Possession with Purpose: an introduction and some explanations

Please welcome to our little team of analysists and helpers Chris Gluck, whose PWP is going to be added to the metrics table this year—a solid instrument in telling us how teams have performed in turning possession into goals. Currently it’s one of the best open-source metrics out there to tell us such things. I hope you’ll enjoy his contributions as much I will – Harrison

First things first — my thanks to Harrison Crow and Matthias Kullowatz for the opportunity to post my Possession with Purpose Introduction on American Soccer Analysis.

If you’ve been following me this past year through Columbian Newspaper–out of southern Washington–you’ll know that I’ve been researching statistics in Major League Soccer. My intent has been to develop a simplified (Strategic) set of team performance indicators that may help others better understand soccer and how the outcome of a game may be better understood based on the primary inputs to the game.

Data for my research comes from documenting and analyzing all 646 MLS Regular Season games in 2013; the source data originates with OPTA and is displayed on the MLS Chalkboard and the MLS Statistics Sheet found through www.mlssoccer.com.

With that here’s my introduction on Possession with Purpose…

To first understand the context, I offer that this is one of the End States of my effort: create a simplified approach and documented method for measuring team performance where the output is an Index that (while excluding points) comes close to matching results in the MLS League Table.

Beginning with that End State in mind here is the End State product:


Observations from the Diagram…

Note that 9 of the top 10 teams in this Index made the MLS Playoffs last year with the Houston Dynamo finishing 12th in the Index.

For comparison, in benchmarking whoscored.com their Index only had 8 of their top 10 teams make the Playoffs, while http://www.squawka .com matched my 90% success rating, but the team they missed in the top 10 (New England) finished 16th in their Index.

From a strategic standpoint, the End State objective has been met; create a simplified approach and documented method for measuring team performance where the output is an Index that (while excluding points) comes close to matching results in the MLS League Table.

Defining the PWP Attacking and Defending Processes…

Here are the six steps in the PWP Strategic Attacking Process:

  1. Gain possession of the ball,
  2. Retain possession and move the ball,
  3. Penetrate & create goal scoring opportunities,
  4. Take shots when provided goal scoring opportunities,
  5. Put those shots taken on goal,
  6. Score the goal.

Here are the six steps in the PWP Strategic Defending Process:

  1. Minimize opponent gaining possession of the ball,
  2. Minimize opponent retaining possession and moving the ball,
  3. Minimize opponent penetrating and creating goal scoring opportunities,
  4. Minimize opponent taking shots when provided goal scoring opportunities,
  5. Minimize opponent putting those shots on goal,
  6. Minimize opponent scoring the goal.

Every step is this process has an average success rate (percentage) based upon data gathered from all 646 MLS Regular Season games.

Understanding the context of these steps versus other conditions and activities that influence the outcome of a game…

In case you missed it I call these Processes and the Indices “Strategic” to separate their value/meaning relative to other things that can influence the outcome of a game.

For me I have two other ways to classify information that can influence the outcomes in those steps. I have Operational conditions and Tactical metrics; provided below are some examples of each:

  • Operational conditions: Scheme of maneuver a team uses in setting up their system, such as flat-back four, flat-back three, double-pivot midfield, single-pivot midfield, bunkering with counterattacking, pressing high, direct attacking, possession-oriented attacking, weather conditions, location of the game (home/away), conference foe, non-conference foe, etc…
  • Tactical metrics: Locations of shots taken, shots on goal, and goals scored; penalty kicks, free kicks, crosses, headers won/lost, tackles won/lost, interceptions, clearances, blocked crosses, blocked shots, etc.

The diagram below shows the PWP Strategic Attacking Process with the average percentage of success rate in MLS for 2013. A more detailed explanation of each step is provided below the diagram.


Step 1: Gain possession of the ball: The intent behind this basic step should be clear; you can’t win the game if you don’t possess the ball to some extent. A second consideration about this step is that the more you possess the ball the less your opponent possesses the ball.

