Saturday, September 13, 2008
A request
In the comments to MF's Rates 'n' States post about the value of even strength points per 60 minutes (EV/60), I mentioned that I actually like the PPG stat explicitly because it does take into account ice-time. My contention was that, while players may be capable of putting up equivalent rates of scoring per minute of ice time (EV/60), they may not be capable of maintaining those equivalent rates when given different icetime (i.e., there are good reasons the coach plays one player for only 5 minutes a game and while giving another guy 15 minutes of ice, despite the fact that they have equivalent EV/60).
MF's response, quite sensibly, was that he doesn't "want to take the coach's word for it as to who is 'capable' of what, [he's] interested in results."
Fair enough.
So, talking with the good folks on the #8 bus yesterday, we thought it might be a good idea if some clever individual could plot EV/60 against Actual Icetime and see what the correlation was. We would be especially interested in looking at those individuals who defy expectations at both ends of the curve -- that is, individuals who look like they are getting way more ice-time than is justified by their EV/60 (i.e., guys who, according to the coach, add value beyond their scoring), and those who look like they are getting way less ice-time than is justified by their EV/60 (i.e., guys whose ability to score doesn't offset their liabilities on the ice).
Once we see the magnitude of the difference, it would be interesting to track the different reasons for those discrepancies by using the various Outscoring stats, Quality of Opposition stats, etc.
My suspicion is that we will find out that coaches aren't such dummies afterall, and that there are pretty solid reasons for their decisions, but I might be wrong.
MF's response, quite sensibly, was that he doesn't "want to take the coach's word for it as to who is 'capable' of what, [he's] interested in results."
Fair enough.
So, talking with the good folks on the #8 bus yesterday, we thought it might be a good idea if some clever individual could plot EV/60 against Actual Icetime and see what the correlation was. We would be especially interested in looking at those individuals who defy expectations at both ends of the curve -- that is, individuals who look like they are getting way more ice-time than is justified by their EV/60 (i.e., guys who, according to the coach, add value beyond their scoring), and those who look like they are getting way less ice-time than is justified by their EV/60 (i.e., guys whose ability to score doesn't offset their liabilities on the ice).
Once we see the magnitude of the difference, it would be interesting to track the different reasons for those discrepancies by using the various Outscoring stats, Quality of Opposition stats, etc.
My suspicion is that we will find out that coaches aren't such dummies afterall, and that there are pretty solid reasons for their decisions, but I might be wrong.
# posted by sacamano @ 8:44 am 
Comments:
First of all, although I don't think this is what you were concluding exactly, I don't think coaches are dummies, and I doubt that this investigation would demonstrate it one way or another anyway.
Second, I suppose the MOST interesting thing to find out would be if production actually declines past a certain level of ice time (or conversely, if keeping ice time at a particular low-ish level maximizes production), but that might be as difficult a project as one could imagine. It's just too damn hard to control for all the other variables.
Lastly, although I'm sure I'd be interested in seeing the results of what you're proposing (the outliers are always what fascinates), I'm skeptical at the quality of conclusions you could draw. So much of why a coach uses Forward #5 the way he does depends directly on who his other 11 are, or who he thinks they are. Why did Tortorella play Brad Richards so damn much? Because he thought it was better than the alternatives. I'm not sure if you'd be able to draw any conclusions with broader implications out of that, apart from some better data on *coaches* as opposed to players (which might be interesting in its own right).
We might get into some related issues in later parts of this series, so stay tuned I suppose.
So much of why a coach uses Forward #5 the way he does depends directly on who his other 11 are
Any more than how a player's EV/60 depends directly on who his line mates are, quality of opposition, etc.?
My conclusion to the post should have been simply that because it builds in ice time (admittedly dependent on the coach's opinion of the player's quality and the quality of his teammates), I suspect PPG is a reasonably robust proxy of what a player is actually capable of giving you.
I dunno, we can all think of counter-examples of players whose games exploded after moving to another coach/team, but how many are there really? How many players actually increase their prduction in any meaningful way when they are suddenly given more icetime -- as opposed to a gradual increase with experience, age, etc.
