NHL Goaltenders and Strength of Opponents

[Doctor No]

I've updated the site for strength of schedule metrics, for the NHL's 1982-83 and 1983-84 seasons. (I've also updated the variability metrics for the same two seasons, and the game-by-game data).

No goaltending pairs make the list (above) of "top 15 sheltered goaltenders".

Mike Palmateer looks to have been sheltered during his second stint with the Leafs - this surprised me somewhat.

Happy hunting!

 

[Doctor No]

Since I developed the SRS estimates (so far) for 2014-15, I figured I'd do a quick update here.

Looking at NHL goaltenders with 12 or more appearances this year, here are the five with the toughest opponent set thus far:

Ray Emery, +0.33
Semyon Varlamov, +0.25
Jonas Hiller, +0.19
James Reimer, +0.17
Calvin Pickard, +0.16

And here are the five with the easiest opponent set thus far:
Karri Ramo, -0.30
Ryan Miller, -0.20
Jonathan Quick, -0.17
Frederik Andersen, -0.16
Steve Mason, -0.15

There are some interesting teammate pairs in that mix. For teams where both goalies have played in 12 or more games, here are the largest spreads between goaltender pairs:

Calgary: 0.49 (Hiller +0.19, Ramo -0.30)
Philadelphia: 0.48 (Emery +0.33, Mason -0.15)
Toronto: 0.18 (Reimer +0.17, Bernier -0.01)
St. Louis: 0.15 (Elliott +0.10, Allen -0.05)
Winnipeg: 0.13 (Hutchinson +0.07, Pavelec -0.06)
Colorado: 0.09 (Varlamov +0.25, Pickard +0.16)
Edmonton: 0.08 (Scrivens +0.00, Fasth -0.08)
Arizona: 0.07 (Smith -0.07, Dubnyk -0.14)
Minnesota: 0.05 (Backstrom +0.05, Kuemper -0.00)
Ottawa: 0.03 (Anderson +0.12, Lehner +0.09)
Buffalo: 0.01 (Enroth +0.12, Neuvirth +0.11)

If I included Eddie Lack (11 games played), then Vancouver's gap would be 0.37 (Lack +0.17, Miller -0.20).

Some of the larger differences are quite interesting (at least to me), and do explain some of the traditional statistics that have been generated thus far. I'm currently working on an article related to this, but strength of schedule is well correlated with shots/game, save percentage, and GAA (each independently, and each in the expected direction).

For instance, based solely on data so far this season, a goaltender with a -0.3 strength of schedule would have an expected 92.0% save percentage on 28.9 shots/game, for a GAA of 2.31. A goaltender with a +0.3 strength of schedule would have an expected 91.1% save percentage on 29.6 shots/game, for a GAA of 2.64.

This metric is inclusive of playing at home or on the road; currently I measure home-ice advantage at 0.307 goals, so a goaltender who played exclusively at home (against "average" opponents) would have a schedule of -0.15, and one who played exclusively on the road (against "average" opponents) would have a schedule of +0.15.

 

[Doctor No]

Looking at the Mason/Emery split further, it looks like Emery has only played in five home games, but nine road games (although two of those road games were in relief).

 

[Doctor No]

I realize now that I've had four additional years of data added to my website since I last updated, so here's more of a data dump than anything.

Note that for the tables below, if a goaltender switched teams, then I didn't bother combining their teams here. I used a minimum 25 games played threshold for all tables.

Five Strongest Schedules, 1981-82 NHL:

Goaltender | Team | SoS Eddie Mio |New York Rangers|

0.36​

|
Greg Millen |Hartford Whalers|

0.25​

|
Pete Peeters |Philadelphia Flyers|

0.16​

|
Don Edwards |Buffalo Sabres|

0.13​

|
Denis Herron |Montreal Canadiens|

0.12​

|

Five Strongest Schedules, 1980-81 NHL:

Goaltender | Team | SoS Denis Herron |Montreal Canadiens|

0.23​

|
Pierre Hamel |Winnipeg Jets|

0.17​

|
Mike Liut |St. Louis Blues|

0.10​

|
Markus Mattsson |Winnipeg Jets|

0.08​

|
Rogie Vachon |Boston Bruins|

0.07​

|
Mike Veisor |Hartford Whalers|

0.07​

|

Five Strongest Schedules, 1979-80 NHL:

