Category: Women’s World Cup 2015

Gluck: What adds more value? Goal Scored or Goal Prevented?

With soccer statistical analysis growing daily, a longer headline might be: 

What do the tea leaves show about team performance measurements in Major League Soccer?  Does the goal prevented show greater value, relative to points earned, than the goal scored?

Even that’s a bit wordy though… maybe it’s…

Soccer Statistics:  What does “right” look like now?

If you read The Numbers Game: Why Everything You Know About Soccer Is Wrong – July 30, 2013 by Chris Anderson (Author), David Sally

There is a section called “On the Pitch, which explains how the game is a balance of strategies.  Preventing a goal is more important to earning points than scoring one, the game is about managing turnovers, and the game can be controlled by both tiki taka as well as keeping the ball out of play longer than the average team does.”  Sourced from this article written here https://www.forbes.com/sites/zachslaton/2013/07/30/everything-we-know-about-soccer-is-wrong/#686a7ab47831

My analysis shows:

Goals scored have more value (relative to points earned) than goals prevented.

Furthermore, I don’t just see the game as a balance of strategies, I see it as a balance of team statistics driven by team operations, strategies and tactics.

In the last four years the balance between how well a team attacks, versus how well the opponent attacks against that team, has more value (relative to points earned) than simply goals scored or prevented.

Finally, what shows as a valuable (balanced) team performance measurement for one team does not hold true as a valuable (balanced) team performance measurement for all teams; either home or away.

Composite Possession with Purpose (CPWP) Indices:

The CPWP index is generated by subtracting team attacking statistics (APWP) from opponent team attacking statistics (DPWP).  This is my way of ensuring I capture a teams’ balanced performance (with and without the ball).

Intimate details on my PWP formulas can be seen in my academic paper published with the International Research Science and Soccer II, published in 2016.   “Possession with Purpose: A Data-Driven Approach to Evaluating Team Effectiveness in Attack and Defense C. Gluck and T. Favero”.

Breaking News:  An abstract on the use of Possession with Purpose Index as a tool for predicting team standings in Professional Soccer has just been approved for presentation (as a poster) at the World Conference on Science and Soccer – Rennes, France 2017.

General information and other relevant articles published, stemming from my research include:

Over the last four years I’ve measured these leagues/competitions using PWP analysis:

  • Major League Soccer 2013, 2014, 2015, 2016,
  • English Premier League 2014,
  • Bundesliga 2014,
  • La Liga 2014,
  • UEFA Champions League 2014,
  • Men’s World Cup 2014, and
  • Women’s World Cup 2015.
  • The lowest correlation this index has had, to the league table, was in MLS 2016 (.75).  The highest correlation this index has was for the EPL and La Liga of 2014 (.94).
  • I’d put the lower correlation in MLS 2016 down to increased parity across the league, but I’ll leave how my index can be used to measure parity, in a league, for another day.

In this analysis I’ve evaluated 18 MLS teams that have played 34 (17 home and away) games in each of the last three years (2014, 2015, and 2016).  This equates to 1003 games of data or 2006 total game events for home and away teams.

My analysis excludes New York City FC, Orlando City FC, Chivas USA, Minnesota United FC, and Atlanta United FC as these teams have not played 34 games in each of the last three years.

Data will be presented in three separate categories, total games, away games, and home games.

In addition to evaluating team performance using my standard PWP Indices I have added three additional families of indices to my analyses.  They are:

  • Composite Possession with Purpose Indices Enhanced with Crossing Accuracy (CPWP-CR),
  • Composite Possession with Purpose Indices Enhanced with Clearances (CPWP-CL),
  • Composite Possession with Purpose Indices Enhanced with Crossing Accuracy and Clearances (CPWP- CR/CL), and
  • My benchmark for passing the common sense ‘giggle check’ is, as always, Goal Differential.

Data arrays:

Total Games

Total game observations for consideration:

In every instance goal differential had the strongest correlation to points earned in the league table.

In every instance a CPWP index had the second and third highest correlation to points earned in the league table.

Best, in order of frequency for correlation to points earned, is provided below:

  • Goal Differential – 18 times 1st *benchmark
  • CPWP Index – 14 times 2nd or tied for 2nd
  • CPWP-CL Index – 6 times 2nd or tied for 2nd
  • CPWP-CR Index – 3 times 2nd or tied for 2nd
  • CPWP-CRCL Index – 3 times 2nd or tied for 2nd

Teams not fitting the norm (PWP Index solely being 2nd best) were: Colorado Rapids, Columbus Crew, LA Galaxy, Montreal Impact, New England Revolution, Portland Timbers, Real Salt Lake, San Jose Earthquakes, Sporting Kansas City, Seattle Sounders, and Toronto FC.

