MLS: More Possession, More Winning?


This column provided by Northwestern University seniors Sam Stejskal and Shashank Churukanti and the good folks at soccer analytics start-up, Chimu Solutions. Take a trip on over to Chimu Solutions (kind of like an Opta on steroids.)

We see possession numbers thrown around all the time in the soccer world.

It’s easy to understand why: the possession stat is simple to calculate and even simpler to understand.

Another reason for the number’s ubiquity: It seems like a good measure of which team will win a match. Hold the ball longer than your opponent and the chances of you scoring more goals – and thus winning – would seem to go up.

But is this actually true? Do teams that hold the ball longer really have a better chance of winning? Do possession stats tell us anything at all?

To find out, we studied possession numbers from every single MLS game this season. We gathered data from all 287 regular season matches through October 12th, first tabulating the goal differential for each game. Here are the results of that tabulation:

We then proceeded to calculate the difference in possession of the two teams.

For draws, we were simply interested in the absolute value of difference in possession, whereas in matches with winners, we subtracted the possession percentage of the losing team from the percentage of the winning team.  A positive number means that the winning team held the possession advantage, while a negative number indicates that the losing team held more of the ball.

After this, we calculated the mean value of the possession difference for all margins for victory.

If having the ball more often really does lead to scoring more goals, then winning by a larger margin should mean a larger advantage in possession.

Simply put, we expect a positive correlation between possession difference and margin of victory.

The following graph disproves this hypothesis:

Apart from the sole 5-goal game (New York’s thrashing of Toronto FC in early July), it is evident that such a positive correlation has not been present thus far in the MLS regular season.  In fact, after omitting the lone outlier, the data shows that the only margin of victory that has flirted with the 50 percent possession mark this year is the two-goal win.

However, for our purposes, there is a problem with using mean values. While they disprove any positive correlation between higher margin of victory and having the majority of possession, mean values are too general. They don’t give us any insight about specific possession data measured against wins and losses.

The below graphs give us this specific data, showing that winning teams have had less possession than their opponents in 112 of the 184 games (61 percent) that have ended in a win.

The following diagrams show the possession numbers for games with one- and two-goal margins of victory.

*remember that a negative number implies that the losing team had an advantage in possession (i.e. if the winning team had a 40-60 disadvantage in possession, the measure for the game would be -20)

The data for games with three- and four-goal margins of victory is similar.

All of this data means that the popular notion that having the ball more increases the likelihood of winning hasn’t held true in the 2011 MLS regular season.

We see this with each of the current top four teams in the league: L.A., Seattle, Salt Lake and Dallas. Each of these teams holds less of the ball in wins than in ties or losses, with fourth-place FC Dallas sticking out the most, holding the ball just 45.6 percent of the time in their 14 wins while having it 53 percent of the time in their 11 losses and 46.8 percent of the time in their seven draws.

The possession data for all of the top four teams is shown below:

Real Salt Lake is a particularly interesting case. They didn’t lose the possession battle until their 21st game – a 3-0 win over New York on August 6th. Since that game, RSL has had less of the ball than their opponents five times – all wins – and more of the ball in six matches – all losses.

All of this data – along with RSL’s curious case – begs the question: why do teams win more when they are out-possessed?

There are several possible reasons. First and foremost, teams that grab a large early lead might tend to sit back and defend, allowing their opponents to hold more of the ball after they’ve built a comfortable advantage.

A good example of this is D.C.’s 4-1 victory over Real Salt Lake on September 24th. D.C., powered by Dwayne De Rosario’s record-breaking first half hat trick, took a 4-0 lead into the half. They didn’t do much attacking after the break, sitting back and allowing RSL to dictate the game in the second half. This tactic – which, of course, is totally reasonable and completely understandable – was partially responsible for RSL’s huge 66.1-33.9 percent possession advantage.

Other possible reasons why MLS teams win more when they have less of the ball could be counterattacking play and discrepancies in finishing ability.

Really, there could be any number of reasons for this interesting phenomenon. The point is, holding an advantage in possession is not a good predictor of which team will win a game.

