Sabermetrics applying to Soccer

[beineke]

Originally posted by kenntomasch
Or would that be because the player who wins the first set has a higher chance of being the better player?


Yeah, they adjusted for that. (Incidentally, the data was from elite players going head-to-head in the era of Lendl, Connors, and Edberg).

I'm afraid I can't locate the paper -- I tried -- but the guys who wrote it were occasional gamblers. They noticed that in those days, if you bet on a 3-0 sweep, you could get very good odds because the betting line was based on the assumption that the outcomes of sets were independent. In the acknowledgements part of the paper, they thanked the casinos for funding their research. :)

[kenntomasch]

Okay, fair enough. If you have two players of relatively comparable ability, and winning the first set gives you a significant probability of winning the second, maybe there is something there.

I don't know if it's a best-of-three or best-of-five scenario (if it's men's, I'm assuming it's best of five), but in a best-of-three, if you drop the first set, you've got to be pretty mentally tough to deal with the fact that you're very close to being eliminated.

[beineke]

There might also be day-to-day fluctuations in form. Intuitively, that seems a little bit different from momentum, but it would also result in streakiness (and the studies in basketball and baseball hitting didn't detect it).

Voros, have you seen any public studies of this kind of phenomenon in baseball pitchers?

[kenntomasch]

I would think an individual sport would be more subject to fluctuations in the participant's day-to-day form. A team could potentially cover for a player who's having an off day, while if it's just you and you don't have it that day, for whatever reason, it's you and you alone.

[superdave]

Originally posted by beineke
In the acknowledgements part of the paper, they thanked the casinos for funding their research. :)
That's hilarious.

[microbrew]

Originally posted by beineke
Goals most certainly are not iid, but that's not the point. What matters is that pretending they're iid gets you a pretty good approximation to reality.

One other note: it's true that psychological perceptions of streakiness are no good. Voros mentioned a study about baseball, and Tversky did something similar with free throw shooting in basketball. At the same time, tennis does show some degree of streakiness ... the player who won the first set has a substantially higher chance of winning the second. So we shouldn't write off the existence of streakiness, just because our intuition is a poor judge of it.

I agree, that while soccer goals aren't really iid, they are pretty close to being iid. In other words, it's a good simplying approximation that makes life easier. I still want to read a paper on this though.

Tennis paper:
"Are Points in Tennis Independent and Identically Distributed? Evidence From a Dynamic Binary Panel Data Model" http://www1.fee.uva.nl/pp/klaassen/index_files/iid.pdf

My hypothesis for soccer: most soccer teams, whether they're tied, one goal up or down, don't change how they play for most of the game.

As for streakiness: unless there is some underlying cause, a player being "hot" or "cold" is a meaningless to me.

[Nutmeg]

Originally posted by microbrew
My hypothesis for soccer: most soccer teams, whether they're tied, one goal up or down, don't change how they play for most of the game.

That's why I disagree with your assertion that goals are unrelated, isolated events, or iid. Soccer teams do change the way they play - significantly - based on the scoreboard.

Originally posted by beineke
Goals most certainly are not iid, but that's not the point. What matters is that pretending they're iid gets you a pretty good approximation to reality.

"Pretending [goals] are iid gets you a pretty good approximation to reality" is an interesting theory, but it is no more proven than the theory that a two-goal lead is the most dangerous in soccer. What evidence is there that supports that theory, and more importantly, how much compelling evidence is there to the contrary?

kenntomasch and superdave
In other news, on the basis of one game from May of 2002, I have determined that Mike Cameron is the greatest home run hitter of all time.

And on the basis of two games in 1938, I can prove Johnny Vandermeer is the greatest pitcher of alltime.

And on the basis of one post, I can prove jri is the greatest poster on bigsoccer.

OK, so I offered one example of a game within the past week. Before I spend any time finding a boatload of others, let me ask, do you guys really believe that the momentum of the game has little to no impact in relation to goals?