When viewing the DPWP, seven teams showed stronger correlations to points earned (preventing the opponent from scoring goals).  They were:  Chicago Fire, Colorado Rapids, Houston Dynamo, Montreal Impact, New York Red Bulls, Philadelphia Union, and Sporting Kansas City.

Meaning 11 teams showed the APWP indices as having higher correlation to points earned; i.e. scoring goals was more important than preventing goals scored.

Away Games

Away game observations for consideration:

In every instance, but one, goal differential had the strongest correlation to points earned in the league table.  The outlying team, where goal differential was not the best correlation to points earned, was Colorado Rapids.

  • I think this exception is worth noting.
  • For me, goal differentials’ correlation to points earned has been THE benchmark in determining whether or not my team performance indices ‘make sense’.
  • Exceeding the benchmark, even once, confirms for me as a soccer analyst, that my approach adds value when looking for ways to help explain the game better.

In every instance a CPWP index had the second and third highest correlation to points earned in the league table.

Best, in order of frequency for correlation to points earned, is provided below:

  • Goal Differential – 17 times 1st *benchmark
  • CPWP Index – 9 times 2nd
  • CPWP-CL Index – 7 times 2nd
  • CPWP-CR Index – 2 times 2nd or tied for 2nd
  • CPWP-CRCL Index – 1 time 2nd or tied for 2nd

Teams not fitting the norm (PWP Index solely being 2nd best) were: Colorado Rapids, Columbus Crew, Chicago Fire, FC Dallas, Houston Dynamo, Montreal Impact, New England Revolution, Portland Timbers, Real Salt Lake, San Jose Earthquakes, and Toronto FC.

When viewing the DPWP indices, ten teams showed stronger correlations to points earned (preventing the opponent from scoring goals).  They were: Columbus Crew, Chicago Fire, Colorado Rapids, FC Dallas, Houston Dynamo, Montreal Impact, New York Red Bulls, Portland Timbers, Philadelphia Union, and Toronto FC.

Meaning ten teams showed the APWP indices as having a higher correlation to points earned; i.e. scoring goals was just as important as preventing goals scored.

Home Games

Home game observations for consideration:

In every instance goal differential had the strongest correlation to points earned in the league table.

In every instance a CPWP index had the second and third highest correlation to points earned in the league table.

Best, in order of frequency for correlation to points earned, is provided below::

  • Goal Differential – 17 times 1st *benchmark
  • CPWP Index – 10 times 2nd
  • CPWP-CL Index – 7 times 2nd
  • CPWP-CR Index – 2 times 2nd or tied for 2nd
  • CPWP-CRCL Index – 1 time 2nd or tied for 2nd

Teams not fitting the norm (PWP Index solely being 2nd best) were: Colorado Rapids, Columbus Crew, Chicago Fire, FC Dallas, Houston Dynamo, Montreal Impact, New England Revolution, Portland Timbers, Real Salt Lake, San Jose Earthquakes, and Toronto FC.

When viewing the DPWP indices six teams showed stronger correlations to points earned (preventing the opponent from scoring goals).  They were: Chicago Fire, Colorado Rapids, Montreal Impact, Philadelphia Union, Sporting Kansas City, and Vancouver Whitecaps.

Meaning 12 teams showed the APWP indices as having a higher correlation to points earned; i.e. scoring goals was more important than preventing goals scored.

Summary:

The CPWP indices are not perfect but they do show very strong, consistent, correlation to points earned in the league table.

In every instance the balance of a teams’ success in possession, passing accuracy, penetration, shot creation, shots taken, shots on goal, and goals scored AND preventing the opponent from doing the same, exceeds either APWP (scoring goals) or DPWP (preventing goals scored).

The same CPWP index was not the best CPWP index for every team relative to points earned in the league table.

Teams playing in away games had different CPWP indices (showing greater correlations to points earned) than games played at home.

The DPWP indices did not, consistently, have a greater correlation to points earned than the APWP indices.

Colorado, Columbus, Montreal, New England, Portland, Real Salt Lake, San Jose, and Toronto consistently showed CPWP-CR and CL indices had greater correlation than the standard CPWP index.

Correlation of all indices, to points earned, differed between home and away games.

Final correlations to points earned for all teams measured (combined) the last three years in MLS were:

  • Goal differential =  .87
  • APWP   = .53 // DPWP = -.51 // CPWP = .74
  • APWP-CR = .52 // DPWP-CR = -.50 // CPWP-CR = .72
  • APWP-CL = .49 // DPWP-CL = -.46 // CPWP-CL = .66
  • APWP-CR/CL = .49 // DPWP-CR/CL = -.47 // CPWP-CR/CL = .66
  • Goals Scored = .63

Conclusions:

The balance of attacking, versus stopping the attack of the opponent, has more value in measuring team performance (relative to points earned in the league table) than goals scored or prevented.