Chicken and egg -- do they get more ice time because they are capable of putting up more points, or are they putting up more points simply because they are getting more ice-time. Obviously a combination, but I'm not sure EV/60 helps sort that out, unless it can be somehow tied to ice time.
I think we can all agree that a player's EV productivity is dependent upon, at minimum:
- his ability
- the ability of his linemates
- the ability of his competition
- the team's strategy/philosophy
- his ice time
- luck
- lots more
EV/60 takes out only one of those items, which is great, so long as the results aren't over interpreted. My problem isn't with the stat, per se, so much as with the tendency of people to use the stat to say that two players are equally "productive" just because they have equivalent rates of production over 60 minutes.
At the end of the day, rates of production aren't what matter -- it is total production. I'm not convinced that EV/60 is a better measure of "potential for total production" than PPG, but I'm willing to be convinced.
And by signing myself 'Sarah', I obviously meant 'Sacamano'. Stupid new blogger
I don't think there's much to be gained by discussing this at length at the moment, but suffice it to say I think it makes complete sense to use rates first, and then look at all the other factors (context).
Hopefully when I get to that stage we can establish a bit of consensus or some rules of thumb in that regard.
I tend to agree with Sacamano, if a player is putting up points at a good rate he'll usually get his ice time. But if he can't be trusted to play against good players he won't get as much.
Teaching to fish:
* cut and paste the table from Desjardins into a spreadsheet (there is one that has both pts/60 and qualcomp columns)
* knock off the top two or three rows that are extraneous.
* block the POS column, press ctrl+H and replace "D" with "Z".
* block the page and sort by the POS column, ascending.
* in a blank cell enter a formula to multiply TOI/60 by PTS/60 and divide by 60.0 from the same row (this will be points-per-game)
* block the POS column and find the first Z (ctrl+F)
* block the rows from there (the last forward) to the top and sort by GP (games played), descending.
* scroll down and decide where to cut it as far as games-played. Forwards who only played a handful of games can have some really screwy Desjardins QualComps, the dust didn't have time to settle. You'll see that and draw the line somewhere, be it 3 games played or 10 games played or wherever.
*
That shouldn't take more than 15 minutes, even if you are unfamiliar with spreadsheets. And half of that if you are.
From there just do as you like. Pearson correlations (=correl()) aren't something you can take further. Neither talent nor icetime is normally distributed and there are a lot of other variables in play. It's still a handy lazyman's indicator.
Still, for EV/60 to points/game: r=.95
EV/60 to TOI/60: r=.66
QualComp to points/game: r=.48
QualComp to TOI/60: r=.36
... and just for shits and giggles:
QualComp to on-ice/off-ice +/-: r=.23 if you draw the line at min 20 games played, as I did.
Then you can scroll through the list of the guys who aren't getting the TOI as much as their EV/60 suggests that they should ... and it's Matt Cullen (he was a really underrated player for a lot of years I thought, age must be catching up though, just can't play the minutes any more I suspect) and about 20 other guys who are in their first three years in the league, and are sporting generally lower qualcomps than a bunch of others on their team, or it was at the point I stopped scrolling through the names. That and some guys who played on really good offensive teams (COL and DET). And I don't know who Dave Bolland is, I'm assuming a rookie though.
So you can draw from that what you will. People from a baseball stats background are going to be drawn towards concluding that for forwards, playing tough minutes (as defined by QualComp) helps them score more and improves their +/- relative to teammates. And a deeper, intellectually honest analysis will confirm that.
I'm thinking that good forwards play against good forwards a lot in this league. Not so much on the Oilers last season of course, but overall. And of course we see that all the time. Have forever, from Gilmour vs Gretzky to Zetterberg vs Crosby.
I wandered off topic a bit:
Isn't it fair to use pts/game as a starting point? It is rarely a bad idea to assume that the coaches in this league are smarter than us, after all.
And from there apply common sense if the situation is appropriate?
So we can look at Miettenin and his impressive EV pts/60. And go from there, asking the questions sacamano has listed above:
- the ability of his linemates
- the ability of his competition
- the team's strategy/philosophy
- his ice time
- luck
I'd add in "strength of team" as well, at least for an explanation of why he got less icetime. Surely he would have gotten a lot more if he played for the Lightning.
