented, or even the amount of possession that one team has. And while there may be a nominal correlation between possession and defensive stats, my experience tells me that something may be wrong with that correlation (it doesn’t pass the smell test).

There are many reasons for this. One is in the definitions of the stats themselves. Tackles are largely tied to dribbles. If your opponent never tries to dribble, you can’t “tackle.” It literally doesn’t matter how many passes a team makes, tackles are one-v-one duels. Moreover, possession stats aren’t actually possession, they don’t tell us the number of times anyone did anything except to pass the ball. Did they pass the ball in an attacking way, so as to expose themselves to an interception? Or did they van Gaal the ball? Passing it sideways in a frustrating game of keepaway?

That last bit exposes another truth about football: style matters. Sam Allardyce’s teams were famous for their supposed tenacious defensive style but did you know.. despite having the lowest possession they also had among the lowest tackles, interceptions, and even fouls committed numbers? Their game plan was to negate football entirely. The perfect result for Allardyce was if the game was delayed for 89 minutes, they got one chance to score a goal from a long punt, and either they won or the match ended 0-0. The same thing is happening right now in Burnley. Last season Burnley were among the lowest in the league in “possession” 43.4% and also the lowest in tackles. This chart illustrates what I mean:


I’m sorting here by possession and defense (defensive actions that win the ball back). Higher possession teams are in the upper half of the table: higher defensive teams are on the right of the table.

Clusters appear:

  • Leicester, Brighton, Southampton, Crystal Palace, and Everton were right in the middle.
  • The W teams (Watford, West Ham, Wolves, and Whuddersfield) were all high defensive action teams and as you can tell by the big grey band, median possession.
  • Cardiff were super low possession, high defense.
  • Man U were slightly high possession with high defense.
  • Man City are probably no surprise: high possession, low defensive actions. Liverpool are there also.
  • Newcastle, Bournemouth and Burnley are all low P low D teams.
  • rsq for last season was a whopping 0.16 and even if I drop off the super outliers (Cardiff, etc.) it only goes up to 0.24

So, what gives here?

  • Possession doesn’t really mean what we think it means. There are an equal number of possessions per game for each opponent.
  • Passing the ball more does indicate (slightly – ~0.2 r^2) that a team will play less defense. But consideration needs to be given to a team’s playing style. Burnley and Bournemouth are not going to put in a ton of tackles no matter how much they have of the ball.
  • What is going on with Man United? A high median passing team and high median defensive team?
  • Arsenal are a high possession team but also straying into the mid-pack defensive team? Is this possibly due to Emery’s counter-punching style?
  • Bournemouth: he seems to be trying to play Man City style on a budget. I wonder what would happen if Howe and Guardiola were swapped like Parent Trap? What if Howe had the resources of a Guardiola and Guardiola the resources of Howe?

I watched the North London Derby again with this idea of play moving back and forth in mind and saw the game in a totally different way than I ever have. I saw possession changing hands a lot, through various methods, but ultimately both teams had the same number of chances to do something with the ball. What they did with the ball is what really matters!

This could explain why the only stat I have ever really cared about is shots. My subconscious told me that goals are the purpose of possession, shots are the way to score goals (usually), thus shots are the way that you want possession to end. When I consciously know that both teams will have the same number of possessions per game, this preference for shots over everything makes sense.

Now for some more stats!

Ok, so Trent Alexander-Arnold has lost possession 113 times so far this season that’s 23 times more than 2nd worst Paul Pogba (90). Andy Robertson is 2nd on Liverpool with 80 possessions lost. Feels like teams should pressure these two players? (Source: https://www.sofascore.com/tournament/football/england/premier-league/17)

However… Possession lost includes all bad passes, being dispossessed, offsides, etc. Anything that turns the ball over to the opponent. T A-A does have 113 possessions lost, however…
60 of them are inaccurate passes (24 long, 36 short)
36 are inaccurate crosses
6 were dispossessed
4 failed dribbles
5 shots
And 2 more mystery turnovers that I can’t find in my sources (probably failed throw-ins).

Meanwhile Paul Pogba is 2nd in the League with 90 possessions lost, how do his numbers compare?
45 are inaccurate short passes
14 bad long passes
1 inaccurate cross
22 were either bad touch or ball dawdling
6 failed dribbles
7 shots

Some things like crosses/passes and failed dribbles/dispossessed may be counting twice here, sources are mixed (I have to use whoscored and sofascore here), but the point is that Pogba is turning the ball over at an alarming rate and not because he’s putting in crosses or something, like T A-A.