  • From a defensive standpoint there are any number of ways a team can work to gain possession of the ball; they include, but are not limited to, tackling, intercepting, clearing the ball, winning fifty-fifty duels on the ground or in the air, or simply gathering a loose ball based upon a deflection or bad pass.
  • For this Process the measurement of success is the percentage of possession a team has in a given game; note that in Soccer, the primary method for measuring possession is to add up the number of passes made in a game and divide into that the amount of passes one team makes (create a ratio percentage of possession); the opposing team has the difference between 100% and the percentage of possession that the other team has.
  • It’s not perfect but it provides a simplified ratio to compare one team versus another…

Step 2: Retain possession and move the ball: It shouldn’t be a secret to many that in most cases the team possessing the ball will need to move the ball in order to penetrate the opponents Defending Third and score a goal.

  • This is not to say a team has a minimum number of passes they need to complete to score a goal; for teams winning possession deep in the opponents Defending Third there may be times where the only thing needed is a quick shot on goal.
  • By and large, however, most teams – when they gain possession of the ball – do so in their own Defending Third and then move the ball (eventually forward) in a position where a teammate can create a goal scoring opportunity for another team member to take a shot.
  • For this process, the measurement of success is the team’s passing accuracy percentage across the entire pitch; passes completed divided into passes attempted.
  • It’s not perfect, but it provides a simplified ratio to compare one team versus another; statistically speaking there are weaknesses in how this percentage is measured by the big data folks.
    • Throw-ins, for example, move the ball across the pitch from one player to another yet they are not officially counted as passes.
    • Successful crosses are also not counted as a successful pass even though the ball moves successfully from one player to another.
    • Oddly enough, when evaluating the data provided on the MLS chalkboard, an Unsuccessful cross is included as a Pass attempted (?!)
    • For the purposes of this analysis I had to count all successful crosses as successful passes; therefore my final pass completions totals will be slightly higher than what Opta provides. It is what it is…
  • I should also point out here that there are occasions when a team wins possession of the ball and takes a shot where no pass was completed. Like I said, this measurement method is not perfect but it is ‘equal’ in ignoring that exception for all teams.
  • Therefore the measurement itself has value in tracking the majority (bell curve) of activities that normally occur in a game of soccer. And as a reminder, these are Strategic steps in PWP; by definition a Strategic step will not measure to a level of granularity; that is where Tactical metrics come into play based upon an Operational condition where the team is applying pressure higher up the pitch.

Step 3: Penetrate and create goal scoring opportunities: Most know that a pitch is divided into three parts; the Defending Third, Middle Third, and Attacking Third. For the purposes of this effort, Penetration is associated with entering the opponent’s Defending Third with the intent to score.

  • For this Process, penetration is measured by dividing the volume of passes a team completes within the opponent’s Defending Third into the volume of passes a team completes across the entire pitch.
  • It’s not perfect but it creates a ratio that treats all teams fairly, and given the overall accuracy of the End State Index (90%), it’s a reasonable way to measure this step.
  • In order to measure this step I first had to manually filter, for all 646 games, every pass attempted and completed using the MLS Chalkboard; my thanks to MLS and OPTA for providing us ‘stats’ guys the opportunity to do that. With Golazo stats now available, that task will be easier next year. As a stats guy, it would have been inappropriate to switch measurement methods ¾’s of the way through the year.

Step 4: Take shots when provided goal scoring opportunities: This is, by far, the hardest indicator to measure, given how current data sites really lack granularity in how they identify/define ‘created goal scoring opportunities.’

  • I define a ‘created goal scoring opportunity’ as any pass, successful or not, that may have ended with another teammate taking a shot. That’s hard to quantify, but an example, if you will:
    • A fullback overlapping down the right side puts in a wicked cross that gets cleared at the last minute by a center-back, with his head. With OPTA and other data companies that wicked cross, though unsuccessful, is not quantified as a goal scoring opportunity created; it’s merely tracked as a clearance and an unsuccessful pass.
    • I disagree; the fullback did their job in putting in that wicked cross – what really happened is the defender also did their job in clearing it – therefore a “potential” for the attacking team to complete a created goal scoring opportuinty and take a shot was denied.
    • Both the attacking team and defending team should be statistically credited for doing what they are expected to do. Others may disagree…
    • But as a Head Coach, I would put to memory that the fullback did what was supposed to happen; create the chance – therefore in my books that player created a goal scoring opportunity.
  • For this Process, the step is measured by counting the number of Shots Taken compared to the number of completed passes in the opponent’s Defending Third.
  • It’s not perfect, but it’s measured in an unbiased manner for every team, and there will be instances where a shot can be taken without a completed pass or originate from a defensive error.
  • In going back to the example, as a Head Coach I would call that effort a “failed assist.” I think there is value in knowing the number of “failed assists” as much as there is in knowing “assists.”
  • By tracking “failed assists” it provides a pure, statistical way, to track individual player performance (tactical metric) that can influence team performance.
  • Bottom line on this one, as contentious as it may be for some, recall the End State of this Final Index… create a simplified approach and documented method for measuring team performance where the output is an Index that (while excluding points) comes close to matching results in the MLS League Table.
  • Given the accuracy rating of 90% in matching the top 10 Playoff teams this year I feel and think the approach to measure this indicator works.
  • If OPTA, or another data compilation agency starts to track “failed assists”, could an Index like this reach 100% accuracy?