Bolland would be a rookie, or at least it should be safe to say he is in his first couple seasons in the league. Played on the Schremp/Perry London Knights Mem Cup team if I recall correctly.
Interesting debate as well, though I would guess it's a case where the numbers will end up supporting the common sense found in "I tend to agree with Sacamano, if a player is putting up points at a good rate he'll usually get his ice time. But if he can't be trusted to play against good players he won't get as much."
The issue of sustain is an unacceptable thought. Most teams have 45 min of EV; 7.5 min PP; 7.5 min of PK.
Average ice time:
Top 90 Fwd 14:07 TOI
50sec/Shift 17.0 shifts/GM
91-180 Fwd 12:34 TOI
47.0sec/Shift 16.0 shifts/GM
181-270 Fwd 11:14 TOI
45.5sec/Shift 14.8 shifts/GM
As you can see there is not much of a difference in time /shift. though first line does show a 6% greater length in shift. Where the real diffrence is in the number of EV shifts. This is more a reflection of coaches choice than ability to play the extra minutes.
This is why I trust the EV/60 numbers with the Comp faced being a key factor in determining player quality.
Post a Comment
<< Home
First of all, although I don't think this is what you were concluding exactly, I don't think coaches are dummies, and I doubt that this investigation would demonstrate it one way or another anyway.
Second, I suppose the MOST interesting thing to find out would be if production actually declines past a certain level of ice time (or conversely, if keeping ice time at a particular low-ish level maximizes production), but that might be as difficult a project as one could imagine. It's just too damn hard to control for all the other variables.
Lastly, although I'm sure I'd be interested in seeing the results of what you're proposing (the outliers are always what fascinates), I'm skeptical at the quality of conclusions you could draw. So much of why a coach uses Forward #5 the way he does depends directly on who his other 11 are, or who he thinks they are. Why did Tortorella play Brad Richards so damn much? Because he thought it was better than the alternatives. I'm not sure if you'd be able to draw any conclusions with broader implications out of that, apart from some better data on *coaches* as opposed to players (which might be interesting in its own right).
We might get into some related issues in later parts of this series, so stay tuned I suppose.
So much of why a coach uses Forward #5 the way he does depends directly on who his other 11 are
Any more than how a player's EV/60 depends directly on who his line mates are, quality of opposition, etc.?
My conclusion to the post should have been simply that because it builds in ice time (admittedly dependent on the coach's opinion of the player's quality and the quality of his teammates), I suspect PPG is a reasonably robust proxy of what a player is actually capable of giving you.
I dunno, we can all think of counter-examples of players whose games exploded after moving to another coach/team, but how many are there really? How many players actually increase their prduction in any meaningful way when they are suddenly given more icetime -- as opposed to a gradual increase with experience, age, etc.
Chicken and egg -- do they get more ice time because they are capable of putting up more points, or are they putting up more points simply because they are getting more ice-time. Obviously a combination, but I'm not sure EV/60 helps sort that out, unless it can be somehow tied to ice time.
I think we can all agree that a player's EV productivity is dependent upon, at minimum:
- his ability
- the ability of his linemates
- the ability of his competition
- the team's strategy/philosophy
- his ice time
- luck
- lots more
EV/60 takes out only one of those items, which is great, so long as the results aren't over interpreted. My problem isn't with the stat, per se, so much as with the tendency of people to use the stat to say that two players are equally "productive" just because they have equivalent rates of production over 60 minutes.
At the end of the day, rates of production aren't what matter -- it is total production. I'm not convinced that EV/60 is a better measure of "potential for total production" than PPG, but I'm willing to be convinced.
And by signing myself 'Sarah', I obviously meant 'Sacamano'. Stupid new blogger
# posted by 
: September 14, 2008 9:59 am
: September 14, 2008 9:59 am I don't think there's much to be gained by discussing this at length at the moment, but suffice it to say I think it makes complete sense to use rates first, and then look at all the other factors (context).
Hopefully when I get to that stage we can establish a bit of consensus or some rules of thumb in that regard.
I tend to agree with Sacamano, if a player is putting up points at a good rate he'll usually get his ice time. But if he can't be trusted to play against good players he won't get as much.
Teaching to fish:
* cut and paste the table from Desjardins into a spreadsheet (there is one that has both pts/60 and qualcomp columns)
* knock off the top two or three rows that are extraneous.
* block the POS column, press ctrl+H and replace "D" with "Z".
* block the page and sort by the POS column, ascending.
* in a blank cell enter a formula to multiply TOI/60 by PTS/60 and divide by 60.0 from the same row (this will be points-per-game)
* block the POS column and find the first Z (ctrl+F)
* block the rows from there (the last forward) to the top and sort by GP (games played), descending.
* scroll down and decide where to cut it as far as games-played. Forwards who only played a handful of games can have some really screwy Desjardins QualComps, the dust didn't have time to settle. You'll see that and draw the line somewhere, be it 3 games played or 10 games played or wherever.
*
That shouldn't take more than 15 minutes, even if you are unfamiliar with spreadsheets. And half of that if you are.
From there just do as you like. Pearson correlations (=correl()) aren't something you can take further. Neither talent nor icetime is normally distributed and there are a lot of other variables in play. It's still a handy lazyman's indicator.
Still, for EV/60 to points/game: r=.95
EV/60 to TOI/60: r=.66
QualComp to points/game: r=.48
QualComp to TOI/60: r=.36
... and just for shits and giggles:
QualComp to on-ice/off-ice +/-: r=.23 if you draw the line at min 20 games played, as I did.
Then you can scroll through the list of the guys who aren't getting the TOI as much as their EV/60 suggests that they should ... and it's Matt Cullen (he was a really underrated player for a lot of years I thought, age must be catching up though, just can't play the minutes any more I suspect) and about 20 other guys who are in their first three years in the league, and are sporting generally lower qualcomps than a bunch of others on their team, or it was at the point I stopped scrolling through the names. That and some guys who played on really good offensive teams (COL and DET). And I don't know who Dave Bolland is, I'm assuming a rookie though.
So you can draw from that what you will. People from a baseball stats background are going to be drawn towards concluding that for forwards, playing tough minutes (as defined by QualComp) helps them score more and improves their +/- relative to teammates. And a deeper, intellectually honest analysis will confirm that.
I'm thinking that good forwards play against good forwards a lot in this league. Not so much on the Oilers last season of course, but overall. And of course we see that all the time. Have forever, from Gilmour vs Gretzky to Zetterberg vs Crosby.
I wandered off topic a bit:
Isn't it fair to use pts/game as a starting point? It is rarely a bad idea to assume that the coaches in this league are smarter than us, after all.
And from there apply common sense if the situation is appropriate?
So we can look at Miettenin and his impressive EV pts/60. And go from there, asking the questions sacamano has listed above:
- the ability of his linemates
- the ability of his competition
- the team's strategy/philosophy
- his ice time
- luck
I'd add in "strength of team" as well, at least for an explanation of why he got less icetime. Surely he would have gotten a lot more if he played for the Lightning.
Bolland would be a rookie, or at least it should be safe to say he is in his first couple seasons in the league. Played on the Schremp/Perry London Knights Mem Cup team if I recall correctly.
Interesting debate as well, though I would guess it's a case where the numbers will end up supporting the common sense found in "I tend to agree with Sacamano, if a player is putting up points at a good rate he'll usually get his ice time. But if he can't be trusted to play against good players he won't get as much."
The issue of sustain is an unacceptable thought. Most teams have 45 min of EV; 7.5 min PP; 7.5 min of PK.
Average ice time:
Top 90 Fwd 14:07 TOI
50sec/Shift 17.0 shifts/GM
91-180 Fwd 12:34 TOI
47.0sec/Shift 16.0 shifts/GM
181-270 Fwd 11:14 TOI
45.5sec/Shift 14.8 shifts/GM
As you can see there is not much of a difference in time /shift. though first line does show a 6% greater length in shift. Where the real diffrence is in the number of EV shifts. This is more a reflection of coaches choice than ability to play the extra minutes.
This is why I trust the EV/60 numbers with the Comp faced being a key factor in determining player quality.
Post a Comment
<< Home
