NHL Goaltenders and Strength of Opponents

Doctor No · Sep 22, 2013

Another in the long weekend preseason category...

Along with the other numbers that I added yesterday (Variation in Goaltender Performance), I have added two metrics to compare a goaltender's strength of schedule.

Currently, I have it implemented for the NHL (1984-85 to present), although I'll be able to add in the AHL over the next few weeks (this was the main thing I was interested in, and so I want to clean up a few other things before I finish that).

The work on the page stems from this draft discussion: http://hfboards.mandatory.com/showthread.php?t=1326065

Ultimately, I decided to focus upon two different metrics, as per the last post of that thread:

Do some goaltenders play a disproportionate share of their games against better opponents (through coaching or random fluctuation)?
Do some goaltenders face stronger shooters than other goaltenders?

To answer the first, I developed an estimate of each team's strength - using their entire (regular season plus postseason data), starting with each team's goal differential (GF minus GA), then adjusting for schedule (each team's average opponent's goal differential). This is an iterative process, but does converge to a metric that estimates how many goals better (or worse) a team is compared to average during the season. The top teams in the league are typically about +1, and the bottom teams in the league are typically about -1 (although some of the early-1990 expansion teams hovered around -2). I also calculate separate ratings for home/away, just in case a team only plays their backup goaltender on the road (for instance).

I then calculate SoS, or the Strength of Schedule, as the (shots-weighted) average strength of opponent.

To answer the second, I took each team's (regular season plus postseason) non-empty net shooting percentage. I take the (shots-weighted) average team shooting percentage, and then subtract it from one. Why? Because this puts it on a save percentage scale, so that one can compare it to the goaltender's actual performance. This is shown on the website as OpS% (opponent-weighted expected save percentage). Note that this *isn't* the average save percentage of the opponents.

Just as with the other metric, these appear on a goaltender's bio page, under "REGULAR SEASON STATISTICS" and under "POSTSEASON STATISTICS". For instance:

http://hockeygoalies.org/bio/belfour.html
http://hockeygoalies.org/bio/lundqvist.html
http://hockeygoalies.org/bio/nabokov.html

And it's also on the team-by-team summary pages, so that you can compare a goaltender to his backups:

http://hockeygoalies.org/bio/nhl/toronto.html
http://hockeygoalies.org/bio/nhl/boston.html
http://hockeygoalies.org/bio/nhl/losangeles.html

Enjoy!

Doctor No · Sep 22, 2013

A few things that I either feel the need to point out, or that I've noticed so far.

SoS balances to zero, or at least in the regular season. One would expect a team to have a positive SoS in the postseason, since they are facing (by definition) above-average teams.

OpS% balances to the league-wide save percentage over the season.

Great teams will (on average) have SoS slightly below zero, since they don't have to play themselves (naturally). Vice-versa for poor teams.

Since I weight these things on a "shots faced" basis, I think that backup goaltenders' numbers are inflated somewhat. Why? Because it's against quality teams that backups are most likely going to enter the game in relief. I still prefer it this way, since it's a measure of what actually happened (if Glenn Healy faces 500 shots in a year, but 100 of those are against juggernauts in relief, then I want to know that when I'm evaluating his performance).

It does appear to pass the "sniff test", at least as far as I can tell. I'll save observations that I've noted for Brodeur, Roy and Hasek in particular, mainly because I want others to discover them as well (I've often been accused of bias in this.

).

Doctor No · Sep 23, 2013

Lunchtime playing around...and to give you a sense of how much these things vary in practice...

Top Ten Strengths of Schedule (regular season, 1984-85 to present, minimum 500 shots faced):

Now, Brian Hayward's expansion Sharks were bad - and they were the only team in the league that year not to get to face themselves.

Guy Hebert with the (same season) Blues I found quite interesting. 14 of his 24 appearances were on the road that year, including dates in Quebec, Pittsburgh, Detroit (three times), Toronto (twice), Chicago, Buffalo, and Washington. Dragging the Blues to an 8-8-2 record for that schedule? That's impressive.

