HFBoards

Go Back   HFBoards > General Hockey Discussion > By The Numbers
Mobile Hockey's Future Become a Sponsor Site Rules Support Forum vBookie Page 2
By The Numbers Hockey Analytics... the Final Frontier. Explore strange new worlds, to seek out new algorithms, to boldly go where no one has gone before.

With or Without You: Mario Lemieux

Reply
 
Thread Tools
Old
07-09-2012, 11:53 PM
  #26
Czech Your Math
Registered User
 
Czech Your Math's Avatar
 
Join Date: Jan 2006
Location: bohemia
Country: Czech_ Republic
Posts: 3,538
vCash: 50
Quote:
Originally Posted by overpass View Post
Don't worry about the French article, it was just a general presentation of the concept, not a detailed example of my method. Going from memory of what I did over 2 years ago...

I used the exponent 2. You raise some interesting points there, but I kept it simple.

Yes, I calculated the effect for each season, I weighted each season's effect by the square of either GP (with) or GP (without), whichever was lower. So seasons in which the player played roughly half the team games are weighted the most, and seasons where they played almost all or almost no games receive a very low weight.
I definitely concur with keeping it simple, especially in a study with limited samples from which limited conclusions can be drawn.
However, one should be careful not to draw conclusions about "clutch" ability from differential in actual vs. pythagorean expected win%, since the exponent is important in such calculations.

Czech Your Math is offline   Reply With Quote
Old
07-11-2012, 10:24 AM
  #27
plusandminus
Registered User
 
Join Date: Mar 2011
Posts: 868
vCash: 500
Like I've said before, I've done this for every player during the last 25 seasons. I also studied certain players more in detail. (It received basically no interest.) I also took it further, by looking at "expected game outcome" (i.e. based on which teams they played against, and the starting goalies), but didn't finish it.

As for MacInnis, it could be of interest to look for when Pronger was injured. If I remember right, they had similar effect, not to mention when both were away. Without both, St Louis was among the worst teams. With both, I think they were best or at least top 3.
As for Forsberg, yes about +.10 percentage units in win percentage. Similar to Sakic.
As for 1992-93 Gretzky, his presense during the regular season didn't seem to help his team win more. Corey Millen, which I can understand, seemed to add significantly more. The 1992-93 LAK is the team I probably studied the most. A very interesting season for them statistically, with Kurri starting out producing lots of points without Gretzky, before suddenly fading, and Sandstrom seemingly being a player significantly helping the team win.
As more Mario, the 2002-03 Pittsburgh is very special, as they started the season great having several star players. But as those players left one by one, the team started to lose. Very big difference between PIT at start of season, and PIT during last half of the season. This affects stats a lot too.
Doing per season listings, and not adjusting for opposition, goalies and injuries of other players, the results partly "makes sense" (i.e. "great" players atop), and partly not (i.e. non-great players atop).
As I've written before, Pavol Demitra - like it or not - stands out as a player whose teams usually had considerably better stats with him than without him (and it's true for several of the teams he played on). Like it or not, but it was a consistant pattern, and he often changed team and often was injured.

I think Nicklas Lidstrom came out atop of any other player (and this was done before the 2011-12 season), perhaps with +.20 or so. But one should be aware to handle games late in the season wisely, as for example Detroit sometimes rested players (including Lidstrom, thus "biasing" his stats as Detroit fairly often lost those games).

Hasek stood out among goalies, generally and specifically speaking (having also considered the opposition). If I remember right, Roy also had good stats.

I think +.10 percentage units should be regarded as very good. Turning a team from .45 (missing playoffs) to .55 (making playoffs) is quite valuable. +.15 is great, potentially turing an average team into one of the elite teams. (That said, the percentages one gets doesn't necessarily be totally reflecting the "true value" of the player, as things like randomness, circumstances, etc plays a part.)

plusandminus is offline   Reply With Quote
Old
07-11-2012, 01:14 PM
  #28
Czech Your Math
Registered User
 
Czech Your Math's Avatar
 
Join Date: Jan 2006
Location: bohemia
Country: Czech_ Republic
Posts: 3,538
vCash: 50
Quote:
Originally Posted by plusandminus View Post
Like I've said before, I've done this for every player during the last 25 seasons. I also studied certain players more in detail. (It received basically no interest.) I also took it further, by looking at "expected game outcome" (i.e. based on which teams they played against, and the starting goalies), but didn't finish it.

As for MacInnis, it could be of interest to look for when Pronger was injured. If I remember right, they had similar effect, not to mention when both were away. Without both, St Louis was among the worst teams. With both, I think they were best or at least top 3.
As for Forsberg, yes about +.10 percentage units in win percentage. Similar to Sakic.
As for 1992-93 Gretzky, his presense during the regular season didn't seem to help his team win more. Corey Millen, which I can understand, seemed to add significantly more. The 1992-93 LAK is the team I probably studied the most. A very interesting season for them statistically, with Kurri starting out producing lots of points without Gretzky, before suddenly fading, and Sandstrom seemingly being a player significantly helping the team win.
As more Mario, the 2002-03 Pittsburgh is very special, as they started the season great having several star players. But as those players left one by one, the team started to lose. Very big difference between PIT at start of season, and PIT during last half of the season. This affects stats a lot too.
Doing per season listings, and not adjusting for opposition, goalies and injuries of other players, the results partly "makes sense" (i.e. "great" players atop), and partly not (i.e. non-great players atop).
As I've written before, Pavol Demitra - like it or not - stands out as a player whose teams usually had considerably better stats with him than without him (and it's true for several of the teams he played on). Like it or not, but it was a consistant pattern, and he often changed team and often was injured.

I think Nicklas Lidstrom came out atop of any other player (and this was done before the 2011-12 season), perhaps with +.20 or so. But one should be aware to handle games late in the season wisely, as for example Detroit sometimes rested players (including Lidstrom, thus "biasing" his stats as Detroit fairly often lost those games).

Hasek stood out among goalies, generally and specifically speaking (having also considered the opposition). If I remember right, Roy also had good stats.

I think +.10 percentage units should be regarded as very good. Turning a team from .45 (missing playoffs) to .55 (making playoffs) is quite valuable. +.15 is great, potentially turing an average team into one of the elite teams. (That said, the percentages one gets doesn't necessarily be totally reflecting the "true value" of the player, as things like randomness, circumstances, etc plays a part.)
I'm not sure all are using the same metric, as at least during Sakic's prime, there was virtually no difference w/ or w/o him.

I like the simplicity of the metric and it's directness in measuring a player's impact. However, as you point out, there are many factors that could influence the results. Another problem is that it only measures the effect in seasons in which the player was injured. This almost or does exclude many seasons for each player and biases the sample for those seasons when the player was injured significantly and therefore may not be at his best.

It's asked ITT about Crosby. The team did worse without him in years previous to the last, but the effects weren't on the scale of many other forwards of the past 30 years, even in those years. One might also consider that while Crosby's point production was outstanding in his brief return, after such a long layoff, he may not have been fully "back" in terms of impact, esp. given that he must be wary of taking another big hit. In general, I think there's enough data to probably conclude that the team was worse without him, but not nearly as much as some would think. This may be influenced by Malkin's presence, but the effect is still there, although I wouldn't make a definitive conclusion on such limited evidence.

I would be hesitant to use Lidstrom in such a study, since he only missed more than 6 games in a season in his final year, and missed only 44 games in his career. You mention he was frequently rested (near the playoffs?) and that being an important factor. That's why it is usually best if the players misses at least a few games in multiple seasons, so that the sample is less likely to contain such biases.

