As you know, ResultsPage calculates various types of league tables on a round-by-round basis, one of them being the Pythagorean table. We’ve isolated the code that calculates the Pythagorean expectation and we now expose it to the outside world as its own API call.

In case you don’t know what I’m talking about, API is short for Application Programming Interface, and it is a collection of programming instructions that allow applications to talk to each other.  It provides a means to access and interact with data and information from multiple sources in an automated and creative way.

Programs interact with APIs in a variety of ways.  Some pass chunks of XML data back and forth.  We decided to design our API on REST principles.  The specifics aren’t that important — what is important is that you can access the Soccer Pythagorean through a URL address.

So here’s the Soccer Pythagorean API call:

http://fmrdlight.herokuapp.com/analytics/pythagorean?matches=XX&scored=XX&allowed=XX

All you have to do is add numbers where the XXs are for number of matches played, goals scored and allowed. That’s it.

What do you get in return?  Here’s an example:

http://fmrdlight.herokuapp.com/analytics/pythagorean?matches=24&scored=50&allowed=25

returns the following:

{
    "points": 47,
    "match": 24,
    "loss": 5,
    "allowed": 25,
    "win": 14,
    "draw": 5,
    "scored": 50
}

The expected point total is in the “points” field, while “win”, “draw”, and “loss” express on possible league record that would result in that point total (calculated from win/draw probabilities).

Now all of this is work in progress, so it’s likely that the URL root (fmrdlight.herokuapp.com) will change as the API matures.  For now, it’s an open API, so don’t be a jerk.  We’ll add some helper functions for those who don’t want to type a URL into a browser. We can create one in Python; maybe others can create their own.

So play around with the Pythagorean API. We hope this feature encourages its wider use in the football analytics community.

Bill Gerrard is someone who I’ve wanted to meet since I started Soccermetrics.  Prof. Gerrard is a pioneer in statistical performance analysis in sport and a leader in bringing an evidence-based approach to coaching in football, rugby league and rugby union.  Perhaps he’s better known in North America for his work with Billy Beane of the Oakland A’s, while in Britain he’s best known for his work with Leeds United and Saracens rugby.

Bill and I conducted the longest interview of this series yet one that covered a lot of areas in sports analytics.  I hope you enjoy it.

[Interview conducted 4 April 2013.]

(Howard) So Bill, tell us a couple of sentences about yourself.

(Bill) I’m a Scot living in Leeds, a big sports fan whose first love was the beautiful game but I follow most sports. I’ve always supported Celtic and, like many Scots of my generation, Leeds United is my English team because so many Scots including Billy Bremner played for the great Revie team. My day job is a university prof at Leeds University Business School. The rest of my time is spent working in sports analytics.

How did you get started in sport analytics?

I started in the mid-90s analysing the economics of the football transfer market. That academic research led naturally into the practical problem of how to value footballers. I developed a valuation algorithm for valuing footballers which I first applied for a stockbroker in late 1997 to analyse the efficiency of Tottenham Hotspur in their transfer dealings.

I noticed that one of your first consulting jobs in football was performing valuation for Leeds United. How did you get that job?

Initially most of the interest in my consultancy work was for financial institutions who were interested in putting a market value on a club’s playing assets as security for various debt instruments. I undertook a squad valuation for a company that provided debt finance for Leeds United. Later the club asked for an updated squad valuation which I subsequently discovered was used in the securitisation deal that raised £60 million.

As a Leeds United supporter, did that job give you a greater appreciation for how football clubs are managed? And if so, in what way?

Running a pro sports team must always be about combining passion and prudence. As a business school prof and a sports fan, I guess I was in a good position to understand both aspects of the sports business. And Leeds United provided a very sad lesson of what happens when passion over-rules prudence in the boardroom.

Do you think the rest of the football establishment has learned any lessons in the years since Leeds United’s relegation?

Leeds United was a victory of vanity over sanity and shows what happens when you chase “The Dream” with little regard for a sustainable business model. In three years Leeds United went from last four in the Champions League to relegation from the Premiership. And a decade on, Leeds United are still struggling to get back into the Premiership. It could have been very different if the senior management of the club had invested in the stadium to grow the revenues to support the wage costs of a higher quality playing squad. Arsenal have shown the right way to do things. Wenger versus Ridsdale – no contest when it comes to constructing a sustainable business model for a football club.

Let’s talk about Arsenal for a moment.  How have Wenger et al. gone about creating their business model and does there exist any tension between it and the desire to win trophies?