Step 5: Put those Shots Taken on Goal: For the most part this is an individual statistic that is added up to create a team performance indicator.

  • For this process, the step is measured by dividing the number of Shots on Goal by the number of Shots Taken.
  • It’s one of the easier indicators to measure, and if you watch any level of soccer, it’s pretty self explanatory – if the Shot can come anywhere within the dimensions of the Goal, it is considered a Shot on Goal. One of two things happens; it goes in or it doesn’t.

Step 6: Score the Goal: One critical objective of the game.

  • I say ‘one’ because indications, I see, lead me to offer that this game is not all about scoring goals.
  • In my research it appears to me that teams who defend better seem to take more points in games than teams that don’t defend very well.
  • A recent example in my End of Season analysis of Vancouver: in Western Conference competition, they scored 35 goals and gave up 35 goals; all told they took just 26 of 72 possible points – clearly, in this example, scoring goals did not result in wins…
  • Prozone, a noted professional sporting analysis company, offers the following in the article: “Using data from the last ten seasons of the Premier League, Anderson and Sally compared the value of a goal scored and the value of a goal conceded. They found that scoring a goal, on average, is worth slightly more than one point, whereas not conceding produces, on average, 2.5 points per match. Goals that don’t happen are more valuable than goals that do happen.”

In closing…

  • It’s not perfect, but it provides reasonable information in a reasonable format that has reasonable value when comparing the End State output to how the MLS League Table finished.
  • For those interested the PWP Strategic Attacking Index and Defending Index are provided below:



  • In looking at these two Indices, note the number on the left; the difference between the Index number in the Attacking Index and the Defending Index is the number that appears to the left in the Final Strategic Index at the beginning of this article.
  • That may help explain why some teams finished above zero, as opposed to below zero in the Final Index.
  • Teams finishing above zero had team attacking percentages that exceeded their team defending percentages; in other words they were better in their attack against the opponents than the opponent’s were in attacking them.
  • Team success rates in these six steps will be used next year to begin to analyze how well the team is performing as the new season starts compared to performance the previous year.

Follow Chris on twitter at https://twitter.com/ChrisGluckPTFC, and keep up with his PWP metric all season! 


27 thoughts on “Possession with Purpose: an introduction and some explanations

  1. I’m so impressed I want to cry. This is really incredible. I’ve been thinking of a similar “process” for the work I want to do but this blows it away. I will probably continue to work on my more simplified approach but know that I am defeated 🙂 I am surprised you don’t factor in the location of the shot as an important factor in creating value. More passes in the final third should lead to a better shot and therefore more goals. Also, if you look at the score of the game when the shot was taken you can get a proxy for defensive pressure being applied in the final 3rd. There’s a great post on 11tegen11.net that works through that. Could be something to add in the future. But this is awesome.

    • Thanks for your comments and great to hear you enjoyed it as much as you did. 🙂

      Your thoughts on location are important to consider – there are many folks out there that already do great work on analyzing shot locations and I think you may see some pretty cool stuff on this site in the next few months as well. For me, and this effort, Shot locations falls under Tactical Metrics so the location itself won’t appear as Strategic measurement. As the PWP effort is expanded the shot location will become a feeder into the larger success rate percentages. Not sure how well that reads but think of it as an onion – the whole onion are the 6 steps and within each of those steps there are layers that influence the outcome – sometimes tactical metrics like ‘tackles’ will influence more than one step – shot locations should influence the final three steps so they will be reviewed over time to see their influence in those six steps. I hope that helps?