Goaltender | Team | SoS Steve Baker |New York Rangers|

0.16​

|
Pierre Hamel |Winnipeg Jets|

0.15​

|
Hardy Astrom |Colorado Rockies|

0.13​

|
Dan Bouchard |Atlanta Flames|

0.08​

|
Phil Myre |Philadelphia Flyers|

0.05​

|

Five Strongest Schedules, 1978-79 NHL:

Goaltender | Team | SoS Jim Bedard |Washington Capitals|

0.44​

|
Denis Herron |Pittsburgh Penguins|

0.32​

|
Bernie Parent |Philadelphia Flyers|

0.31​

|
Wayne Stephenson |Philadelphia Flyers|

0.27​

|
Ed Staniowski |St. Louis Blues|

0.25​

|

Five Easiest Schedules, 1981-82 NHL:

Goaltender | Team | SoS Vincent Tremblay |Toronto Maple Leafs|

-0.22​

|
Gilles Meloche |Minnesota North Stars|

-0.23​

|
Ron Low |Edmonton Oilers|

-0.26​

|
Rollie Melanson |New York Islanders|

-0.28​

|
Ed Staniowski |Winnipeg Jets|

-0.33​

|

Five Easiest Schedules, 1980-81 NHL:

Goaltender | Team | SoS Eddie Mio |Edmonton Oilers|

-0.20​

|
Doug Soetaert |New York Rangers|

-0.20​

|
Dan Bouchard |Quebec Nordiques|

-0.24​

|
Rejean Lemelin |Calgary Flames|

-0.29​

|
Richard Sevigny |Montreal Canadiens|

-0.31​

|

Five Easiest Schedules, 1979-80 NHL:

Goaltender | Team | SoS Gary Edwards |Minnesota North Stars|

-0.13​

|
Bob Sauve |Buffalo Sabres|

-0.14​

|
Pat Riggin |Atlanta Flames|

-0.18​

|
Bill McKenzie |Colorado Rockies|

-0.21​

|
Pete Peeters |Philadelphia Flyers|

-0.24​

|

Five Easiest Schedules, 1978-79 NHL:

Goaltender | Team | SoS Gary Bromley |Vancouver Canucks|

-0.07​

|
Tony Esposito |Chicago Black Hawks|

-0.16​

|
Michel Larocque |Montreal Canadiens|

-0.18​

|
Phil Myre |St. Louis Blues|

-0.18​

|
Greg Millen |Pittsburgh Penguins|

-0.34​

|

Anyhow, I'm sure exactly what I intend for you to do with these data - it's nice to see Jim Bedard get some recognition; I always felt that he was underappreciated in his day.

The workload split between (rookie) Greg Millen and (veteran) Denis Herron, I've written about on the History forum.

Pete Peeters' appearance on the "easier" list in 1979-80 was also his rookie season; I have a theory about rookie usage that I haven't been able to fully test yet. Peeters/Myre's workload split is similar to the above, although not as extreme.

Of course, Millen and Peeters would both end up on the "strongest" list by the end of this exercise, which suggests that they gained something from their limited exposure period.

Eddie Mio went from the easy list (1980-81 Oilers) to the strong list (1981-82 Rangers). I would have expected his statistics to suffer, although they did not. I do calculate a separate schedule strength metric (average opponents' shooting percentage relative to the league) which should roughly influence save percentage (that's the intent at least), and although Mio's schedule got harder, the shooting percentages got easier.

Hands in the air if you expected to see an early-1980s Leafs goaltender on the easy list. It looks like they used Larocque and Tremblay in a sheltering fashion, with Tremblay getting the easier end of it.

Do you remember Richard Sevigny's stellar 20-4-3 record for the Canadiens in 1980-81? His schedule helped, while Denis Herron (6-9-6) took the tough matchups. It's rare for a team to simultaneously have the goaltender with the toughest schedule and the goaltender with the easiest schedule, but Montreal did in 1980-81.

Speaking of schedule disparity, Ed Staniowski (the lowest on the 1981-82 list) would have also been the lowest on the list in 1980-81, if I'd expanded the criteria to 19 games minimum. Mike Liut (+0.10) and Staniowski (-0.64) is perhaps the largest difference I've seen between a club's top two.

 

[Colt.45Orr]

I read it and find it interesting.

What's stopping people from replying most likely, is, we have very little to add!

Same here.