Goals scored, on average, (APWP) have more value (relative to points earned in the league table) than goals prevented (DPWP).

The correlation of team measurements, relative to points earned, varies from team to team, both home and away.

Therefore the value of individual player statistics (used to create those team statistics) varies from player to player, both home and away..

For example:  The CPWP-CR and CPWP-CL indices showed 2nd best for correlation to points earned for Colorado, Columbus, Chicago, FC Dallas, Houston, Montreal. New England, Portland, Real Salt Lake, San Jose, and Toronto (in away or home games) over the last three years:

Therefore, the players who play on those teams should have their individual statistics (for crosses and/or clearances) weighted differently than players who play on the other teams; because the value of their successful crosses/clearances had greater weight relative to those teams earning points.

Last but not least, what the other leagues/competitions offered after one season/competition:

  • EPL // APWP = .92 // DPWP = -.88 // CPWP =.94
  • La Liga // APWP = .93 // DPWP = -.90 // CPWP = .94
  • UEFA Champions League // APWP = .74 // DPWP = -.66 // CPWP = .81
  • Bundesliga // APWP =.89 // DPWP = -.84 // CPWP = .93
  • Men’s World Cup 2014 // APWP = .58 // DPWP =-.77 // CPWP = .76
  • Women’s World Cup 2015 // APWP = .63 // DPWP = -.77 // CPWP = .76

Both the Men’s and Women’s World Cup competitions saw the value of the goal prevented greater than the goal scored.  In all other instances the balance between the two showed greater correlation.

Anderson and Sally weren’t wrong at all; it’s more about what right looks like depending on what league/competition is being evaluated.

Best, Chris

You can follow me on twitter @chrisgluckpwp

COPYRIGHT: All Rights Reserved.  PWP Trademark

NOTE:  All the data used in my analysis is publicly available with the exception of the Women’s World Cup 2015 data; my thanks to OPTA for providing me that data last year.

Gluck: Updated Possession with Purpose and the New Total Soccer Index

Much has transpired in the world of soccer statistics over the past four years since I first published: Possession with Purpose – An Introduction and some Explanations.

CLICK this link for my NEW simplified power point presentation update of Possession with Purpose the Total Soccer Index

  • The .pdf version should make it easier to print and use as reference material.

Within you’ll find:

  • Definition of TSI
  • Purpose of TSI
  • Premise of TSI
  • Parts of TSI
  • Leagues / competitions analyzed
  • Application of TSI and its parts
  • The data for leagues / competitions analyzed
  • Observations & conclusions by league / competition as well as reviewing TSI across leagues / competitions

My thanks to all for your support and kind words throughout the years.

In Summary:

  • The sum of the parts has greater correlation to points earned than the parts independent of each other.
  • Player A, from Team A, within any given league, has a different correlation to points (performance/outcome) than Player B, Team B, Player C Team C, etc in that same league.  In other words outcomes of individual player statistical analyses are NOT EQUAL from team to team and league to league.
      • Said differently, clearances or crosses (used as a measurement in fantasy soccer) for one player, on one team, DO NOT have the same weight/value of clearances or crosses for a different player on a different team.
      • Same can be said for passes or shots taken, etc.
      • Therefore, Calculations such as Expected Goals are not an apples to apples comparison between teams within the same league.  Yes, it’s a predictive tool, but flawed/
  • The lower the overall correlation of the Total Soccer Index to points earned the greater the parity within the league or competition; this also intuits those are less predictable.

Best, Chris

@ChrisGluck1

Possession with Purpose – Prozone – and more…

No detailed statistics today – just a narrative to pass on a few tidbits as I prepare my End of Season analysis for Europe.

The news:

The European Season is ending.

  • There’s the winners, the losers, and those that stay afloat to live another year.
  • I’ll peel back the results on the English Premier League, Bundesliga, La Liga, and UEFA Champions League in the next few weeks.
  • For now, in La Liga the PWP Composite Index has a .94 correlation coefficient (r) to points earned in the league table; the Bundesliga sits at .92, the English Premier League sits at .94, while the UEFA Champions League sits at .87.
  • All incredibly strong and far stronger than MLS (.61) this year; last year MLS finished at .87.
  • Speaking of MLS, does a league, where winners display more characteristics of counterattacking, versus just possession-based attacking, detract from predictability?
  • In other words does the lower correlation support a League’s ability to achieve “parity” in professional soccer?
  • If so, is that style/type of football attractive enough to continue to grow footy in the States?
  • If not – does that mean the business model currently set up in the States won’t ever achieve a league “status” that matches the “prestige” most seem to attach to the top leagues in Europe?
  • More to follow…

I think these two video presentations by Hector Ruiz and Paul Power, from Prozone, are worth listening to.