Caprice and Brodeur got the opportunity to play division rivals Edmonton and Calgary a lot.

The list is similar (with goalies on bad teams) but different (all from the 1980s) if we look at opponent-weighted expected save percentage:

Hail to the "King" in this case (probably not the crown that he wanted, although us vintage Canuck fans still love him).

You'll note that opponent-adjusted expected save percentage doesn't normalize across eras - unlike strength of schedule. Leaguewide save percentages were 0.875 in 1984-85 and 0.874 in 1985-86, so we'd expect more low numbers for those seasons.

Since this metric makes the most sense in the context of its actual season, it is best used to compare against what a goaltender actually did that season. For instance, Brodeur was expected to put up about a 0.871 in his games; he actually posted an 0.855 (still pretty low). On the other hand, it's far better than you or I could have done, facing the Oilers and Flames on a semi-nightly basis.

Doctor No · Sep 23, 2013

On the flip side of the coin, these ten goaltenders have faced the easiest competition since 1984-85:

One can see that 1992-93, in the throes of expansion, had its "haves" and "have nots" - so the top three here are all from that season.

The top three all have something going on with them - let's start with Riendeau. His tandemmate, Tim Cheveldae, played 67 games (to Riendeau's 22), with a strength of schedule of +0.04. So what was the difference? This doesn't look to be a case of Cheveldae only playing road games (Riendeau's home/road was about 50/50), but his opponents were clearly not the best and the brightest. In fact, Riendeau went 13-4-2 while putting up an 87.7% save percentage. GMs weren't fooled by the win/loss record, and Riendeau began his slow march out of the NHL.

Similar case of shielding with Reese - Mike Vernon's strength of schedule was -0.15, despite playing the vast majority of the games. Reese managed a 14-4-1 record despite league-average save percentage.

Ah, and there's Red Light in the third spot. Similar situaton to the above, with Patrick Roy facing a slightly above-average (0.03) opponent on average, and Racicot getting the dregs. By this point in the story, you won't be surprised to see that Racicot went 17-5-1 in his efforts, and was out of the league a year later.

(At this point, can we put to bed the notion of win-loss record as a reliable measure of goaltender competence? It's great for teams, yes - that's the point of playing. But teams can't follow the strategy employed here: play only the worst opponents. Maybe in the Big 12 conference.

).

Looking at opponent-weighted expected save percentage, we see the opposite effect from above: a list filled with today's best and brightest:

The reason for the high amount of recent goaltenders here? 2011-12 has the highest league-wide save percentage in history (0.9135), so deviations higher than that are more likely to top this list (2010-11 is second-highest at 0.9129).

This list may be obvious at this point, so I'll mention someone who just missed the list - Sebastien Caron of the 2002-03 Penguins, who - given the opposition he faced - would have been expected to stop 91.5% of shots. The league average that year was about 90.9%, so this is a big difference - and as a rookie (not playing his first game until January of that season), he fared well, stopping 91.6% of shots. When compared to what he should have stopped, we see his performance as league-average (still remarkable for a rookie).

Doctor No · Sep 23, 2013

Cam Ward (2005-06) appearing on the list of easiest schedules brings out one thing that I'd ultimately like to test.

Are goaltenders sheltered in their rookie seasons? For the ones who are, how does it affect their future career paths?

Beyond that, what types of goaltenders are sheltered? Do certain coaches have a greater tendency to shelter?

Doctor No · Sep 23, 2013

It also looks like the Maple Leafs realized that Glenn Healy was done - besides the 1999-00 season above (strength of schedule -0.42), his next (and final) season would have made the list (SoS -0.50) if he'd played enough games.

That -0.5 is a huge number for a typical opponent - what it means is that an average team would be favored by 0.5 goals on neutral ice.