BTW --was there a more uneven split than Rask/Johnson last year?

Isles fans are disgusted with him, but he got so much rest AND all the softball teams last year.

 

[Doctor No]

Great point about Rask (-0.02) and Johnson (-0.28); last year was also the first year that Rask had the easier schedule of Boston's goaltenders:

http://www.hockeygoalies.org/bio/nhl/boston.html

(This year, Rask is at -0.05 and Svedberg is at -0.21, so he's taking the "senior" role once more).

Looking at pairs where both goaltenders played at least twenty games apiece, the Rask/Johnson schedule split (0.26) was the third-largest out there. Not quite historical levels (see post #12), but then again, we're in an era where the best teams and the worst teams are closer together, so it's hard to be "historical".

Other notable differences last year (2013-14):
Montreal: Budaj (+0.16), Price (-0.15) - total spread of 0.31
Minnesota: Backstrom (+0.20), Kuemper (+0.03), Harding (-0.10) - total spread of 0.30
Philadelphia: Emery (+0.16), Mason (-0.09) - total spread of 0.25
NY Islanders: Poulin (+0.18), Nabokov (-0.06) - total spread of 0.24
Carolina: Peters (+0.11), Ward (+0.01), Khudobin (-0.10) - total spread of 0.21
Winnipeg: Montoya (+0.17), Pavelec (-0.01) - total spread of 0.18
Columbus: McElhinney (+0.07), Bobrovsky (-0.08) - total spread of 0.15

It shouldn't be surprising that on the list of those facing easier schedules, many performed well (the glaring exception would be Pavelec).

 

[Doctor No]

Here are the 2014-15 NHL goaltenders who have played at least ten games with a team, and my current best estimate of their schedule strength (using a Simple Rating System). Team strengths are based on the values in the other thread.

The final column represents the greatest difference between one team's goaltenders.

Results here are through January 14, 2015.

Team | Goaltender One | 2014-15 Strength | Goaltender Two | 2014-15 Strength | Goaltender Three | 2014-15 Strength | Schedule Diff
CGY| Jonas Hiller |+0.13| Karri Ramo |-0.30|||+0.43
VAN| Ryan Miller |-0.14| Eddie Lack |+0.26|||+0.40
PHI| Steve Mason |-0.13| Ray Emery |+0.22|||+0.35
LOS| Jonathan Quick |-0.16| Martin Jones |+0.12|||+0.28
BOS| Tuukka Rask |-0.02| Niklas Svedberg |-0.29|||+0.28
SJS| Antti Niemi |+0.04| Alex Stalock |-0.21|||+0.24
MIN| Darcy Kuemper |+0.00| Niklas Backstrom |+0.22|||+0.22
COL| Semyon Varlamov |+0.12| Calvin Pickard |+0.13| Reto Berra |-0.09|+0.21
STL| Jake Allen |-0.12| Brian Elliott |+0.08|||+0.21
DET| Jimmy Howard |-0.09| Petr Mrazek |-0.23|||+0.14
EDM| Ben Scrivens |+0.07| Viktor Fasth |-0.06|||+0.13
CHI| Corey Crawford |+0.00| Antti Raanta |+0.11|||+0.10
CAR| Cam Ward |+0.04| Anton Khudobin |+0.14|||+0.10
TBL| Ben Bishop |-0.05| Evgeni Nabokov |-0.15|||+0.10
ARI| Mike Smith |-0.03| Devan Dubnyk |-0.12|||+0.09
TOR| Jonathan Bernier |+0.03| James Reimer |+0.11|||+0.08
FLO| Roberto Luongo |-0.11| Alvaro (Al) Montoya |-0.05|||+0.06
OTT| Craig Anderson |+0.05| Robin Lehner |+0.09|||+0.04
CBJ| Sergei Bobrovsky |+0.05| Curtis McElhinney |+0.02|||+0.03
NYR| Henrik Lundqvist |-0.02| Cameron Talbot |+0.00|||+0.02
BUF| Jhonas Enroth |+0.09| Michal Neuvirth |+0.08|||+0.02
NYI| Jaroslav Halak |+0.09| Chad Johnson |+0.11|||+0.01
WPG| Ondrej Pavelec |-0.04| Michael Hutchinson |-0.03|||+0.01
ANA| Fredrik Andersen |-0.15|||||
DAL| Kari Lehtonen |+0.01|||||
MON| Carey Price |+0.07|||||
NJD| Cory Schneider |+0.07|||||
NSH| Pekka Rinne |-0.06|||||
PIT| Marc-Andre Fleury |-0.05|||||
WAS| Braden Holtby |-0.05|||||

Results are sorted from greatest single-team difference to smallest single team-difference.