Arsenal’s top turnover machine is Ainsley Maitland-Niles with 65 possessions lost. His numbers are:

37 are failed passes (11 long, 26 short)
3 inaccurate crosses (of 7, which is odd)
12 are possessions lost through bad touch (7) or dispossessed (5)
2 are shots
5 failed dribbles

Weirdly, I can’t account for the others. There must be something that isn’t reported in any of my sources. This is especially perplexing because I think failed dribbles and dispossessed often count the same. I went back through every match and counted all of his failed passes (throws, crosses, everything) and he has 46. That actually squares everything.

I want to be very clear that I’m not “hammering” A M-N. Just reporting the numbers as they exist today.

Anyway, this has been my rambling Monday report of nothing. Have a great day!

Qq

*UEFA actually count the time with the ball, which usually results in some pretty comical ball in play times. The ball is out of play a lot more than people realize, about 1/3 of the game is dead ball time.

15 comments

      1. It would be cool if you could let us know what matches you will be watching and maybe we can have a discussion on these matches in the comment section with other people as well.

  2. I subscribe to the idea that football is a game about space; creating it, exploiting it, destroying it. Not the ball. Mourinho had that right, in part. You don’t need a lot of possession if you are efficiently creating and exploiting space when you do have the ball. Conversely, if you are effective at limiting and destroying your opponent’s space you don’t need a lot of defensive actions, they’ll end up turning the ball over to you. Atletico v. Barcelona in the Champions League a couple of years ago – Atletico put up an impenetrable wall, but I remember very few actual tackles or interceptions, just Barca players dribbling into blind alleys and losing the ball. I remember Klopp’s Dortmund teams would sometimes demolish teams with 35-40% possession stats, but they were so strong at running into space and making passes in behind the defense.

    Reading Guardiola’s biographies, he’s given a lot of thought to where you can/cannot lose possession and how to mitigate that and hence his adoption of the inverted fullbacks that squeeze in to seal off the middle when they have the ball. He’s a ball first manager, but he respects that you have to well positioned in case you lose the ball to destroy any available space in the middle. City defensively pressure and limit space and the opposition turn the ball over, needing few defensive actions to accomplish winning possession back.

    It’s a great discussion.

  3. I’m not sure I’d expect possession against unevenly matched teams to trend towards 50-50. Witness the late Wenger years. Arsenal usually had pretty good possession stats, but a lot of that was sterile sideways and backwards passing. That still happens some. And it will frequently be the case when very strong teams play weaker teams. Unless the weaker team is very good at pressing, there’s little upside for them playing more openly as opposed to Big Sam mode. If you’re a good team, the question is can you translate the possession into a win. Sometimes even really good teams can’t do that(see Barcelona-Chelsea Champs semi in 2012).
    So I agree generally that a straight possession stat is meaningless.

  4. what is significant about possession is the philosophical approaches to play. most teams that are confident they’ll win a game will want more of the ball, believing they are more likely to score. they may also see it as a form of defense meaning that if they have the ball, they are less likely to concede.

    another philosophy is to concede possession, defend the space in the final third, and hit you on the break. once again, it’s a philosophical approach so. trust the data as numbers don’t lie; there is unlikely to be any statistical correlation between defensive actions and possession. for most, it’s about philosophy and ability.

  5. Great post Tim

    In some sports such as baseball its very easy to understand and interpret stats and in most cases in baseball the stats will be consistently accurate for determining a players effectiveness. However the same is not true of European football and your post accurately points of just how easily these sort of stats can be misinterpreted, especially the defensive, passing and creativity stats. A player who puts in a lot of tackles is not necessarily a good defender in fact the opposite is sometimes true. Tackling can be a last ditch effort when the opposition player is about to beat you and the best defenders stay on their feet as much as possible. Someone who tackles a lot might actually be a bad defender. Ball possession and pass completion percentages can also mislead. Players like Denilson and Elneny have fantastic pass completion percentages but I don’t think any good team would want either of them in regular starting 11. Chances created is another potentially misleading stat. Believe it or not Lucas Perez was the second leading chance creator in Spain while he was on loan from Arsenal to Deportivo in 17/18. However, Deportivo did not keep him after the loan spell despite his chance creation rate. Arsenal didn’t want him, he could not get minutes with West Ham last year and now he is back in Spain and does not get regular minutes for Alaves which is a bottom half of the team in La Liga. Quite obviously the chances created stat has almost zero correlation with this players actual effectiveness. Another example stats being misleading is Arsene’s use of the Statdna program to highlight potential transfer targets. I understand how Xhaka, Mustafi, Elneny and Lucas Perez would all score well on statdna and Arsene spent close to $100M on those players but quite obviously their strong statistical scores were highly misleading in regard to how effective the players would be in an Arsenal shirt.