11 of Healy's 15 games that season were on the road, but these were his opponents (strength in parentheses):

Islanders at home (-1.183)
Buffalo on road (+0.490)
Carolina on road (-0.200)
Tampa Bay at home (-1.140)
NY Islanders on road (-0.939)
NY Islanders on road (-0.939)
Boston on road (-0.185)
Buffalo on road (+0.490)
Atlanta on road (-0.908)
Columbus at home (-0.598)
Philadelphia on road (+0.325)
Chicago on road (-0.266)
Calgary on road (-0.275)
Tampa Bay on road (-0.896)
Florida at home (-0.764)

Ladies and gentlemen, that's a bad set of opponents.

By comparison, Curtis Joseph's average opponent strength in those two years was +0.03 and -0.01. This should be example #1 in the "sheltered backup" exhibit.

Doctor No · Sep 23, 2013

I do promise - ultimately I'll post something in this thread that someone else finds interesting enough to talk about. Usually "Glenn Healy was overrated" is guaranteed to get a few responses. :laugh:

Where my homies at?

seventieslord · Sep 23, 2013

The whole world knew Healy stunk in 2000! Man, was he bad. Almost as bad as Beaupre in the 2006 and 2007 seasons for the Leafs. We had bad luck with backups there for a while. Kidd was not much better.

(i meant 1996 and 1997, obviously)

Doctor No · Sep 23, 2013

I always felt bad for Beaupre at the end of his career - I think the Ottawa time really hurt him mentally. Not a lot of 5-foot-8 guys left in the league, even then, so I always rooted for Bobo.

He's an interesting case - when he came over from Ottawa in January 1996, he did look terrible. But he played eight games - three against (record setting) Detroit, one against Colorado, one against Chicago, and then San Jose, Washington, and Hartford.

I don't know what Felix Potvin's strength of schedule was just for the time Beaupre was in town, but his full season SoS was +0.03. Beaupre's was +0.86 (which is amazingly high). He might genuinely have been done, but the Leafs didn't do him any favors.

The next year? Even worse than that (before being demoted) - and he got a cupcake schedule to boot.

Filatov2Kovalev2Bonk · Oct 7, 2013

Taco MacArthur said:
I do promise - ultimately I'll post something in this thread that someone else finds interesting enough to talk about. Usually "Glenn Healy was overrated" is guaranteed to get a few responses.

Where my homies at?

It looks very interesting but I'm still a little puzzled. A negative schedule bearing indicates the opponent wasn't "favoured" or "quality" during that span, is that correct?
And of course we'd need context for the save percentages...I was going in my head "Hmmm, .877 might look really bad but if the average save percentage for that era is .880 or .885 (probably higher, much higher I imagine) then it doesn't look quite so bad." Is there a reference I can check for that sort of stuff?

Thanks for doing this by the way, but sadly when I bring up even "basic" advanced stats (especially when dealing with non-Mirtle media) it usually ends up not being part of their pre-conceived "Your team sucks, mine doesn't and I work for an outlet that really needs the team's support so I can't give an honest opinion, but thanks for reading!" :p

By the way, not sure we can mention outside sites here, but have you heard of the brodeurisafraud one? He's taken a fairly deep look at non-win statistics and broken them down. I especially liked the breakdown he did of Roy's team strength during his Cup win years. Anyway, getting the hairy eyeball from the spouse...

Final edit: Clicked on the Belfour link above, oh gawd I'm going to be spending way too much time on your site.

(

Doctor No · Oct 7, 2013

Thanks! I'll take the questions in reverse...

Love the Brodeur/Fraud site - it's one of the best goaltending analytics sites out there (Hockey Prospectus snapped him up, so you can find him more readily around there these days). He actually posts here (used to be a regular, but not as often any more). I'll let him announce himself if he chooses to do so (since it's none of my business :laugh:

).