The major aspect of Arsenal’s business plan in recent years has been to build the Emirates Stadium This involved a very high level of debt financing but it was prudent given the cash flow projections. By increasing capacity from 37,000 at Highbury to 60,000 in their new stadium, Arsenal have more than doubled their matchday income. This has given them the financial resources to afford the top players. There has been some unrest amongst fans that Arsenal have not yet reinvested these resources into the playing squad but the resources are there for Wenger to compete in the transfer market. And I think we can see evidence of a greater willingness to spend big in the contract extension agreed with Walcott.

I’d like to ask a few questions about your work on competitive balance. One of your more recent findings was that even though North American sports leagues are more regulated than their European counterparts, the competitive balance as measured by “win dispersion” is better in the European leagues. Could you explain that?

It seems to me that competitive balance is quite a complex concept that has lots of different aspects to it. Win dispersion measures the variation in sporting performance between teams in a league. We would expect a lower degree of variation in win percentages between teams in leagues that are more competitively balanced. Given that North American leagues have tended to take a more proactive approach to managing competitive balance through, for example, salary caps and player drafts, you would expect lower win dispersion than in the less-regulated European football leagues. Surprisingly you tend to find win dispersion in lower in European football. One reason for this is because of tied games with around a quarter of football games tending to be tied.

What about measuring competitive balance by competitions won, which I imagine would be how most people think of competitive balance?

Win dispersion is only one aspect of competitive balance. The concentration of competition winners – prize dispersion – is another aspect. And yet another aspect of competitive balance is whether or not teams tend to have similar levels of sporting performance from year to year.

And how variable is that performance among teams in professional football?

When it comes to performance persistence (i.e. the year-to-year stability of team performance), you find that the leading European football leagues tend to have similar levels as the North American major leagues with one exception, the NFL which has a very low level of performance persistence. But this is no real surprise given that the NFL is probably the most proactive sports league in the world in terms of trying to ensure competitive balance.

There’s a lot of anxiety among football fans about competitive balance, especially when comparing who wins the Premier League now to who won the First Division in the 70s and 80s. But then I read papers by people like Peter Sloane who state that there were the same concerns about a two-tiered competition in the 60s! Are fans right to feel uneasy today, or has professional football always had issues with competitive balance?

My sense is that concerns about competitive balance have grown significantly in recent years across European football. And I would say the concerns are about the emergence of a three-tier not two-tier competitive structure. As well as the concerns about the super-rich top tier pulling away from other clubs in the top leagues, there are also concerns about the increasing gap between the top leagues and the rest of the football family. The latest TV deal will widen the gap between the Premiership and the rest of the English football pyramid making it more and more difficult for promoted clubs to survive in the Premiership. Whether or not UEFA’s soon-to-be implemented financial fair play rules will reverse the trends is questionable but at least there is acknowledgement of the problems and the need to act. But it’s a difficult tightrope to tread by the governing bodies because if they go too far in trying to regulate the super rich clubs they face a real threat of these clubs setting up on their own independent European Super League.

That’s a very good point.  I’ve always felt that UEFA might not want to push too hard on Financial Fair Play for similar reasons.  So how do you anticipate UEFA enforcing FFP in practice?

Perhaps when it comes to politics I’m too cynical but I wouldn’t be surprised if “temporary” exceptions were allowed to UEFA’s FFP if it is found that some of the biggest European clubs are going to be excluded from the Champions League. It will hurt UEFA financially if those clubs are excluded and it will intensify the threats of a breakaway European Super League.

You had a chapter on competitive balance and the sports media rights (The Economics of Sport and Media by Jeanrenaud and Késenne) where you wrote, “Unlike standard textbook industries, there is good reason to believe that rights market deregulation in the professional team sports industry may be against the consumer interest.” Has the experience of domestic competitions in Spain and Italy provided support to your point?

I am a strong believer in the collective selling of league media rights for a number of reasons, one of which is that teams are effectively monopoly suppliers to their own fans. Market deregulation creates rather than weakens market power in this case. I also think that collective selling of TV rights provides an effective revenue redistribution mechanism. The experiences of Spain and Italy show the problems created if you remove this redistribution mechanism.

It seems that football analytics is being asphyxiated by lack of data for outside analysts, lack of expertise from internal staff at football clubs, and lack of interest (if not strong cultural resistance) from the football community. Is that perception accurate?

