APBRmetrics

[Chicago76]

I've seen a couple of articles recently on Billy Beane and his fascination with statistics and soccer. I'm curious if anyone has done any work in soccer on here. If you have, I'd be interested to here what you've done.

http://soccernet.espn.go.com/columns/story?id=495270&root=mls&cc=5901

There probably aren't too many soccer fans or former players around here, but for those of you with more than the obligatory "I played until I was 12" experience, this may be of interest.

Soccer is leaps and bounds more difficult to quantify than traditional American sports.

-less defined possession attributes
-less north/south and more east/west possession movements
-+/- won't work because of substitution limitations
-fewer things to record: goals, shots, corner kicks, passes completed, passes mis hit, tackles won/lost. That's about it on the ball.
-off the ball movement is tremendously important and difficult to record without hours of film analysis. The concept of quick triangles and passing, and vacating space to open an area up for exploitation by a teammate is critical to getting a defense disorganized.
-stretches of possessions can be entirely uneventful.

What makes this promising to me is that there probably is not another sport in the world where this could be more beneficial. Soccer teams abroad are run in a much more capitalistic manner than major American sports (no draft, no cap, relegation to other leagues for poor performance can lead to major $$$ hits, a very liquid transfer market for players, limited/no revenue sharing). As a result, there is probably more financial incentive for this type of analysis in soccer. This could lead to some interesting analyses that may be applicable to basketball.

A few comparisons/applications (thinking out loud):

-space and player movement. Basketball and hockey are as close as it gets to soccer in the sporting world, but I'd argue space is most important in soccer. This might lead to some innovative ways of thinking about motion, and the success of isolations and little three-man "games within the game".

-the concept of chance creation/usage. This is soccer's Iverson question. In soccer, a team without a creative genius has a difficult time breaking down a defense (see: USA). The genius might fail 19 times out of 20, but things pay off on the 20th attempt. The alternative is to play it safe and not risk surrendering possession. What is the cost of failure in terms of counterattacks? What is the value of success in terms of shots created? How are the costs/value different depending upon where the actions occurred?

-the opposite: the safe play AKA not all assists are created equal. Some players are really skilled at gaining possession, slowing the game down, and making the safe pass (see: Claudio Reyna). This has value in soccer. There was an index for the English league that attempted to capture this (Actim for a while, OPTEX before that). The problem is, the safe pass is safe because the defense deems it to be the most harmless one. It's not always the best pass. Players attempting only safe passes can be artificially rewarded with high completion rates.

-the "someone's got to get it" value of a rebound, or in soccer, a tackle. If you have a designated midfielder/central defender making a sizable chunk of tackles, is he really that valuable? If you slid a more technically skilled player into that position, would he be able to replicate that success simply because that is where a lot of tackles are made?

-roles. Basketball requires players to fill different roles at different times of the game. Soccer is the same way. Attackers who lose possession may quickly become on the ball defenders or tacklers. A defender making a run may become a "point guard".

[HoopStudies]

I read this a couple days ago. Soccer is a sport where, as mentioned in the article, you need to come up with different metrics. For instance, "possessions" don't really help. They exist and you can delineate them and, to some degree, they "work." But they aren't getting at the essence of soccer offense and defense. It's in the same way that possessions don't really help in football -- field position is more relevant.

What I find useful are 2 things:

- Watching soccer with people from different countries. Different countries talk about different things with soccer, how to attack, how to defend, what's most important. I watch and I see what they're talking about. Americans attack almost entirely on the sidelines. Brazilians attack the middle a lot.

- In JQAS, there was a paper published on potential functions in soccer. This was an interesting paper, one that used limited data (extremely) to arrive at an estimate of parts of the field that are valuable.

The first bullet leads to significance about tactics. Those tactics can actually affect the second bullet. Once you develop a potential function, you can start working on individuals. Individuals who move you up/down the potential function with passes or dribbles tend to be more valuable. What I've run into problems with is the issue of how much spacing affects individuals moving around the potential function.

The bigger problem right now is data, actually. The USOC in some of their work does digitize the kind of data you need. But I don't know if anyone else does so. It sounds like MatchAnalysis (the firm referenced in that doc) would do it. That data plus the right analysis can get you somewhere. I think you could also do some work on spacing.

Right now, I think MA may be confounding data with analysis. Comments like, "You can win a header, and if it doesn't fall to your team, that counts as a loss of possession, so it's not 100 percent accurate," are what suggest this to me. Each data point is neutral. Analysis is what gets you to solutions and that statement implies that they're putting them together at a small level and perhaps overinterpreting.