  • In this video (tactical profiling) Hector, who attended my presentation at the World Conference on Science and Soccer last year, talks about his latest efforts that include breaking down the different types of possession in a much greater detail than I ever could with public data.
  • Of note is Hector substantiates my finding that a Head Coach’s tactical approach can be differentiated through tracking possession (passing characteristics) on the pitch.
  • He also helps begin to solve the riddle on measuring which players perform better or worse given those different styles of possession.
  • A soap-box, for me, when looking at my article on ‘Moneyball relative to soccer’, is the inability of modern day soccer statistics to show real value on how well teammates actually influence an individual’s success or failure on the pitch relative to how the team actually plays (what style it works to).
  • Here’s a direct lift from my article referenced above…

Modern day soccer statistics, for the most part, don’t measure the appropriate level of influence teammates, opposing players, and Head Coaching tactics – as such when I say I’m not a Moneyball guy when it comes to soccer it really means I don’t buy all that crap about tackles, clearances, goals scored, etc…

I value players relative to team outputs and I strongly feel and think the more media and supporters who understand this about soccer the less frustration they will (have) in blaming or praising one individual player over another player.

  • In the next video (game intelligence) Paul takes a similar approach in analyzing team behavior like PWP – separating out defensive characteristics from attacking characteristics while also modeling a ‘defensive press’ that measures success or failure in passing based upon whether or not a defender is hindering the attacker.
  • This topic has been one that I have also touched on last year – here’s a direct quote from my article on Hurried Passes.

So what is missing from the generic soccer statistical community to account for the void in Unsuccessful Passes?  Is it another statistic like Tackles Won, Duals Won, Blocked Shots, or Recoveries?

I don’t think so – none of them generated a marked increase in the overall correlation of those three activities already identified.  I think (it) is the physical and spatial pressure applied by the defenders as they work man to man and zone defending efforts.

  • Likewise, Paul also touches on ‘passing vision’ (in my words it’s not the innate vision many of us think of for players) – it’s more a discussion and analyses (I think) on the ‘windows of passing lanes’ available to players and whether or not they have tendencies to play riskier passes versus safer passes in relation to what the defenders are doing.
  • For me this simply means Paul has taken the same defensive pressure data and flipped it to view the success or failure of a player to find another player to pass to or create a shot given the defensive pressure (lanes/vision) that are blocked or open.
  • In simplistic teams (with new event statistics) you can capture and intuit that success or failure by filtering passes as being ‘open or hindered’ and also apply that same filter to create ‘open or hindered’ shots.  My article on this approach was also published some time ago – New Statistics in Soccer (Open Pass and Open Shot)
  • Finally, Paul also speaks to a game of soccer resembling the behavior of a school of fish; I’m not sure I’m convinced that is the best analogy – especially when he talks about under-loading and overloading, but his view does closely resemble mine where the game of soccer perhaps is best represented by a single-cell Amoeba.

All told – two well crafted presentations that begin to open up and really reinforce some of my soap-box issues with soccer statistics since starting my research three years ago.

To be redundant – soccer is not just about scoring goals – there is more to the game than goals scored; these two presentations continue to support my view that the world of soccer statistics needs to continuously get better…

My back-yard / stubby pencil approach to team performance analysis is soon to be published through Rand.

  • I want to express my sincere thanks to Terry Favero – my Co-Author – who helped me navigate the challenging waters of writing an Academic Paper.
  • Terry added considerable value, as well, in researching other works to help set the stage on the differences of PWP versus other efforts developed and published across the globe.
  • Finally, Terry provided superb editorial support – a challenge in that the writing styles one normally sees in a blog are completely unacceptable when writing an Academic Paper.
  • Great fun and the first of at least two to three more.

Last but not least, the Women’s World Cup is beginning.

  • Last year I applied the principles of PWP to the Men’s World Cup – with good order.
  • I’ll refresh everyone on how that took shape and then begin to chart how PWP takes shape for the Women’s World Cup.
  • I wonder what, if any, differences will show in comparing the women’s game to the men’s game?
  • Will the data show the same trends in quality and quantity?
  • Or will we see a reduction in quantity that may end up driving an increase in quality?
  • More to follow.

Best, Chris

COPYRIGHT – All Rights Reserved.  PWP – Trademark