Statistical Player Ratings: USA vs MEX

[NoSix]

As for the Sweden match, I have tracked every touch for each USA player in the recent match versus Mexico. Once I have collected sufficient data (4-5 matches worth), I will calculate statistical player ratings with different types of touches weighted according to their importance to scoring and conceding goals. In the interim, I am publishing a simplistic, provisional rating that is a function of both the quantity and quality of each player's touches per 90 minutes (all touches weighted equally). Given the obvious limitations of the provisional rating, comparisons between players are probably only useful for players who play the same position.

As always, I welcome feedback if you think the ratings for a particular player are way too high or too low, just be sure to say why you think that is the case. Such feedback is quite helpful in fine tuning the model that will produce the final ratings.


rating player
105 Feilhaber
104 Adu
82 Dempsey
61 Corrales
56 Bradley
51 Clark
49 Convey
45 Bocanegra
42 Donovan
41 Edu
33 Moor
30 Altidore
28 Howard
23 Onyewu
2 Lewis

[izha]

As for the Sweden match, I have tracked every touch for each USA player in the recent match versus Mexico. Once I have collected sufficient data (4-5 matches worth), I will calculate statistical player ratings with different types of touches weighted according to their importance to scoring and conceding goals. In the interim, I am publishing a simplistic, provisional rating that is a function of both the quantity and quality of each player's touches per 90 minutes (all touches weighted equally). Given the obvious limitations of the provisional rating, comparisons between players are probably only useful for players who play the same position.

As always, I welcome feedback if you think the ratings for a particular player are way too high or too low, just be sure to say why you think that is the case. Such feedback is quite helpful in fine tuning the model that will produce the final ratings.


rating player
105 Feilhaber
104 Adu
82 Dempsey
61 Corrales
56 Bradley
51 Clark
49 Convey
45 Bocanegra
42 Donovan
41 Edu
33 Moor
30 Altidore
28 Howard
23 Onyewu
2 Lewis


Everything that gives Corrales more than -1000 is a total crap. And that's all what I have to say about your rating.

[TC4life]

This proves that stats mean absolutly nothing. can you at least say of how these stats were tallied? Or does the lower the number mean the better they played?

[NoSix]

Everything that gives Corrales more than -1000 is a total crap. And that's all what I have to say about your rating.

Corrales led the team in touches, and the quality of his touches was about the team average. The rating accurately reflects those facts.

[PVancouver]

Let me first say that the idea of coming up with an objective rating system is a laudable one, but in reality, subjectivity must enter the rating scheme or the data become highly polluted.

Your concept of "touches per minute" being valuable is highly questionable. I suspect this is why Lewis is given such a low score, because it wasn't because he turned the ball over. In fact he barely touched the ball at all. In the Sweden match you have Corrales a high score because he had the most touches per minute with an average touch quality score.

Likewise Onweyu's passing was excellent, although he rarely went long. His only poor play that I could tell was a couple of avoidable fouls.

I am interested to know how you rated Adu and Feilhaber so highly. Not that the rest of the world doesn't agree with you, but I thought Adu's performance in particular was subpar, and Feilhaber's was average at best.

If Adu dribbles the ball forward, and makes 15 touches with his feet as he dribbles forward, do you count that as 15 positive touches? Adu did have a couple of good dribble runs, but nevertheless he had more turnovers per minute than any US player. He did have one well taken corner, although it did not find a US target.

Dempsey did some spectacular things but overall had only an average at best game. He is another dribbler so perhaps this is skewing your results.

I could see giving Corrales the score you did for his touches on the ball but his defense was not very good, and that is being kind. Are you trying to objectively rate defense, too?

The rating I would be most interested in hearing details on is Drew Moor. Many would agree with you that he had a poor game but I rated his passing as among the best on the team, and he had a considerable number of touches.

Your goalkeeper ratings should just be eliminated, because they are nonsensical.

There are a lot of plays in soccer where you don't have the time or opportunity to direct the ball to a place where your team is likely to maintain possession--e.g. clearances, 50/50 balls, working out of pressure. Losing possession of the ball in these situations should not be counted against the player.

I would not even count Onyewu's "misclearance" to Rosicky in the World Cup as a bad touch. It wasn't a great clearance but only 1% of the time would a clearance like that have resulted in a goal. If you don't feel his clearance was good enough to count as a positive touch, just don't count it at all.