Goaltender links are active (and if you click on them, you can dig down into the game logs, and I can say that more people are visiting my site).

 

[Doctor No]

One interesting thing about Svedberg's schedule so far - even though most of his games have been on the road (9 of 12), his opponent set has been pretty bad (featuring Buffalo twice, Columbus twice, Colorado, New Jersey).

 

[Doctor No]

Here's an 2014-15 NHL update, with goaltenders through the all-star break:

Team | Goaltender One | 2014-15 Strength | Goaltender Two | 2014-15 Strength | Goaltender Three | 2014-15 Strength | Schedule Diff
CGY| Jonas Hiller |+0.14| Karri Ramo |-0.31|||+0.45
VAN| Ryan Miller |-0.11| Eddie Lack |+0.28|||+0.39
PHI| Steve Mason |-0.11| Ray Emery |+0.22|||+0.33
COL| Semyon Varlamov |+0.18| Calvin Pickard |+0.12| Reto Berra |-0.15|+0.33
MON| Carey Price |+0.09| Dustin Tokarski |-0.23|||+0.32
BOS| Tuukka Rask |-0.03| Niklas Svedberg |-0.32|||+0.29
SJS| Antti Niemi |+0.04| Alex Stalock |-0.19|||+0.22
LOS| Jonathan Quick |-0.12| Martin Jones |+0.08|||+0.21
MIN| Darcy Kuemper |+0.01| Niklas Backstrom |+0.18|||+0.17
EDM| Ben Scrivens |+0.09| Viktor Fasth |-0.05|||+0.15
ARI| Mike Smith |+0.05| Devan Dubnyk |-0.10|||+0.15
STL| Jake Allen |-0.08| Brian Elliott |+0.04|||+0.13
FLO| Roberto Luongo |-0.16| Alvaro (Al) Montoya |-0.04|||+0.12
CAR| Cam Ward |+0.01| Anton Khudobin |+0.12|||+0.11
PIT| Marc-Andre Fleury |+0.00| Thomas Greiss |+0.10|||+0.10
TBL| Ben Bishop |-0.10| Evgeni Nabokov |-0.17|||+0.08
WPG| Ondrej Pavelec |-0.08| Michael Hutchinson |+0.00|||+0.08
CHI| Corey Crawford |+0.04| Antti Raanta |-0.02|||+0.05
OTT| Craig Anderson |+0.06| Robin Lehner |+0.09|||+0.03
BUF| Jhonas Enroth |+0.09| Michal Neuvirth |+0.06|||+0.02
TOR| Jonathan Bernier |+0.06| James Reimer |+0.08|||+0.02
CBJ| Sergei Bobrovsky |+0.05| Curtis McElhinney |+0.04|||+0.02
DET| Jimmy Howard |-0.10| Petr Mrazek |-0.12|||+0.02
NYI| Jaroslav Halak |+0.10| Chad Johnson |+0.09|||+0.01
NYR| Henrik Lundqvist |-0.01| Cameron Talbot |-0.01|||+0.01
ANA| Fredrik Andersen |-0.14|||||
DAL| Kari Lehtonen |+0.03|||||
NJD| Cory Schneider |+0.07|||||
NSH| Pekka Rinne |-0.05|||||
WAS| Braden Holtby |-0.05|||||

Montreal and Pittsburgh enter the "interesting" list, now that Tokarski and Greiss have appeared in ten games apiece.

The general rule is that - unless coaches are doing something deliberate - these differences get closer to zero as the season wears along.

If you see a goaltender on the above list and this list, and you're wondering why he moved more than you think he should, remember that goaltender's strength of schedule can change for two reasons: (1) they played additional opponents, and (2) the SRS measurement of their opponents changed with additional data. This iteration *does* include the adjustments to my SRS formulas, which turned out to have a very minor impact.

 

[Doctor No]

One thing to add - since it looked odd enough that I had to check it manually - is Devan Dubnyk's Minnesota Wild career.

