Moen and Halpern high QoC stats are mainly from specific defensive assignments and spot duty on the top two lines as wingers. Gomez's .484 Corsi RelQoC is pretty standard among players on the second matchup unit and its fair to say that is what he was sent out against most games.
He did? That would be some positive news if true. I hope the guy comes back hungry and puts up some points because he's got to be embarrassed after the season he just had. If he doesn't do anything in the off-season to step it up, I'm not sure how he can look at himself in the mirror.
You brought up the QoC argument that it was 7th on the team, I was pointing out with that they were still consistant with that of a 2nd line matchup (maybe more towards 2nd with a little bit of 3rd line). The best place to get an easy visual look at who he was matching up with his here on Oliver's most excellent blog (use google translate for English) http://enattendantlesnordiques.blogs...bel-outil.html
Gomez's competition trends much more towards the higher than the lower.
You brought up the QoC argument that it was 7th on the team, I was pointing out with that they were still consistant with that of a 2nd line matchup (maybe more towards 2nd with a little bit of 3rd line). The best place to get an easy visual look at who he was matching up with his here on Oliver's most excellent blog (use google translate for English) http://enattendantlesnordiques.blogs...bel-outil.html
Gomez's competition trends much more towards the higher than the lower.
I don't take that seriously because there's no definition of what is considered a group X player. He pretty much says he toyed with the stats without any real direction.
It also says AK46 had a tougher job than Pleks, which is obviously a joke. According to this, Gomez had the toughest opposition along with Gionta, which clearly wasn't the case.
Nice attempt by the guy, but it's inconclusive at best.
I don't take that seriously because there's no definition of what is considered a group X player. He pretty much says he toyed with the stats without any real direction.
It also says AK46 had a tougher job than Pleks, which is obviously a joke. According to this, Gomez had the toughest opposition, which clearly wasn't the case.
Nice attempt by the guy, but it's inconclusive at best.
Keep reading, the explaination is at the end. Essentially they are determined largely the same way CorsiRelQoC is by taking the RelCoris of opposing players. CorsiRelQoC is probably the most reliable QoC measure out there (although it certainly has its hits and misses) so its a valid way of doing things.
I have no idea where you got that Kostitsyn had a tougher job than Plekanec according to that chart. The table shows Plekanec having the toughest job there is for a Habs forward. Plekanec had the most events against the top group by a long shot, 598 to 470 for AK with less total events. The guys with the most porportion of events against the top and top 2 groups are Pacioretty, Plekanec, Cammaleri, Gionta, Gomez, Kostitsyn in that order which is pretty much what I would expect from watching the games.
Keep reading, the explaination is at the end. Essentially they are determined largely the same way CorsiRelQoC is by taking the RelCoris of opposing players. CorsiRelQoC is probably the most reliable QoC measure out there (although it certainly has its hits and misses) so its a valid way of doing things.
I have no idea where you got that Kostitsyn had a tougher job than Plekanec according to that chart. The table shows Plekanec having the toughest job there is for a Habs forward. Plekanec had the most events against the top group by a long shot, 598 to 470 for AK with less total events. The guys with the most porportion of events against the top and top 2 groups are Pacioretty, Plekanec, Cammaleri, Gionta, Gomez, Kostitsyn in that order which is pretty much what I would expect from watching the games.
He also says he wasn't sure it was the best way to do it.
But that's regardless. I just checked the last board and it makes a bit more sense. Basically, Gomez got **** on by good players and outchanced scrubs. Sounds about right.
But that just adds on to the +/- per 60. It's pretty consistent with it. But once again, I'm not sure how he determines the groups. He'd need more groups to be more accurate considering his values go from -3 to 3.
But that just adds on to the +/- per 60. It's pretty consistent with it. But once again, I'm not sure how he determines the groups. He'd need more groups to be more accurate considering his values go from -3 to 3.
Not being sure that it's the best way to do it is not a sign of weakness in the model.
Having more groups would cause too many cases where the number of events would be too low for the result to be relevant. The same thing already affect a guy like White who only played 63 events against Group E. That's not a lot to make any kind of conclusion.
If you read the comments, at one point he mentions someone else proposal to use 4 groups, for 4 lines. It does make some intuitive sense.
In the end, how you separate these groups can only be arbitrary. The same way that a survey would arbitrary separate the 24-28 and the 30-36 years old. It's arbitrary, but it doesn't mean it doesn't lead to useful results.
Not being sure that it's the best way to do it is not a sign of weakness in the model.