[kenntomasch]

Originally posted by Nutmeg
"Pretending [goals] are iid gets you a pretty good approximation to reality" is an interesting theory, but it is no more proven than the theory that a two-goal lead is the most dangerous in soccer. What evidence is there that supports that theory, and more importantly, how much compelling evidence is there to the contrary?

A bunch. I did study this one. The 2-0 lead teams won outright 90 percent of the time. I guess I took "the most dangerous lead in soccer" to mean "one that gets blown a lot". But it doesn't.


OK, so I offered one example of a game within the past week. Before I spend any time finding a boatload of others, let me ask, do you guys really believe that the momentum of the game has little to no impact in relation to goals?

No. You just said that one game was "plenty of evidence to the contrary." One game can't be plenty of evidence of anything, which is why we busted your balls about it.

I don't think we have any idea how strong "momentum" really is or if it really exists, much less if it has little or not impact in relation to goals. But if you can do an actual study on it, be our guest.

[voros]

Originally posted by beineke
There might also be day-to-day fluctuations in form. Intuitively, that seems a little bit different from momentum, but it would also result in streakiness (and the studies in basketball and baseball hitting didn't detect it).

Voros, have you seen any public studies of this kind of phenomenon in baseball pitchers?
Not directly, but there was a study done about "pitching to the score." The concept of pitching to the score, like soccer, is the belief that some pitchers are able to control how many runs they allow based on how many the team is scoring for them, thereby winning more games than you'd expect from just looking at their runs allowed and run support totals.

The study, as expected, showed that pitching to the score either does not happen more often with some pitchers, or doesn't happen to a meaningful extent. That the pitchers effect on winning seems to be based strictly on how many runs he gives up (and components to that extent) and not on how efficiently he distributes those runs. Don't remember who did it. I think either Keith Woolner or David Grabiner. The problem with Sabermetric studies is that it's not like psychology or physics or something where there's journals this stuff appears in. It's scattered everywhere.

And the Gilovich example was basketball not baseball. In fact the Tversky study was in fact Gilovich, Vallone and Tversky doing that study: "The Hot Hand in Basketball: On the misperception of random sequences. Appeared in Cognitive Psychology."

[mpruitt]

Originally posted by kenntomasch
A bunch. I did study this one. The 2-0 lead teams won outright 90 percent of the time. I guess I took "the most dangerous lead in soccer" to mean "one that gets blown a lot". But it doesn't.


Did you calculate any numbers at how often the lead stayed 2-0 or was equalised only to have the original leading team go on to win? Because that would seem to be more relavent to me. It's pretty hard to score three unanswered goals on someone (just my assumptive oppininion) but I wonder if people have developed that the perception that a 2-0 lead is the most dangerous lead in soccer because.

1. If the team that's down comes back to score people are evidently aware that the equalizer could come at any moment, when they had already thought that the game was sealed up. Where as a 1-0 lead is obvious that the game is still within reach.

2. Most importantly, if a team jumps out to a 2-0 lead that means that at some stretch of the game they were doing a good job scoring. They may have then developed a bunkering style and gotten caught in that. My question really is, after surrending an equalizer on a 2-0 or 2 goal lead how often does the original team go on to win?

[kenntomasch]

Yes. I have the whole thing somewhere, but not handy at the moment. I had a category for "Blown leads", which is 2-0 leads which became 2-2 or 2-X, whether the team lost, ended up with a draw, or then won despite blowing the 2-0 lead.

And teams that went up 2-0 at home were almost bulletproof, if I remember correctly.

Keep in mind, this was only a study of MLS, there aren't many years in the study, and I had to consider shootout wins and losses a couple of different ways for the first few years' worth of data. But given how it wasn't a namby-pamby bunch of results, I concluded that, at least from this limited study (and without being able to check, say, different leagues in different countries game-by-game with boxscores), that 2-0 leads aren't blown quite so often as you would be led to believe by the term "Most Dangerous Lead in Soccer".