Czech Your Math is offline   Reply With Quote
Old
07-11-2012, 03:47 PM
  #29
plusandminus
Registered User
 
Join Date: Mar 2011
Posts: 868
vCash: 500
Quote:
Originally Posted by Czech Your Math View Post
I'm not sure all are using the same metric, as at least during Sakic's prime, there was virtually no difference w/ or w/o him.
I've explained my method earlier. It combines factual game outcome (win, loss, draw, etc - I think it may be called "decision" in English) with pythagoran GF and GA (probably using 2.0 as the expondent). I also had a proven way of dealing with low numbers of games, sort of adjusting them to "average". I myself trust it, and think that the research I've done (remember I have done this for every player during all the last 25 or so seasons) supports it.

Anyway, the seasons 1997-98, 1998-99 and 1999-2000 are prime examples of where Sakic seemed to raise Colorado by just above +.10. With him, his team appeared quite consistantly about .59, and without him just below .500. (That in itself is a bit interesting considering the well-known players Colorado had, and the fact that they had above average goaltending.)

Quote:
Originally Posted by Czech Your Math View Post
I like the simplicity of the metric and it's directness in measuring a player's impact. However, as you point out, there are many factors that could influence the results. Another problem is that it only measures the effect in seasons in which the player was injured. This almost or does exclude many seasons for each player and biases the sample for those seasons when the player was injured significantly and therefore may not be at his best.
Yes it has many limitations. I don't think it can be used for ranking players based on how "good" or valuable they were. But it does give knowledge into the effect different players presense (or injury) can have on teams. And it can help in getting the overall picture. Things like +/- are very depedant on the strength of the teammates, and is also affected by how a player is used (for example against top opponents). You probably know that I did spend hundreds of hours focusing on +/- and similar things. Combining that with this kind of studies adds information.

Quote:
Originally Posted by Czech Your Math View Post
It's asked ITT about Crosby. The team did worse without him in years previous to the last, but the effects weren't on the scale of many other forwards of the past 30 years, even in those years. One might also consider that while Crosby's point production was outstanding in his brief return, after such a long layoff, he may not have been fully "back" in terms of impact, esp. given that he must be wary of taking another big hit. In general, I think there's enough data to probably conclude that the team was worse without him, but not nearly as much as some would think. This may be influenced by Malkin's presence, but the effect is still there, although I wouldn't make a definitive conclusion on such limited evidence.
It seems I got 2010-11 Pittsburgh with Crosby at about .63, during first half of the season. During the second half, without him, they were about .49. So it does seem he helped them significantly. (Him missing half the season is sort if pretty "ideal" for this typ of study.)

I agree completely with players sometimes not being "in top shape" (or how to say it in English) when they play. There are many players being affected by injuries that aren't being publicly spoken about. A player may receive pretty much criticism for "underperforming", while in fact maybe - under the circumstances - doing quite heroic performances.
Injuries is a very sad part of hockey. There are so many great players whose careers have been badly affected. Orr, Mario. I remember news articles about Gretzky being a shadow of his former self (due to injuries, I think mainly his back?), while still continuing to lead the league in assists. Forsberg is another case. Lindros.

Quote:
Originally Posted by Czech Your Math View Post
I would be hesitant to use Lidstrom in such a study, since he only missed more than 6 games in a season in his final year, and missed only 44 games in his career. You mention he was frequently rested (near the playoffs?) and that being an important factor. That's why it is usually best if the players misses at least a few games in multiple seasons, so that the sample is less likely to contain such biases.
Yes, Lidstrom was sometimes rested during the last, or last two, games of the regular season. I developed an "application" for keeping track of such things too, instantly seeing for every game which players on a team was playing or not.
Of course I agree with what you wrote (it was even I who mentioned Lidstrom, and missing games late in the season, as a "bias").

plusandminus is offline   Reply With Quote
Old
07-11-2012, 04:09 PM
  #30
Czech Your Math
Registered User
 
Czech Your Math's Avatar
 
Join Date: Jan 2006
Location: bohemia
Country: Czech_ Republic
Posts: 3,538
vCash: 50
Quote:
Originally Posted by plusandminus View Post
I've explained my method earlier. It combines factual game outcome (win, loss, draw, etc - I think it may be called "decision" in English) with pythagoran GF and GA (probably using 2.0 as the expondent). I also had a proven way of dealing with low numbers of games, sort of adjusting them to "average". I myself trust it, and think that the research I've done (remember I have done this for every player during all the last 25 or so seasons) supports it.
I think the pythagorean GF/GA data is important in to some degree supporting the effect on win % and interesting in detailing the effect and its reasons. The matter of exponent is likely a small complication in this, if one is calculating expected win% from such data.

IIRC, some your work was more within games themselves, and used per-minute type data to some degree. If this is correct, I'm not sure how comparable this would be to metrics using whole games, but could be interesting nonetheless.

Quote:
Originally Posted by plusandminus View Post
Anyway, the seasons 1997-98, 1998-99 and 1999-2000 are prime examples of where Sakic seemed to raise Colorado by just above +.10. With him, his team appeared quite consistantly about .59, and without him just below .500. (That in itself is a bit interesting considering the well-known players Colorado had, and the fact that they had above average goaltending.)
This is a good point, that individual seasons should also be examined. In Sakic's case, the effect from '98-00 is still less than others, and is neutralized by his '97 and '03 data.


Quote:
Originally Posted by plusandminus View Post
Yes it has many limitations. I don't think it can be used for ranking players based on how "good" or valuable they were. But it does give knowledge into the effect different players presense (or injury) can have on teams. And it can help in getting the overall picture. Things like +/- are very depedant on the strength of the teammates, and is also affected by how a player is used (for example against top opponents). You probably know that I did spend hundreds of hours focusing on +/- and similar things. Combining that with this kind of studies adds information.
I agree with you mostly. I think adjusted plus-minus is the better metric, given it's wider availability and much larger sample of data. I understand that players are utilized differently, but the metric captures something very important: how much better the team was with or without the player on the ice at even strength. There are complicating factors, as in most metrics. In such cases, it is important to look at other data and look at similar players to see if what the metric suggests is reinforced or contradicted by other data and/or the same metric for other similar players.

Quote:
Originally Posted by plusandminus View Post
It seems I got 2010-11 Pittsburgh with Crosby at about .63, during first half of the season. During the second half, without him, they were about .49. So it does seem he helped them significantly. (Him missing half the season is sort if pretty "ideal" for this typ of study.)
I have .63 with and .56 without, but the difference could be in how we count SO/OT W/L. I counted all W/L the same, none as ties, since it would take much more time to differentiate between the two (no info about SO/OT in player game logs on HR).

Quote:
Originally Posted by plusandminus View Post
I agree completely with players sometimes not being "in top shape" (or how to say it in English) when they play. There are many players being affected by injuries that aren't being publicly spoken about. A player may receive pretty much criticism for "underperforming", while in fact maybe - under the circumstances - doing quite heroic performances.
This is true, and one reason per-game data and per-minute data must often be taken with a grain of salt. It's one thing for a player to miss a few games or 1/4 season on occasion and perform at a similar level in the past/future. It's another when a player suddenly performs at a higher level for a half season or less (see Kariya, Bure, Crosby, etc.). Forsberg too, had by far his best season after a full season off, which must be considered IMO. In the end, we can most fairly evaluate what players did, and much less so how they would have done, although external circumstances and differing environments should also be considered.