We continue the conversation with Bill Gerrard in the upcoming edition of the Soccermetrics Newsletter.  Only subscribers will be able to access the full interview, so subscribe now.  You are a subscriber, right?

If you follow Aaron’s blog or his tweets, you will be inundated by an avalanche of match statistics from over 70,000 players representing over 10,000 clubs in over 60 leagues around the world.  It is, quite simply, a sports statistical masterpiece.  Sports statistics is Aaron’s life, and he’s been working in this field for close to 20 years.  His 20-year statistical almanac of the English Premier League is an example of his work.

Aaron and I have formed a mutual admiration society over the past two years, and we have worked together on a few projects.  Now we’ll be more closely joined as we integrate his massive data sources into our analytics infrastructure.  We’ll also make use of his experience with the various segments of the football industry to develop and deliver analytics-related products, services, and content of value to prospective customers.

Please welcome Aaron to the Soccermetrics team!

I don’t need to tell you that the Big Five European leagues are pretty much decided, and have been decided for a while.  English Premier League, Spanish Primera, and German 1.Bundesliga are dead, dead, dead.  Serie A and Ligue 1 are still alive, but Juventus have pulled away and PSG are starting to do the same.  There is one exception, though, and it’s outside the Big Five: the Dutch Eredivisie.  Make that two exceptions — Turkey’s Süper Lig is tightening up as well.

So what are we seeing in the Pythagorean records of the Big 5 leagues?  I think it can be distilled in a few observations:

• England and Spain should be closer than they are.  If you look at the difference in expected points between Barcelona and Manchester United and their closest challengers, only two or three points separate them from second place.  In fact, both sides are playing at 10 and 14 points, respectively, above their statistical expectations — numbers we have not seen in a league-winning side since FC Twente in 2010. Neither side’s defense has been outstanding by their own high standards; Manchester United has the third-best scoring defense in the Premier League while Barcelona ranks fourth.  The difference in United’s case is that a number of challengers aren’t playing at a high level this season, while Barcelona’s offensive exploits (and especially Lionel Messi’s) more than compensate for its defense.
• PSG are ahead in Ligue 1, pretty much as expected.  PSG have been a perennially underachieving side for several years but thanks to Qatari billions they have a serious chance of winning Ligue 1.  Their -3 Pythagorean residual is within the range of uncertainty of the Pythagorean expectation.
• Juventus and Bayern Munich are the truly dominant sides in domestic competition.  Not only are both sides leading by large margins (Juventus by nine points, and Bayern by 20), their performances are also in line with their statistical expectations.  Their low Pythagorean residuals are the result of their defensive records; Bayern’s residual was 0 or +1 for much of the season thanks to their streak of clean sheets and high scoring.
• Eredivisie is a three-way race, but it shouldn’t be.  The Dutch top flight is going to be a fight between the Big Three teams (PSV, Ajax, Feyenoord) for the title, but if you’re going by goal statistics and expectations, PSV should be well in front.  They aren’t because of a significant difference in home and away performance and a strong performance by Feyenoord.
• Goal difference is a proxy for expected points.  The nice thing about the ResultsPage app is that one can observe the relationship between goal difference and expected points quite easily.  In almost all cases, sorting on goal difference would yield a league table in the same order as the Pythagorean table.

So who are the overachievers at this point of the season, on the basis of Pythagorean expectation?  In addition to Manchester United and Barcelona, I would include Marseille (+10), Feyenoord (+8), Stuttgart (+8*), Hamburg (+7), Olympiacos (+7), Lazio (+6), Getafe (+8), and Rayo Vallecano (+7).  The asterisk is attached to Stuttgart because their expected point total would put them in the relegation zone yet they are performing well enough to be safe.  The same applies to Bastia (+5) in Ligue 1.

Who are the biggest underachievers at this point of the season?  They would be Liverpool (-7), Saint-Étienne (-6), Troyes (-7), Eskişehirspor (-7), Mersin (-7*), AZ Alkmaar (-7), Roda JC (-5*), and Orduspor (-5*).  I attach the asterisks to Roda, Mersin and Orduspor because their expected point total would place them in a safe position yet because of their underperformance they are in the relegation zone.

And of course, you can check all of this out yourself at ResultsPage, our league results and table application (and guinea pig).

CORRECTION: There was an error in the database in which the names of Sochaux and Saint-Étienne were swapped.  I’ve corrected the database and the text above.