      You are also correct about the score-line influencing activities on the pitch – for me, the score-line change is an Operational condition of the game when it occurs. In earlier work on PWP I noted that as the scoreline changed a number of activities were influenced and if down 1-nil; 2-nil with 10-15 minutes to go a team that is possession-oriented is likely to apply more direct attacking… this condition will slightly skew the data so it’s important to know when that happens. As I’m only one guy I don’t have the time to get into that level of detail yet :).

      Another great challenge for many of us stats guys is that the open-source data folks make it very hard for us to separate out simple stats given a score-line change; perhaps that will change in the near future? All the best and keep up your research! Chris

    • As Chris allueded to, ASA will be presenting a lot of really exciting and detailed information with shot locations and game states that will only further influence this type of work by people like Chris.

  2. Would “failed assists” be the same as Opta’s “key passes,” thereby making “chances created” the metric you are looking for in Step 4? That still wouldn’t cover times where the shot never went off, like in your example. Your formula of Shots Taken/Completed Passes in the Attacking Third is a measure of how direct that team is playing. I expected to see Chances Created somewhere in Step 3 or 4 and am wondering how you managed to avoid that.

    • Phil, Great question(s) and this may take a few minutes to answer – “failed assists” includes more than just Opta Key Passes – at least the key passes that have been offered up and defined in the MLS Chalkboard. I also believe Opta makes a judgment call ‘after’ a pass is made to determine whether or not it was ‘key’; I disagree with this judgmental decision – any pass into the box is ‘key’ – what makes it more or less key is how well the defenders versus attackers manage that pass… does that make sense?

      In a regular football environment, for me as a coach, any cross offered up into the 18 yard box is an opportunity for a teammate to take it and score from it – in counting key passes in Opta they are also not included in that definition of a key pass. And interestingly enough crosses that are successful are not even counted as ‘successful passes’ in Opta either. But as noted in the article unsucessful crosses are counted as ‘unsuccessful passes’.

      So without having detailed information on the ‘stream of passes working towards a shot taken’ the only strategic recourse is to look for a general relationship between passes made (successful and unsuccessful as appropriate) to shots taken. At least that way the approach is consistent for all the teams in MLS.

      ‘Chances created’ is not quite the same as ‘goal scoring opportunities created’; for me a ‘chance created’ is where the teammate has latched onto the ball and taken a shot – I am looking at more than just that – I also want to be able to glean how effective the defense was in preventing that GSO from becoming a ‘chance created’ – for me a ‘chance created’ would be derived from minusing out ‘defensive activities like defensive clearances.

      So in essence I didn’t manage to avoid that I simply can’t collect it at this time given the data that is available. But also note that my calculation includes ‘successful passes’ so by using that I am separating out defensive activities that result in an unsuccessful pass.

      If you read some of my earlier articles on possession with purpose, on my blog site, the first 15 – 20 games or so I did specifically collect that data for Portland – and there is clear evidence that possession, penetration, creation of goal scoring opportunities (as I’d like to define them), defensive clearances, shots taken, shots on goal and goals scored had a very strong inter-relationship with each other – my R2 for that curvelinear relationship was .987. The R2 at that time for the simple ‘goals scored’ hovered around .6000 or so – clearly reinforcing that the ‘collective analysis of multiple data points in a stream of possession’ outweigh the value of just one data point (goals scored).

      But that data was hand collected for every Timbers match and is not/was not available across all MLS teams – so I had to adjust to establish an equal playing field to analyze in order to create the Indices I had intended on creating.

      Hence the decision to work with shots taken versus completed passes. I do suggest that this adjusted approach is not perfect 🙂 as I statistically already know this given my previous (single focused) analysis on the Timbers.

      So in answer to your last consideration is chances created somewhere in step 3 or 4, yes – with my individual research on Portland last year step 3 is ‘physical penetrations’ (counts) and step 4 is ‘goal scoring opportunities created’ (counts) (inbetween there are the defensive activities/clearances that occur prior to the shot being taken).