Here are your ratings from the Sweden game an obviously some names jump right out:


rating player
122 Donovan
119 Moor
90 Corrales
88 Altidore
85 Davis
62 Edu
61 White
55 Kljestan
55 Clark
45 Noonan
43 Conrad
41 Twellman
38 Parkhurst
38 Robinson
34 Guzan
32 Rolfe
20 Goodson


This comment explains a lot:

I think I have already addressed this issue. The provisional ratings weight all touches equally, so Robinson's goal counts no more in the ratings than Twellman's initial touch to kick off the match. The final ratings will weight touches according to their importance in scoring and conceding goals, so that Robinson's touch will count more.

Twellman's initial touch to kick of the match should have no value at all. No professional soccer player should mishit a simple kick-off. Conversely, a "quality" goal like Robinson's should count for far more than just a simple pass.

Along the same lines, all of those unpressured possession passes along the back line by Mexico and sometimes by the US should count for nothing or practically nothing because it is very unlikely for a professional soccer player to mishit such a pass. And even if the pass is mishit, there is usually enough margin for error that the pass can be recovered. Counting these as positive touches doesn't seem very helpful.

I could go on, but I'm sure you get the idea.

[sidefootsitter]

Come along, boys and girls.

A short pass - 1 point.

A long pass (completed) - 3 points.

A short pass (lost) - minus 2.

A long pass (lost) - minus 1.

Shot on goal (missed barely, outside the box) - 3 points.

Shot on goal (missed, inside the box) - 1 point.

A bad cross (out of touch, or into a defender) - minus 2

A good cross (placed between the PK spot and the 6-yard line) - 2.

Et cetera.

Once you create a system, you can can juggle the grades per event until it reflects the reasonable consensus value.

I know it can be done.

[SUDano]

Come along, boys and girls.

A short pass - 1 point.

A long pass (completed) - 3 points.

A short pass (lost) - minus 2.

A long pass (lost) - minus 1.

Shot on goal (missed barely, outside the box) - 3 points.

Shot on goal (missed, inside the box) - 1 point.

A bad cross (out of touch, or into a defender) - minus 2

A good cross (placed between the PK spot and the 6-yard line) - 2.

Et cetera.

Once you create a system, you can can juggle the grades per event until it reflects the reasonable consensus value.

I know it can be done.

Your point system has nothing to do with the situation at hand. Some crosses perfectly placed between PD and 6 yard line is a bad cross when you have a near post and a far post run. A Shot on goal missed inside box may be the greatest play ever if a perfect pass then touch, then great pass is a great save by GK. A rolling ball into the 6 with a slipped GK and it's put over the goal is not worth 2 points. Your system is simply objective when we know everything in soccer is situational based and subjective.

[sidefootsitter]

... Your system is simply objective when we know everything in soccer is situational based and subjective. First of all, you can grade it subjectively.

Second, if you have a solid system in place, then a bad good shot (or a good bad shot, if you will) will even out after a while.

Third, you can always adjust the award points.

[NoSix]

Your concept of "touches per minute" being valuable is highly questionable. I suspect this is why Lewis is given such a low score, because it wasn't because he turned the ball over. In fact he barely touched the ball at all.

Lewis had both the lowest quantity and the lowest quality of touches on the team, therefore the lowest rating. Seems perfectly reasonable to me.

Likewise Onweyu's passing was excellent, although he rarely went long. His only poor play that I could tell was a couple of avoidable fouls.


The quality of Onyewu's offensive touches was the best on the team; the quality of his defensive touches was slightly below the team average. With the proper weighting, I would anticipate that Onyewu will be in the top half of the final rankings.



I am interested to know how you rated Adu and Feilhaber so highly. Not that the rest of the world doesn't agree with you, but I thought Adu's performance in particular was subpar, and Feilhaber's was average at best.

If Adu dribbles the ball forward, and makes 15 touches with his feet as he dribbles forward, do you count that as 15 positive touches? Adu did have a couple of good dribble runs, but nevertheless he had more turnovers per minute than any US player. He did have one well taken corner, although it did not find a US target.

Dempsey did some spectacular things but overall had only an average at best game. He is another dribbler so perhaps this is skewing your results.


Yes, obviously the provisional ratings value dribble touches too highly, but the final ratings will weight them properly and Adu's and Dempsey's ratings relative to the rest of the team will be lower as a result.


I could see giving Corrales the score you did for his touches on the ball but his defense was not very good, and that is being kind. Are you trying to objectively rate defense, too?


Yes, I tabulate both offensive and defensive touches and plan to incorporate both in the final ratings. The quality of Corrales offensive touches was above the team average, while the quality of his defensive touches was below the team average. I would anticipate that relative to his teammates, Corrales's rating will drop in the final ratings.

The rating I would be most interested in hearing details on is Drew Moor. Many would agree with you that he had a poor game but I rated his passing as among the best on the team, and he had a considerable number of touches.