The way that I interpret the two stats (so far - I'm still learning the nuances of them myself, but I'm writing up everything I've figured out in a paper):

Strength of Schedule (SoS) - how many goals better (or worse) the average opponent is than an average opponent. This one doesn't try and distinguish between being a good offensive team and being a good defensive team, so (for instance) if Team A wins all of their games 6-5, and Team B wins all of their games 1-0, they'll both be about 1 goal above average. Typically, the best teams in a season will be about +1, and the worst teams in a season will be about -1. Edmonton in the 1980 was consistently about +1.0 to +1.5, while the expansion Senators were about -2.0 or so. So if you played exclusively against the expansion Senators, your strength of schedule would be about -2.0.

The main reason I built this was to see instances where a goaltender is being protected (or the opposite). Home ice advantage is worth between 0.2 and 0.4 goals per season, so if a goaltender is played exclusively on the road, then their expected strength of schedule would be between +0.1 and +0.2.

In the playoffs, you'd expect a goaltender to have an above-average strength of schedule, since (by definition) the teams that make the playoffs are above average.

The way that I use the other measure - OpS% - is as an expected benchmark for a goaltender's save percentage. If a goaltender had a save percentage of 0.920, but their opponents' average shooting percentage was 8%, then they didn't excel (as it would seem on the surface) - they did about what was expected. So I think of OpS% as "par" for goaltenders. Unlike strength of schedule, I don't adjust OpS% for home/road - although it's something that I could do at some point (cutting the data in half again would make it less credible, so I haven't considered it yet).

For instance, look at Tim Cheveldae and Nikolai Khabibulin for the 1994-95 Winnipeg Jets:

www.hockeygoalies.org/bio/nhl/winnipeg.html

The rookie, Khabibulin, had superficially better statistics - a higher save percentage and a lower goals-against average. However, Cheveldae was playing against opponents that were about 0.15 goals/game better than Khabibulin. On the other hand, Khabibulin faced slightly better offenses (0.898 expected save percentage, versus 0.899 for Cheveldae). The latter measure isn't adjusted for home/road, so it may be that Chevy played more road games.

Ultimately, both goaltenders did worse than "par" - which is also borne out in the other advanced metrics; GD tells us that Cheveldae gave up 16 goals more than an average goaltender, while Khabibulin gave up 4.2 goals more than an average goaltender.

I've got a glossary here:

http://hockeygoalies.org/stats/glossary.html

That attempts to explain everything on the site. If anything there doesn't make sense, definitely speak up and I can write it more clearly.

Doctor No · Nov 7, 2013

Pulling from this thread:
http://hfboards.mandatory.com/showthread.php?t=1532773

I ran the numbers, and Bannerman/Skorodenski has the largest differential in strength of schedule since 1984-85, with Skorodenski's opponents about 0.6 goals better than Bannerman's.

Here's the top fifteen differences since 1984-85 (only counting teams where two goaltenders played at least 25 games apiece):

Season | Team | Main Goaltender | SoS | Second Goaltender | SoS | Diff
1984-85|Chicago| Murray Bannerman |-0.41| Warren Skorodenski |+0.19|0.60
1992-93|Montreal| Patrick Roy |+0.03| Andre Racicot |-0.48|0.51
1984-85|Detroit| Greg Stefan |-0.39| Corrado Micalef |+0.11|0.50
2000-01|Pittsburgh| Jean-Sebastien Aubin |-0.39| Garth Snow |+0.11|0.50
2000-01|Dallas| Ed Belfour |+0.12| Marty Turco |-0.37|0.49
1984-85|Vancouver| Richard Brodeur |+0.29| Frank Caprice |-0.19|0.48
1984-85|Quebec| Mario Gosselin |-0.32| Dan Bouchard |+0.15|0.47
1992-93|Calgary| Mike Vernon |-0.15| Jeff Reese |-0.60|0.45
1998-99|St. Louis| Grant Fuhr |+0.11| Jamie McLennan |-0.31|0.42
1985-86|Los Angeles| Bob Janecyk |-0.23| Darren Eliot |+0.18|0.41
1985-86|Detroit| Greg Stefan |-0.22| Mark LaForest |+0.18|0.40
1995-96|Edmonton| Bill Ranford |+0.14| Curtis Joseph |-0.23|0.37
2005-06|Carolina| Martin Gerber |-0.05| Cam Ward |-0.41|0.36
1988-89|Hartford| Peter Sidorkiewicz |-0.16| Mike Liut |+0.20|0.36
1985-86|St. Louis| Rick Wamsley |-0.36| Greg Millen |+0.00|0.36