    1. I really like xA and big chances created as a metric for chance creation. It tells us a little bit more about how good those chances were rather than just bulk numbers of passes to a shooter.

  6. Joshuad

    I did not get back to you yesterday but I agree that someone who can finish or be creative at age 20 can still finish and be creative at age 30. However, that does not mean they are still as effective or as good at age 30 as they were at age 25 or 26. . Giroud has hardly scored for either club or country and in the last 3 years and he has lost his starting spot on his last clubs because he did not score enough. Its hard to image you would think that Giroud voluntarily decided to stop scoring for Arsenal and Chelsea when it cost him his regular club minutes and both clubs needed someone to score goals. The reality is Giroud is just not good at scoring anymore. When you watch Ozil play its quite clear that his is no longer even close to the dynamic dangerous play maker in the attacking 1/3 that was 3-4 years ago.

    Lacazette is the one example where I agree that he could probably score more goals if he was more selfish but I don’t think he is the prolific scorer who would challenge for a golden boot the way Aubameyang can.

    1. if over the next ten games arsenal scored 32 goals but an ever-present lacazette only scored 2, would you drop lacazette?

      btw, your claim that giroud no longer scores for france or chelsea is wrong. according to espn fc, in his last 8 starts, giroud has 7 goals (3 for france). this isn’t including the europa league final where he contributed a goal, an assist, and won a penalty contributing in 3 of the 4 goals chelsea scored that day.

      the point is there’s more to good center forward play than just goals. that’s the point wenger was making about the criticality of firmino at liverpool. from a biological perspective, giroud may not be atp but he’s the enzyme that catalyzes the france attack.

  7. Tim, as a longtime Houstonian and fan of the Houston Astros, I’ve watched their evolution since they burned it down in 2010, eschewing the Moneyball standard and building the Astroball monster. Every facet of this organization is ultra-efficient analytically. The largest strides (IMO) were in draft and prospect evaluation. The Astros went from losing 400+ games 2010-2013– to winning 100+ 4 of last 5 years– by restocking a depleted minor league system to the point where they have to offload talent because 1-2 levels of position players (in the minors) are blocked from making it to the majors.

    In this fashion, the Astros have, each of the last 3 seasons, dealt many top-100 prospects for 3 of the top-10 MLB starting pitchers.

    Apologies for the long lead-in. But it goes to the point about baseless football stats being forced into an equation for the sake of it. To attempt to create standards for football that just can’t be made with any depth– of stat tracking.

    Here’s where I say I agree with the simplicity of the x-shot stats that you are a proponent of. It’s not perfect. But as close as you can get to the reality of an outcome or– expected outcome.

    Where analytics are relevant in football? Are how the Astros have refined selecting their draft picks or signing prospects. Data-driven video analytics based on historical data being graded at a much deeper level (than those stat models bought or leased). Data gathered as far back as 2-3 decades (or as far back as a query requires).

    I followed the progress of Arsenal’s StatDNA project for as long as I could locate information on it publicly between 2012-2017 (when AFC went dark on it). With this latest group of youth players, and including Holding, Mavropanis, Guendouzi (and Bramhall)? My hunch is that Arsenal have been successful in creating a system that is now yielding results.

    Best evidence to disregard WhoScored’s player ratings?
    When their system rated Shkodran Mustafi in their PL team of the season 2 seasons ago.

  8. Josh

    I understand there is more to playing CF then just scoring, however scoring is easily the single most important thing a CF can do for a team. You can get creativity from all over the pitch but the CF is in by far the best position of any player on the pitch to score goals. Very few teams in the world have the luxury of goal scoring firepower in the rest of their squad that France does and for less fortunate teams having a low scoring CF makes its going to be much more difficult to score enough goals. Its simple math. A low scoring CF such as Giroud is a huge opportunity cost that almost no team in the world not named France can afford. Arsene loves creative CF and yet he dropped Giroud to the bench and then sold him because he didn’t have the firepower in his squad and he couldn’t afford to use Giroud. Giroud doesn’t start regular matches for Chelsea for the same reason. Your suggestion that Giroud could still score regularly but decided to be unselfish in order to be a facilitator for Arsenal and Chelsea seems poorly thought out when the evidence suggests that’s not what his manager wanted and it cost his spot in the starting line up on both teams.

  9. Could clearances maybe account for some of the discrepancy? Maybe a clearance is a possession lost for the attacking team but not one”gained” for the defending team.

Comments are closed.

Related articles

Read more stories on our homepage