Czech Your Math is offline   Reply With Quote
Old
07-11-2012, 05:47 PM
  #31
overpass
Registered User
 
Join Date: Jun 2007
Posts: 3,517
vCash: 500
Quote:
Originally Posted by plusandminus View Post
I think Nicklas Lidstrom came out atop of any other player (and this was done before the 2011-12 season), perhaps with +.20 or so. But one should be aware to handle games late in the season wisely, as for example Detroit sometimes rested players (including Lidstrom, thus "biasing" his stats as Detroit fairly often lost those games).
Going back to the OP, Mario Lemieux in 1995-96 was another interesting case. Many of his missed games were the second half of road back to backs, where he was resting. As those games are the most difficult to win, his "without" games that season were far from representative.

I'm not aware of other players being used in the same way, but when players sit games out in this way it causes major problems for a WOWY analysis.

overpass is offline   Reply With Quote
Old
07-11-2012, 06:12 PM
  #32
plusandminus
Registered User
 
Join Date: Mar 2011
Posts: 868
vCash: 500
Quote:
Originally Posted by overpass View Post
Going back to the OP, Mario Lemieux in 1995-96 was another interesting case. Many of his missed games were the second half of road back to backs, where he was resting. As those games are the most difficult to win, his "without" games that season were far from representative.

I'm not aware of other players being used in the same way, but when players sit games out in this way it causes major problems for a WOWY analysis.
Good point. I've seen other cases too, maybe not as extreme, but anyway. Certain goalies. I wrote about it in some thread, probably during a discussion with Hockey Outsider (?), but I don't remember which thread.

Edit: The above is also a prime example of where it would be of value to pay attention to schedule (opposition, home/away), to see how "easy" or "hard" the games might have been expected to win.

plusandminus is offline   Reply With Quote
Old
07-11-2012, 06:14 PM
  #33
plusandminus
Registered User
 
Join Date: Mar 2011
Posts: 868
vCash: 500
Quote:
Originally Posted by Czech Your Math View Post
I think the pythagorean GF/GA data is important in to some degree supporting the effect on win % and interesting in detailing the effect and its reasons. The matter of exponent is likely a small complication in this, if one is calculating expected win% from such data.
In a way, I think factual won-loss stats are what should be used. A win is a win, no matter if it's by 2-1 or 7-2.
In another way, I like the nuances one gets by using GF and GA to calculate pythagoran win %.
So I combined the two, making them equally important, rather than doing two separate studies. (I had already done some "boring" research of more mathematical nature, sort of like I tried to suggest in your stickied thread, so I felt confident I knew what I was doing.)


One can think more, deeply think more, about the above two things. On a team level, focus is basically only on it's about factual win %. On a player level, most stats focus on GF and GA related things. For example, +/- is GF and GA related, while points scored is GF related.
So our apprach to stats may be seen as partly contradictory. Player stats focus a lot on team GD (GF-GA). The better team goal difference, the more beneficial for the players' +/-, and vice versa. While on the other hand, on a team level, goal difference is (especially in the NHL) made relatively unimportant.
The "best" for a player (skater), statistically, is when his team gets as good goal difference as possible, because that will be benificial to his +/-, and the more GF his team scores, the more beneficial to his scoring stats. But a team doesn't necessarily care for those things, because they just want to win, or get into a penalty shootout, and to gain as many points as possible. If a team leads by several goals, they may bench(?) certain players.
Okay, the above two things often may go hand in hand. But basically we often treat player stats (like +/-) as if the main goal of the NHL teams is to gain as good goal difference as possible, when it's actually not.


Regarding the exponent, I think the key is to know what one is doing and what effects different exponents have. This too includes "boring" mathematical studies. For example, it matters how far from "average" the results ends up. I can't really express this properly in English, but let's have two series where one has a wider spread from average than the other. Then the highest results from that series will be higher than the highest from the other series, and the lowest results will be lower. (Oh, if this is not understandable, please ignore it, as I can't express myself propoerly.)

Quote:
Originally Posted by Czech Your Math View Post
IIRC, some your work was more within games themselves, and used per-minute type data to some degree. If this is correct, I'm not sure how comparable this would be to metrics using whole games, but could be interesting nonetheless.
No, I used whole games.

But now that you mention it, I'm open to experiment with including icetime as well. But it's not a priority, I'm not sure about it's real meaningfulness, and I won't do that within any near coming future


Quote:
Originally Posted by Czech Your Math View Post
This isa good point, that individual seasons should also be examined. In Sakic's case, the effect from '98-00 is still less than others, and is neutralized by his '97 and '03 data.
With examining individual seasons, I not only meant looking at the results for those seasons. I rather meant taking the time to also manually look up what the roster looked like for each game. Like a table with player names on one axis, and game numbers on the other, so that one can easily see which players were out simultanously, when players got traded, etc. I can get that info within 2-3 seconds, just having to enter season and team and push a button, and I hope a site like hockeyreference will soon show that kind of info too.

(There are so many things that could be done, and I've thought about starting a stats website, providing things that aren't very easily conducted/found, but decided it would probably not be worth the effort and cost. A boring Stanley Cup playoffs, followed by an exciting Football European Cup/Championship even further convinced me.)


Quote:
Originally Posted by Czech Your Math View Post
I agree with you mostly. I think adjusted plus-minus is the better metric, given it's wider availability and much larger sample of data. I understand that players are utilized differently, but the metric captures something very important: how much better the team was with or without the player on the ice at even strength. There are complicating factors, as in most metrics. In such cases, it is important to look at other data and look at similar players to see if what the metric suggests is reinforced or contradicted by other data and/or the same metric for other similar players.
I've actually become more and more skeptical towards +/-, including traditional attempts at adjusting it. Anyway, I think that if one wants to adjust +/-, one should start by trying to get rid of the effect of goaltending.

If one should manage to "adjust away" the influence of goaltending, then team stats would look differently. The team GF might perhaps be affected by a goal or so, but the team GA would be affected tremendously!

Colorado, taking away the effect of Roy, would perhaps suddenly look pretty average, rather than like the constant elite team they were. That despite having guys like Sakic, Forsberg, and a bunch of great defencemen (Blake, Bourque, Foote...). Forsberg's and Hejduk's great league leading +/- from 2002-03 would not be as impressive.

The above is another case that I think needs "boring" mathematical research. One needs to methodously(?) first learn to isolate the goalie effect, something that would likely take lots of time and effort and not resulting in very publicly appealing results.
Of course, it might be a very difficult study too. For example, to what extend was a certain team's goaltending stats affected by its playing system and skater performance, and vice versa?

IF one would manage to "adjust away" goalie influence, I said that team stats would look differently. And here one might also want to think more deeply about that. For example, we again can focus on the thing at the beginning of this reply, namely factual win % vs pythagoran win % (based on GF and GA), and also the "conflict" between what teams want and what players "want" regarding their stats. We likely would need to face arbitrariness, as well as the fact that we would be looking at "goalie riddened" stats from a World that was included goalies as an important part of the game.
Example: "See, Colorado without Roy actually were quite average. That is, their skaters as a whole performed average. And Forsberg (bad example, but anyway) actually looks only slightly better than the average player, rather than like a candidate for the best/MVP player." But, things were as they were, and we actually don't know how things would have turned if they had an average goalie instead of Roy.


Quote:
Originally Posted by Czech Your Math View Post
I have .63 with and .56 without, but the difference could be in how we count SO/OT W/L. I counted all W/L the same, none as ties, since it would take much more time to differentiate between the two (no info about SO/OT in player game logs on HR).
I think I used decimals. 2 for 60min win, 0 for 60min loss. Maybe 1.333 for overtime win , 0.667 for overtime loss. Maybe 1 for OT draw, thus ignoring the shootout result (although I did include it when examining the goalies).