Having more groups would cause too many cases where the number of events would be too low for the result to be relevant. The same thing already affect a guy like White who only played 63 events against Group E. That's not a lot to make any kind of conclusion.
If you read the comments, at one point he mentions someone else proposal to use 4 groups, for 4 lines. It does make some intuitive sense.
In the end, how you separate these groups can only be arbitrary. The same way that a survey would arbitrary separate the 24-28 and the 30-36 years old. It's arbitrary, but it doesn't mean it doesn't lead to useful results.
As I said, it's consistent with the +/- per 60 and obviously the other stats he used to make the model.
It pretty much shows he was outmatched against top players (don't know what his score compares to group E, which is frustrating), was par against group D (2nd liners??? About same value???), and did decent against lower end competition.
But overall was the least effective of all the top 6 forwards in outchancing the opposing team.
He also says he wasn't sure it was the best way to do it.
But that's regardless. I just checked the last board and it makes a bit more sense. Basically, Gomez got **** on by good players and outchanced scrubs. Sounds about right.
But that just adds on to the +/- per 60. It's pretty consistent with it. But once again, I'm not sure how he determines the groups. He'd need more groups to be more accurate considering his values go from -3 to 3.
Its pretty rough but the results largely make sense. It also jives with how Martin has always worked in Montreal with the top 6 matching other team's top six with occasitional help from some defensive specialists. This is why Moen and Halpern got top six time on Gomez and Plekanec's wings as those lines were to be used in defensive matchups against top lines.
Groups are basically formed from their ranking of a player on their respective team by their CorsiRelQoC and taken as the standard deviation from the team average.
The values are decimal numbers between -3 to 3 according to how many standard deviations they were from the team mean value with a player one standard deviation greater than the mean being at 1 (would be at about the 83rd percentile of players), or a player two standard deviations below the mean at -2 (about 5th percentile of players.) Numbers greater that 3 are not needed as that would be in the upper fraction of one percent. This effectively makes a league wide list of which players a team considers their best matchup players by how much and is used to form groups an even 5 groups from top to bottom. The method is pretty noisy because of the many different line use strategies teams employ but the chance data shows that it is a fair approximation of quality of player with enough players.
Gomez was break even or better on chances for everyone except the very top group and 47% isn't that bad. The really reason for the bad +/- is the hideously bad chance conversion when he was on ice, not just from his own stick but everyone else on ice.
Like I've said elsewhere, Gomez's case looks an awfully lot like Patrick Marleau of 07-08 where a rediculously low on ice team shooting percentage of 5.12%, turned his even strength point total to 22 from 41 the year before. Compare to Gomez's on ice team shooting percentage dropping from 8.51 (around league average) to 4.73 taking his ES point total from a solid 40 to a dismal 20.
The next season Marleau's team shooting percentage regressed to the mean like it usually does to 8.64 (above average) and he got 46 even strength points. Its likely that Gomez's next season will follow the same pattern.
The winner of the thread title contest is Paul Dipietro with "Rancho Relaxo"
For those who didn't win (and there were some great ones), you know there'll be a 5th edition soon enough, so keep thinking! (Keep it atleast quasi PC though, the admins are watching me)
Its pretty rough but the results largely make sense. It also jives with how Martin has always worked in Montreal with the top 6 matching other team's top six with occasitional help from some defensive specialists. This is why Moen and Halpern got top six time on Gomez and Plekanec's wings as those lines were to be used in defensive matchups against top lines.
Groups are basically formed from their ranking of a player on their respective team by their CorsiRelQoC and taken as the standard deviation from the team average.
The values are decimal numbers between -3 to 3 according to how many standard deviations they were from the team mean value with a player one standard deviation greater than the mean being at 1 (would be at about the 83rd percentile of players), or a player two standard deviations below the mean at -2 (about 5th percentile of players.) Numbers greater that 3 are not needed as that would be in the upper fraction of one percent. This effectively makes a league wide list of which players a team considers their best matchup players by how much and is used to form groups an even 5 groups from top to bottom. The method is pretty noisy because of the many different line use strategies teams employ but the chance data shows that it is a fair approximation of quality of player with enough players.
Gomez was break even or better on chances for everyone except the very top group and 47% isn't that bad. The really reason for the bad +/- is the hideously bad chance conversion when he was on ice, not just from his own stick but everyone else on ice.