      I also tracked what side (left right or middle) where the penetration occured and then (left right or middle) where the GSO occured and what type of GSO was used – cross, free kick, corner, throughball, switch, long ball from square 1, etc)

      That being said, and the lack of global data for all MLS teams using that criteria, I had to adjust in considering the overall output (the Index generated). Even in adjusting there remains a statistical relationship that can and does have relevance to an End State in comparison to the final league tables. 90% accuracy is on the same par as having an R2 of .9000.

      I would offer that if statistical analyses are to be leveraged to better understand teams then the statistical data collected and offered freely to the public needs to get better.

      Others besides myself, Ted Knutson, have noted that the Opta data could be a whole lot better… so I didn’t manage to avoid that approach – I simply couldn’t execute that approach outside of just analyzing the Timbers…

      I hope that answers your question and it may provide more than you want to know 🙂 Feel free to contact me anytime Phil – always love chatting about trying to get better, more relevant data, from Opta…

    • Phil, One other important consideration here is that ‘chances created’ and then ‘shots taken’ is geared more to a ‘tactical success’ rate of a particular player as opposed to the strategic team performance index that GSO’s look at – for example – there are superb stats out there on whose taking shots, putting them on target and then scoring – what is missing is the ‘team’ creation of those opportunities that can come from anywhere by anyone – even the goal keeper on a long ball (assist or failed assist) – to support that ball striker on a collective basis. Put another way – within this strategic approach there are ‘operational considerations’ that influence the game and then ‘tactical details’ that influence the game – how well an individual striker executes that tactical step falls up under the difference between GSOs created and those that are stopped by team defense (clearances, interceptions etc) as well as the individual ability of the striker… guess what I am trying to say is consider the phrase ‘chances created’ falls up under ‘tactical metrics’ that support or can be influenced by operational metrics that tuck up under ‘strategic indicators’ – does that help clarify?

  3. Pingback: Portland Timbers – 2014; Could it be six pieces of Silver? | Portland Timbers

  4. Pingback: Portland Timbers – 2014; Could it be six pieces of Silver? | Clogs

  5. Pingback: Seeing Red – Bulls draw 1-1 in Fiery Chicago | New York Sports Hub

  6. Pingback: MLS PWP: Team Performance Index through Week 4 | American Soccer Analysis

  7. Pingback: USMNT – My thoughts after 2-2 Draw with Mexico | American Soccer Analysis

  8. Pingback: Assessing the Impact of Montreal as Red Bulls draw 2-2 | New York Sports Hub

  9. Pingback: MLS Possession with Purpose – The best and worst of Week 5 | American Soccer Analysis

  10. Pingback: Possession with Purpose: My MLS thoughts and the Indices through Week 5 | American Soccer Analysis

  11. Pingback: MLS PWP through 6 Weeks: Does the wheat begin to separate from the chaff? | American Soccer Analysis

  12. Pingback: Expected Wins… MLS Soccer | American Soccer Analysis

  13. Pingback: Real Salt Lake Slides Past Timbers 1-Nil | Soccer Yanks | Chris Gluck

  14. Pingback: Cumulative PWP: Four-Horse race to the Shield? | American Soccer Analysis

  15. Pingback: PWP – Who’s best and Worst after 8 Weeks | American Soccer Analysis

  16. Pingback: PWP-Pick-List Week 10 – weaving Expected Wins into my predictions this week… | American Soccer Analysis

  17. Pingback: Red Bulls Dominate and Defeat Dynamo 4-nil in Dynamic Fashion… | New York Sports Hub

  18. Pingback: Dynamo Dynamic in Attack and Bulls Bullish on Defense – Week 9 Ends in MLS | American Soccer Analysis

  19. Pingback: PWP: Chicago lights up New York while Montreal feels the Impact of Sporting KC | American Soccer Analysis

  20. Pingback: Impact of Montreal on MLS | Statistical Analysis | Chris Gluck

  21. Pingback: Expected Wins #2 – After 184 MLS Events (92 Games) | American Soccer Analysis

  22. Pingback: You say you want a Revolution? Possession with Purpose From a Different Angle | American Soccer Analysis

  23. Pingback: Expected Wins 2 – After 184 ‘events’ (92 games) | Possession with Purpose

Leave a Reply

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

WordPress.com Logo

You are commenting using your WordPress.com 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