Moor had the 4th least number of touches per 90 min on the team, and the quality of his touches was only about average, so I think his low rating is deserved. Moor is a particularly interesting case, because it was the two touches that he didn't get that were most distressing to USA fans, and those will only indirectly be reflected in the ratings as a missed opportunity for positive touches. It may be that defensive touches in themselves are not adequate to explain the variability in goals conceded, but if that is the case, it will be reflected in the results of the analysis.

Your goalkeeper ratings should just be eliminated, because they are nonsensical.


Once saves are weighted properly relative to other defensive touches, I think they will be fine.

There are a lot of plays in soccer where you don't have the time or opportunity to direct the ball to a place where your team is likely to maintain possession--e.g. clearances, 50/50 balls, working out of pressure. Losing possession of the ball in these situations should not be counted against the player.

I would not even count Onyewu's "misclearance" to Rosicky in the World Cup as a bad touch. It wasn't a great clearance but only 1% of the time would a clearance like that have resulted in a goal. If you don't feel his clearance was good enough to count as a positive touch, just don't count it at all.


Based on feedback received after the Sweden match, I have decided to count clearances from inside the defensive penalty area to outside the penalty area as successful touches.

Twellman's initial touch to kick of the match should have no value at all. No professional soccer player should mishit a simple kick-off. Conversely, a "quality" goal like Robinson's should count for far more than just a simple pass.

Along the same lines, all of those unpressured possession passes along the back line by Mexico and sometimes by the US should count for nothing or practically nothing because it is very unlikely for a professional soccer player to mishit such a pass. And even if the pass is mishit, there is usually enough margin for error that the pass can be recovered. Counting these as positive touches doesn't seem very helpful.


Well, I have seen a goal scored directly from the kick-off, so it is not always the case that the first touch is inconsequential. I prefer to let the data determine which types of touches are most valuable, rather than make arbitrary assumptions apriori.

I also track where on the field a touch takes place, so whether or not, say, touches in the defensive half have no value in explaining goals scored is a testable hypothesis. If they are valuable I will keep them in, if not then I can eliminate them from the analysis (and not have to track them in future matches - yay!)

In any case, thanks for the input. Depending on the Poland result, I may be able to calculate the first set of "final" ratings after the next match.

[NoSix]

Come along, boys and girls.

A short pass - 1 point.

A long pass (completed) - 3 points.



The problem I see with your proposal is that the relative weights of different touches are arbitrary and subjective. In my approach, I hypothesize a relationship between different types of touches and goals scored and conceded and then determine the best fit weights from the data using standard regression techniques.

[PVancouver]

Isn't your analysis likely to indicate that the fewer touches by the defense and goalkeeper, the better?

[Mr Martin]

I commend the effort to rate each player's touches and contributions.

But something isn't right when Feilhaber and Adu, who only played the final 30 minutes (fresh legs against tired legs) and participated in a grand total of zero goals, scored over 100 "points" to top the squad. Adu's best play was drawing a free kick, and Feilhaber solidified the midfield, but hardly transformed it.

Meanwhile, Altidore and Onewu, the goal scorers and who each played the full 90 minutes against a very high-quality opponent, rated below Moor. Altidore was involved in three dangerous scoring chances, one called back for an iffy foul on Marquez, one of the world's best defenders. Oneywu was putting out fires in the back all game long, as well as being a dangerous force on set pieces. Meanwhile, Moor was widely understood to have been the primary US player responsible for allowing Mexico's goals, and a reason for many of the fires Oneywu had to put out.

It doesn't pass the "smell test". Something just doesn't add up here...

[miked9]

How does this system compare to the one that poster: illinoiszissou uses? Where is he on this thread?

[NoSix]

[QUOTE=Mr Martin;13857871]But something isn't right when Feilhaber and Adu, who only played the final 30 minutes (fresh legs against tired legs) and participated in a grand total of zero goals, scored over 100 "points" to top the squad. Adu's best play was drawing a free kick, and Feilhaber solidified the midfield, but hardly transformed it.

Meanwhile, Altidore and Onewu, the goal scorers and who each played the full 90 minutes against a very high-quality opponent, rated below Moor. Altidore was involved in three dangerous scoring chances, one called back for an iffy foul on Marquez, one of the world's best defenders. Oneywu was putting out fires in the back all game long, as well as being a dangerous force on set pieces. Meanwhile, Moor was widely understood to have been the primary US player responsible for allowing Mexico's goals, and a reason for many of the fires Oneywu had to put out.
QUOTE]

Feilhaber and Adu both had a lot of touches in their 28 minutes of action; their high provisional ratings reflect that. Obviously, Altidore's and Onyewu's goal scoring touches were more important than any of Feilhaber's or Adu's touches, but the provisional ratings don't capture that, because they weight all touches equally. The final ratings will weight more important touches more heavily, so Altidore and Onyewu will move up in the ratings relative to Feilhaber and Adu.