Exec Summary: We perform a cluster analysis to identify the collections of talent that have played in Major League Soccer over the last six seasons.  We find that salary and minutes played are related to player performance, college draftees suffer from a lack of both and take longer to reach parity with peers who enter the league through other channels.  College draftees can be and have been successful in MLS, but the uneasy feelings that many club executives have toward them are not entirely misplaced.

—-

Major League Soccer (MLS) is a unique sports league within the scope of North American major sports leagues. It exhibits practices that are common among other sports leagues on the continent:

• such as a single permanent division of franchised teams in the USA and Canada,
• a salary cap with limited exceptions,
• a playoff system,
• and a draft allocation scheme.

Unlike the other North American sports leagues, it is not the dominant domestic competition for its sport in the world, and unlike the NFL and MLB and to some extent the NBA and NHL, it must coexist within a global network of domestic soccer leagues. The domestic soccer leagues in Europe, and especially the leagues in the “Big Five” countries (England, Spain, Italy, Germany, and France), strongly influence finances, practices, and the culture of the professional game.

MLS also operates in a soccer culture that is unique to other soccer-playing countries in that the local game has been dominated in recent decades by youth and college soccer players drawn from the middle to upper-middle classes. These two factors present implications for talent recruitment in MLS.

Since the league’s founding, MLS officials have had an uneven relationship with the college soccer system.  To be sure, MLS has relied on the college system as a source of its base players and coaches, such as Clint Dempsey, Brian McBride, Bruce Arena, Sigi Schmid, and Bob Bradley. Yet many in the league have expressed dissatisfaction with the overall quality of the talent coming out of the college ranks and the restrictions imposed on player development by the NCAA. At the Soccer Analytics session at the 2012 MIT Sloan Sports Analytics Conference, Seattle Sounders co-owner Drew Carey stated when discussing the use of data to predict player development, “I don’t think the college system is good enough for MLS.”

The objective of this work is to examine Carey’s assertion by asking the following: what are the contributions of college draftees to Major League Soccer and how do they compare to their counterparts who are either free agents or participants in Generation adidas?  We start by performing a cluster analysis of match, salary, and demographic data of the players between the 2007-2012 MLS seasons.  We use 2007 as a start date for two reasons: it was the first year that salary data was published on the MLS Players Union website, and it was also the first year of the Designated Player Rule. We add a variable that describes a player’s entry point into Major League Soccer:

A total of 2628 salary records and 2451 statistical records from 1149 players are incorporated into the dataset.  As we’ve stated in previous posts that used MLS Players Union salary data, these data aren’t definitive because of discrepancies with official League data, or because of players being released before salary surveys were sent out.  Seventy-three (73) players have no salary data and are excluded from the dataset.  Also, players in the MLS Pool (not contracted to a specific team but available on-call in case of lack of eligible players) are excluded from the dataset.

For those who care about the analysis details, we apply a k-means clustering algorithm to the dataset using the R programming language.  You can find more specifics about the routine at this link.

Goalkeeper Analysis

Ok, so now let’s get into the results. We identify five distinct groups among those goalkeepers who have played in MLS between 2007-2012, whose characteristics are presented in the table below.  The principal factors that differentiate the five groups are base salary, minutes played, matches played, and age.

If we take a slice of the statistical data across ages, we see significant differences between the Draftee and Generation adidas cohorts.   Generation adidas pays each player a salary that is large enough to compensate for leaving college early or bypassing it altogether.  In general, goalkeepers who enter the league via Generation adidas are paid more, play more minutes, and record more shutouts than those who enter via the college draft.  (The spike in minutes played and shutouts at age 21 is the result of Ryan Meara’s performance for the NY Red Bulls during the 2012 season.)

Averaged base salaries of the Draftee and Generation adidas cohorts at the goalkeeper position. Data for MLS regular season, 2007-2012.

Average minutes played of Draftee and Generation adidas cohorts at the goalkeeper position. Data from MLS regular season, 2007-2012.

Average shutouts of the Draftee and Generation adidas cohorts at the goalkeeper position. Data from MLS regular season, 2007-2012.

Field Player Analysis

We also identify five distinct groups among the field position players who have played in MLS between 2007-2012, whose characteristics are presented in the table below.  The principal factors that differentiate the five groups are base salary, minutes played, matches played, and goals scored.