Anyway, one similarity to basketball is the need to come up with a way to share credit. For passes from one location to another, the credit needs to be split between the passer and receiver. I've only barely done anything on this so far in soccer.

Your points are good. I hadn't heard the Claudio Reyna argument before actually. Given that I didn't play soccer until I was 12 and I really started picking it up recently, I have a lot to learn to make my math and theory really work. But the potential to do work in soccer, to use work from other sports, and to feed back into other sports is big. I wish I had more time for it.
_________________
Dean Oliver
Author, Basketball on Paper
http://www.basketballonpaper.com

[schtevie]

Isn't there any low-hanging fruit here?

My impression is that a high proportion of goals scored come off of set pieces. And these are discrete events where folks have easily measurable skills.

I would like someone who knows soccer to tell me why an optimal strategy of the English national team shouldn't be to have its players try to carom the ball off opposing defenders across the endline, with Beckham taking the corners and Crouch finishing.

[Carlos]

Well, most teams try to force corner shots, even if they don't have a Beckham-Crouch like duo. The problem is that it's not so easy to force a corner shot (defenders are actively trying to avoid it, you know).
Two areas of interest that I would like to see addressed are; a) relation between field goal attempts and scoring (essentially FG%). It's much more variable than in basketball (and so scoring itself is very variable) but I wonder if the number of shots at goal wouldn't have some predictive power. The other would be scoring and game time. My impression is that last minute goals are more frequent than they should and I've wondered if this could be verified.

[Rasta978]

Ah, soccer. My second love. Thanks to Fox Soccer, I usually watch 2 or 3 games each weekend. So I'm very interested in discussing the characteristics of soccer. What we learn from soccer can hopefully further our analysis in basketball. Thanks for starting this thread.

I've done a lot of theorizing about soccer, but unfortunately (as mentioned earlier) the available data is sparse. The 'free' data, anyway. In any event, the most basic statistical concept in soccer is goal differential, and its high correlation with winning. Working against us are smaller sample sizes (shorter seasons), lower scoring games, and of course the possibility of the game ending with a draw. Even without these limitations, where do we start when it comes to sharing credit for goals scored, and blame for goals allowed? I don't know the answer.

There is an enormous variety in formations and tactics employed from team to team. This, in turn, dictates different roles and responsibilities for players that are classified as the same position. Sometimes this occurs on the same team. For example, from the Englush national team, Michael Owen is small and quick, and uses his pace and technical ability to score goals. Peter Crouch is 6'6", and assumes the role of target man, using his height advantage on headers in the box. Different players, different attributes, both effective goal scorers. But are they comparable statistically?

On the other hand, consider the Steven Gerrard and Frank Lampard. Individually, at their respective teams, they are outstanding midfielders. However, when asked to play side-by-side for the national team, the results were horrible: the players were too similar. The English team played better when Gareth Barry, by all accounts a lessor player, replaced Lampard in the lineup.

Further, to make matters worse, consider how many different positions there are in soccer, each with different roles to play. It isn't just defender, midfielder, and forward. Instead, from back to front, you have goalkeepers, centerbacks, fullbacks, wingers, central midfielders (usually one attacking and one defensive), and several varieties of strikers. In order to truly compare soccer players, you have to classify them at least to this level.

Lots of issues, and a lack of available statistical detail, we're looking at a huge undertaking. I'm sure the big clubs, who have the financial wherewithal to spend $30 million on a single player, have devoted significant resources to studying this. I expect Billy Beane will do likewise, although with a lower budget. For us fans, I don't know if the discussion can go beyond theoretical at this point.

Instead, let me mention a simple alternative: match form. What is match form? Its a rating between 1 and 10 for each player on the pitch determined subjectively by the spectators, commentators, or pundits. That doesn't sound like much, but if you poll several hundred match observers the results will be interesting. Their collective knowledge should tell us something useful. BBC Sports does this for every game.

http://news.bbc.co.uk/sport2/hi/football/teams/a/arsenal/live_text/6173791.stm

[Jacob]

Very interesting discussion. As a European, soccer is my "native" sport and I have spent many years and euro's developing models for quantifying player performance PER-style. It was a frustrating time, but with a happy end - I sold my work to a professional Dutch team... And they promply started winning. (At least until this season, haha.)