At Chimu Solutions, we recognize this – and are looking to come up with metrics that better predict performance. We use a network model to statistically analyze player and team performance, calculating the importance of each player within the context of their team’s network and properly rewarding players for their contributions to attacking and defending movements.

21 responses to this post.

  1. Posted by Joe on 2011/10/17 at 3:11 PM

    Results may not be applicable for Barcalona


  2. I think at that point you have to add a talent variable.

    I would argue this (without the data which is sacrilegious in this column). That on average the more talented team wins.

    You play over the top ball with Barca and they still easily beat 15 of the 20 La Liga teams.


  3. Posted by Geeps on 2011/10/17 at 4:23 PM

    Graphs tell me Possession Difference for 1-goal games is normally distributed (an expected outcome). In 2-goal games, it slightly favors team with higher possession (what I’d expect). To do this right, you need to do multi-variate analysis (that is, many variables can influence the outcome, not just possession). With proper analysis, some variables will prove to have stronger correlation to wins. In Barcelona’s case, they not only have 70% ball possession, they complete 700+ passes, and win back possession quickly. They also avoid long-balls (higher probability of a bad pass or loss of possession). Hard to measure, but Barca have “smarter” players. Be that pass selection, body position, and supporting movement to enable safe passes / possession. MLS lacks is all these categories. In fact, I’d venture to bet that MLS current playing style will not show clear correlation between possession and wins. That’s because they don’t play intelligent, smart soccer that Barca does.


  4. Posted by Geeps on 2011/10/17 at 4:25 PM

    If I look at youth soccer (my son plays competitive), they can easily lose a game even if they dominate. Why? Because they don’t know how to win and how to do things (smart play) to avoid big mistakes. They remind me of MLS.

    MLS is physical first and foremost; not technical.


  5. Posted by Jake Claro on 2011/10/17 at 4:26 PM

    Yeah re:Barcelona: I’m more interested in how this applies across leagues. We may be seeing a bias in the way in which the American game is played that is not reflected in other leagues. Also, would be even more interesting to see at the international level. Lastly, this is a cross-sectional analysis of MLS. A longitudinal study would be more definitive, and warrant the conclusion you make at the end. Otherwise, you’ve simply looked at one point in time in the MLS universe and identified a potential trend that demonstrates possession is not correlated with winning.

    Interesting nonetheless.


  6. Posted by Tim on 2011/10/17 at 4:35 PM

    Very interesting article, but it doesn’t take into account that once a team is ahead they typically play more defensively in order to preserve the win. Playing defensively usually means less possession.


    • Posted by matthewsf on 2011/10/17 at 4:39 PM

      A fair point, although what I think would be interesting would be to see if that is in fact true.

      I would say–again, non analytically–that team’s don’t really bunker in the 1st half, so that’s 45 minutes got there. But then you also need to account for if the team came from behind. Say a team is losing 2-0 and comes back to win 3-2, they should have the possession advantage still, right?

      Lots of variables to account for.


    • Posted by Ufficio on 2011/10/17 at 5:45 PM

      I would recommend rereading the fifth- and fourth-to-last paragraphs.


  7. Posted by Jimmy on 2011/10/17 at 7:30 PM

    What it shows is that counter-attacking is more effective against MLS defenses. The kind of extended possession one sees in MLS can lead to bad giveaways and counters as often as goals (if not more so).


  8. It would be interesting to see if this holds true across leagues. MLS is a league with a more even distribution of talent with a generally Northern Europe/English influence (with more of a Latin blend at times). With the salary budget, there may be a lack of elite finishers when compared to higher budget leagues.

    The American soccer infrastructure is fairly young— there are only a handful of “coaching trees” within the league. A good chunk of the league either played for or assisted for Bruce Arena/Bob Bradley.


    • Posted by Jared on 2011/10/18 at 3:11 AM

      Bunker Bob and Bruce definitely have had a major affect on the style of MLS. The question with that though is whether they play that way because of the talent or would they do that anyways? As we’re seeing with Klinsmann, it’s not easy to play the possession game if you don’t have talented playmakers.