In all cases, I listed the goaltender with more minutes played first. I also embedded links to the goaltenders' biographies if you want to dig in further.

The first thing that stands out to me is that most of these instances are in the mid-1980s (and the first few years where I have data). I wonder if this is a consequence of divisional schedules then, or if it's a consequence of how goaltenders were used by coaches then (prior to the mega contracts, it was probably easier to justify not playing a top goaltender in certain situations). I haven't perused the numbers to see which coaches show up more than once here, but if they do, it might suggest a strategy. Greg Stefan makes the list twice, with a different backup each time.

In most cases (but not all), the more established goaltender is playing the more difficult schedule, which corresponds with our notions of how goaltenders "should be" used. Of course, the cases that don't conform to this notion are the more interesting ones - starting at the top of the list, I still can't explain Bannerman and Skorodenski.

Anyhow, hopefully you all find this useful on some level - something to explore further.

Doctor No · Nov 7, 2013

Well, at least I thought it was interesting... :laugh:

The most curious example is Ranford/Joseph in 1995-96 - on the face of it, you'd say "oh, Ranford played the tough games, and Joseph the easy ones."

However, Ranford played nearly all of the Edmonton games in the first half of the year - he was traded to Boston on January 11th, and Curtis Joseph didn't play his first game in Edmonton until January 13th.

It genuinely seems like Edmonton's schedule just got easier in mid-January (and I can't explain why that would have happened).

TBIF · Nov 11, 2013

I read it and find it interesting.

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

Doctor No · Nov 11, 2013

TBIF said:
I read it and find it interesting.

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

Thanks! That's probably true - although I'm trying to think of new things to look at, so if anyone has ideas that could be tested, I'm game.

One thing on my list is to check to see how different coaches handle goaltending tandems - for instance, we know that Mike Keenan is famous for pulling goaltenders earlier than other coaches. Does he also spot start goalies?

hatterson · Nov 11, 2013

Ok, finally had a chance to read this in detail.

First off, great work. Really interesting to see some of the splits.

One might look at W/L stats and say that Roy (31-25-5) had a terrible 92/93 season and that Racicot was more valuable (17-5-1) but when you look at comparative Strength of Schedules (2nd biggest) and expected save percentage you can see that although that season wasn't up to normal Roy standard, he was still clearly the better goalie. Not that rational people actually needed to be told that, but it serves as a good example.

Second:

Chalupa Batman said:
You'll note that opponent-adjusted expected save percentage doesn't normalize across eras - unlike strength of schedule. Leaguewide save percentages were 0.875 in 1984-85 and 0.874 in 1985-86, so we'd expect more low numbers for those seasons.

I think I'll slightly disagree, or at least ask for proof, on the point that strength of schedule normalizes across eras. Although I'd expect it to be close and obviously you're still going to balance to zero, intuition tells me that the higher scoring the era is, the wider range you'd expect teams to have.

To take it to an absurd extreme to hopefully establish my point, lets consider two seperate eras. One in which the average save percentage is in around 98% and one in which the average save percentage is around 50%. Going by last years shot differentials teams ranged from +6.5 to -4.8. That would mean in the first era if everything was average goaltending you'd have a range of +.13 to -.096. In the second era you'd have a range +3.25 to -2.4. Now obviously shots doesn't correspond directly to goals, but just using an eye test is seems like there was a bigger range in the mid 80s than the last few years or the early 00s.