He doesn't have ten games played for Minnesota yet, so he doesn't appear on the chart, but his strength of schedule is really low so far:

Buffalo on the road (-1.434)
Arizona at home (-1.260)
Columbus at home (-0.750)
Detroit on the road (+0.456) - only played about half of this one

I weight these by minutes played, and Dubnyk currently comes in with a Minnesota strength of schedule of -0.94.

Expect his statistics to get closer to his true level of ability once his opportunities even out.

 

[Canadian Canuck]

Eddie Lack always faces tough teams

 

[Doctor No]

Eddie Lack always faces tough teams

Sure looks that way, and he's got a small road game "advantage" as well.

 

[Doctor No]

One thing to add - since it looked odd enough that I had to check it manually - is Devan Dubnyk's Minnesota Wild career.

He doesn't have ten games played for Minnesota yet, so he doesn't appear on the chart, but his strength of schedule is really low so far:

Buffalo on the road (-1.434)
Arizona at home (-1.260)
Columbus at home (-0.750)
Detroit on the road (+0.456) - only played about half of this one

I weight these by minutes played, and Dubnyk currently comes in with a Minnesota strength of schedule of -0.94.

Expect his statistics to get closer to his true level of ability once his opportunities even out.

Even on the road, playing Edmonton tonight's not exactly going to help Dubnyk's strength of schedule.

He might have a real shot at the records in post #4 (although to be fair, I'd probably combine his Arizona and Minnesota numbers for that comparison).

 

[Caeldan]

So for Ottawa Lehner's actually had the tougher schedule. He's getting rather infrequent starts, which makes it seem odd that he wouldn't be sheltered - especially since he hasn't looked good in them too.

Do you have the granularity to refine by back-to-backs to just see who gets the tougher of the two games?

 

[Doctor No]

So for Ottawa Lehner's actually had the tougher schedule. He's getting rather infrequent starts, which makes it seem odd that he wouldn't be sheltered - especially since he hasn't looked good in them too.

Exactly - although the difference is slight.

I calculate these on a minutes-weighted basis for all games (not just starts), so my gut tells me that a backup goaltender would have a tougher schedule, all else being equal (of course, all else is never equal). Rationally, what games are you more likely to need the backup? Against tougher opponents.

Looking at it, though, it doesn't seem to bear out (at least for the last few years). Definitely bears more studying.

Do you have the granularity to refine by back-to-backs to just see who gets the tougher of the two games?

I've got that, and it's on my list to code it in an algorithm, but right now I have to do it manually. Here's what I believe are the Senators' back-to-back games, and the goaltender schedules involved:

(Please correct if something looks funky)

October 25 - Lehner vs. New Jersey (-0.585)
October 26 - Anderson at Chicago (+1.052)

November 8 - Anderson vs. Winnipeg (+0.177)
November 9 - Lehner vs. Toronto (-0.255)

November 28 - Anderson at Florida (-0.331)
November 29 - Lehner at Tampa Bay (+0.638)

December 6 - Anderson at Pittsburgh (+0.627)
December 7 - Anderson vs. Vancouver (+0.070)

December 19 - Anderson vs. Anaheim (+0.108)
December 20 - Lehner at Montreal (+0.553)

January 3 - Anderson at Boston (+0.208)
January 4 - Lehner vs. Tampa Bay (+0.341)

Looks reasonably split (tough vs. easy) in the back-to-backs, but I didn't actually pull out the calculator to composite them.

 

[Caeldan]

Now here's my other random thought I was considering, and I wonder if there's any worth to doing this...

But taking the strength of schedule and using it to normalize the GAA and SV% of each goalie? I wonder if it'd end up actually showing anything of note, or just be a bunch of extra effort to show things we already know/can tell other ways.

That and of course, be able to compare back to backs, or possibly even look at in backup cases what kind of opponents they're facing on >7 days of rest.

 

[Doctor No]

Now here's my other random thought I was considering, and I wonder if there's any worth to doing this...

But taking the strength of schedule and using it to normalize the GAA and SV% of each goalie? I wonder if it'd end up actually showing anything of note, or just be a bunch of extra effort to show things we already know/can tell other ways.

That and of course, be able to compare back to backs, or possibly even look at in backup cases what kind of opponents they're facing on >7 days of rest.