[NoSix]

Isn't your analysis likely to indicate that the fewer touches by the defense and goalkeeper, the better?

No, I would expect the more successful touches and the fewer unsuccessful touches the better, on both offense and defense.

[NoSix]

How does this system compare to the one that poster: illinoiszissou uses? Where is he on this thread?

I see two main differences. My system is complete in that all touches are tracked, and the weights for different types of touches are determined from the data rather than being chosen arbitrarily and subjectively.

[Mr Martin]

But something isn't right when Feilhaber and Adu, who only played the final 30 minutes (fresh legs against tired legs) and participated in a grand total of zero goals, scored over 100 "points" to top the squad. Adu's best play was drawing a free kick, and Feilhaber solidified the midfield, but hardly transformed it.

Meanwhile, Altidore and Onewu, the goal scorers and who each played the full 90 minutes against a very high-quality opponent, rated below Moor. Altidore was involved in three dangerous scoring chances, one called back for an iffy foul on Marquez, one of the world's best defenders. Oneywu was putting out fires in the back all game long, as well as being a dangerous force on set pieces. Meanwhile, Moor was widely understood to have been the primary US player responsible for allowing Mexico's goals, and a reason for many of the fires Oneywu had to put out.


Feilhaber and Adu both had a lot of touches in their 28 minutes of action; their high provisional ratings reflect that. Obviously, Altidore's and Onyewu's goal scoring touches were more important than any of Feilhaber's or Adu's touches, but the provisional ratings don't capture that, because they weight all touches equally. The final ratings will weight more important touches more heavily, so Altidore and Onyewu will move up in the ratings relative to Feilhaber and Adu.

So why post the provisional ratings if they are essentially meaningless without the weights?

Also, in your opening post, you wrote that comparisons may only be useful among players in similar positions. But even that doesn't work in the case of Convey and Donovan, who were playing the mirror image positions along the left and right midfield respectively. Convey gets more provisional points than Donovan, and there isn't anyone who would even conceive of rating Convey's poor performance remotely similar to Donovan's quality performance (particularly his excellent play in the 1st half, where he was the US's most dynamic attacker).

I look forward to seeing the weighted ratings.

[sidefootsitter]

The problem I see with your proposal is that the relative weights of different touches are arbitrary and subjective... Yes. The weights are arbitrary.

But if you don't think that the post-game totals make sense, then you can change the weights - should there be more for a clearance from inside own penalty box? ... should there be more for a tackle? more for an outside shot on goal? more for a goal? ... - etc.

Mind you, that this is how Opta gives out their awards/ratings. They have 25 various categories - both in offense and defense - to which the weights are assigned.

The reason for the above is due to the fact that different players do different things on the field and you can't based everything on their passing.

[NoSix]

So why post the provisional ratings if they are essentially meaningless without the weights?

Also, in your opening post, you wrote that comparisons may only be useful among players in similar positions. But even that doesn't work in the case of Convey and Donovan, who were playing the mirror image positions along the left and right midfield respectively. Convey gets more provisional points than Donovan, and there isn't anyone who would even conceive of rating Convey's poor performance remotely similar to Donovan's quality performance (particularly his excellent play in the 1st half, where he was the US's most dynamic attacker).

Because I don't think the provisional ratings are meaningless. Convey and Donovan are an excellent example. The raw data show that Donovan's quantity of touches was higher than Convey's, but Convey's quality of touches was higher. Since Convey only played 69 minutes, on a per 90 min basis his quantity of touches was also higher than Donovan's. Since the relative proportion of different types of touches seems to be about the same between Convey and Donovan, I don't think the weighting will affect their relative ratings much. I, for one, think the convential wisdom overrates Donovan's performance and underrates Convey's. An objective evaluation suggests, if anything, that Convey's performance was slightly better.

[soccertom]

As for the Sweden match, I have tracked every touch for each USA player in the recent match versus Mexico. Once I have collected sufficient data (4-5 matches worth), I will calculate statistical player ratings with different types of touches weighted according to their importance to scoring and conceding goals. In the interim, I am publishing a simplistic, provisional rating that is a function of both the quantity and quality of each player's touches per 90 minutes (all touches weighted equally). Given the obvious limitations of the provisional rating, comparisons between players are probably only useful for players who play the same position.



Nice work as usual NoSix. Thanks for sharing!