Like I've said elsewhere, Gomez's case looks an awfully lot like Patrick Marleau of 07-08 where a rediculously low on ice team shooting percentage of 5.12%, turned his even strength point total to 22 from 41 the year before. Compare to Gomez's on ice team shooting percentage dropping from 8.51 (around league average) to 4.73 taking his ES point total from a solid 40 to a dismal 20.
The next season Marleau's team shooting percentage regressed to the mean like it usually does to 8.64 (above average) and he got 46 even strength points. Its likely that Gomez's next season will follow the same pattern.
I still don't get how that affects his defense (not to mention the low S% while he's on the ice can very well be a direct result of his play).
It was putrid. Stats say it. The show he put on the ice say it.
As I said, it's consistent with the +/- per 60 and obviously the other stats he used to make the model.
It pretty much shows he was outmatched against top players (don't know what his score compares to group E, which is frustrating), was par against group D (2nd liners??? About same value???), and did decent against lower end competition.
But overall was the least effective of all the top 6 forwards in outchancing the opposing team.
He was probably the weakest overall at outchancing Kostisyn usually had the benefit of playing with the best and 2nd or 3rd best forward while Gomez was usually counted on to be a leader on his line so Gomez may have been better than him. By outchancing against top level guys he was still better than anyone in the bottom six except the curious case of Desharnias who might have been playing over his head over a short period of time.
I compared this to last season Gomez was the chance% leader at even strength for centers and behind only Cammaleri and Gionta for forwards. Looks like Gomez had a somewhat down season in play and a huge down season in luck.
Would you say that the math is flimsy because it's an attempt to depict something that isn't describable by math, or because it's a failed attempt at depicting something that could, indeed, be understood trough (correct) maths?
FWIW, I'm the guy who put this whole thing together and I'm really interested to hear criticism of my work. That's why I put it out there, after all.
He was probably the weakest overall at outchancing Kostisyn usually had the benefit of playing with the best and 2nd or 3rd best forward while Gomez was usually counted on to be a leader on his line so Gomez may have been better than him. By outchancing against top level guys he was still better than anyone in the bottom six except the curious case of Desharnias who might have been playing over his head over a short period of time.
I compared this to last season Gomez was the chance% leader at even strength for centers and behind only Cammaleri and Gionta for forwards. Looks like Gomez had a somewhat down season in play and a huge down season in luck.
Saying 'he had the chance', when Cammalleri is far from being good on D. Kostitsyn's stats seem skewed though, because he's better against top end players than Gomez, yet his % is thrown off by Group B forwards. I assume that's from playing with offensively inept players on his trips to the lower lines.
Once again, 'luck'. Find me legit top 6 players that end up with both third line-type production and the lowest +/- per 60 on their team, all with top 6 ice-time and usually some top 6 linemates. Luck is an excuse for people who don't try.
Quote:
Originally Posted by MathMan
That's the problem. You say his defense was "putrid". The stats actually say no such thing.
I'm sorry. He's average to below average. The general jist about Gomez is that he's solid defensively. So he was putrid by that standard.
Would you say that the math is flimsy because it's an attempt to depict something that isn't describable by math, or because it's a failed attempt at depicting something that could, indeed, be understood trough (correct) maths?
FWIW, I'm the guy who put this whole thing together and I'm really interested to hear criticism of my work. That's why I put it out there, after all.
I entirely read your post, once, and think I understood the majority of it.
First, congratulation, that's a lot of work. I didn't saw obvious flaw. I'm curious about the tools you used (beyond your data sources), as I sincerely hope you did not compute all that by hand.
I think the only weak point of your model (as far as I understand it) is still how you compare different teams CQRC, normalized or not. I'm not convince it's possible to compare players from different team with a microstats that is purely team-centric at its base, although that over the run of a season, it may not make a big difference overall.
Amusingly, I once used standard-deviation calculation based on the "typical" hockey stats (G, A, P, etc.) but in the context of a fantasy hockey leagues where players statistics had to change depending on their seasons performances... obviously our model was way simpler, and didn't have the benefit the modern hockey microstats bring.
Nobody "cast aside" last year performances: we are just trying to explain it.
When you look at his microstats, the explanation that appears isn't that Gomez markedly slowed down. He's still producing more or less as many chances as he used to. But for some reason, his shot percentage and those of his teammates were noticeably lower last year than their respective average. And since shot percentage is a statistic that tend to vary widely for no other reason than pure randomness, there is some justified expectation that he and his wingers will probably bounce back.
Not entirely correct. Quality of shots counts. And good quality chances come from good passes, and yes, also good anticipation and positioning by wingers.
Both were lacking on Gomer's lines for long stretches last year.