I'm working on a paper related to this. Without giving away too much, there's good correlation between strength of schedule (as I'm currently defining it) and each of the three metrics:

• Shots faced per game
• Save percentage
• Goals-against average (which is basically the first times the complement of the second)

Clarifying - each of the three are separately correlated (in the "expected" direction) with strength of schedule. Based on the 2014-15 NHL data through the all-star break, opponent strength has the following r^2 relationships with these:

• Shots faced per game - 78%
• Save percentage - 51%
• Goals-against average - 95%

(To deal with smaller data sizes on each end, I'm bucketing everything into shifting 0.4-unit intervals, with 0.1-unit increments. So I look at save percentage for all teams with strength -0.2 to 0.2, then -0.1 to 0.3, then 0.0 to 0.4, et cetera)

So it's a nice shortcut for modern seasons (where we have more granular information available, like even strength save percentage, or power plays faced).

For earlier seasons, where those data aren't available, it's a big step (my opinion). I've got this on my site back to 1978-79 so far, and I've found it reasonably illuminating.

 

[hatterson]

I'm shocked that the r^2 of opp strength vs GAA is so high. I mean, intuitively, I would have expected a not insignificant relationship, but that blows my expectations out of the water.

How does that compare to past seasons? Is the first half (ish) of 2014-2015 an aberration or do past seasons also show this incredibly high relationship?

 

[Doctor No]

My first impression was that I agreed that it seemed high, however it's probably a bit circular:

My "team strength" metric is a reflection of goal differential, adjusted for schedule. Teams can have a good goal differential by one of two ways: they can score a lot of goals, or they can allow a small number of goals.

So now that I think about it, I'm not as surprised at the correlation between (A*B) and opponent strength, where A = shots/60 and B = shooting percentage. I'm more curious about the separate relationships with A and B, but the nice thing about this stuff is that there's always plenty to look at.


Various r^2 values:
For 2013-14: shots/60 93%, sv% 77%, GAA 91%
For 2012-13: shots/60 70%, sv% 76%, GAA 86%
For 1993-94: shots/60 79%, sv% 67%, GAA 88%
For 1983-84: shots/60 69%, sv% 79%, GAA 95%

I picked 1993-94 because it's one of my favorite seasons, and picked 1983-84 because (1) it's ten years earlier, and (2) there were more spread between high and low teams.

So 2014-15 in progress does appear to be more correlated (in total) than most seasons (roughly defined as "I picked four").

 

[hatterson]

Yea, it makes sense that there'd be a non-trivial relationship, but I'm still surprised it's so high given that SoS isn't just based on the opponents goals for. Although as I glance through the stats, I am noticing a fairly decent correlation between goals for per game and goal differential per game on the team level. The same also holds for GA/g (R^2 for both were in the .6-.7 range on the season by season basis for the 4 seasons I checked [00-01, 03-04, 06-07 and 13-14]).

What that seems to imply to me is that as much as we think of teams that are great on one side and terrible on the other, it just doesn't end up happening all that much in reality. Same with the idea that teams frequently sacrifice offensive performance to get better defensively (or vice vera). For the most part, offensive numbers were independent of defensive numbers (r^2 of .1-.15 for ga vs gf) so although you'll get the occasional team who is great offensively and not great defensively, teams are "expected" to be middle of the pack in both metrics so improving one leads to an improved goal differential since the other isn't expected to drop.

Although since my brain is on this path, it may be interesting to break out strength of schedule into offensive and defensive components. Defensive strength of schedule being how talented (offensively only) the teams you faced were, and offensive being the opposite.

The reason I'm thinking that is because I'd expect a goalies stats to be much better (excluding wins, and assuming equal talent) if he played the 07-08 Rangers (2.5g/g) 82 times versus the 07-08 Senators (3.15g/g) despite both teams have a very similar goal differential.

I would expect (based on a guess, not data) that GAA would be very strongly correlated with defensive strength of schedule which would also mean a strong correlation to overall strength of schedule, but not necessarily a strong correlation with offensive strength of schedule. However, given what I said above about the separate correlations between goals for and differential and goals against and differential, that may not make too much of a difference over a combined SoS metric, but it would be interesting to see nonetheless.

Forgive me if some of this is a little rambly or doesn't make sense, I'll blame it on being hopped up on cold meds this morning

 

[Doctor No]

Yea, it makes sense that there'd be a non-trivial relationship, but I'm still surprised it's so high given that SoS isn't just based on the opponents goals for. Although as I glance through the stats, I am noticing a fairly decent correlation between goals for per game and goal differential per game on the team level. The same also holds for GA/g (R^2 for both were in the .6-.7 range on the season by season basis for the 4 seasons I checked [00-01, 03-04, 06-07 and 13-14]).