The key to cracking soccer's code is to think in terms of "chances at scoring a goal", and then work backwards, to figure out how a team creates those chances, and assign credit to the players instrumental in this. You can do the same the other way around - see how the defense prevents the opposing team from increasing the scoring potential of their possessions, and assign credit for that.

Of course, tactics and player synergy are even more important than in basketball, and player values - especially for the midfielders - fluctuate immensely depending on the system they play in.

[Chicago76]

[Harold Almonte]

I think the low hanging fruits are for goalkeepers who have a lot of privileges that the rest of players don't have. That makes the game more difficult to play and to track stats, and a little bit different than basketball here. How many stats are tracked in hockey?

The value of possessions are so meaningless that probably the only things that matter is the quantity of things you can do (specially stops, passes and potential passes to the area, fouls, out of plays, etc), and the distance from the goal you (and your shot) can reach, I think the action inside the goal little area must be weighted different or tracked appart from long distance plays with the capacity to reach the goal with the ball. But the midfield game has no possession value until the game is ending (tied or by 1).

[gabefarkas]

[schtevie]

Gabe,

Your limited gastronomic interests aside, I don't understand the remark: "consequences". What consequences?

When offensive players get the ball in the deep corners and attack toward the box confronted by a defender, there is typically a deke dance that goes on, where the offender is apparently trying to get by said defender and/or create space for a centering pass.

To my untrained eye, it doesn't typically appear that the third option, of caroming the ball off the defender, setting up a corner kick, is considered a primary option.

Now, my possible misimpression aside, maybe players do behave rationally, in terms of the relative prospective success rates of goals coming off the direct attack vs. corners. Then again, maybe they don't.

Knowing what this ratio is would be an interesting - and easily calculated - piece of evidence however.

Whatever the case, I don't understand what "consequences" are in play here.

[kjb]

One possible consequence could be fewer scoring opportunities. Miss "caroming" the ball off a defender and it's likely a goal kick for the defense. Or perhaps you hit the defender and the ball doesn't go over the endline. Or the defender controls the blast.

Another could be increased physical play from the opponent. They may view it as a chickenpoop tactic and take exception to being blasted point blank with the ball. And they could show their ire with harder tackles, more aggressive tripping, etc.

Another possible consequence could be sportsmanship penalties, yellow cards, red cards. Repeatedly and intentionally hammering opponents with the ball could be interpreted as dangerous play.
_________________
My blog

[schtevie]

kjb, I think your suggested list basically begs the question. Whether or not the strategy is optimal is the entire point.

I don't think the chickenpoop angle will cut it. It is entirely common when battling for control on the sidelines for players to try to kick the ball off their opponents. Not only that, they don't seem to mind blasting kicks that come perilously close to opponents' heads off of free kicks. Furthermore, blasting the ball off the defenders body is both very difficult to do at close quarters (as in the situation I am describing) and probably relatively ineffective and unnecessary.

Since the last post, I made a google foray with the intent of finding any corner kick stats, and to my surprise none were found. In fact, at the premier league site there is no breakdown of goals scored in the stat sheet at all. I find this rather shocking. Never mind corners, how about penalties at least. Very strange. Very, very strange, given the low cost of tabulating such stats. I wonder how interesting fantasy soccer can be?

[kjb]

I definitely think it's worth counting these things and looking at the outcomes and relative advisability of different strategies. (I'm not going to do it because I'm just not that in to soccer.) I was just making a list of possible consequences to the strategy you're proposing.
_________________
My blog

[gabefarkas]

[schtevie]

Let me just repeat my conjecture about a game I have never played nor watched terribly attentively.

The only point I am making is that it MIGHT be profitable (depending on apparently unmeasured, hence apparently unknown, risks and rewards) for attacking players (typically forwards) to increase the relative frequency that they try not to get past the interposed defenders but rather to oblige the latter to touch the ball last before crossing the end line.

Suppose that my conjecture is true. This necessarily (I think) implies that it is a net positive gain for the offense. How would defenses react then as a consequence? Well, they would be obliged to give the attackers more space, so as to diminish the increased number of corner kicks. This then necessarily increases the ability of the offensive player to get a conventional cross in, or work whatever magic they would otherwise do.

Of course there would be a response from the defense, but how could it imply anything but a net offensive gain when all is said and adjusted?

Such an hypothetical net positive offensive (and defensive) innovation has historical counterparts in all sports, including basketball.