[mpruitt]

yeah that would seem pretty self evident, but that's interesting that your numbers seem to back it up pretty concretely. like i said it's probably all in the correct preception of a game being pretty well in hand vs. the fear and suprise when it's sudenly not, or lost. i'd imagine it's like anything, defeats are a lot more vivid in your memory than successes.

[kenntomasch]

Well, it's just like the "How many times do you see it, a guy makes a great defensive play to end an inning, and then is the first guy up in the next inning" thing in baseball. It probably doesn't happen all that often, but every time it does, you can be sure it'll be brought up as if it's some mystical thing. It's just a coincidence.

Just like, again attempting to draw a connection between discrete events, we see a team blow a 2-0 lead and we assume it's because they let down with a seemingly "big" lead. But it rarely happens.

Just human nature.

[voros]

Originally posted by kenntomasch
Well, it's just like the "How many times do you see it, a guy makes a great defensive play to end an inning, and then is the first guy up in the next inning" thing in baseball. It probably doesn't happen all that often, but every time it does, you can be sure it'll be brought up as if it's some mystical thing. It's just a coincidence. Au contraire. It happens roughly 11% of the time a great play is made to end an inning. :)

[microbrew]

Here's my back of the envelope calculation for the two-goal lead issue:

Assuming we're using a Poisson Process:
Blowing a two-goal lead consists of combining the probabilities of a team scoring a goal, then another goal, and then not giving up any more goals.

[mpruitt]

Originally posted by voros
Au contraire. It happens roughly 11% of the time a great play is made to end an inning. :)

Mr. Know It All

[Nutmeg]

Originally posted by kenntomasch
No. You just said that one game was "plenty of evidence to the contrary." One game can't be plenty of evidence of anything, which is why we busted your balls about it.

Fair enough.

Originally posted by kenntomasch
I don't think we have any idea how strong "momentum" really is or if it really exists, much less if it has little or not impact in relation to goals. But if you can do an actual study on it, be our guest.

And yet statements like "pretending goals are iid gets us much closer to reality" are being made and accepted here.

I don't think they are. I don't believe that pretending they are gets us any closer to reality, because in my experience momentum has a huge impact in relation to goals, and there is no evidence that would support the assumption that pretending soccer goals are iid gets us closer to reality is true.

Anyway, I'll go back to quietly reading along.

[kenntomasch]

Originally posted by voros
Au contraire. It happens roughly 11% of the time a great play is made to end an inning. :)

Roughly, assuming each slot in the order has an equal chance of coming up first the next inning (I don't think it does, actually - those at the top of the order tend to get more plate appearances than those at the bottom, increasing the chance that someone in the top third of the order will be up next inning rather than someone in the middle or bottom third) and that each spot has a player likely to make a great defensive play (not completely likely, either - or at least not equal. 11% of American League lineups make no defensive plays at all, and a catcher or firstbaseman are less likely to make a great defensive play than a center fielder or a shortstop, who may or may not bat near the top of the order).


And yet statements like "pretending goals are iid gets us much closer to reality" are being made and accepted here.

Not by me. My eyes glaze over at the whole iid/Poisson discussion, so I choose to deal with the things I understand. :)

[microbrew]

I haven't seen a study that says soccer goals are iid, or are close to it. Noone, except maybe me, said it's absolute truth.

However, when an iid assumption is made, the math works out nicely. This assumption is made in some of the academic papers listed in this thread. The papers use a Poisson distribution, of which one of the requirements is that the events (goals) be iid. Remember, it's only an approximation, but one that works well.

The arguement for momentum, whatever it is, might be true, but I think it's effect doesn't make the assumption of iid a bad assumption. In general, I'm just skeptical of the concept of momentum, irregardless of sport, unless it's mass times velocity.

As for my hypothesis: well, Icould be completely wrong- I haven't thought it through, and I haven't tested.

[beineke]

Originally posted by Nutmeg
There is no evidence that would support the assumption that pretending soccer goals are iid gets us closer to reality is true.


As JG noted earlier in this thread, the Poisson model worked very well in predicting a team's win % from its totals of goals scored and goals allowed. In fact, it did better than any approximation to the Pythagorean formula.