What that seems to imply to me is that as much as we think of teams that are great on one side and terrible on the other, it just doesn't end up happening all that much in reality. Same with the idea that teams frequently sacrifice offensive performance to get better defensively (or vice vera). For the most part, offensive numbers were independent of defensive numbers (r^2 of .1-.15 for ga vs gf) so although you'll get the occasional team who is great offensively and not great defensively, teams are "expected" to be middle of the pack in both metrics so improving one leads to an improved goal differential since the other isn't expected to drop.

Although since my brain is on this path, it may be interesting to break out strength of schedule into offensive and defensive components. Defensive strength of schedule being how talented (offensively only) the teams you faced were, and offensive being the opposite.

The reason I'm thinking that is because I'd expect a goalies stats to be much better (excluding wins, and assuming equal talent) if he played the 07-08 Rangers (2.5g/g) 82 times versus the 07-08 Senators (3.15g/g) despite both teams have a very similar goal differential.

I would expect (based on a guess, not data) that GAA would be very strongly correlated with defensive strength of schedule which would also mean a strong correlation to overall strength of schedule, but not necessarily a strong correlation with offensive strength of schedule. However, given what I said above about the separate correlations between goals for and differential and goals against and differential, that may not make too much of a difference over a combined SoS metric, but it would be interesting to see nonetheless.

Forgive me if some of this is a little rambly or doesn't make sense, I'll blame it on being hopped up on cold meds this morning

No, I think that you're exactly spot on. I've talked to a few people about best practices for compiling offensive and defensive SRS components separately, and I still haven't nailed it yet. When I was getting my doctorate, I had a classmate who would (sincerely, and often) say "It's easy once you figure it out," and that's how I sort of feel about that.

Part of what might get us there is the opponent-weighted shooting percentage that I talk about earlier in this thread (the one that you made major contributions in making it make more sense). That's because I was working on adjusting the methodology to improve it (I finally made two adjustments yesterday, although haven't refreshed the database yet, and have one more adjustment to make that will impact postseason numbers). I also weighted this one by shots faced, whereas the SRS method I (still) weight by time played.

(As an aside, those adjustments reflect (1) a better way - in my opinion - to deal with the small sample size issues in playoff numbers, and (2) the fact that some/much of the reason scoring is lower in the playoffs is because you've got a select sample of above-average goaltenders).

On some level, I wonder how much the opponent-weighted shooting percentage reflects what we're looking at? Definitely a lot more to think about.

 

[Doctor No]

Updated with games played through February 1.

Team | Goaltender One | 2014-15 Strength | Goaltender Two | 2014-15 Strength | Goaltender Three | 2014-15 Strength | Schedule Diff
NSH| Pekka Rinne |-0.04| Carter Hutton |+0.42|||+0.46
MON| Carey Price |+0.10| Dustin Tokarski |-0.34|||+0.44
VAN| Ryan Miller |-0.15| Eddie Lack |+0.27|||+0.42
COL| Semyon Varlamov |+0.22| Calvin Pickard |+0.10| Reto Berra |-0.14|+0.36
PHI| Steve Mason |-0.14| Ray Emery |+0.21|||+0.35
BOS| Tuukka Rask |-0.02| Niklas Svedberg |-0.31|||+0.29
CGY| Jonas Hiller |+0.00| Karri Ramo |-0.28|||+0.28
SJS| Antti Niemi |+0.06| Alex Stalock |-0.19|||+0.25
NJD| Cory Schneider |+0.06| Keith Kinkaid |+0.30|||+0.24
STL| Jake Allen |-0.08| Brian Elliott |+0.10|||+0.19
LOS| Jonathan Quick |-0.09| Martin Jones |+0.08|||+0.16
MIN| Darcy Kuemper |+0.04| Niklas Backstrom |+0.18|||+0.13
FLO| Roberto Luongo |-0.17| Al Montoya |-0.05|||+0.12
ARI| Mike Smith |+0.04| Devan Dubnyk |-0.08|||+0.12
EDM| Ben Scrivens |+0.05| Viktor Fasth |-0.06|||+0.11
CAR| Cam Ward |+0.04| Anton Khudobin |+0.14|||+0.09
PIT| Marc-Andre Fleury |+0.01| Thomas Greiss |+0.08|||+0.08
WPG| Ondrej Pavelec |-0.05| Michael Hutchinson |+0.02|||+0.07
CHI| Corey Crawford |+0.06| Antti Raanta |+0.00|||+0.06
TBL| Ben Bishop |-0.12| Evgeni Nabokov |-0.18|||+0.06
TOR| Jonathan Bernier |+0.01| James Reimer |+0.07|||+0.05
NYI| Jaroslav Halak |+0.10| Chad Johnson |+0.15|||+0.05
DET| Jimmy Howard |-0.11| Petr Mrazek |-0.08|||+0.03
NYR| Henrik Lundqvist |-0.01| Cameron Talbot |+0.01|||+0.03
BUF| Jhonas Enroth |+0.07| Michal Neuvirth |+0.05|||+0.02
CBJ| Sergei Bobrovsky |+0.05| Curtis McElhinney |+0.08|||+0.02
OTT| Craig Anderson |+0.04| Robin Lehner |+0.03|||+0.01
ANA| Fredrik Andersen |-0.10|||||
DAL| Kari Lehtonen |+0.05|||||
WAS| Braden Holtby |-0.04|||||