  9. […] Guardian runs some numbers and asks: Is there a correlation between possession and winning in […]


  10. Posted by Jim on 2011/10/18 at 7:28 AM

    Possession as an indicator of winning depends on playing style (e.g., Barca). For teams like Stoke City or Chivas USA, possession isn’t their style. Doing regression analysis on numerous variables where data is available can be done fairly easily (assuming you know stats and have Excel or other analytical tool).

    Correlation is where I think this sort of analysis makes sense. That is, how closely does one variable (say possession) correlate (influence) another (such as scoring). You can do that across multiple variables (multivariate analysis).

    Unlike American football or baseball, soccer is free flowing as opposed to set play after set play and exacting patterns. Players can interchange positions, are all over the pitch, people coming at you all the time, switch back and forth between offense and defense constantly . . . it’s organized chaos.

    As such, difficult to measure as little nuiances that arent’ measured – like “understanding” among players, eye contact, body positioning, timing of run, clever pass, and so on . . . . influence scoring and winning. Remember this mental ability in soccer has to be done at speed, little time on the ball, people coming at you full speed, dynamics of the specific play are continually changing as people are constantly moving, and ability to adjust to imprecisions of passing or crosses. Difficult stuff!!! And that’s why I love football/futbol/soccer soooooo much! American football and baseball are mind numbing. Like watching Napoleon putting his army through troop movement: slow, methodical, and boring.

    In the end, soccer is too complex to analyze statistically to a deep level. Collectively, factors such as ball possession (percentage), passing accuracy, number of passes, and shots on goal may be biggest indicators. But of course, there are occasional antidotes: such as when Inter beat Barca in Champions League.


  11. Posted by Steve H on 2011/10/18 at 9:58 AM

    Would it be possible to run the numbers based on the score? In other words, for the top four teams, how is their possession like when the game is tied? If they have more possession, are they more likely to score? Less likely? Any correlation at all?

    Similarly, if a team is down or up by a goal, is there any correlation between percentage of possession and likelihood of scoring next?


  12. Posted by Jim on 2011/10/18 at 12:34 PM

    Steve H,
    The answer is yes. As long as there’s data, regression can be performed between any two or more variables. Whether they are correlated or not will also come out in the analysis. In the end, you could determine which variables had most impact (correlation) to winning or scoring.


  13. Posted by Damon on 2011/10/18 at 1:21 PM

    I think it would be more interesting to look at what the possession stats were up until the first goal of the game.


  14. Posted by chad on 2011/10/18 at 2:41 PM

    I often hear that stats are of less use in analysing soccer, since the game is so free flowing. No doubt. But why aren’t there more set plays in soccer (even when not in a dead ball situation)? Isn’t the flip side of playing against an organised defense that you should be able to anticipate where they will be playing when the ball is in certain positions? Anybody out there played high level soccer? Are set pieces really the only time that teams run plays? On a somewhat similar note, it seems to me that very few of the US Nats have a good understanding with each other no matter how talented they are. When I watched Italy in 2006, the forwards knew when to start their runs b/c they could anticipate that over the top 9-iron chip from Pirlo, even if Pirlo was facing the sideline and not making eye contact. Other than Jermaine Jones to Jozy in Chicago (I think), we’ve missed that sort of quick hitting strike based on 2 players understanding each other…. almost like a set play.


  15. Advantage in possession needs to be read in conjunction with territorial advantage – having lots of possession with a large territorial advantage equals attacking pressure. Having possession in your own half does not. So time of possession in the attacking third might be a better measure. Would be interesting to look at possession and territorial advantage in smaller slices of the game, such as the 10 minutes before a goal is scored.

    Also gets at the types of goals scored in MLS. MLS has few players that have the skill set to unpick a tightly packed defense, so a lot of goals are scored just after possession is exchanged – that would be something to look at too – the average time after change of possession before a goal is scored.


  16. […] Soccer Statistically and Chimu Solutions (here and here) have looked at data from the MLS, and found that teams are actually more successful in matches […]


  17. […] Shin Guardian: More Possession, More Winning? […]


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: