Mobile Hockey's Future Become a Sponsor Site Rules Support Forum vBookie Page 2
 Notices Please do not post or solicit links to illegal game streams.

 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.

08-21-2011, 06:38 PM
#176
Registered User

Join Date: Jan 2006
Location: bohemia
Country:
Posts: 4,842
vCash: 500
Quote:
 Originally Posted by plusandminus I've studied correlations between ES icetime and ESGF+ESGA and they don't necessarily correspond very well, perhaps especially regarding forwards. For example, during 2002-03, I guy like Kowalchuk (who had a reputation as an offensive minded player) had about 1.5 times more ESGF+ESGA played than the average player. So here we thus have our first bias. ESGF=60 and ESGA=50, will give higher "ice time" than ESGF=50 and ESGA=40, even if real ice time is the same. ?
I'm not sure ice time is so important. If a player is on ice for a much higher % of ES GF+GA than his % of his ice time, that's okay. If he's able to perform at or above the GF/GA ratio of the team as a whole, then the more volume the better for the team. If he's at a worse GF/GA level, then the high volume will negatively effect the player portion of the metric even more.

Quote:
 Originally Posted by plusandminus I'm a bit against dividing ESGF by ESGA. As i wrote in another reply some day ago, I think there are better ways to do it. To use the above example, I would say that 60-50 and 50-40 is equally good, despite the latter one getting a slightly higher ratio (if I understand you right).
I don't know why you are so against calculating GF/GA ratios. They shouldn't be used randomly, but in this case they are the primary basis of the pythagorean win% calculation, so of great importance.

Whether 60/50 or 50/40 is better may depend most on context (all other things being equal). On a bad team, the extra 10/10 might be helpful, while on a good team, it may be hurtful.

Quote:
 Originally Posted by plusandminus I googled, and according to wikipedia I got the impression that rather 1.8 was the "right exponent", at least in baseball? Wouldn't shootout goals be excluded from the stats?
I saw more than one study for hockey. If 1.8 or whatever number is deemed a solid number, I have not attachment to 2.0 as exponent. It does vary by sport though, I think mainly due to differing scoring levels.

Quote:
 Originally Posted by plusandminus That's 94-38 = +56 with Forsberg. And 84-87 = -3 without. Not only did he have the league's by far best ES+/-, and scored the highest amount of ESpts, on a team that without him (and the guys who were on the ice with him) had negative +/-. Adding ESGF+ESGA, we get 94+38=132 for him. 178+125=303 for the team. That's an "ice time" of 43.56 % according to ESGF+ESGA. I got 43.56 %, so at least one of us (perhaps I) may be wrong. In reality, the correct answer seems to be 28.84 %. (if my data is correct) So, the estimated percentage is about 1.5 times higher.
You are correct. I have realized yet another error in this hastily put together study. I had calculated player's % of team's GF and GA separately and then summed them, instead of summing them before calculating the player's % of team. I came up with this idea a couple days ago and used some existing player data for the basis of most of the calculations, but that doesn't excuse my sloppiness.

Nice job of checking my math!

Quote:
 Originally Posted by plusandminus And I think three players with identical EStime, having ESGF-ESGA of 40-40 and 30-30 and 20-20 contributes equally much.

Quote:
 Originally Posted by plusandminus It would be interesting to this stat listed for all the players on a team. (I can do it myself, but not right now.) With the risk of being called an idiot, does the sum of all players equal 100??
I don't have calculations for an entire team. I think the totals would not exactly balance, but should not be way off either.

The total for all players on the team should be somewhere around:

5 * 82 * (team's ES pythagorean win%)

note: 5 is number of skaters per goal

For a team with equal ES GF and GA, should be ~205 "ES wins"

Quote:
 Originally Posted by plusandminus Pittsburgh were a bit special that year, starting the season with some very good players, just to see them drop off one by one. So Mario's stats sank deeper and deeper during the season. (If I remember right.)
Yeah, Pittsburgh was real "special" for a few years there post-Jagr.

In '94 and '96, Lemieux's R-On was slightly less than R-Off, although a big reason for that is Jagr being such a large part of the R-Off. His R-On/R-Off is great in '97 (1.97) and very good in 2001 (1.39), but a lot of that was due to those being the 1.5 seasons he played with Jagr at even strength. After that, he was mostly weak.

Quote:
 Originally Posted by plusandminus A thought I have, is that one might want to seperate forwards and defencemen, since they may not be easily comparable. An additional way of improving (or not) the method, could be to include ESpts in the calculations, to estimate how much different players contributed to their ESGF. Now I'm mainly thinking of doing it for forwards, to help seperate the offensive contributions of linemates, although it might be useful to apply (perhaps in a differnt form) to defencemen as well. To do something similar for ESGA would of course be basically impossible (unless one apply an assumption like "defencemen being more responsible for ESGA, while forwards being more responsible for ESGF").
Yes, separating forwards and defensemen is one possibilty. Would rather not do that... and what about players like Coffey that could almost be classified as either? I like using a player's ES points as a % of ESGF, but this requires even more data and doesn't address ESGA. I think the latter is probably the better way to go, or it may be better to just live with a "pure" but flawed metric.

Quote:
 Originally Posted by plusandminus Finally, which you likely are aware of, this stat only tells us about players' contributions during ES. So the "rankings" here are ES only. Creating similar stats for PP and SH would rank players differently. For example. While Forsberg had "much better" ES stats than Naslund in 2002-03, Naslund had better PP stats.
Of course this, like adjusted plus-minus, isn't an all-encompassing metric. It's meant to shed light on even strength value. Still, about ~75% of goals occur at even strength and it's even strength play that leads to penalties, so ES play is crucial to overall value.

It's not meant to measure all aspects of the game.

08-22-2011, 09:36 AM
#177
plusandminus
Registered User

Join Date: Mar 2011
Posts: 980
vCash: 500
Quote:
 I don't know why you are so against calculating GF/GA ratios. They shouldn't be used randomly, but in this case they are the primary basis of the pythagorean win% calculation, so of great importance.
I'll try to explain.

If I was asked to rank the following seasonal ES stats for players, without paying any attention at all to context, I would rank them as follow (with a tie for 2nd best):

 GF-GA GD GF/GA GF+GA (GF/GA)*(GF+GA) 72-50 +22 1.440 112 161.28 60-40 +20 1.500 100 150 40-20 +20 2.000 60 120 45-30 +15 1.500 75 112.5 7- 4 + 3 1.750 11 19.25 3- 1 + 2 3.000 4 12
GD=GF-GA (goal difference). GS=GF+GA (goal sum).

1. The guy with a GF/GA of 3.000 looks far too good compared to the others.
2. The lower numbers, the more extreme GF/GA. (It's a bit like pts per game. The fewer games played, the more extreme points per game.)

That's why I generally think one should be careful with using GF/GA.

Also:
3. Player ice time share during ES vary a lot between players. So does the amount time the player was not on the ice. No matter if one use real ice times, or take GF+GA, the differences are big.
4. Thus, when comparing "with" and "without", we would be comparing for example a GF/GA based on very low numbers, with a GF/GA based on very high numbers.

I'm not convinced yet regarding how good the win formula, and other formulas are at handling the things I mentioned above. Maybe they are great.

Quote:
 Originally Posted by Czech Your Math I'm not sure ice time is so important. If a player is on ice for a much higher % of ES GF+GA than his % of his ice time, that's okay. If he's able to perform at or above the GF/GA ratio of the team as a whole, then the more volume the better for the team. If he's at a worse GF/GA level, then the high volume will negatively effect the player portion of the metric even more.
The above seems based a lot on GF/GA, and I think GF/GA can "lie".

Let's say we have a 2-3 result without player on ice (GF/GA=0.400).
Player on ice doing 5-3 will make his team win 7-6, despite GF/GA=1.667.
Player on ice doing 2-1will only make his team draw 4-4, despite GF/GA of 2.00.

As I said, maybe the win formula and other formulas have methods to guard for such contradictions.

Quote:
 Whether 60/50 or 50/40 is better may depend most on context (all other things being equal). On a bad team, the extra 10/10 might be helpful, while on a good team, it may be hurtful.
By themselves, I think both are equal. Context may make one look better, but I'm not sure GF/GA is the best way to determine that.
I may be wrong.

08-22-2011, 01:05 PM
#178
Registered User

Join Date: Jan 2006
Location: bohemia
Country:
Posts: 4,842
vCash: 500
Quote:
Originally Posted by plusandminus
If I was asked to rank the following seasonal ES stats for players, without paying any attention at all to context, I would rank them as follow (with a tie for 2nd best):

 GF-GA GD GF/GA GF+GA (GF/GA)*(GF+GA) 72-50 +22 1.440 112 161.28 60-40 +20 1.500 100 150 40-20 +20 2.000 60 120 45-30 +15 1.500 75 112.5 7- 4 + 3 1.750 11 19.25 3- 1 + 2 3.000 4 12
GD=GF-GA (goal difference). GS=GF+GA (goal sum).

1. The guy with a GF/GA of 3.000 looks far too good compared to the others.
2. The lower numbers, the more extreme GF/GA. (It's a bit like pts per game. The fewer games played, the more extreme points per game.)

That's why I generally think one should be careful with using GF/GA.
The metric I've been tinkering with does not use any formula akin to (GF/GA)*(GF+GA). Neither does Overpass' adjusted plus-minus. Also, I agree that small sample sizes tend to lead to skewed results. That is why taking the best X seasons or career numbers are going to be more reliable for almost any metric.

In your 60/40 vs. 40/20 example, context is very important. First, it tells you in what environment the data was created. Second, it tells you what impact the player's performance is going to have. Since one player has 20/20 more than the other, if the R-Off of his team was > 1.0, then his performance did not help his team, while if it was < 1.0 it did help his team.

Quote:
 Originally Posted by plusandminus Also: 3. Player ice time share during ES vary a lot between players. So does the amount time the player was not on the ice. No matter if one use real ice times, or take GF+GA, the differences are big. 4. Thus, when comparing "with" and "without", we would be comparing for example a GF/GA based on very low numbers, with a GF/GA based on very high numbers. I'm not convinced yet regarding how good the win formula, and other formulas are at handling the things I mentioned above. Maybe they are great.
Again, that's why need multiple seasons to have any real solid data.

Maybe I should focus more or solely on the player's portion of the formula. I came up with the distribution of team's ES wins while thinking of some way to address Overpass' concern that players on great teams are hampered by the team's strong R-OFF. Honestly, getting credit for "just showing up" is not that great, although it's actually "playing a lot for great teams", and usually it's very good players who get lots of ice time over many years on great teams.

The player's portion is calculated by deducting his ES goals for/against from the team totals. The better the ratio and the more goals he was on ice for, the more impact it will have on the estimated win% differential, but it's more complex than multiplying the GF/GA ratio by the sum of ES GF + GA.

Quote:
 Originally Posted by plusandminus The above seems based a lot on GF/GA, and I think GF/GA can "lie". Let's say we have a 2-3 result without player on ice (GF/GA=0.400). Player on ice doing 5-3 will make his team win 7-6, despite GF/GA=1.667. Player on ice doing 2-1will only make his team draw 4-4, despite GF/GA of 2.00. As I said, maybe the win formula and other formulas have methods to guard for such contradictions.
All stats can "lie", 76% of statisticians can attest to that.

I'm not using GF/GA ratio as an absolute metric. In referring to Overpass' adjusted plus-minus, I do think R-ON/R-OFF is a valuable metric. It tells you in % terms how much more effective the team was with that player on the ice than without him on the ice, and that's a valuable piece of information.

Quote:
 Originally Posted by plusandminus By themselves, I think both are equal. Context may make one look better, but I'm not sure GF/GA is the best way to determine that. I may be wrong.
They are similar, but not equal in most cases. An extra 10 GF and 10 GA is outstanding on the '75 Capitals and rather weak on a dynasty team.

08-22-2011, 01:38 PM
#179
Registered User

Join Date: Jan 2006
Location: bohemia
Country:
Posts: 4,842
vCash: 500
Quote:
 Originally Posted by overpass Notice that the first list has 9 of the top 10 with an R-OFF below 1. On the second list, 9 of the top 10 have an R-OFF above 1.
Thanks for posting those, I like seeing the "pure" list.

On second list, six players in top 50 with R-OFF < 1:

Bourque, Jagr, M. Howe, Lindros, Thornton, Selanne

Rather underrated group for the most part.

While Dionne, Forsberg and Lindros all helped linemates make the list, it's especially impressive to see Jagr help two separate centers (Francis, Nylander) make the list.

Perhaps I should stay with a "pure" list and just use estimated win% (player portion of ES value)? I think that would look similar to your second (100% adjusted) list.

Last edited by Czech Your Math: 08-22-2011 at 01:52 PM.

08-22-2011, 03:46 PM
#180
plusandminus
Registered User

Join Date: Mar 2011
Posts: 980
vCash: 500
Quote:
 Originally Posted by Czech Your Math ...
It seems I got a bit misunderstood when I included a (GF/GA)*(GF+GA) column. That column was just there as an example of what results it would show. It was the other columns that were the important ones. Below I have deleted that column.

 GF-GA GD GF/GA GF+GA 72-50 +22 1.440 112 60-40 +20 1.500 100 40-20 +20 2.000 60 45-30 +15 1.500 75 7- 4 + 3 1.750 11 3- 1 + 2 3.000 4
GD=GF-GA (goal difference). GS=GF+GA (goal sum).

Quote:
 Originally Posted by Czech Your Math Also, I agree that small sample sizes tend to lead to skewed results. That is why taking the best X seasons or career numbers are going to be more reliable for almost any metric.
Yes. But has it been tested out how much more reliable?
I may try to examine that a bit more.

I also intend to look at game by game to see just what ES result the player had in the game ("with"), and what ES result the team had with him off ice ("without").
I know this can only be done for recent seasons, unless one wants to rely on estimated ES stats, but I still think it would be interesting to see what results it will produce. I'll get GP W D L GF-GA Pts for the players ("with") and for "without" them, and can then compare the two.

Quote:
 In your 60/40 vs. 40/20 example, context is very important. First, it tells you in what environment the data was created. Second, it tells you what impact the player's performance is going to have. Since one player has 20/20 more than the other, if the R-Off of his team was > 1.0, then his performance did not help his team, while if it was < 1.0 it did help his team.
I understand your example here. But I so far don't like GF/GA to be used, for reasons I have tried to put forward.

Quote:
 The player's portion is calculated by deducting his ES goals for/against from the team totals. The better the ratio and the more goals he was on ice for, the more impact it will have on the estimated win% differential
I understand that. That's what I call "with" and "without". And "with" may be unproportional compared to "without".

I will experiment a bit on my own to see how much it may affect the results.

Quote:
 All stats can "lie", 76% of statisticians can attest to that.
That was a funny statement.

Quote:
 They are similar, but not equal in most cases. An extra 10 GF and 10 GA is outstanding on the '75 Capitals and rather weak on a dynasty team.
Yes, I understand that.

08-22-2011, 06:59 PM
#181
Registered User

Join Date: Jan 2006
Location: bohemia
Country:
Posts: 4,842
vCash: 500
Quote:
 Originally Posted by plusandminus IYes. But has it been tested out how much more reliable? I may try to examine that a bit more.
The "pure" part is the player portion. Taking the differential of the estimated pythagorean win% based on performance. What part of that do you disagree with? I understand the exponent is a bit difficult to pinpoint, but I don't think it makes much difference when comparing players, since all would be affected similarly.

The assignment of some portion of the team's success based is somewhat arbitrary and perhaps not necessary at all.

Quote:
 Originally Posted by plusandminus I also intend to look at game by game to see just what ES result the player had in the game ("with"), and what ES result the team had with him off ice ("without"). I know this can only be done for recent seasons, unless one wants to rely on estimated ES stats, but I still think it would be interesting to see what results it will produce. I'll get GP W D L GF-GA Pts for the players ("with") and for "without" them, and can then compare the two.
Are you talking about ES data in games the player did not play? If you have that data, it would be interesting. However, if that's what you mean, why not just look at total results (record, GF, GA). I've calculated the actual vs. expected win% for a few players, and could also calculated expected win % using pythagorean based on GF/GA.

If you're bringing ice time into the picture, I don't consider that very important in comparison.

Quote:
 Originally Posted by plusandminus I understand that. That's what I call "with" and "without". And "with" may be unproportional compared to "without". I will experiment a bit on my own to see how much it may affect the results.
With and without, yes it's a fairly simple concept.
I don't understand what you mean by unproportional.

08-22-2011, 07:31 PM
#182
plusandminus
Registered User

Join Date: Mar 2011
Posts: 980
vCash: 500
Below is an example to illustrate the principle of looking at every single game to determine ES contributions. Everything in the table is ES only. 2002-03 season.

x=without player. tot (or t)=with player+without player
W D L = won draw loss
Pts=Pts, win=2, draw=1, loss=0
I need to keep table names short. Ask if unclear.

I think only games where the player were on the ice on a goal (no matter what type of goal) are counted. The missing games are all 0-0 for the player.
(Better would be to include all the games the player participated in, but unfortunately that informations seems to be depleted because of a disk crash several years ago.)

Just an example.

 Team Pos Name Game +/- x+/- Pts xPts totPts Pts-x tot-x W D L xW xD xL tW tD tL COL C PETER FORSBERG 68 55 1 97 61 91 36 30 40 17 11 21 19 28 39 13 16 COL R MILAN HEJDUK 73 49 -1 98 72 94 26 22 37 24 12 27 18 28 41 12 20 LA R ZIGMUND PALFFY 71 24 -26 86 53 71 33 18 30 26 15 17 19 35 25 21 25 DAL R JERE LEHTINEN 64 33 21 89 72 86 17 14 34 21 9 27 18 19 34 18 12 DAL D DERIAN HATCHER 75 27 27 90 85 99 5 14 36 18 21 32 21 22 40 19 16 TB R MARTIN ST. LOUIS 70 7 -19 78 54 67 24 13 29 20 21 16 22 32 25 17 28 COL L ALEX TANGUAY 69 38 10 88 76 89 12 13 35 18 16 31 14 24 38 13 18 DET D NICKLAS LIDSTROM 80 36 2 95 84 97 11 13 37 21 22 31 22 27 40 17 23 NJ R JAMIE LANGENBRUNNER 59 16 14 66 62 75 4 13 25 16 18 22 18 19 32 11 16
Forsberg was +55 during ES, without him Colorado was +1 during ES. Forsberg was 40-11-17 (W-L-D) during ES. That would have resulted in 97 pts in 68 games. It seems he (and his units) helped getting Colorado 30 more "ES points" (91 instead of 61). Again, everything is ES only, and only games where player where on the ice on a goal is counted.

Sorted another way, it would look like:

 Team Pos Name Game +/- x+/- Pts xPts totPts Pts-x tot-x W D L xW xD xL totW totD totL CBJ R DAVID VYBORNY 59 15 -63 71 32 37 39 5 24 23 12 9 14 36 14 9 36 COL C PETER FORSBERG 68 55 1 97 61 91 36 30 40 17 11 21 19 28 39 13 16 LA R ZIGMUND PALFFY 71 24 -26 86 53 71 33 18 30 26 15 17 19 35 25 21 25 CAR D SEAN HILL 66 5 -44 69 42 52 27 10 22 25 19 11 20 35 20 12 34 COL R MILAN HEJDUK 73 49 -1 98 72 94 26 22 37 24 12 27 18 28 41 12 20 MTL D ANDREI MARKOV 71 18 -23 88 62 66 26 4 32 24 15 20 22 29 25 16 30 TB R MARTIN ST. LOUIS 70 7 -19 78 54 67 24 13 29 20 21 16 22 32 25 17 28 CBJ L GEOFF SANDERSON 71 -1 -56 70 47 42 23 -5 20 30 21 13 21 37 14 14 43 LA L ALEXANDER FROLOV 58 16 -16 69 48 59 21 11 28 13 17 19 10 29 22 15 21 TB C VACLAV PROSPAL 70 7 -21 81 60 67 21 7 27 27 16 16 28 26 25 17 28

The difference between Vyborny on the ice, and Vyborny off the ice, is hugh. +15 with, -63 without. With him, 71 points in 59 games, without him only 32 points in 59 games. His contributions however only helped Columbus get 5 more points more than if he had been +/- 0 in every game. Same data and limitations as above.

Just an example. Just intended as something to add to the debate.
I know important stats are missing for older data. Again, just an example.

Pts, xPts and totPts can be used to tell us about pts per game:
 Team Pos Name Game ES+/- xES+/- Pts xPts totPts Pts-xPts totPts-xPts ATL C KAMIL PIROS 2 4 -2 2.000 0.000 1.500 2.000 1.500 WAS D MICHAEL FARRELL 1 1 -1 2.000 0.000 1.000 2.000 1.000 NAS D TOMAS KLOUCEK 1 1 -1 2.000 0.000 1.000 2.000 1.000 VAN R PAT KAVANAGH 2 2 -2 2.000 0.000 1.000 2.000 1.000

I personally find totals more useful.

Results does seem more useful if for example only taking those with minimum of 41 games:
 Team Pos Name Game ES+/- xES+/- Pts xPts totPts Pts-xPts totPts-xPts COL C PETER FORSBERG 68 55 1 1.426 0.897 1.338 0.529 0.441 COL R MILAN HEJDUK 73 49 -1 1.342 0.986 1.288 0.356 0.301 LA R ZIGMUND PALFFY 71 24 -26 1.211 0.746 1.000 0.465 0.254 NJ R JAMIE LANGENBRUNNER 59 16 14 1.119 1.051 1.271 0.068 0.220 DAL R JERE LEHTINEN 64 33 21 1.391 1.125 1.344 0.266 0.219 PHO L LADISLAV NAGY 62 24 -10 1.210 0.952 1.145 0.258 0.194 LA L ALEXANDER FROLOV 58 16 -16 1.190 0.828 1.017 0.362 0.190 COL L ALEX TANGUAY 69 38 10 1.275 1.101 1.290 0.174 0.188 DAL D DERIAN HATCHER 75 27 27 1.200 1.133 1.320 0.067 0.187 STL R ERIC BOGUNIECKI 43 11 -10 1.186 0.721 0.907 0.465 0.186

I prefer the totals (first two tables in this post) more than the averages.

Perhaps, though, "Pts as a total" might be wisely combined with "xPts per game".

Dividing, or dividing with square root, would give:
 Team Pos Name Game ES+/- xES+/- Pts PtsG xPts totPts totPts2 totPts3 COL C PETER FORSBERG 68 55 1 97 1.426 0.897 1.338 108.13114 102.41445 LA R ZIGMUND PALFFY 71 24 -26 86 1.211 0.746 1.000 115.20754 99.538178 COL R MILAN HEJDUK 73 49 -1 98 1.342 0.986 1.288 99.361111 98.678208 CBJ R DAVID VYBORNY 59 15 -63 71 1.203 0.542 0.627 130.90625 96.407176 MTL D ANDREI MARKOV 71 18 -23 88 1.239 0.873 0.930 100.7741 94.170744 STL D AL MACINNIS 78 19 -3 92 1.179 0.962 1.038 95.680000 93.821959 DET D NICKLAS LIDSTROM 80 36 2 95 1.188 1.050 1.212 90.476190 92.710506 TB C BRAD RICHARDS 75 8 -17 82 1.093 0.840 0.973 97.619047 89.469334 STL D BARRET JACKMAN 74 17 -6 87 1.176 0.959 1.014 90.676056 88.819012 TB R MARTIN ST. LOUIS 70 7 -19 78 1.114 0.771 0.957 101.11111 88.806906 NYR C ERIC LINDROS 75 15 -3 87 1.160 0.960 1.053 90.625 88.794003 WAS D SERGEI GONCHAR 75 17 -5 86 1.147 0.960 1.040 89.583333 87.773382
TotPts2=Pts/(xPts/Games). TotPts3=Pts/sqrt(xPts/Games)

I'm not saying this last table above gives the best results, just that it uses another way of combining "with" and "without".

Maybe the results of the "win formula" and/or overpass' method would be fairly similar (or not).

I think this is an interesting way to look at things. I know one needs some data that are not available for "old" seasons, but still. And the results here may be compared to the other methods in this thread, to see how they correspond to each other.

Edit:
The above should basically work for all eras, no matter how high or low scoring. Only thing to adjust for would be GP per season (for teams, i.e. 82 nowadays, 80 during Gretzy's prime).
It should also take care of injuries. If a player misses a game, he simply gets 0 pts for that game. It should be very easy to aggregate different seasons to get career totals (if it wasn't for necessary data missing).

Last edited by plusandminus: 08-23-2011 at 07:47 AM. Reason: spelling

08-23-2011, 11:40 AM
#183
seventieslord
Student Of The Game

Join Date: Mar 2006
Location: Regina, SK
Country:
Posts: 30,647
vCash: 500
Quote:
 OK. I've studied correlations between ES icetime and ESGF+ESGA and they don't necessarily correspond very well, perhaps especially regarding forwards. For example, during 2002-03, I guy like Kowalchuk (who had a reputation as an offensive minded player) had about 1.5 times more ESGF+ESGA played than the average player. So here we thus have our first bias. ESGF=60 and ESGA=50, will give higher "ice time" than ESGF=50 and ESGA=40, even if real ice time is the same. ?
I'm surprised CYM didn't mention/ask this, but... did you mean he had about 1.5 times the "average" player? or the average of his own team? there are only so many minutes to go around, and I find it unlikely that he would be that far off of his own team's average. If you compare him to another team, sure, his icetime figures would look wonky. but that's not what a model that uses GF & GA to calculate icetime does.

Also, Kovalchuk is about the most extreme example of a "high risk, high reward" type player. Watching the guy play it is clear he is going to cause both more goals for and more goals against while on the ice. If there was ever a player who would be an outlier whose GF/GA figures might cause an estimation model to lie, it would be him. With that said, that's no reason to throw out such a model.

Quote:
 If I was asked to rank the following seasonal ES stats for players, without paying any attention at all to context, I would rank them as follow (with a tie for 2nd best): GF-GA GD GF/GA GF+GA (GF/GA)*(GF+GA) 72-50 +22 1.440 112 161.28 60-40 +20 1.500 100 150 40-20 +20 2.000 60 120 45-30 +15 1.500 75 112.5 7- 4 + 3 1.750 11 19.25 3- 1 + 2 3.000 4 12 GD=GF-GA (goal difference). GS=GF+GA (goal sum). Comments: 1. The guy with a GF/GA of 3.000 looks far too good compared to the others. 2. The lower numbers, the more extreme GF/GA. (It's a bit like pts per game. The fewer games played, the more extreme points per game.) That's why I generally think one should be careful with using GF/GA.
I have an even more compelling reason to throw out the bottom player - any calculation done on him would be based on an obscenely low number of game situations. It's a poor sample size and is practically meaningless. Of course, if it got further in the season and he was 40-13, then we'd have something to talk about, and whether he was outperforming the 72-50 player would be a worthy question to ask.

08-23-2011, 01:46 PM
#184
plusandminus
Registered User

Join Date: Mar 2011
Posts: 980
vCash: 500
Quote:
 Originally Posted by seventieslord I'm surprised CYM didn't mention/ask this, but... did you mean he had about 1.5 times the "average" player? or the average of his own team? there are only so many minutes to go around, and I find it unlikely that he would be that far off of his own team's average. If you compare him to another team, sure, his icetime figures would look wonky. but that's not what a model that uses GF & GA to calculate icetime does.
Thanks for asking. I compare only with his team's totals.

Below are what it looks like. I have only included players with GF+GA >= 20. (Expect even larger differences for the other players.)
I think the raw data I assembled and corrected should be near 100 % correct.
The rightmost column shows GF+GA per 60 minutes. I have divided all those values by league average, to make it easier to compare.

Highest:
 Team Name pos ESTOI EStoi% ESGFGA% diff ESGFGA normESGFGAper60min CAR BRUNO ST. JACQUES D 297 0.081 0.138 0.057 35 1.577 CHI BURKE HENRY D 250 0.065 0.094 0.029 29 1.552 ATL ILYA KOVALCHUK F 1153 0.305 0.379 0.074 130 1.509 NYR REM MURRAY F 398 0.105 0.155 0.050 44 1.479 EDM MIKE COMRIE F 910 0.241 0.328 0.086 100 1.470 CHI ERIC DAZE F 722 0.188 0.257 0.070 79 1.464 LA ADAM DEADMARSH F 265 0.071 0.106 0.035 29 1.464 COL PETER FORSBERG F 1091 0.288 0.419 0.131 119 1.459 BOS JOE THORNTON F 1285 0.340 0.433 0.093 140 1.458 ATL KIRILL SAFRONOV D 414 0.110 0.131 0.022 45 1.454 TOR ALEXANDER MOGILNY F 943 0.258 0.348 0.090 102 1.447 TOR KAREL PILAR D 245 0.067 0.089 0.022 26 1.420 BOS GLEN MURRAY F 1400 0.371 0.458 0.087 148 1.414 ATL DANY HEATLEY F 1164 0.308 0.359 0.051 123 1.414 NYI ROMAN HAMRLIK D 1281 0.355 0.460 0.106 134 1.400 PIT MARIO LEMIEUX F 1082 0.286 0.387 0.101 113 1.397 MTL NIKLAS SUNDSTROM F 384 0.098 0.129 0.031 40 1.394 PHO LANDON WILSON F 328 0.089 0.125 0.036 34 1.387 SJ NICHOLAS DIMITRAKOS F 244 0.065 0.089 0.023 25 1.371 BOS MIKE KNUBLE F 1061 0.281 0.334 0.053 108 1.362

Lowest:
 Team Name pos ESTOI EStoi% ESGFGA% diff ESGFGA normESGFGAper60min WAS BRIAN SUTHERBY F 554 0.146 0.096 -0.05 27 0.652 PIT IAN MORAN F 1053 0.278 0.175 -0.10 51 0.648 CGY STEVE BEGIN F 436 0.118 0.078 -0.04 21 0.644 STL TYSON NASH F 532 0.142 0.087 -0.05 25 0.629 TOR RICHARD JACKMAN D 535 0.146 0.085 -0.06 25 0.625 FLA IGOR ULANOV D 823 0.215 0.146 -0.07 38 0.618 MIN JASON MARSHALL F 462 0.119 0.084 -0.04 21 0.608 TB ALEXANDER SVITOV F 511 0.132 0.083 -0.05 23 0.602 CAR JAROSLAV SVOBODA F 549 0.150 0.095 -0.05 24 0.585 TOR PAUL HEALEY F 458 0.125 0.068 -0.06 20 0.584 BUF VACLAV VARADA F 525 0.139 0.082 -0.06 22 0.561 CGY STEVE MONTADOR D 627 0.170 0.096 -0.07 26 0.555 STL SHJON PODEIN F 540 0.144 0.073 -0.07 21 0.520 COL SERGE AUBIN F 700 0.185 0.095 -0.09 27 0.516 PIT IAN MORAN D 1053 0.278 0.134 -0.14 39 0.496

Biggest differences between real ES%, and ES% estimated by GF+GA:
 Team Name pos ESTOI EStoi% ESGFGA% diff ESGFGA normESGFGAper60min PIT IAN MORAN D 1053 0.278 0.134 -0.14 39 0.496 COL PETER FORSBERG F 1091 0.288 0.419 0.131 119 1.459 NYI MATTIAS TIMANDER D 1159 0.321 0.213 -0.11 62 0.716 NYI ROMAN HAMRLIK D 1281 0.355 0.460 0.106 134 1.400 PIT MARIO LEMIEUX F 1082 0.286 0.387 0.101 113 1.397 PHO OSSI VAANANEN D 1056 0.287 0.192 -0.10 52 0.659 PIT IAN MORAN F 1053 0.278 0.175 -0.10 51 0.648 STL PAVOL DEMITRA F 1151 0.308 0.406 0.098 116 1.348 TB VINCENT LECAVALIER F 1134 0.293 0.390 0.097 108 1.274 BOS JOE THORNTON F 1285 0.340 0.433 0.093 140 1.458 WAS SERGEI GONCHAR D 1584 0.417 0.509 0.092 143 1.208 TOR ALEXANDER MOGILNY F 943 0.258 0.348 0.090 102 1.447 CGY STEPHANE YELLE F 1054 0.285 0.200 -0.09 54 0.686 COL SERGE AUBIN F 700 0.185 0.095 -0.09 27 0.516 BOS GLEN MURRAY F 1400 0.371 0.458 0.087 148 1.414 EDM MIKE COMRIE F 910 0.241 0.328 0.086 100 1.470 COL MILAN HEJDUK F 1219 0.322 0.405 0.083 115 1.262 TB MARTIN ST. LOUIS F 1123 0.290 0.372 0.082 103 1.227 FLA OLLI JOKINEN F 1168 0.305 0.385 0.080 100 1.146 BOS SEAN O'DONNELL D 1150 0.305 0.229 -0.08 74 0.861 VAN TODD BERTUZZI F 1223 0.334 0.413 0.079 117 1.280 DET MATHIEU DANDENAULT D 1203 0.314 0.392 0.078 122 1.357 PHO PAUL MARA D 1129 0.307 0.384 0.077 104 1.233 FLA IVAN NOVOSELTSEV F 960 0.250 0.327 0.077 85 1.185 MIN MARIAN GABORIK F 1078 0.278 0.355 0.077 89 1.105 ATL ILYA KOVALCHUK F 1153 0.305 0.379 0.074 130 1.509

This makes, in my opinion, GF+GA a bit unreliable when using it to determine ice time. Maybe there are steps involved in the calculations that takes care of that, so that it really doesn't matter that much. But I felt a need to point it out.

Quote:
 Also, Kovalchuk is about the most extreme example of a "high risk, high reward" type player. Watching the guy play it is clear he is going to cause both more goals for and more goals against while on the ice. If there was ever a player who would be an outlier whose GF/GA figures might cause an estimation model to lie, it would be him. With that said, that's no reason to throw out such a model.
Agree about Kowalchuk. But there were lots of other players like him. And plenty of the opposite kind too, i.e. players with "low risk getting scored on, but also low risk for their opponents to be scored on".

Quote:
 I have an even more compelling reason to throw out the bottom player - any calculation done on him would be based on an obscenely low number of game situations. It's a poor sample size and is practically meaningless. Of course, if it got further in the season and he was 40-13, then we'd have something to talk about, and whether he was outperforming the 72-50 player would be a worthy question to ask.
I agree. Unfortunately, the lower numbers the more "extreme" results. And the higher numbers, the less "extreme" results. But problem is that that rule continues to be true even for high numbers. The 40-13 case will still be slightly more vulnerable than the 72-50 case (I think, but that case it might be negligable).

Anyway, I don't like dividing GF by GA, because to me it only makes sense when
GF+GA is equally big for every player.
(But maybe there is something built in in overpass' method, that takes care of that.)

I look at it like this:
A: Player being 40-20 actually helped his team improve by +20.
B: Player being 30-15 actually helped his team improve by +15.
Both have the same GF/GA (2.0). But I think what matters is GF-GA.
Let's say B was 30-12 instead, raising his GF/GA to 2.5. That may make him look like a better player when looking at GF/GA. But he would have helped his team improve by +18.
During the whole season, I think a player's goal is to help their team improve by as many goals as possible, by getting as good GF-GA as possible.

Last edited by plusandminus: 08-23-2011 at 02:21 PM.

 08-23-2011, 02:26 PM #185 plusandminus Registered User   Join Date: Mar 2011 Posts: 980 vCash: 500 I had to edit the part about (and with) the tables in my last post. (I made a change before posting on the board, so that you would see if player played L, C, R, but that flawed the results, since some forwards played on more than one position, so I had to call all forwards F. If there is some guy playing both D and F, which is rare, the stats for that guy may be unreliable.) Last edited by plusandminus: 08-23-2011 at 02:34 PM.
 08-23-2011, 03:01 PM #186 overpass Registered User   Join Date: Jun 2007 Posts: 3,935 vCash: 500 plusminus, I'll try to address your concerns about the use of GF/GA ratios. First, my adjusted plus-minus metric does not use the player's on-ice GF/GA ratio (R-ON) in the calculations. Only the difference is used, as in regular plus-minus. The R-OFF ratio is used to calculate the team adjustment component, in conjunction with the sum of GA and GF. So the adjustment scales with the number of GF and GA, just as with regular plus-minus I really don't think your concerns about the use of ratios apply to this metric. The ratios are useful for a quick overview of the on/off ice relationship of the player and team, but are incomplete without a measure of quantity. Adjusted plus-minus includes a measure of quantity.
08-23-2011, 04:02 PM
#187
plusandminus
Registered User

Join Date: Mar 2011
Posts: 980
vCash: 500
Quote:
 Originally Posted by overpass plusminus, I'll try to address your concerns about the use of GF/GA ratios. First, my adjusted plus-minus metric does not use the player's on-ice GF/GA ratio (R-ON) in the calculations. Only the difference is used, as in regular plus-minus.
Thanks for clarifying.

Quote:
 The R-OFF ratio is used to calculate the team adjustment component, in conjunction with the sum of GA and GF. So the adjustment scales with the number of GF and GA, just as with regular plus-minus
I think I may understand. I think I perhaps need to try it out myself to be sure.

Quote:
 I really don't think your concerns about the use of ratios apply to this metric. The ratios are useful for a quick overview of the on/off ice relationship of the player and team, but are incomplete without a measure of quantity. Adjusted plus-minus includes a measure of quantity.
OK.

Sorry if I may have gotten things wrong. I think I need to re-produce myself what you are doing, to be sure I've understood things right.

08-23-2011, 04:08 PM
#188
seventieslord
Student Of The Game

Join Date: Mar 2006
Location: Regina, SK
Country:
Posts: 30,647
vCash: 500
Quote:
Originally Posted by plusandminus
Thanks for asking. I compare only with his team's totals.

Below are what it looks like. I have only included players with GF+GA >= 20. (Expect even larger differences for the other players.)
I think the raw data I assembled and corrected should be near 100 % correct.
The rightmost column shows GF+GA per 60 minutes. I have divided all those values by league average, to make it easier to compare.

Highest:
 Team Name pos ESTOI EStoi% ESGFGA% diff ESGFGA normESGFGAper60min CAR BRUNO ST. JACQUES D 297 0.081 0.138 0.057 35 1.577 CHI BURKE HENRY D 250 0.065 0.094 0.029 29 1.552 ATL ILYA KOVALCHUK F 1153 0.305 0.379 0.074 130 1.509 NYR REM MURRAY F 398 0.105 0.155 0.050 44 1.479 EDM MIKE COMRIE F 910 0.241 0.328 0.086 100 1.470 CHI ERIC DAZE F 722 0.188 0.257 0.070 79 1.464 LA ADAM DEADMARSH F 265 0.071 0.106 0.035 29 1.464 COL PETER FORSBERG F 1091 0.288 0.419 0.131 119 1.459 BOS JOE THORNTON F 1285 0.340 0.433 0.093 140 1.458 ATL KIRILL SAFRONOV D 414 0.110 0.131 0.022 45 1.454 TOR ALEXANDER MOGILNY F 943 0.258 0.348 0.090 102 1.447 TOR KAREL PILAR D 245 0.067 0.089 0.022 26 1.420 BOS GLEN MURRAY F 1400 0.371 0.458 0.087 148 1.414 ATL DANY HEATLEY F 1164 0.308 0.359 0.051 123 1.414 NYI ROMAN HAMRLIK D 1281 0.355 0.460 0.106 134 1.400 PIT MARIO LEMIEUX F 1082 0.286 0.387 0.101 113 1.397 MTL NIKLAS SUNDSTROM F 384 0.098 0.129 0.031 40 1.394 PHO LANDON WILSON F 328 0.089 0.125 0.036 34 1.387 SJ NICHOLAS DIMITRAKOS F 244 0.065 0.089 0.023 25 1.371 BOS MIKE KNUBLE F 1061 0.281 0.334 0.053 108 1.362

Lowest:
 Team Name pos ESTOI EStoi% ESGFGA% diff ESGFGA normESGFGAper60min WAS BRIAN SUTHERBY F 554 0.146 0.096 -0.05 27 0.652 PIT IAN MORAN F 1053 0.278 0.175 -0.10 51 0.648 CGY STEVE BEGIN F 436 0.118 0.078 -0.04 21 0.644 STL TYSON NASH F 532 0.142 0.087 -0.05 25 0.629 TOR RICHARD JACKMAN D 535 0.146 0.085 -0.06 25 0.625 FLA IGOR ULANOV D 823 0.215 0.146 -0.07 38 0.618 MIN JASON MARSHALL F 462 0.119 0.084 -0.04 21 0.608 TB ALEXANDER SVITOV F 511 0.132 0.083 -0.05 23 0.602 CAR JAROSLAV SVOBODA F 549 0.150 0.095 -0.05 24 0.585 TOR PAUL HEALEY F 458 0.125 0.068 -0.06 20 0.584 BUF VACLAV VARADA F 525 0.139 0.082 -0.06 22 0.561 CGY STEVE MONTADOR D 627 0.170 0.096 -0.07 26 0.555 STL SHJON PODEIN F 540 0.144 0.073 -0.07 21 0.520 COL SERGE AUBIN F 700 0.185 0.095 -0.09 27 0.516 PIT IAN MORAN D 1053 0.278 0.134 -0.14 39 0.496

Biggest differences between real ES%, and ES% estimated by GF+GA:
 Team Name pos ESTOI EStoi% ESGFGA% diff ESGFGA normESGFGAper60min PIT IAN MORAN D 1053 0.278 0.134 -0.14 39 0.496 COL PETER FORSBERG F 1091 0.288 0.419 0.131 119 1.459 NYI MATTIAS TIMANDER D 1159 0.321 0.213 -0.11 62 0.716 NYI ROMAN HAMRLIK D 1281 0.355 0.460 0.106 134 1.400 PIT MARIO LEMIEUX F 1082 0.286 0.387 0.101 113 1.397 PHO OSSI VAANANEN D 1056 0.287 0.192 -0.10 52 0.659 PIT IAN MORAN F 1053 0.278 0.175 -0.10 51 0.648 STL PAVOL DEMITRA F 1151 0.308 0.406 0.098 116 1.348 TB VINCENT LECAVALIER F 1134 0.293 0.390 0.097 108 1.274 BOS JOE THORNTON F 1285 0.340 0.433 0.093 140 1.458 WAS SERGEI GONCHAR D 1584 0.417 0.509 0.092 143 1.208 TOR ALEXANDER MOGILNY F 943 0.258 0.348 0.090 102 1.447 CGY STEPHANE YELLE F 1054 0.285 0.200 -0.09 54 0.686 COL SERGE AUBIN F 700 0.185 0.095 -0.09 27 0.516 BOS GLEN MURRAY F 1400 0.371 0.458 0.087 148 1.414 EDM MIKE COMRIE F 910 0.241 0.328 0.086 100 1.470 COL MILAN HEJDUK F 1219 0.322 0.405 0.083 115 1.262 TB MARTIN ST. LOUIS F 1123 0.290 0.372 0.082 103 1.227 FLA OLLI JOKINEN F 1168 0.305 0.385 0.080 100 1.146 BOS SEAN O'DONNELL D 1150 0.305 0.229 -0.08 74 0.861 VAN TODD BERTUZZI F 1223 0.334 0.413 0.079 117 1.280 DET MATHIEU DANDENAULT D 1203 0.314 0.392 0.078 122 1.357 PHO PAUL MARA D 1129 0.307 0.384 0.077 104 1.233 FLA IVAN NOVOSELTSEV F 960 0.250 0.327 0.077 85 1.185 MIN MARIAN GABORIK F 1078 0.278 0.355 0.077 89 1.105 ATL ILYA KOVALCHUK F 1153 0.305 0.379 0.074 130 1.509

This makes, in my opinion, GF+GA a bit unreliable when using it to determine ice time. Maybe there are steps involved in the calculations that takes care of that, so that it really doesn't matter that much. But I felt a need to point it out.
To be honest, I think this reveals how effective a model that uses GF/GA can be.

they actaually do include an adjustment based on the line or pairing the player was on, because as you can see, goals are more frequent on a per-minute basis when the higher minute players are out there. So this will partially mitigate the effect that you're seeing, an effect that is pretty small to begin with.

I mean, if you want to try to approximate icetime across the board from 1967-onwards, be my guest. But can you improve on what exists enough that it is worth the time it would take?

As for Kovalchuk, considering two other Thrashers are in your top-20, he doesn't appear to really be that much of an outlier on his team. He just barely makes the bottom list that you provided, and only because it is based on a "raw" difference and not a percentage difference (if it was based on that, he would not be anywhere near the top)

Quote:
 I agree. Unfortunately, the lower numbers the more "extreme" results. And the higher numbers, the less "extreme" results. But problem is that that rule continues to be true even for high numbers. The 40-13 case will still be slightly more vulnerable than the 72-50 case (I think, but that case it might be negligable). Anyway, I don't like dividing GF by GA, because to me it only makes sense when GF+GA is equally big for every player. (But maybe there is something built in in overpass' method, that takes care of that.) I look at it like this: A: Player being 40-20 actually helped his team improve by +20. B: Player being 30-15 actually helped his team improve by +15. Both have the same GF/GA (2.0). But I think what matters is GF-GA. Let's say B was 30-12 instead, raising his GF/GA to 2.5. That may make him look like a better player when looking at GF/GA. But he would have helped his team improve by +18. During the whole season, I think a player's goal is to help their team improve by as many goals as possible, by getting as good GF-GA as possible.
Keep in mind that generally when we talk about r-on and r-ff we are talking about very large sample sizes - seasons at the very least, and generally blocks of seasons, or careers. Like you, I don't see any value in the micro aspect, these 10-20-game segments and single game scenarios you are providing.

overpass, as usual, did a much better job than I could have done, fuddling my way through descriptions of statistics in "normal people" words and terms.

08-23-2011, 06:02 PM
#189
plusandminus
Registered User

Join Date: Mar 2011
Posts: 980
vCash: 500
Quote:
Originally Posted by overpass
Glossary of Terms:

SFrac: Season Fraction. 1.00 is a full season. I prefer it to games played because it gives a 48 game season, a 74 game season, an 80 game season or an 82 game season the same weight.
\$ESGF: Even-strength goals for, normalized to a 200 ESG scoring environment and with estimated SH goals removed.
\$ESGA: Even-strength goals against, normalized to a 200 ESG scoring environment and with estimated SH goals removed.
R-ON: Even strength GF/GA ratio when the player is on the ice.
R-OFF: Even-strength GF/GA ratio when the player is off the ice.
XEV+/-: Expected even-strength plus-minus, which is an estimate of the plus-minus that an average player would post with the same teammates. The calculation is described above.
EV+/-: Even –strength plus-minus, which is simply plus-minus with estimated shorthanded goals removed and normalized to a 200 ESG environment.
AdjEV+/-: Adjusted even-strength plus-minus, which is even-strength plus-minus minus expected even-strength plus-minus. This is the final number.
The following three stats evaluate special teams play and are not related to adjusted plus-minus. I’m including them in the table for a quick reference to the player’s contributions outside of even-strength play.
PP% : The % of the team’s power play goals for that the player was on the ice for.
SH%: The % of the team’s power play goals against that the player was on the ice for.
\$PPP/G: Power play points per game, normalized to a 70 PPG environment and with pre-1988 PP assists estimated.

Results
Here are the top 60 in career adjusted even-strength plus-minus, as well as the players in the HOH Top 100 and several others who were strongly considered for voting.

 Rk Player SFrac \$ESGF/G \$ESGA/G R-ON R-OFF XEV+/- EV+/- AdjEV+/- /Season PP% \$PPP/G SH% 1 Ray Bourque 20.30 1.17 0.85 1.37 0.96 -62 524 586 29 88% 0.45 58%
Sorry, but I'm still a bit confused.
The text says \$ESGF and \$ESGA are adjusted to 200 ESG scoring environment.
The \$ prefix makes me think "adjusted to 200 ESG scoring environment".
The text says EV+/- is also adjusted to 200 ESG scoring environment. ?

If it had been named "\$EV+/-", I would have been pretty sure it's \$ESGF-\$ESGA, because \$ tells me that it's "adjusted to 200 ESG scoring environment".
My question is... Is EV+/- = \$ESGF-\$ESGA ?

The text also says:
Quote:
 To calculate the adjusted plus-minus, I take the player’s on-ice total goals for and against as given. I calculate an expected plus-minus for the player, based on his team’s off-ice performance.
"As given", does that mean they are not "adjusted to 200 ESG scoring environment"?

Example, using made up seasonal stats for one player and his team.

Everything is ES only.
"without" or "w" means when player was off the ice.
GD (goal difference) = GF-GA.

 Lge aver ESGF per team teamGD teamGF teamGA playerGD playerGF playerGA withoutGD withoutGF withoutGA pGF/pGA wGF/wGA 100 +20 60 40 +10 24 14 +10 36 26 1.714 1.385

 Lge aver ESGF per team \$teamGD \$teamGF \$teamGA \$playerGD \$playerGF \$playerGA \$withoutGD \$withoutGF \$withoutGA pGF/pGA wGF/wGA 200 +40 120 80 +20 48 28 +20 72 52 1.714 1.385

Are the \$ values above true? Is that how the ""adjusted to 200 ESG scoring environment" values are calculated?

If the above is correct, then how do we use the different variables to calculate the missing columns?
R-On? R-Off?

Can someone write down the formula's for calculating the values, using the variables of the tables above?

And can we write "rOn" and "rOff" instead of "R-On" and "R-Off", to make it easier to understand the formulas? ("-" can otherwise be interpreted as a minus sign. But we don't take R minus On, and R minus Off, right? If, I'm even more confused.)

Quote:
 The expected plus-minus is calculated using the off-ice performance regressed partially to even, as a player should be expected to play somewhat better than a set of bad teammates or worse than a set of good teammates.
I understand the regression thing. But I'm confused about other things, including what exactly to regress (I guess it's "wGF/wGA"). ? Is rOff = "wGF/wGA" regressed to even?

Quote:
 I then calculate an actual plus-minus, which differs from official NHL plus-minus in that it is normalized to a scoring environment of 200 even-strength goal per season and does not include shorthanded goals. I subtract the “expected plus-minus” from the “actual plus-minus” to generate an adjusted plus-minus number.
I think I need an example to understand.

08-23-2011, 06:17 PM
#190
overpass
Registered User

Join Date: Jun 2007
Posts: 3,935
vCash: 500
Quote:
 Originally Posted by plusandminus Sorry, but I'm still a bit confused. The text says \$ESGF and \$ESGA are adjusted to 200 ESG scoring environment. The \$ prefix makes me think "adjusted to 200 ESG scoring environment". The text says EV+/- is also adjusted to 200 ESG scoring environment. ? If it had been named "\$EV+/-", I would have been pretty sure it's \$ESGF-\$ESGA, because \$ tells me that it's "adjusted to 200 ESG scoring environment". My question is... Is EV+/- = \$ESGF-\$ESGA ? The text also says: "As given", does that mean they are not "adjusted to 200 ESG scoring environment"?
All numbers are adjusted to the 200 ESG per team-season scoring environment. Yes, EV+/- is simply \$ESGF-\$ESGA. "As given" still means scoring-adjusted numbers.

Quote:
Originally Posted by plusandminus
Example, using made up seasonal stats for one player and his team.

Everything is ES only.
"without" or "w" means when player was off the ice.
GD (goal difference) = GF-GA.

 Lge aver ESGF per team teamGD teamGF teamGA playerGD playerGF playerGA withoutGD withoutGF withoutGA pGF/pGA wGF/wGA 100 +20 60 40 +10 24 14 +10 36 26 1.714 1.385

 Lge aver ESGF per team \$teamGD \$teamGF \$teamGA \$playerGD \$playerGF \$playerGA \$withoutGD \$withoutGF \$withoutGA pGF/pGA wGF/wGA 200 +40 120 80 +20 48 28 +20 72 52 1.714 1.385

Are the \$ values above true? Is that how the ""adjusted to 200 ESG scoring environment" values are calculated?
All correct. I don't scoring-level adjust the off-ice numbers, since I only use them to calculate a ratio, but if you did want to adjust for scoring level that would be correct.

Quote:
 Originally Posted by plusandminus If the above is correct, then how do we use the different variables to calculate the missing columns? R-On? R-Off? Can someone write down the formula's for calculating the values, using the variables of the tables above? And can we write "rOn" and "rOff" instead of "R-On" and "R-Off", to make it easier to understand the formulas? ("-" can otherwise be interpreted as a minus sign. But we don't take R minus On, and R minus Off, right? If, I'm even more confused.)
R-ON = \$ESGF/\$ESGA. For single seasons, multiple seasons, or careers.

R-OFF = (TeamESGF-PlayerESGF)/(TeamESGA-PlayerESGA) for a single season. For multiple seasons, take the sum of the XEV+/-, \$ESGF, and \$ESGA, and calculate by turning around the formula for XEV+/-

XEV+/- = (\$ESGF+\$ESGA)/(1+R-OFF^0.65)*R-OFF^0.65 - (\$ESGF+\$ESGA)/(1+R-OFF^0.65)

EV+/- = \$ESGF - \$ESGA

AEV+/- = (EV+/-) - (XEV+/-)

Quote:
 Originally Posted by plusandminus I understand the regression thing. But I'm confused about other things, including what exactly to regress (I guess it's "wGF/wGA"). ? Is rOff = "wGF/wGA" regressed to even? I think I need an example to understand.
Adjusted plus-minus applies the regression to the ratio rOff in the XEV calculation. It's simply rOff^0.65, which regresses rOff toward 1. If I understand your terms correctly, wGF/wGA = rOff as I present it. I have never presented the regressed rOff value alone.

Full example:

Player X has 60 ESGF, 45 ESGA. His team has 140 ESGF, 155 ESGA. League scoring level is 150 ESGA/team.

\$ESGF = 60*200/150 = 80
\$ESGA = 45*200/150 = 60

rOn = 60/45 = 1.33
rOff = (140-60)/(155-45) = 80/110 = 0.73

XEV+/- = (80+60)*(0.73^0.65)/(1+0.73^0.65) - (80+60)/(1+0.73^0.65)
= 140*0.81/1.81 - 140/1.81
= 63 - 77
= -14

EV+/- = 80 - 60 = 20

AEV+/- = 20 - 14 = 34

Last edited by overpass: 08-23-2011 at 11:10 PM.

08-24-2011, 01:42 PM
#191
plusandminus
Registered User

Join Date: Mar 2011
Posts: 980
vCash: 500
Quote:
 Originally Posted by overpass (Formulas.)
OK, I post my results for 2002-03 season using overpass' formulas. I have not adjusted to era, since I only look at one season. I used .65 as the "regress to even" number. Let me know if numbers are wrong. (Please don't overfocus on the number of decimals shown. I want them there for verifying purposes.)

Lowest expected +/-:
 Team Pos Name TOIshare +/- +/- without player rOn rOff exp+/- adj+/- PIT F MARIO LEMIEUX 0.2856 -15 -49 0.7656 0.5702 -20.4062 5.4062 CBJ D LUKE RICHARDSON 0.4065 -26 -41 0.6709 0.6204 -20.3193 -5.6806 CBJ D JAROSLAV SPACEK 0.3544 -19 -48 0.7206 0.5966 -19.4554 0.4554 ATL F DANY HEATLEY 0.3079 -1 -50 0.9839 0.6296 -18.3552 17.3552 PIT F ALEXEI KOVALEV 0.2415 -4 -60 0.9149 0.5420 -17.6831 13.6831 PIT D DICK TARNSTROM 0.2631 -7 -57 0.8600 0.5547 -17.5984 10.5984 CBJ F DAVID VYBORNY 0.2859 15 -82 1.5172 0.4810 -17.0432 32.0432 CBJ F GEOFF SANDERSON 0.2760 -1 -66 0.9767 0.5417 -16.7163 15.7163 PIT F MARTIN STRAKA 0.2428 -6 -58 0.8723 0.5573 -16.5250 10.5250 CAR D SEAN HILL 0.3440 5 -54 1.1471 0.5385 -14.4917 19.4917
(Comment: Dominated by Pittsburgh and Columbus players.)

Highest expected +/-:
 Team Pos Name TOIshare +/- +/- without player rOn rOff exp+/- adj+/- COL F JOE SAKIC 0.2269 9 45 1.2500 1.5696 11.7839 -2.7839 PHI D ERIC WEINRICH 0.3580 15 30 1.3750 1.4688 11.8073 3.1926 PHI D KIM JOHNSSON 0.3605 16 29 1.3902 1.4603 11.9996 4.0004 DAL D RICHARD MATVICHUK 0.2767 -5 57 0.8611 1.8028 12.6784 -17.6784 COL D ROB BLAKE 0.3652 19 35 1.4524 1.4795 13.0408 5.9591 DAL D DERIAN HATCHER 0.4141 27 25 1.5870 1.4098 13.2289 13.7710 VAN D BRENT SOPEL 0.3782 -10 35 0.8246 1.4861 13.3167 -23.3168 COL D ADAM FOOTE 0.3842 27 27 1.5510 1.4091 13.8747 13.1252 DAL D SERGEI ZUBOV 0.3819 17 35 1.4048 1.5385 14.0487 2.9512 COL F STEVEN REINPRECHT 0.2496 -9 63 0.8125 1.9403 18.4572 -27.4572 COL D GREG DE VRIES 0.4051 16 38 1.2759 1.6667 21.7152 -5.7152

 Team Pos Name TOIshare +/- +/- without player rOn rOff exp+/- adj+/- COL F PETER FORSBERG 0.2885 55 -1 2.7188 0.9880 -0.4688 55.4688 COL F MILAN HEJDUK 0.3223 49 5 2.4848 1.0610 2.2119 46.7881 DET D NICKLAS LIDSTROM 0.3878 36 1 1.7500 1.0112 0.4793 35.5207 LA F ZIGMUND PALFFY 0.3235 24 -28 1.6316 0.7228 -10.5125 34.5125 PHO F LADISLAV NAGY 0.3002 25 -26 1.8929 0.7593 -7.2309 32.2309 COL F ALEX TANGUAY 0.3104 38 16 2.1875 1.1928 5.8372 32.1627 CBJ F DAVID VYBORNY 0.2859 15 -82 1.5172 0.4810 -17.0432 32.0432 DAL F JERE LEHTINEN 0.2803 33 19 2.4348 1.2262 5.2278 27.7722 STL F ERIC BOGUNIECKI 0.2453 24 -10 1.8889 0.9083 -2.4386 26.4385 BOS F MIKE KNUBLE 0.2811 22 -11 1.5116 0.9027 -3.5934 25.5934 PHO F DAYMOND LANGKOW 0.3225 19 -20 1.5588 0.8039 -6.1608 25.1607 MTL D ANDREI MARKOV 0.3408 18 -26 1.4865 0.7869 -7.1518 25.1517

Comment: Similar to the method I used and posted yesterday (post #184). Forsberg atop here too. Hejduk higher, as is Tanguay, which I think is not so good. Lidstrom higher. Vyborny lower, I think he should be higher, as I think +15 with vs -82 "without" is a huge difference. Palffy, Lehtinen, Boguniecki on my list too.
Martin St Louis, who was high on my list(s), is missing here. He's 25th here. Perhaps it has to do with how he distributed his ESGF and ESGF game by game.

Best adjusted (if not "regressing R-Off to even"):
 Team Pos Name TOIshare +/- +/- without player rOn rOff exp+/- adj+/- COL F PETER FORSBERG 0.2885 55 -1 2.7188 0.9880 -0.7212 55.7212 COL F MILAN HEJDUK 0.3223 49 5 2.4848 1.0610 3.4023 45.5976 CBJ F DAVID VYBORNY 0.2859 15 -82 1.5172 0.4810 -25.58120 40.5812 LA F ZIGMUND PALFFY 0.3235 24 -28 1.6316 0.7228 -16.0919 40.0919 PHO F LADISLAV NAGY 0.3002 25 -26 1.8929 0.7593 -11.0842 36.0842 DET D NICKLAS LIDSTROM 0.3878 36 1 1.7500 1.0112 0.7374 35.26257 COL F ALEX TANGUAY 0.3104 38 16 2.1875 1.1928 8.9670 29.0330 MTL D ANDREI MARKOV 0.3408 18 -26 1.4865 0.7869 -10.9724 28.9724 NYI D ROMAN HAMRLIK 0.3548 16 -15 1.2712 0.8256 -12.8025 28.8025 PHO F DAYMOND LANGKOW 0.3225 19 -20 1.5588 0.8039 -9.4565 28.4565 STL F ERIC BOGUNIECKI 0.2453 24 -10 1.8889 0.9083 -3.7500 27.7500 BOS F MIKE KNUBLE 0.2811 22 -11 1.5116 0.9027 -5.5256 27.5255 ATL F DANY HEATLEY 0.3079 -1 -50 0.9839 0.6296 -27.9545 26.9545 CAR D SEAN HILL 0.3440 5 -54 1.1471 0.5385 -21.9000 26.9000 TB D DAN BOYLE 0.3600 13 -22 1.2889 0.7755 -13.0229 26.0229

Comment: At first glance it looks "better", in that linemates appear a bit more separated. But experience has shown me that regressing "when player off the ice" to even usually gives better results. (By the way, 6 Europeans atop.)

Although I like to compare the results to "my" mentioned method (see post #184), my method is not perfect. I need to find a way to include games where the player played but wasn't on the ice on any ES goals either way. Plus think more about it. Good things with it are that it doesn't need to pay attention to ice time, and doesn't have to adjust to "different GPG in different eras".

Now that I may know how overpass have done the calculations, and have been able to (hopefully) reproduce the results, my impression is that overpass' technique gives good results considered how relatively simple it is.
By simple, I mean that it only depends on ESGF and ESGA (and adjustment for ESGF per season), and that basically the only "tricky" part was the formula to calculate the expected +/-. That formula seem to produce interestingly good results.
(I did however not understand how it was done until it was written down in detail.)

The things I "don't like" about it, may be mostly present when looking at single seasons. When aggregating seasons, it might become alright (as players will play on other lines, on teams differently strong, etc). That is also what overpass have said.

Thanks overpass for explaining more in detail. It currently seems as if most of my "suspicions" regarding your method may have been a bit overstated.

Hopefully I will be able to reproduce czechyourmath's method too... eventually.

Last edited by plusandminus: 08-26-2011 at 06:55 PM.

08-24-2011, 04:28 PM
#192
plusandminus
Registered User

Join Date: Mar 2011
Posts: 980
vCash: 500
Quote:
 Originally Posted by Czech Your Math I have an alternative that might be fairer to players on great teams without using somewhat arbitrary regressions to the mean. It's not exactly comparable to adjusted plus-minus, but it uses much of the same methodology. For lack of a better term, I might call it "even strength value". It has two primary components: 1. Player's share of team success at even strength 2. Player's marginal (additional) success at even strength Once you calculate each component, simply add them together. Player's share of team success at ES is calculated as: 82 * (Team Exp. ES Win %) * (Player's ESGF + ESGA) / (Team's ESGF + ESGA) where Team Expected ES Win % = (ESGF)^N / (ESGF^N + ESGA^N) this is the pythagorean win formula; N = 2 (or another number if supported by data) Player's marginal contribution to team success is calculated as follows: Subtract the player's ESGF and ESGA from the team's totals. Recalculate the ES Win % from the new numbers (this is ES Win % without player). Subtract Team Exp. ES Win % from ES Win % without player. Multiply the difference in Win % by 82 to yield player's marginal contribution. Then add player's share of team success ES and player's marginal contribution to team success at ES to get "ES Value" (whatever is proper term). The results are in the same ballpark as plus-minus.
Is the above still true? Or did you come up with changes to improve further?

Anyway, I tried to calculate according to how I interpreted the instructions, and below are my results for the 2002-03 season.
Columns starting with "x" is "without" player, i.e. teamStat - playerStat. GS = goal sum (GF+GA).
I haven't multiplied anything by 82. Should be OK anyway, right?

You are welcome to check my math for errors and/or misunderstandings.

 Team Pos Name GF GA GS xGS xGF xGA +/- x+/- playerWin xWin teamWin pmargCont Win COL F PETER FORSBERG 87 32 119 109 82 83 55 -1 0.262331 0.286398 0.683506 0.419013 0.681344 COL F MILAN HEJDUK 82 33 115 103 87 82 49 5 0.247891 0.276771 0.683506 0.404928 0.652819 DAL D DERIAN HATCHER 73 46 119 79 86 61 27 25 0.204417 0.307920 0.688292 0.447368 0.651785 COL D ADAM FOOTE 76 49 125 81 93 66 27 27 0.194943 0.300838 0.683506 0.440139 0.635082 COL D GREG DE VRIES 74 58 132 70 95 57 16 38 0.168469 0.317685 0.683506 0.464787 0.633256 COL F ALEX TANGUAY 70 32 102 92 99 83 38 16 0.221417 0.245484 0.683506 0.359154 0.580571 WAS D SERGEI GONCHAR 80 63 143 32 68 70 17 -2 0.063001 0.281536 0.553229 0.508895 0.571896 DET D NICKLAS LIDSTROM 84 48 132 73 90 89 36 1 0.144899 0.262009 0.617310 0.424436 0.569335 DAL D PHILIPPE BOUCHER 60 38 98 74 99 69 22 30 0.191479 0.253581 0.688292 0.368420 0.559899 DAL D SERGEI ZUBOV 59 42 101 69 100 65 17 35 0.178541 0.261343 0.688292 0.379697 0.558238 PHI D KIM JOHNSSON 57 41 98 61 92 63 16 29 0.162122 0.260459 0.672411 0.387350 0.549472 COL D ROB BLAKE 61 42 103 73 108 73 19 35 0.175689 0.247891 0.683506 0.362675 0.538364 PHI D ERIC WEINRICH 55 40 95 60 94 64 15 30 0.159465 0.252486 0.672411 0.375493 0.534958 DAL F MIKE MODANO 55 30 85 77 104 77 25 27 0.199242 0.219942 0.688292 0.319547 0.518789 DAL F JERE LEHTINEN 56 23 79 85 103 84 33 19 0.219942 0.204417 0.688292 0.296991 0.516933 PHI D ERIC DESJARDINS 54 29 83 70 95 75 25 20 0.186042 0.220593 0.672411 0.328062 0.514104 OTT D WADE REDDEN 62 44 106 60 101 77 18 24 0.136208 0.240635 0.644722 0.373238 0.509446 BOS F GLEN MURRAY 83 65 148 29 84 91 18 -7 0.047945 0.244688 0.534016 0.458203 0.506148 DET D MATHIEU DANDENAULT 70 52 122 55 104 85 18 19 0.109170 0.242160 0.617310 0.392282 0.501452 COL D DEREK MORRIS 55 34 89 75 114 81 21 33 0.180503 0.214197 0.683506 0.313379 0.493882

Forsberg atop here too. But I think there are far too much Colorado dominance at the top. Basically, it seems to list the players with highest ESGF+ESGA on the teams.
We see some familiar names from the other two methods, like Hejduk, Lehtinen, Tanguay, but also many new.

Have I missed something in my calculations?

While I think "my" method and overpass' method ended up with quite similar results, I think this method gives the most "different" results. That does not necessarily have to bad, but looking at the table it does not seem to care much about "how good" the player played. Guys like Foote and DeVries don't look special +/- wise when comparing them to how Colorado did when they were off the ice.
The list is very much dominated by defencemen.
The only forwards on the list are: Forsberg-Hejduk-Tanguay, Modano-Lethinen and G.Murray. Among forwards will soon follow Bertuzzi-Naslund-Morrison (in between them are a few other forwards), all close to each other.

Shoudn't there be some consideration paid to GF-GA, or GF/(GF+GA), or even GF/GA?
Maybe I've missed something?

Edit: By the way, some guys ended up with slightly negative numbers. Is that OK?
Worst:
 Team Pos Name GF GA GS xGS xGF xGA +/- x+/- playerWin xWin teamWin pmargCont Win CBJ F KENT MCDONELL 0 1 1 -68 120 186 -1 -66 -0.064606 0.000950 0.291681 0.003256 -0.061350 CBJ F MATHIEU DARCHE 0 1 1 -68 120 186 -1 -66 -0.064606 0.000950 0.291681 0.003256 -0.061350

Edit:

I experimented a bit more.
teamWin = team win formula
xWin = appplying win formula but with "without" stats instead of team stats. ("Without"=team-player.)
Then the differences between the two.
playerWin = applying win formula but with player stats instead of team stats. Gives strange results for players with low numbers.
The results below looks far "better" than the ones above.
One thing I suspect is still missing, is to add something more to it. I think we know below much "difference" the player did, but I think there might be something more added? (Perhaps something to do with (playerGF+playerGA) / (teamGF+teamGA)?? I'm very tired now, by will continue probably tomorrow.

 Team Pos Name GF GA GS xGS xGF xGA +/- x+/- teamWin xWin Diff Diff% playerWin COL F PETER FORSBERG 87 32 119 109 82 83 55 -1 0.683506 0.493939 0.189567 1.383786 0.880833 COL F MILAN HEJDUK 82 33 115 103 87 82 49 5 0.683506 0.529559 0.153947 1.290707 0.860616 LA F ZIGMUND PALFFY 62 38 100 20 73 101 24 -28 0.485404 0.343142 0.142262 1.414586 0.726928 PHO F LADISLAV NAGY 53 28 81 24 82 108 25 -26 0.496310 0.365673 0.130637 1.357250 0.781797 DET D NICKLAS LIDSTROM 84 48 132 73 90 89 36 1 0.617310 0.505586 0.111724 1.220979 0.753846 CBJ F DAVID VYBORNY 44 29 73 -52 76 158 15 -82 0.291681 0.187898 0.103783 1.552336 0.697155 PHO F DAYMOND LANGKOW 53 34 87 18 82 102 19 -20 0.496310 0.392573 0.103737 1.264248 0.708448 NAS D JASON YORK 49 35 84 4 72 96 14 -24 0.460379 0.360000 0.100379 1.278830 0.662162 NYI D ROMAN HAMRLIK 75 59 134 17 71 86 16 -15 0.503436 0.405322 0.098114 1.242064 0.617724 STL F ERIC BOGUNIECKI 51 27 78 38 99 109 24 -10 0.548834 0.452033 0.096801 1.214145 0.781081 COL F ALEX TANGUAY 70 32 102 92 99 83 38 16 0.683506 0.587237 0.096269 1.163935 0.827143 TB D DAN BOYLE 58 45 103 4 76 98 13 -22 0.467543 0.375552 0.091991 1.244948 0.624234 MTL D ANDREI MARKOV 55 37 92 10 96 122 18 -26 0.474210 0.382406 0.091804 1.240069 0.688438 MIN F PASCAL DUPUIS 46 30 76 17 80 95 16 -15 0.503984 0.414910 0.089074 1.214682 0.701591 CAR D SEAN HILL 39 34 73 -44 63 117 5 -54 0.313326 0.224770 0.088556 1.393984 0.568173 LA F ALEXANDER FROLOV 49 33 82 12 86 106 16 -20 0.485404 0.396951 0.088453 1.222831 0.687965 DAL F JERE LEHTINEN 56 23 79 85 103 84 33 19 0.688292 0.600566 0.087726 1.146072 0.855661 BOS F MIKE KNUBLE 65 43 108 33 102 113 22 -11 0.534016 0.448970 0.085046 1.189424 0.695587 TB F MARTIN ST. LOUIS 57 46 103 2 77 97 11 -20 0.467543 0.386556 0.080987 1.209509 0.605591 STL D AL MACINNIS 69 50 119 33 81 86 19 -5 0.548834 0.470086 0.078748 1.167518 0.655694

Results looks much more similar to the other methods (those "by overpass" and "by me").

Dividing instead gives different results, see below. But those above are "better", right? ?

 Team Pos Name GF GA GS xGS xGF xGA +/- x+/- teamWin xWin Diff Diff% playerWin CBJ F DAVID VYBORNY 44 29 73 -52 76 158 15 -82 0.291681 0.187898 0.103783 1.552336 0.697155 LA F ZIGMUND PALFFY 62 38 100 20 73 101 24 -28 0.485404 0.343142 0.142262 1.414586 0.726928 CAR D SEAN HILL 39 34 73 -44 63 117 5 -54 0.313326 0.224770 0.088556 1.393984 0.568173 COL F PETER FORSBERG 87 32 119 109 82 83 55 -1 0.683506 0.493939 0.189567 1.383786 0.880833 PHO F LADISLAV NAGY 53 28 81 24 82 108 25 -26 0.496310 0.365673 0.130637 1.357250 0.781797 COL F MILAN HEJDUK 82 33 115 103 87 82 49 5 0.683506 0.529559 0.153947 1.290707 0.860616 CBJ F GEOFF SANDERSON 42 43 85 -68 78 144 -1 -66 0.291681 0.226845 0.064836 1.285816 0.488236 PIT F ALEXEI KOVALEV 43 47 90 -68 71 131 -4 -60 0.290868 0.227051 0.063817 1.281069 0.455643 NAS D JASON YORK 49 35 84 4 72 96 14 -24 0.460379 0.360000 0.100379 1.278830 0.662162 FLA D ANDREAS LILJA 36 29 65 -31 75 120 7 -45 0.356902 0.280898 0.076004 1.270575 0.606457 PHO F DAYMOND LANGKOW 53 34 87 18 82 102 19 -20 0.496310 0.392573 0.103737 1.264248 0.708448 ATL F DANY HEATLEY 61 62 123 -52 85 135 -1 -50 0.354528 0.283889 0.070639 1.248826 0.491870

Last edited by plusandminus: 08-24-2011 at 05:47 PM. Reason: adding more text

 08-24-2011, 05:50 PM #193 plusandminus Registered User   Join Date: Mar 2011 Posts: 980 vCash: 500 I added text to the previous, perhaps confused, post. Not sure if it became less or more confused. Very tired now.
 08-21-2012, 01:07 AM #194 OrrNumber4 Registered User     Join Date: Jul 2002 Country: Posts: 9,458 vCash: 500 This is just a fantastic thread. Thought I would bump it up so other's could see it. Should be pinned.
 10-01-2012, 09:07 AM #195 Sixbladeknife Registered User   Join Date: Oct 2011 Country: Posts: 36 vCash: 500 This is amazing stuff, very insightful! Do you have / are you willing to share the year-to-year spreadsheets? A friend of mine and I are trying to rank the best players since 1990, and this information would be very useful. Last edited by Sixbladeknife: 10-01-2012 at 09:37 AM. Reason: Typo
10-11-2012, 07:16 PM
#196
overpass
Registered User

Join Date: Jun 2007
Posts: 3,935
vCash: 500
Quote:
 Originally Posted by Sixbladeknife This is amazing stuff, very insightful! Do you have / are you willing to share the year-to-year spreadsheets? A friend of mine and I are trying to rank the best players since 1990, and this information would be very useful.
Just posted it here.

http://hfboards.hockeysfuture.com/sh....php?t=1270041

I don't fully endorse the single season plus-minus ratings as significant. There's a lot of random variation still at the season level. But I do find it useful for looking at groups of seasons like, say, Gretzky's Edmonton years compared to his LA years. Or for looking peak seasons (over several years) for any player.

11-30-2015, 07:17 PM
#197
LeBlondeDemon10
BlindLemon Haystacks

Join Date: Jul 2010
Country:
Posts: 3,256
vCash: 500
Quote:
Originally Posted by overpass
 Rk Years Player Seasons \$F/G \$A/G R-ON R-OFF XEV+/- EV+/- AEV+/- /Season PP% SH% 1 68-76 Bobby Orr 7.4 1.84 0.85 2.18 1.10 51 600 549 75 98% 64% 2 80-89 Wayne Gretzky 9.7 1.67 1.08 1.54 1.10 66 463 397 41 86% 37% 3 92-01 Jaromir Jagr 9.2 1.40 0.96 1.46 0.95 -27 337 365 39 68% 12% 4 83-92 Ray Bourque 9.2 1.20 0.79 1.51 0.94 -30 307 337 37 88% 54% 5 93-02 Eric Lindros 7.2 1.39 0.86 1.62 0.95 -20 312 332 46 74% 15% 6 77-86 Bryan Trottier 9.4 1.13 0.60 1.87 1.21 83 406 323 34 66% 24% 7 71-80 Bobby Clarke 9.8 0.94 0.47 1.98 1.20 68 374 305 31 65% 42% 8 81-90 Mark Howe 8.4 1.16 0.76 1.53 0.94 -29 277 305 36 61% 44% 9 78-87 Mike Bossy 9.4 1.08 0.60 1.80 1.17 67 370 302 32 75% 5% 10 95-04 John Leclair 8.4 1.19 0.72 1.64 1.07 31 320 289 34 68% 1% 11 72-81 Guy Lafleur 9.2 1.25 0.62 2.01 1.50 185 473 289 31 74% 5% 12 76-85 Marcel Dionne 9.7 1.04 0.82 1.28 0.80 -107 179 286 30 81% 14% 13 97-07 Peter Forsberg 6.9 1.17 0.66 1.76 1.00 1 287 285 41 74% 17% 14 74-83 Borje Salming 9.2 1.25 1.00 1.25 0.85 -89 190 280 30 72% 57% 15 74-83 Larry Robinson 9.3 1.51 0.83 1.83 1.54 247 522 275 30 50% 52% 16 74-83 Steve Shutt 9.4 1.06 0.51 2.09 1.51 160 424 264 28 42% 1% 17 88-97 Mario Lemieux 6.5 1.46 1.07 1.37 0.89 -51 210 260 40 98% 40%
Very interesting that some of these top seasons occurred because a player was paired with another: Trottier/Bossy, Lafleur/Shutt, Lindros/Leclair and Lemieux/Jagr.

11-30-2015, 07:30 PM
#198
LeBlondeDemon10
BlindLemon Haystacks

Join Date: Jul 2010
Country:
Posts: 3,256
vCash: 500
Quote:
 Originally Posted by Czech Your Math I made the correction to prevent double counting of value. I have run some players' career ES Values with deductions for replacement level win% of 0, .125 and .250 (note: .250 is approx. the worst win% for non-expansion teams). Career ES Value --------------------- Bourque 451 Gretzky 446 Robinson 423 Lidstrom 413 Jagr 362 Potvin 318 Lafleur 282 Clarke 271 Orr 258 Lemieux 255 Lindros 232 Forsberg 210 Bossy 208 Ovechkin 136 Crosby 117 Career ES Value ARP .125 ------------------------ Bourque 365 Robinson 349 Gretzky 355 Lidstrom 336 Jagr 294 Potvin 260 Lafleur 233 Clarke 227 Orr 220 Lemieux 202 Lindros 191 Forsberg 176 Bossy 174 Ovechkin 111 Crosby 95 Career ES Value ARP .250 ------------------------- Bourque 280 Robinson 275 Gretzky 265 Lidstrom 258 Jagr 226 Potvin 201 Lafleur 185 Clarke 183 Orr 182 Lindros 149 Lemieux 148 Forsberg 142 Bossy 141 Ovechkin 85 Crosby 72 I'm not sure that even with a .250 team win% deduction (which is the most I would consider) it would really solve the issue of defensemen with long careers on good teams being possibly over-represented at the top of the list. One solution could be to separate d-men and forwards for comparison purposes. Another might be to use a player's ES points as a % of ESGF on-ice for, to give more talented offensive players their just due.
Wow, no Esposito. Is that an oversight or was he that one dimensional?

12-05-2015, 01:27 PM
#199
Registered User

Join Date: Jan 2006
Location: bohemia
Country:
Posts: 4,842
vCash: 500
Quote:
 Originally Posted by LeBlondeDemon10 Wow, no Esposito. Is that an oversight or was he that one dimensional?
It varied, but overall he wasn't that far above average in terms of overall ES play. The first half of his career he was (likely) good to very good. The second half he was mediocre to weak.

Chicago (Young) Espo- No data for this period. It's likely that a younger Espo, playing with Hull, had positive adjusted +/- during this period.

Peak Espo- He has some very good numbers from '68 to '72, when he was at his peak.

Prime Espo- From '73 to '75, although he was still a scoring machine, his ES data was mediocre at best (more like somewhat weak).

NYR (Old) Espo- Once he left Boston, it was more weak than mediocre.

Forum Jump

 Bookmarks