Carter Hutton hits the 10 game mark with a bang, and Nashville enters the list up high. Nine of Hutton's twelve appearances have been on the road, which doesn't help his schedule.

New Jersey also joins the interesting list with Keith Kinkaid. Kinkaid's home/road split is normal, so it's the actual quality of teams he's facing that produces the high number.

 

[Doctor No]

Additional data to help you decide whether or not you should pay attention to this stuff:

So far in 2014-15 (through February 1), these are the average GAA, save percentage, and shots faced per game when the opponent is within 0.2 of the following levels:

Opp Strength | GAA | Sv% | S/60 | Minutes
-1.00|2.09|0.925|28.0|4,497
-0.50|2.22|0.921|27.9|10,588
+0.00|2.70|0.909|29.8|29,536
+0.50|2.87|0.906|30.4|23,029
+1.00|3.00|0.911|33.9|4,303
R^2 | 94% | 76% | 84%

(So the row 1.0 represents all instances where the opponent was between 0.8 and 1.2). I threw overall R^2 on there while I was at it.

I tried not to cherry pick rows or anything, because I'm genuinely interested in this stuff. Hopefully you all find it intriguing as well.

 

[hatterson]

Additional data to help you decide whether or not you should pay attention to this stuff:

So far in 2014-15 (through February 1), these are the average GAA, save percentage, and shots faced per game when the opponent is within 0.2 of the following levels:

Opp Strength | GAA | Sv% | S/60
-1.00|2.09|0.925|28.0
-0.50|2.22|0.921|27.9
+0.00|2.70|0.909|29.8
+0.50|2.87|0.906|30.4
+1.00|3.00|0.911|33.9
R^2 | 94% | 76% | 84%

(So the row 1.0 represents all instances where the opponent was between 0.8 and 1.2). I threw overall R^2 on there while I was at it.

I tried not to cherry pick rows or anything, because I'm genuinely interested in this stuff. Hopefully you all find it intriguing as well.

Interesting to see the little patterns, although I'm trying not to read too much into these as we're only dealing with 2/3 a season worth of data so there's likely some fluctuation still to be had.

Save percentage consistently drops off as the teams you face get better, but there's that little kick up on the tail side against the strongest team. Although GAA still falls due to the "large" jump in shots faced.

Could you throw how many games fall into each category by chance?

Also, any reason you chose a range of and intervals of .5 instead of a range of .25 with the same intervals? Seems like you're dropping some middle cases there...although perhaps that was intentional to make there more of break between each group.

Edit: Oh, and are these strength metrics as of current, or as of the time the game was played?

 

[Doctor No]

Ah, good point - I'll add minutes faced. Part of the reason for the noise on both ends is what you mention; there's more data in the middle, so if you get one "weird" team on either end, it tweaks the results.

("weird" = a team that messes up the beautiful theoretical pattern that I want to see. )

And the reason for 0.5 increments is largely convenience - I group them up internally by rolling 0.1 increments (so -1.2 to -0.8, -1.1 to -0.7, -1.0 to -0.6, et cetera).