The main difference between the Draftee cohort and the Legacy/Foreign Free Agent group is the lack of match experience as expressed by minutes played, matches played, and age.  There is less of a difference in match experience between the Legacy/Foreign cohort and the elite groups, but significant differences in output.  Another interpretation of the cluster analysis is that MLS has a multiple pay band for its Designated Players — lowest band for players from the mid-level European leagues, middle band for the highest-paid Designated Players who don’t have broad commercial appeal, and the highest band for those DPs who were last playing for major European clubs.

When we take a slice of statistical data across ages, we see that once again there are significant differences in pay levels between the Draftee cohort and either Foreign Free Agents or Generation adidas players.  The first figure below compares the minutes played between Draftees and Foreign Free Agents, and the main feature of the graph is the difference in minutes played between ages 21 and 24.  (The huge spike at age 19 in the Draftee plot is because of Joao Plata’s season in 2011.)  It’s possible that this gap in minutes played accounts for the lag in goals scored and shots produced by members of the Draftee group between these ages, but that gets into the whole causation-vs-correlation issue.

Average minutes played of the Draftee and Foreign Free Agent cohorts for all field players. Data from MLS regular season, 2007-2012.

Average base salary for the Draftee and Foreign Free Agent cohorts for all field players. Data from MLS regular season, 2007-2012.

Draftee players are able to reach parity in minutes played by age 24 or 25, but the gap in on-field performance takes longer to close, and in the case of goals, it appears that it does not close.

Average shots of Draftee and Foreign Free Agent cohorts for all field players. Data from MLS regular season, 2007-2012.

Average goals of the Draftee and Foreign Free Agent cohorts for all field players. Data from MLS regular season, 2007-2012.

Concluding Remarks

There are so many caveats that you can throw at this kind of analysis, and to be honest there is so much more to be done in terms of longitudinal studies of player performance.  We’ve gone through a generation and a half since data tracking became more widely used in football, so we are still crawling along before we can walk, run, and fly.

It does appear that there is a gap in expectations (proxied by base salary and minutes played) and performance of players who enter MLS via the college draft and those who enter it via other channels.  That gap persists over a long term and in some cases it’s never closed.  Now, there are players who have entered the MLS through the college draft and became elite players, but most of those players went on to play overseas within 3-4 seasons.  So there is almost certainly a selection bias present in these results.

So to return to Drew Carey’s assertion about college players and MLS, there does exist a gap in performance, but there are so many explanations for it that a more complete analysis of player performance over a career is necessary.

[Special thanks to Aaron for providing access to his huge soccer database, and thanks to the members of our Data Intelligence team -- Ryan Sonnet, Khaldoon Abu-Hakmeh, Pavel Nekrasov, Billy Marsden, Duane Rollins, Paul Foster, and Tuuwala Lok -- for sourcing and verifying bio/demographic data on the players.]

Sarah Rudd (OnFooty/StatDNA), without whom this meetup wouldn’t have been possible.

Part of the crowd at the Sports Analytics meetup.

I’d like to be able to make interesting remarks about the sport analytics panels at the Conference, but to be honest, I can’t because I spent much of the conference in the corridors meeting people.  If you’re looking for those insights, there are plenty of places to go such as Mitch Lasky, Zach Slaton, Joe Posnanski, and ESPN Daily Dime (Day 1 and Day 2).  If you want an idea of how the SSAC has transformed from a lightly-organized affair to a slick, almost corporate production, you must read Kevin Arnovitz’s piece at TrueHoop.

The sessions that I did got out of my way to attend were the Research Paper finalists and the Evolution of Sport presentations.  It’s not too surprising that I would attend the Research Paper presentations given my analytical background, but there were some EOS talks that were interesting and not as far-fetched as I originally thought.

Nevertheless, my priority was to roam the halls, talk to select exhibitors, and have meetings (scheduled or impromptu) with colleagues and prospects.  The greatest value of these conferences is the serendipity, those chance meetings with people that will change your life.  It was true for me last year as I ended up meeting a future advisor and an investor.  This year I reconnected with a former graduate school colleague who is now a professor, had an extended conversation with one of my recent hires, and got invited to a sponsor’s party.  So it’s very frustrating that the SSAC has made it so difficult to make those serendipitous moments possible.

What do I mean?  Here is a photo of my nametag from the Sloan conference, taken from 18 inches away:

My Sloan conference nametag

And here is my nametag from Startup Riot, an entrepreneur-oriented conference in Atlanta that I attended the week before:

My nametag at Startup Riot

Which nametag can you see from a foot away?  Which one doesn’t require you to have either 20/10 vision or lean forward into someone else’s space?  And would you feel comfortable doing the latter in order to identify some famous person who you’ve wanted to introduce yourself to?