Quote:
Originally Posted by Czech Your Math View Post
This is true, and one reason per-game data and per-minute data must often be taken with a grain of salt. It's one thing for a player to miss a few games or 1/4 season on occasion and perform at a similar level in the past/future. It's another when a player suddenly performs at a higher level for a half season or less (see Kariya, Bure, Crosby, etc.). Forsberg too, had by far his best season after a full season off, which must be considered IMO. In the end, we can most fairly evaluate what players did, and much less so how they would have done, although external circumstances and differing environments should also be considered.
Yes, I think we basically have to judge players based on their (f?)actual performance. Not try to adjust for "healthyness" or "degree of injury".

By the way, Forsberg himself recently stated something like... First he had his techical skills, and his strong desire to fight and win. Then he had a time when he could add an increased understanding of the game, and this coincided with his Hart win and his biggest playoffs scoring success. Then his body got worse, and he had to rely on his understanding of the game. (I'm not sure I agree, as I think he "always" understood the game greatly, but anyway, sort of.)

plusandminus is offline   Reply With Quote
Old
07-11-2012, 06:16 PM
  #34
overpass
Registered User
 
Join Date: Jun 2007
Posts: 3,517
vCash: 500
Quote:
Originally Posted by plusandminus View Post
Good point. I've seen other cases too, maybe not as extreme, but anyway. Certain goalies. I wrote about it in some thread, probably during a discussion with Hockey Outsider (?), but I don't remember which thread.
Right, some goalies have had that usage. Jacques Plante in 1970-71, among others.

overpass is offline   Reply With Quote
Old
07-11-2012, 06:44 PM
  #35
seventieslord
Moderator
 
seventieslord's Avatar
 
Join Date: Mar 2006
Location: Regina, SK
Country: Canada
Posts: 23,877
vCash: 500
Quote:
Originally Posted by plusandminus View Post
I've actually become more and more skeptical towards +/-, including traditional attempts at adjusting it. Anyway, I think that if one wants to adjust +/-, one should start by trying to get rid of the effect of goaltending.

If one should manage to "adjust away" the influence of goaltending, then team stats would look differently. The team GF might perhaps be affected by a goal or so, but the team GA would be affected tremendously!

Colorado, taking away the effect of Roy, would perhaps suddenly look pretty average, rather than like the constant elite team they were. That despite having guys like Sakic, Forsberg, and a bunch of great defencemen (Blake, Bourque, Foote...). Forsberg's and Hejduk's great league leading +/- from 2002-03 would not be as impressive.
All adjusted +/- attempts that I have seen, are in comparison to other players on the team (i.e. the team was x% better when so and so was on the ice compared to when they weren't). So they all have the same help from good goalies and lack of it from bad goalies. Please explain what you mean by adjusting +/- by removing goalie influence.

(if your point is about "puck luck" - one player being the victim of some stinkers by one goalie while the same goalie stands on his head for another player - that is indeed valid and something that CORSI effectively negates. As far as +/- is concerned, it should wash out over time, but maybe it doesn't. I mean, if we were comparing Sakic and Forsberg, and concluded that Sakic was an adjusted +60 and Forsberg an adjusted +20 over the course of 4 seasons, I don't think it would be realistic to conclude that Roy simply played better when Sakic was on the ice, that would be an incredible coincidence)

If your point is not about puck luck then I would like to know just what you mean.

seventieslord is offline   Reply With Quote
Old
07-11-2012, 07:18 PM
  #36
Canadiens1958
Registered User
 
Canadiens1958's Avatar
 
Join Date: Nov 2007
Posts: 10,864
vCash: 500
Goalies WOWY

Quote:
Originally Posted by overpass View Post
Right, some goalies have had that usage. Jacques Plante in 1970-71, among others.
Goalies are somewhat unique in this regard you mention Jacques Plante in 1970-71:

Check the raw data when Plante was replaced due to injury,

by Bob Perreault during the 1955-56 season (small sample space):

http://www.hockey-reference.com/teams/MTL/1956.html

and Charlie Hodge 1960-61(30 out of 70 games):

http://www.hockey-reference.com/teams/MTL/1961.html

Raw data but interesting.


Last edited by Canadiens1958: 07-11-2012 at 09:03 PM. Reason: typo
Canadiens1958 is offline   Reply With Quote
Old
07-11-2012, 08:43 PM
  #37
overpass
Registered User
 
Join Date: Jun 2007
Posts: 3,517
vCash: 500
Corrected WOWY numbers for Mario Lemieux. Thanks to Czech Your Math for pointing out errors and hockey-reference.com for having game logs. (When I originally did this I was working off newspaper archives...very difficult for the 1993-94 season when Mario never played more than three consecutive games.)

Year Team GP W L T GF GA W% GF/G GA/G
1986-87 WithMario 63 28 28 7 246 230 0.500 3.90 3.65
1986-87 WithoutMario 17 2 10 5 51 60 0.265 3.00 3.53
1989-90 WithMario 59 27 28 4 249 264 0.492 4.22 4.47
1989-90 WithoutMario 21 5 12 4 71 95 0.333 3.38 4.52
1990-91 WithMario 26 14 9 3 112 100 0.596 4.31 3.85
1990-91 WithoutMario 54 27 24 3 230 205 0.528 4.26 3.80
1991-92 WithMario 64 34 23 7 291 234 0.586 4.55 3.66
1991-92 WithoutMario 16 5 9 2 54 67 0.375 3.38 4.19
1992-93 WithMario 60 45 10 5 286 196 0.792 4.77 3.27
1992-93 WithoutMario 24 11 11 2 75 73 0.500 3.13 3.04
1993-94 WithMario 24 15 7 2 78 68 0.667 3.25 2.83
1993-94 WithoutMario 60 29 20 11 221 217 0.575 3.68 3.62
1995-96 WithMario 70 44 24 2 331 257 0.643 4.73 3.67
1995-96 WithoutMario 12 5 5 2 31 27 0.500 2.58 2.25
2000-01 WithMario 43 26 14 3 168 134 0.640 3.91 3.12
2000-01 WithoutMario 39 16 17 6 113 122 0.487 2.90 3.13
2001-02 WithMario 24 10 12 2 63 66 0.458 2.63 2.75
2001-02 WithoutMario 58 18 34 6 135 183 0.362 2.33 3.16
2002-03 WithMario 67 22 40 5 160 215 0.366 2.39 3.21
2002-03 WithoutMario 15 5 9 1 29 40 0.367 1.93 2.67
2003-04 WithMario 10 3 4 3 20 29 0.450 2.00 2.90
2003-04 WithoutMario 72 20 47 5 170 274 0.313 2.36 3.81
2005-06 WithMario 25 6 13 6 65 93 0.360 2.60 3.72
2005-06 WithoutMario 57 16 33 8 179 220 0.351 3.14 3.86

Using the same seasons to define "career", "prime", and "peak" as I did in the OP:

Estimated impact: Mario Lemieux (career)

+0.131 Win% (or 21 standings points over an 82 game season)
+0.57 GF/G (or 46 goals added over an 82 game season)
-0.09 GA/G (or 7 goals prevented over an 82 game season)

Estimated impact: Mario Lemieux (prime)

+0.192 Win% (or 31 standings points over an 82 game season)
+1.11 GF/G (or 91 goals added over an 82 game season)
-0.00 GA/G (or 0 goals prevented over an 82 game season)

Estimated impact: Mario Lemieux (peak)

+0.229 Win% (or 38 standings points over an 82 game season)
+1.27 GF/G (or 104 goals added over an 82 game season)
-0.02 GA/G (or 2 goals prevented over an 82 game season)

overpass is offline   Reply With Quote
Old
07-11-2012, 08:53 PM
  #38
Czech Your Math
Registered User
 
Czech Your Math's Avatar
 
Join Date: Jan 2006
Location: bohemia
Country: Czech_ Republic
Posts: 3,538
vCash: 50
Quote:
Originally Posted by plusandminus View Post
In a way, I think factual won-loss stats are what should be used. A win is a win, no matter if it's by 2-1 or 7-2.
In another way, I like the nuances one gets by using GF and GA to calculate pythagoran win %.
So I combined the two, making them equally important, rather than doing two separate studies. (I had already done some "boring" research of more mathematical nature, sort of like I tried to suggest in your stickied thread, so I felt confident I knew what I was doing.)
Factual win% seems superior to GF & GA data over a large enoughsample. What I'm saying is that with very limited samples, the GF & GA data might not be subject to random error (i.e. "luck") as much as W/L, esp. in the case of post-lockout W/L/OT/SO.

Quote:
Originally Posted by plusandminus View Post
One can think more, deeply think more, about the above two things. On a team level, focus is basically only on it's about factual win %. On a player level, most stats focus on GF and GA related things. For example, +/- is GF and GA related, while points scored is GF related.
So our apprach to stats may be seen as partly contradictory. Player stats focus a lot on team GD (GF-GA). The better team goal difference, the more beneficial for the players' +/-, and vice versa. While on the other hand, on a team level, goal difference is (especially in the NHL) made relatively unimportant.
The "best" for a player (skater), statistically, is when his team gets as good goal difference as possible, because that will be benificial to his +/-, and the more GF his team scores, the more beneficial to his scoring stats. But a team doesn't necessarily care for those things, because they just want to win, or get into a penalty shootout, and to gain as many points as possible. If a team leads by several goals, they may bench(?) certain players.
Okay, the above two things often may go hand in hand. But basically we often treat player stats (like +/-) as if the main goal of the NHL teams is to gain as good goal difference as possible, when it's actually not.
Certainly winning would certainly seem to be of primary importance, at least that's one of our assumptions. Technically, hockey is an entertainment business where the team is probably trying to maximize their long-term profit and the player is probably trying to maximize his own LTP. Such LTP may in large part be maximized by maximizing team win%, but This is an example of an assumption we all take for granted, since it saves time, but in reality is a matter of interpretation and can affect player performance for better or worse. I think you know now why we usually skip a lot of assumptions and use a more "ideal" starting point.

If winning is the primary goal, then how do we evaluate individual players within a team environment? There's no individual wins or losses, so we must use that which wins are most dependent on: scoring and preventing goals. Of course you know this (and I assume that's what you are saying the bolded sentence quoted above). We know from pythagorean studies that the relationship between GF-GA differential (actually, to be more precise, GF/GA ratio) and win% has a very close correlation. Since we have various metrics of individual/team performance based on this data (Goals, Points, TGF/A, TGF/A at ES, etc.), this seems the best data we have to attempt to link individual performance with team success.

I'm uncertain why a team wouldn't care for a better GF/GA differential/ratio? Sure, when the game is no longer in doubt, it no longer matters, but how often is this? For instance, if Gretzky scored a larger % of his points when his team was up by 4+ goals with less than 5 minutes in the game, then that may be a reason to discount his point totals. Even then, he likely had a large part in putting them up by such a margin, either directly (points) or indirectly (possession, his presence using the defensive resources of the opponent, etc.). I don't see a better alternative starting point than the data most directly related to team success, unless we are to rely on purely subjective data or abandon such evaluations and comparisons altogether.

Quote:
Originally Posted by plusandminus View Post
Regarding the exponent, I think the key is to know what one is doing and what effects different exponents have.
Yes, this is true. If one isn't making definitive conclusions and merely wants a "ballpark figure" the the standard exponent (2) can be used. However, if one is drawing conclusions such as that a team or individual was clutch or lucky (or vice versa), then it becomes very important.

Quote:
Originally Posted by plusandminus View Post
No, I used whole games.
I must not have seen this study and was thinking of another study you did. Since you have done a comprehensive study over a long period, and many should be interested and be able to understand this type of study, it may be worthwhile to present that data in this thread or in its own thread. You might want consider restricting the sample to better players (e.g. 600+ career games) and/or players for whom the data will be more reliable (e.g., > 40 games missed and more than 2 seasons with > 4 games missed)

Quote:
Originally Posted by plusandminus View Post
With examining individual seasons, I not only meant looking at the results for those seasons. I rather meant taking the time to also manually look up what the roster looked like for each game. Like a table with player names on one axis, and game numbers on the other, so that one can easily see which players were out simultanously, when players got traded, etc. I can get that info within 2-3 seconds, just having to enter season and team and push a button, and I hope a site like hockeyreference will soon show that kind of info too.
If you had it in database form, you might be able to use linear regression, using each player as a binary variable. However, if you start adjusting by schedule and such, the numbers of variables might overwhelm the number of values for the dependent variable (i.e. there are only 82 gm/season).

Quote:
Originally Posted by plusandminus View Post
I've actually become more and more skeptical towards +/-, including traditional attempts at adjusting it. Anyway, I think that if one wants to adjust +/-, one should start by trying to get rid of the effect of goaltending.

If one should manage to "adjust away" the influence of goaltending, then team stats would look differently. The team GF might perhaps be affected by a goal or so, but the team GA would be affected tremendously!
Adjusted plus-minus wouldn't depend on quality of goalie, unless A) the player played a disproportionate amount of time at ES with certain goalies compared to others, and B) the goalies varied significantly in quality. No reason to assume it would vary, other than randomly.

Quote:
Originally Posted by plusandminus View Post
By the way, Forsberg himself recently stated something like... First he had his techical skills, and his strong desire to fight and win. Then he had a time when he could add an increased understanding of the game, and this coincided with his Hart win and his biggest playoffs scoring success. Then his body got worse, and he had to rely on his understanding of the game. (I'm not sure I agree, as I think he "always" understood the game greatly, but anyway, sort of.)
He may have felt that way, but I think having the entire '02 regular season off must have been an important factor. At the very least, it may have given him a better chance of staying healthy in '03.

One thing I've thought about since doing the study on "adjusted adjusted" numbers is that the study was based on PPG. It shows that the better forwards generally had an easier time getting adjusted points in the '90s than in the '80s, e.g. However, what's not really considered is that it seems like it was so much tougher for players to stay healthy in DPE. This would both decrease their point totals obv. and could make it seem "easier" to achieve a higher adjusted PPG due to fewer games played. I'm not sure how to account for these factors simulataneously, although a regression study might do the trick.

Czech Your Math is offline   Reply With Quote
Old
07-11-2012, 09:09 PM
  #39
plusandminus
Registered User
 
Join Date: Mar 2011
Posts: 868
vCash: 500
Quote:
Originally Posted by seventieslord View Post
All adjusted +/- attempts that I have seen, are in comparison to other players on the team (i.e. the team was x% better when so and so was on the ice compared to when they weren't). So they all have the same help from good goalies and lack of it from bad goalies. Please explain what you mean by adjusting +/- by removing goalie influence.
What you describe, sounds like a way of simply measuring how good a player performed +/- wise, compared to his teammates? In that regard, I agree that each player on a certain team was favoured/disfavoured basically equally by the team's goaltending.

That method has advantages, but also disadvantages. Bossy+Trottier on the same line likely helps them significantly in this regard, compared to if splitting them up on two lines. On different lines, they would have had to "compete" with each other. Same - at least in theory - with Crosby and Malkin. Or Zetterberg and Datsyuk. Or Pronger and MacInnis when they played together. Or if we would make CSKA Moscow a team in the NHL, their "green unit" might outperform their other units far more than if splitting them up.
This I assume you - at least in theory - agree with?

But are you sure you haven't seen other attempts at adjusting +/-, or using its components (GF and GA) in other formulas? I know they exist, and have tried some of them out. I even think I have written about it here on this board.

I can add that I meant "+/-" in a broad sense. +/- is about GF and GA. Many formulas use GF and GA, including for example pythagoran ones.


Oh... this takes time to try to explain... And it's 3 AM here. But I give it a try...
It's difficult trying to explain "much" in few words. And there are so many angles.

One approach where one may want to eliminate the influence of goaltending, is when looking at penalty killing stats. (I posted about that some months ago, as I tried to find out which were the statistically best penalty killers after having "eliminated" goaltending influence. Lidstrom and Chelios on Detroit is an example of two defencemen doing well.)

One may also tend to think of the GF part as fairly stable. We know that many players (especially the stars) produce basically the same amount of points from season to season, and they may be on ice for fairly the same GF from season to season. GA appears more unstable. Mario Lemieux produces his points, and were on ice for his GF, no matter if he was like +25 or -25. One might even take it so far, that one would think GA is of slightly less importance for forwards. There is support for that thought.

Another case is where one wants to attribute win points, or win percentages, to players. One method is to start with the teams as a whole, and give them numbers (or just used their factual points gained, or factual win %). Then one can try to determine how much part (percentage) each player may have had in the team's successs, by looking at things like situational icetimes and situational +/-, and how it all came to lead to team earning their points/wins.
A key step in that process, is to try to determine the influence of the goaltending.

I've been away most of the time during the last two months, but aren't there authorities in the field who tries to isolate goaltending influence? I have forgotten the guy's name, but he's sometimes mentioned here.

I think many agree that raw +/- can often seem like an inconsistent stat, while points scored is often more consistent from season to season. (There are exceptions.)


I'm surprised that CzechYourMath seem to also say that goaltending doesn't affect +/-. I think it definitely does.


Quote:
Originally Posted by seventieslord View Post
(if your point is about "puck luck" - one player being the victim of some stinkers by one goalie while the same goalie stands on his head for another player - that is indeed valid and something that CORSI effectively negates. As far as +/- is concerned, it should wash out over time, but maybe it doesn't. I mean, if we were comparing Sakic and Forsberg, and concluded that Sakic was an adjusted +60 and Forsberg an adjusted +20 over the course of 4 seasons, I don't think it would be realistic to conclude that Roy simply played better when Sakic was on the ice, that would be an incredible coincidence)
It wasn't what I thought about.
But now that you mention it, I came to think of another thing. I get the spontanous impression that the defensively poorer players, those "messing things up" or creating serious turnovers, might benifit more by having a prime Hasek/Roy than the players who are more stable defensively. But I'm far too tired now to try to analyze that further. What would you think?


Quote:
Originally Posted by seventieslord View Post
If your point is not about puck luck then I would like to know just what you mean.
I hope you understood me.

plusandminus is offline   Reply With Quote
Old
07-11-2012, 09:18 PM
  #40
overpass
Registered User
 
Join Date: Jun 2007
Posts: 3,517
vCash: 500
Quote:
Originally Posted by plusandminus View Post
One approach where one may want to eliminate the influence of goaltending, is when looking at penalty killing stats. (I posted about that some months ago, as I tried to find out which were the statistically best penalty killers after having "eliminated" goaltending influence. Lidstrom and Chelios on Detroit is an example of two defencemen doing well.)
Before eliminating the influence of goaltending, you'd have to define goaltending.

If you assume that goaltending == save percentage and remove it, you are also assuming that the skaters of a team have no ability to influence save percentage, and defining defensive play as reducing shots against.

Reducing shots against is important, and you might want to measure it, but I don't think it's established that it's 100% of defensive play.

overpass is offline   Reply With Quote
Old
07-11-2012, 09:26 PM
  #41
plusandminus
Registered User
 
Join Date: Mar 2011
Posts: 868
vCash: 500
Quote:
Originally Posted by overpass View Post
Before eliminating the influence of goaltending, you'd have to define goaltending.

If you assume that goaltending == save percentage and remove it, you are also assuming that the skaters of a team have no ability to influence save percentage, and defining defensive play as reducing shots against.

Reducing shots against is important, and you might want to measure it, but I don't think it's established that it's 100% of defensive play.
I have written about this too, on this forum, some months ago. I even started a specific thread where I wanted to discuss the subject. I addressed the same topics as you did now. I also, three hours ago, wrote "Of course, it might be a very difficult study too. For example, to what extend was a certain team's goaltending stats affected by its playing system and skater performance, and vice versa?" here in this thread.
So I agree.

plusandminus is offline   Reply With Quote
Old
07-11-2012, 09:29 PM
  #42
Canadiens1958
Registered User
 
Canadiens1958's Avatar
 
Join Date: Nov 2007
Posts: 10,864
vCash: 500
Winning

Quote:
Originally Posted by Czech Your Math View Post

Certainly winning would certainly seem to be of primary importance, at least that's one of our assumptions. Technically, hockey is an entertainment business where the team is probably trying to maximize their long-term profit and the player is probably trying to maximize his own LTP. Such LTP may in large part be maximized by maximizing team win%, but This is an example of an assumption we all take for granted, since it saves time, but in reality is a matter of interpretation and can affect player performance for better or worse. I think you know now why we usually skip a lot of assumptions and use a more "ideal" starting point.

If winning is the primary goal, then how do we evaluate individual players within a team environment? There's no individual wins or losses, so we must use that which wins are most dependent on: scoring and preventing goals. Of course you know this (and I assume that's what you are saying the bolded sentence quoted above). We know from pythagorean studies that the relationship between GF-GA differential (actually, to be more precise, GF/GA ratio) and win% has a very close correlation. Since we have various metrics of individual/team performance based on this data (Goals, Points, TGF/A, TGF/A at ES, etc.), this seems the best data we have to attempt to link individual performance with team success.

I'm uncertain why a team wouldn't care for a better GF/GA differential/ratio? Sure, when the game is no longer in doubt, it no longer matters, but how often is this? For instance, if Gretzky scored a larger % of his points when his team was up by 4+ goals with less than 5 minutes in the game, then that may be a reason to discount his point totals. Even then, he likely had a large part in putting them up by such a margin, either directly (points) or indirectly (possession, his presence using the defensive resources of the opponent, etc.). I don't see a better alternative starting point than the data most directly related to team success, unless we are to rely on purely subjective data or abandon such evaluations and comparisons altogether.
Winning = keeping your job. True for all the elements - management, coaches, players and support staff that contribute to the final on ice product.

How to optimize winning conditions is open for debate. To a large extent the different philosophies about winning is what analysts try to measure.

A team does not care for a better goal differential because there are too many variables that it does not control. A team or coach/management cares about what they can control. In this case GA. Reducing the GA makes the goals a team score that more valuable and makes it harder for a team to lose. Setting a target of the lowest GA in the league and reaching it guarantees regular season success. The differential between GF/GA does not have the same effect. Since the other team's defense cannot be controlled trying to compensate for defensive weaknesses with GF eventually fails.

Canadiens1958 is offline   Reply With Quote
Old
07-11-2012, 10:43 PM
  #43
shazariahl
Registered User
 
Join Date: Apr 2009
Posts: 1,420
vCash: 500
Quote:
Originally Posted by Canadiens1958 View Post
Since the other team's defense cannot be controlled trying to compensate for defensive weaknesses with GF eventually fails.
Unless you're the 80's Oilers.

shazariahl is offline   Reply With Quote
Old
07-11-2012, 11:47 PM
  #44
Czech Your Math
Registered User
 
Czech Your Math's Avatar
 
Join Date: Jan 2006
Location: bohemia
Country: Czech_ Republic
Posts: 3,538
vCash: 50
Quote:
Originally Posted by plusandminus View Post
What you describe, sounds like a way of simply measuring how good a player performed +/- wise, compared to his teammates? In that regard, I agree that each player on a certain team was favoured/disfavoured basically equally by the team's goaltending.

That method has advantages, but also disadvantages. Bossy+Trottier on the same line likely helps them significantly in this regard, compared to if splitting them up on two lines. On different lines, they would have had to "compete" with each other. Same - at least in theory - with Crosby and Malkin. Or Zetterberg and Datsyuk. Or Pronger and MacInnis when they played together. Or if we would make CSKA Moscow a team in the NHL, their "green unit" might outperform their other units far more than if splitting them up.
This I assume you - at least in theory - agree with?
The bolded is what adjusted plus-minus does. That's why I said that goaltending shouldn't significantly influence that metric.

It filters out countless factors simultaneously, while only adding the complicating factor of teammates used as comparisons. Linemates are still a factor, but they are with the raw metric. This is why it's such an elegant metric. It vastly simplifies the comparison process, while stilling measuring something critical: how much better a team was with the player than without him on the ice. It's much more reliable than "with or without win%", because while this metric is more directly measuring the primary value (winning), it still has countless variables involved and for most players has a very limited sample of missed games that, if at all available, varies tremendously from year to year.

Czech Your Math is offline   Reply With Quote
Old
07-11-2012, 11:58 PM
  #45
Czech Your Math
Registered User
 
Czech Your Math's Avatar
 
Join Date: Jan 2006
Location: bohemia
Country: Czech_ Republic
Posts: 3,538
vCash: 50
Quote:
Originally Posted by Canadiens1958 View Post
Winning = keeping your job. True for all the elements - management, coaches, players and support staff that contribute to the final on ice product.
Generally this is true. I was making the point that if we spell out every assumption, it can be a tedious, almost endless process.

There are cases in sports where teams may not care as much about winning as maximizing profit. Without a profit, they may not have to care about winning at all, since they may no longer have a franchise. If selling tickets/merchandise increases profit more than winning, then they may have incentive to emphasize the former over the latter. OTOH, there are teams on the other end of the spectrum which may care disproportionately about winning and try to "buy a championship." Of course, even when winning may not be first priority, once other constraints are met (budget, entertainment value), there seems no incentive not to maximize winning (unless the team is tanking for a top draft pick... but of course such a thing would never happen...?).

Quote:
Originally Posted by Canadiens1958 View Post
How to optimize winning conditions is open for debate. To a large extent the different philosophies about winning is what analysts try to measure.
Winning is sun which most sports analyses revolve around. I agree that there's debate as to how to maximize winning.

Quote:
Originally Posted by Canadiens1958 View Post
A team does not care for a better goal differential because there are too many variables that it does not control. A team or coach/management cares about what they can control. In this case GA. Reducing the GA makes the goals a team score that more valuable and makes it harder for a team to lose. Setting a target of the lowest GA in the league and reaching it guarantees regular season success. The differential between GF/GA does not have the same effect. Since the other team's defense cannot be controlled trying to compensate for defensive weaknesses with GF eventually fails.
I think we've had this debate before, so don't see reason to have it again. Generally, the greater the difference in GF & GA (and GF/GA ratio), the greater the difference in winning %. Whether this is achieved by increasing GF or decreasing GA is not generally important in most cases. Where we may agree is in the cases of weaker teams, which may generally have incentive to play lower scoring games, since these trend more toward .500 than higher scoring games. IOW, generally the more goals/runs scored, the more likely it is that the better team will win.


Last edited by Czech Your Math: 07-12-2012 at 12:20 AM.
Czech Your Math is offline   Reply With Quote
Old
07-12-2012, 05:28 AM
  #46
Canadiens1958
Registered User
 
Canadiens1958's Avatar
 
Join Date: Nov 2007
Posts: 10,864
vCash: 500
Debatable

Quote:
Originally Posted by Czech Your Math View Post
I think we've had this debate before, so don't see reason to have it again. Generally, the greater the difference in GF & GA (and GF/GA ratio), the greater the difference in winning %. Whether this is achieved by increasing GF or decreasing GA is not generally important in most cases. Where we may agree is in the cases of weaker teams, which may generally have incentive to play lower scoring games, since these trend more toward .500 than higher scoring games. IOW, generally the more goals/runs scored, the more likely it is that the better team will win.
Goes against the history of upsets. 1945 NHL semi finals Montreal/Toronto, 1960 World Series NY Yankees/Pittsburgh, 1967 NHL Toronto Maple Leafs, just a short list of upsets, win the close or low scoring games while losing the the blow-outs. Question of making the few goals or runs that a team scores more valuable by optimizing resources.

http://en.wikipedia.org/wiki/1960_World_Series


Last edited by Canadiens1958: 07-12-2012 at 05:31 AM. Reason: link
Canadiens1958 is offline   Reply With Quote
Old
07-12-2012, 09:25 AM
  #47
plusandminus
Registered User
 
Join Date: Mar 2011
Posts: 868
vCash: 500
Quote:
Originally Posted by Czech Your Math View Post
The bolded is what adjusted plus-minus does. That's why I said that goaltending shouldn't significantly influence that metric.
Well, I tried to give my point of view, and to explain how I looked at it. You seem dead sure that it's only about what you write above. I am familiar with other methods. I'm a bit surprised others aren't. Maybe some kind of misunderstanding somewhere.

Quote:
Originally Posted by Czech Your Math View Post
It filters out countless factors simultaneously, while only adding the complicating factor of teammates used as comparisons. Linemates are still a factor, but they are with the raw metric. This is why it's such an elegant metric. It vastly simplifies the comparison process, while stilling measuring something critical: how much better a team was with the player than without him on the ice. It's much more reliable than "with or without win%", because while this metric is more directly measuring the primary value (winning), it still has countless variables involved and for most players has a very limited sample of missed games that, if at all available, varies tremendously from year to year.
I disagree with you. And I have tried to explain why, and how I think. I think both methods have advantages and disadvantages. I do think I know enough hockey, and statistics, to have a decent understanding of what I'm talking about. And I do think you too knows math and stats well, and knows a lot about hockey.

For example, the best players on a team aren't necessarily the ones with best +/- on the team (or best ES +/- per time unit).


I said yesterday, I have spent very many hours during the last year on NHL stats. While it has broaden my knowledge, and made me learn interesting things, I came to a point where I thought it might be enough for me. I have examined lots of things. There are many more things to examine, but from my point of view it would require a lot of work, and I think I at this point rather stop where I am.

I had hoped I would have been able to better communicate here about some things, but it seems I can't. I've been into stats and hockey since the late 1970s, and been writing here for about 16 months, and would have hoped to sort of have "made a name" here by now (among the regulars). But instead, when I write it's a bit as if the respondents think I have just turned my attention to hockey and its statistics. Instead of getting a feeling they sort of know in general about my previous posts and studies, and all the work I have done in organizing data in order to do studies few (or sometimes none) here are able to currently do, I never get referred to or even mentioned.

While I learn all the time from others, it seems as I basically can't ever make others curious or "learn" about things I write. Seldom anyone writes something like "Good point!" or similar, like is often the case when others write. If (hypothetically) there are 12 things that matter, and I take the time to go through 10 of them, I get replies about the the I didn't mention rather than the 10 I mentioned. (I know I myself often tend to focus on missing things, but tries to balance it.) This isn't news to some here, as this is how I've been experiencing things for a while.