There were a number of people at the conference who I had only known via LinkedIn or Twitter and wanted to meet in person, but never did because that chance meeting in the halls never happened.  And it was even less likely to happen if you had to be within six inches of a person’s chest to identify him (or her, but you’d never make it that far without getting slapped).

This gets into another complaint about the conference: the attendee list wasn’t made available until I checked-in at the conference.  As a result there were people who I wanted to meet who I couldn’t contact at the conference and could only hope to meet by chance (which is more difficult with the nametags on display).  Other conferences, such as the Leaders Sports Summit, not only makes the attendees list available to registered attendees, they also facilitate introductions between interested parties.  The SSAC is becoming more of a sports business conference, so why can’t they make it easier to people to do business?

At the end of every conference, I take a step back and ask myself if the conference is worth my time and money to attend, in terms of technical knowledge gained, contacts made, ideas generated, or sales made.  I was starting to have doubts about the SSAC over the last two years as the price to attend continued to climb, but I still conclude that I need to be at this conference.  The opportunities for cross-pollination are much greater here than at any sports conference, and even though the SSAC is becoming less “geeky”, the level of analytics discourse is higher there than at any other sports conference of its size.  (NESSIS and NCSSORS are very technical, but only 50-80 people attend them.)  It is still value for my money, and I hope to attend and participate as long as I can.

Soccer Analytics panel at 2013 MIT Sloan Sports Analytics Conference. Panelists (L-R): Albert Larcada, Chris Anderson, Blake Wooster, Jeff Agoos. Moderator (far R): Marc Stein.

Returning to moderate the panel was Marc Stein of ESPN, and joining him were Albert Larcada (ESPN), Jeff Agoos (MLS), Blake Wooster (Prozone Sports), and Chris Anderson (Soccer by the Numbers and Anderson/Sally).

Marc ran the session as a continuation of last year’s panel, so he focused on two issues: what is next in soccer analytics, and how will analytics be used throughout the soccer industry.  The intent was good, but it limited the scope of the panel discussions.

Albert and Jeff could talk about how data are used by a media company and a professional sports league office, but Prozone isn’t going to talk about ongoing or future projects at such a forum and even if a sports betting consultancy were on the panel, they wouldn’t talk much about their use of analytics, either.  This left much of the task to Chris who, as someone independent of the rest of the industry, could give a detached perspective on the state of analytics.  His comments made it clear that there’s still a lot of the present in soccer analytics that needs to be matured before we talk about what’s next.

In Chris’ opening comments, he pointed out that football analytics descends from work done in notational analysis by Charles Reep in 1950 and the work that he performed for a number of title-winning teams after that. (I recommend reading this account of Reep’s life in the 2002 Journal of Sports Sciences; even if you read only the first page it provides a good picture of Reep’s remarkable work.)  There is also the work of Mike Hughes (who influenced, among others, Gavin Fleig and Blake Wooster) and the great Dynamo Kiev/Ukraine manager Valeriy Lobanovsky.

What has happened, Chris said, is less of a revolution than it is an evolution, and that’s right, although the generational cycles have shortened thanks to technology.  As a byproduct of technology, data are becoming less expensive and more of a commodity, while a community of armchair (or is it desktop?) analysts has formed outside the realm of the football clubs.  So data are more available than ever before, and more people are comfortable talking about data in sport than ever before.

Albert Larcada (L) and Chris Anderson (R) at 2013 MIT Sloan Sports Analytics Conference

So if data have a higher profile than ever before, what is keeping them from being used more extensively by football clubs?  The panelists addressed this issue from various directions during the session — cultural resistance, contextual challenges, inherent noise in match inference and prediction, and short-termism.  It’s possible that cultural resistance will weaken as a generation of players, executives, and supporters grow up in an era where data are more accepted, and context might be captured better as sensing technology gets cheaper and more pervasive.  But the structure of the game creates a ceiling on the level of precision to which we can analyze players and teams, and the short-term mentality of professional clubs (partly, but not entirely, a product of promotion/relegation rules) limits their capacity to embrace innovation.