I am still sure that I have things of value to contribute with, things that will help bring hockey knowlegde forward. I think I would be a loss for the forum if I stopped writing. But as I said, after about 16 months here it's obvious (in the eyes of the regulars here) I am just where I started, and definitely wouldn't be missed. So it would be unwise for me to participate here due to "the warm climate", "supporting atmosphere" or because "this is a place where I feel appreciated", because to me it's rather the opposite. If I have a strong and true curiosity about learning more about NHL and its stats, then that in itself could be a strong reason for me to spend time here in order to get feedback and inspiration. And so it was. But now my curiosity have started to fade, and I think that if I try to discuss "advanced" things in a half-hearted way, it may not turn out very good.

I also feel I'm at a level now where any discussions like the ones here in this thread, will be "advanced" and take up a lot of time. I'm "done" with most of the more easy studies (like "on/off +/-" or "adjusted points") and would want to "move on" to more advanced studies that builds on the common ones. And that takes time, lots of time, and discussions also take lots of times. And there will be very few, if anyone, to discuss them with. And based from previous experience here, few if any would be interested anyway. Plus that I have a feeling that even very time consuming studies won't help us get rid of different biases anyway.

Maybe I ought to summarize. Thanks for your questions and suggestions. It's hard for me to reply because it seems to require me to explain about my "whole focus", so I tend to write fragments that I feel are being prone to misunderstandings.


Last edited by plusandminus: 07-12-2012 at 09:32 AM.
plusandminus is offline   Reply With Quote
Old
07-12-2012, 09:34 AM
  #48
seventieslord
Moderator
 
seventieslord's Avatar
 
Join Date: Mar 2006
Location: Regina, SK
Country: Canada
Posts: 23,877
vCash: 500
Quote:
Originally Posted by plusandminus View Post
What you describe, sounds like a way of simply measuring how good a player performed +/- wise, compared to his teammates? In that regard, I agree that each player on a certain team was favoured/disfavoured basically equally by the team's goaltending.

That method has advantages, but also disadvantages. Bossy+Trottier on the same line likely helps them significantly in this regard, compared to if splitting them up on two lines. On different lines, they would have had to "compete" with each other. Same - at least in theory - with Crosby and Malkin. Or Zetterberg and Datsyuk. Or Pronger and MacInnis when they played together. Or if we would make CSKA Moscow a team in the NHL, their "green unit" might outperform their other units far more than if splitting them up.
This I assume you - at least in theory - agree with?
Yes, of course. CYM did a better job expanding on it than I would.

Quote:
But are you sure you haven't seen other attempts at adjusting +/-, or using its components (GF and GA) in other formulas? I know they exist, and have tried some of them out. I even think I have written about it here on this board.

I can add that I meant "+/-" in a broad sense. +/- is about GF and GA. Many formulas use GF and GA, including for example pythagoran ones.


Oh... this takes time to try to explain... And it's 3 AM here. But I give it a try...
It's difficult trying to explain "much" in few words. And there are so many angles.

One approach where one may want to eliminate the influence of goaltending, is when looking at penalty killing stats. (I posted about that some months ago, as I tried to find out which were the statistically best penalty killers after having "eliminated" goaltending influence. Lidstrom and Chelios on Detroit is an example of two defencemen doing well.)

One may also tend to think of the GF part as fairly stable. We know that many players (especially the stars) produce basically the same amount of points from season to season, and they may be on ice for fairly the same GF from season to season. GA appears more unstable. Mario Lemieux produces his points, and were on ice for his GF, no matter if he was like +25 or -25. One might even take it so far, that one would think GA is of slightly less importance for forwards. There is support for that thought.

Another case is where one wants to attribute win points, or win percentages, to players. One method is to start with the teams as a whole, and give them numbers (or just used their factual points gained, or factual win %). Then one can try to determine how much part (percentage) each player may have had in the team's successs, by looking at things like situational icetimes and situational +/-, and how it all came to lead to team earning their points/wins.
A key step in that process, is to try to determine the influence of the goaltending.

I've been away most of the time during the last two months, but aren't there authorities in the field who tries to isolate goaltending influence? I have forgotten the guy's name, but he's sometimes mentioned here.

I think many agree that raw +/- can often seem like an inconsistent stat, while points scored is often more consistent from season to season. (There are exceptions.)


I'm surprised that CzechYourMath seem to also say that goaltending doesn't affect +/-. I think it definitely does.




It wasn't what I thought about.
But now that you mention it, I came to think of another thing. I get the spontanous impression that the defensively poorer players, those "messing things up" or creating serious turnovers, might benifit more by having a prime Hasek/Roy than the players who are more stable defensively. But I'm far too tired now to try to analyze that further. What would you think?
If we're talking about "adjusted" +/- (which is the only version of +/- any serious stats guy should be concerned with), then I wouldn't agree. If Patrick Roy was consistently covering up the errors of some mediocre Colorado player, he was also doing the same strong work when Sakic and Forsberg were on the ice, so ultimately they would end up the same relative distance ahead of the mediocre player in adjusted +/-, than if they all had a poor goalie to work with.

seventieslord is offline   Reply With Quote
Old
07-12-2012, 10:20 AM
  #49
plusandminus
Registered User
 
Join Date: Mar 2011
Posts: 868
vCash: 500
Quote:
Originally Posted by seventieslord View Post
If we're talking about "adjusted" +/- (which is the only version of +/- any serious stats guy should be concerned with), then I wouldn't agree.
Just to ask. Do you mean that one is not "a serious stats guy" if one uses methods other than "on/off" to adjust the raw +/-?
Or are you referring to "adjusted +/-" as opposed to raw +/-?

I spent an hour, in the middle of the night, trying to explain to you how I meant, so I would find it a bit disrespectful if you just discarded my attempt to answer you.


Last edited by plusandminus: 07-12-2012 at 10:34 AM.
plusandminus is offline   Reply With Quote
Old
07-12-2012, 10:38 AM
  #50
overpass
Registered User
 
Join Date: Jun 2007
Posts: 3,517
vCash: 500
Quote:
Originally Posted by plusandminus View Post
For example, the best players on a team aren't necessarily the ones with best +/- on the team (or best ES +/- per time unit).
This would be mostly because of the players they play with and the players they play against, correct? Brian MacDonald has calculated fully adjusted plus-minus numbers that account for these factors, but they only cover the last few seasons because they necessary data only exists for those seasons.

Just looking at the data it appears that having 2 or 3 star players together on a line is a big factor in plus-minus, including basic adjusted versions. See Trottier/Bossy, Simmer/Dionne/Taylor, Gretzky/Kurri, Savard/Larmer, Leclair/Lindros, Kariya/Selanne, Heatley/Spezza/Alfredsson, etc. Compare to players like younger Steve Yzerman or Dale Hawerchuk who didn't have star linemates and didn't stand out in plus-minus. Marcel Dionne was only OK in plus-minus in his first few seasons, but when Dave Taylor and later Charlie Simmer joined him they were dominant in plus-minus.

If you want to move beyond a "basic" adjusted plus-minus and start adjusting for teammates, by all means go for it. I haven't done so because the data isn't easily available and I don't want to put the time into merging databases (one of those time-consuming tasks that is often required). But it might be possible to adjust historical (pre-lockout) data for the effect of linemates to some degree by using the number of even strength points players participated in together as a proxy for time spent together.

overpass is offline   Reply With Quote
Reply

Forum Jump


Bookmarks

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off



All times are GMT -5. The time now is 10:10 AM.

monitoring_string = "e4251c93e2ba248d29da988d93bf5144"
Contact Us - HFBoards - Archive - Privacy Statement - Terms of Use - Advertise - Top - AdChoices

vBulletin Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
HFBoards.com is a property of CraveOnline Media, LLC, an Evolve Media, LLC company. 2014 All Rights Reserved.