If there is a place where data are being used, it is in the media organizations, and Albert spoke at length about the data visualizations that are shown on ESPN SportsCenter.  He showed a couple of examples of maps that compared playing tendencies of Lionel Messi and Cristiano Ronaldo and illustrated ESPN’s priorities on analytics: that they be self-contained, communicate the story succinctly, and be accessible to on-air talent.  One sentence of his stuck with me: we — the soccer analytics community — are not the primary SportsCenter audience.  And because we are not the primary audience, understanding for the broader public is critical.  It doesn’t mean that we can’t develop better match predictors or game models or stuff like that.  But until those can be explained in a way that the broader public can understand what is being communicated, those analytics are not ready for public consumption.

I left the ballroom observing that while soccer analytics has grown tremendously since 2010, there remain significant and unique challenges. Unlike baseball and basketball, whose analysts developed much of their work outside the game and with public data, working with proprietary data is the only way an analyst can do any meaningful work in soccer.  The only way one can work with proprietary football data is through the sports data companies or a professional football club.  And would a club be willing to pay for an analyst to deliver insight on match strategy and player performance and present said insight at public conferences?  The question almost answers itself.

We still have a very long way to go.  And if the early work on analytics is any indication, the window of opportunity and receptivity in the professional football world doesn’t stay over indefinitely.

UPDATE: For a bracing example of the limited window of opportunity, read Mitch Lasky’s review of the Soccer Analytics panel.

I warn you that while I tried to capture what the panelists were saying, these are paraphrases and not guaranteed to be 100% accurate. At any rate, here we go:

Panelists:

• Albert Larcada (ESPN)
• Chris Anderson (Soccer by the Numbers)
• Blake Wooster (Prozone)
• Jeff Agoos (MLS)

discussion topics:

• what are next steps?
• how can data be used throughout industry?

* what’s next step?
(Chris) where have we been?  misnomer to describe this as “revolution” — more an “evolution”
soccer has more history in analytics than people think
Reep’s coding system goes back to 1950 (notation system — all events in pitch),  worked with teams in 1950s-1980s

biggest advancement in analytics: cost of data going down, accessibility increasing
gives MCFC Analytics as example

more people writing about it in soccer analytics blogosphere
clubs work hasn’t moved the needle

* why are clubs not keeping pace?
(Blake) cultural challenge exists, persists
regulation driving more interest in due diligence, data usage
generational shift as well
Mike Hughes notational analysis –> Blake, Gavin Fleig
mentions generations –
X: access to technology
Y: making things faster, better
Z: generate insight, making things easier

(Albert) more data available — more data you must visualize
huge on media side — focus on telling stories

(Jeff) more data isn’t always better, ref Silver’s book (The Signal and The Noise)
owners see it as expense or investment

* are analytics more valuable in MLS?
(Jeff) to a point, but there’s cost-benefit equation to consider

(Albert) real-time shot charts, heat maps (Trumedia)
some heat maps actually used on SportsCenter (Messi vs Ronaldo)
shows video

(Jeff) challenge is capturing context
(Blake) diff b/t what’s important, and what’s interesting
(Albert) important to remember that SportsCenter audience =/= audience in room
telling the story
(Chris) technology used to communicate story very important
have moved quickly from zero data –> drowning in data

talking about Mourinho’s use of data
(Blake) discussing about Mourinho’s match preparation through data
no-stat all-stars more common in soccer than you think
“most important things are the hardest to measure”

(Albert) is there a player in Europe who is like Shane Battier?
(Blake) IQ — EQ — CQ (contextual intelligence)

(to Jeff) Do you wish you had more access to data ten yrs ago?
(Jeff) absolutely — new players growing up in culture where data is accepted
more tools, visualization critical

* does soccer world lag behind bball on analytics?
(Chris) fair point — “going 0 to 100 in data is scary to lots of clubs”
hard to remember that we’re not dealing with North American sports
- winners don’t feel need to innovate
- relegation battlers too risk averse
- middle runners feel they should finish in top 6 anyway

(Blake) short-termism rampant
manager lifespans extremely short
75% of managers will coach maximum of 75 matches

(Jeff) management decision — captive to short-term concerns

half of league have invested in analysis / BI teams
advance scouting, sports science, academy

(Albert) what is analytics? is it video? data viz? stats/CS?
use of analytics is very gray

(Chris) proving the ROI is very hard
element of trust hugely important
element of randomness is HUGE in soccer

(Blake) we need success stories more than anything else

(Chris) we’re now on Liverpool 2.0 now
big announcement at beginning of LFC 1.0 not productive long-term

(Albert) nature of game not designed for analysis, lots of noise
baseball “is a PhD’s dream” (me: not sure I agree)

(Blake) mentions case of Michu (one who got away) — Type II error
15 goals in Spain, very low transfer fee for Swansea, missed by six scouts

(Jeff) coaches are doing measurements all the time, just in a different way

(Chris) let’s not pile on coaches
dealing with lots of tasks, limited resources
lots of fads in the industry
just b/c we think it’s demonstrably useful doesn’t mean it will be

(Blake) challenge and opportunity
if coach doesn’t get it, it’s our fault for not communicating it

(Chris) we pretend that coaches know how to win football games
not clear that soccer analytics knows what wins games
what is theory of what produces wins in football?
clubs: intuitive non-verbal theory

* what are new metrics that should fans be more familiar with to help game-watching experience?
(Albert) timely stats, fans get info when they want it
(Chris) best player debate not all that interesting
favorite metrics are ones hardest to measure –> defensive metrics
prefers team performance
(Blake) game intelligence
(Albert) certain skillset required
(Chris) difference between prediction and explanation
prediction: interest from bettors, media
explanation: interest from clubs
(Jeff) for MLS, how do we compare ourselves to other leagues
measure everything –> indices of match performance
(Chris) soccer’s unique in terms of “right way” to play
what does it mean to be an attractive league?
(Jeff) attractive vs compelling

* Bundesliga and MLS have bought data and made available to clubs. Premier League hasn’t.  Why?
(Blake) used to be resistance in PL, less so now
do same in Qatar, Poland, etc
more cost effective to make central investment
common language

I am adding the summary later, but I’ll liveblog the remainder, for what it’s worth.

The panel has come a long way from its first appearance in 2010, when it was called the Emerging Analytics panel and combined panelists from American football and the English Premier League.  Even so, it was a packed session as people staked out seats two hours before the actual session took place.

In 2011, American football and soccer were given separate sessions, each very well-attended and comprised of a combination of professional club, data supplier, and independent analytics talent.  In 2012 there was more of a 50/50 split between American and British participants on the panel, which reflects the two major markets for statistical analysis in sport.  I was pleasantly surprised by last year’s session, especially by Drew Carey’s contributions to the discussion, but nevertheless I left thinking that there remains a very long road for analytics in soccer to travel.

Last year I made the following complaint about the Soccer Analytics panel:

The one charge that could be made against the Soccer Analytics panel is that there are no representatives of the upstream elements of the analytics value chain.  There is no one from a sports data company, nor anyone from a soccer analytics company or blog.  Why isn’t there someone up there like Blake Wooster of Prozone or John Coulson of Opta, or analysts like Chris Anderson, Sarah Rudd, or me?

Sadly, I won’t be participating in this year’s panel, but the Soccer Analytics panel will be sponsored by Prozone.  The panelists represents a wider range of the analytics value chain than ever before — Blake Wooster returns to the panel representing Prozone, Albert Larcada (ESPN) and Jeff Agoos (MLS) represent the users of data in the media and the professional game, and for the first time a member of the soccer analytics blogosphere joins the panel in Chris Anderson (Soccer by the Numbers).  I’m very pleased to say that I’ve gotten to know all of the panelists personally and professionally, and with the exception of Albert, I’ve heard all of the presenters speak about soccer analytics in various forums.

The objective of this year’s panel is to build on the guided questions from last year, inquiring about the strategies and tools used to inform decisions from the playing field to the front office.  Here are last year’s questions:

• What role can analytics play in the world’s favorite sport?
• How can analytics be used to field the best team formation?
• How can analytics help clubs find players that would suit them best?
• What are the challenges that face soccer clubs in adopting analytics and how can they be addressed?

Even though there is no representative from a professional club, Blake and Chris have worked with a number of clubs so they might be able to share some insight on how clubs use data and the current lessons learned.  Chris is writing a book with his business partner David Sally on soccer analytics, so perhaps he will be able to share some highlights from it.  Jeff comes from the perspective of a league organization that is focused more on the business of sport, but player acquisition is a key element of MLS so perhaps he will speak to that.  Albert might talk about the Soccer Power Index and the internal simulations conducted by ESPN Stats and Information, but I’ve learned from last year to keep an open mind.  In a way, it’s too bad that neither Gavin Fleig nor an Opta representative will be on the panel because I’m almost certain that MCFC Analytics will be a topic for discussion.  And if it’s not, it should be.

I look forward to a very thought-provoking Soccer Analytics panel on Saturday.