Does your iterative process take this into account? Or does the difference in practice (as opposed to my absurd example eras) not mean enough to matter? Or does it not fully normalize? (I think those are the three options, please correct me if you can think of another)

Third:
A couple ideas to normalize opponent save percentage across eras.

1.) Look at difference from average. Similar to wins above replacement or other such similar stats you could easily calculate the difference between a goaltenders actual performance and the expected performance of a perfectly average goaltender in the same scenario. Although this doesn't adjusted the expected save percentage it does give you an easier way to compare goalie performances. If expected was .880 and actual was .890 that would be roughly equivalent of expected being .910 and actual being .920

2.) Adjust based on league average save percentage. Calculating difference from league average and then adjusting that to an arbitrary era average should let you see the difference between what a goalie in 1986 was expected to do and a goalie in 2012 was expected.

With both of these options obviously hit the issue that a .010 difference is much more significant at .940 than at .880, but I think that could be solved be either looking at the numbers based on deviations from average or by a similar adjustment, ie. .05 above .950 is the same as .10 above .900 in terms of percentage improvement.

Fourth:
A question regarding OpS%. Have you noticed that schedule differences are large enough to make a difference if a team faces an above average (or below average) set of goaltenders consistently?

Specifically I'm thinking of something like last year. The northeast division had Ottawa at .933, Boston at .923, Toronto at .917, Buffalo at .913 and Montreal at .904. Whereas the Southeast has Washington at .916, Winnipeg at .901, Tampa at .899, Carolina at .897 and Florida at .887.

If a team exclusively played the Southeast their average shooting percentage would be assumed to be higher (and thus their opponents OpS% lower) as compared to exclusively playing against the Northeast.

Now, these numbers are including all phases and empty net, so I'm not sure if the differences really hold up when looking at non-empty net even strength only shots. Nor am I sure if the difference (should it exist) really matters when looking at season long data or if the schedule is balanced enough to make up for it. Nor am I sure how to tell the difference between a below average set of goaltenders and an above average set of shooters (or vice versa)

Either way, it's something I thought I'd mention.

Doctor No · Nov 11, 2013

Dang - great stuff (thanks)! A lot of meat in your post, so I'll respond in parts as I can get data together.

And I'll start with the first one, on which you're absolutely right - the "SoS" (strength of schedule) metric does balance to zero (and so an average team in 1984-85 is going to have the same strength as an average team in 2013-14).

However (and here's where you made a great point), since SoS is supposed to measure how many goals better than an average team a goaltender's set of opponents is, it expands (or contracts) to the scoring environment. The strength metric that I use (when weighting together a goaltender's opponents) starts with goal differential (goals scored minus goals allowed), and then gets adjusted by the average goal differential of a team's schedule - that result then gets fed back into itself, and the process repeats until the team strengths converge (which they will do under reasonable assumptions).

The corollary to this is that in a high-scoring environment, the best teams will have higher strengths (and the worst teams will have lower strengths). Two examples to put this into a more concrete environment:

(1) When I used to run power ratings for the NBA (where scores are typically around 110-95 or so), the best teams' power ratings were routinely in the +10 to +20 range. Top NHL teams in today's era are routinely in the +0.8 to +1.2 range.

(2) My power rating for Chicago last year is (0.816, 1.147), meaning that they would be favored by 1.147 goals at home against an average team, and by 0.816 goals on the road against an average team. If we magically entered a world where goals were worth ten points, then Chicago's power rating would be (8.16, 11.47) - obviously much higher.

Although it certainly seems obvious *now*, it's clear from my writings above that I've never thought about it in this way. It does explain a lot about why most of the extreme results in the tables above are from the higher-scoring 1980s (whether it explains all of the difference, or merely some of the difference, remains to be explored).

Doctor No · Nov 12, 2013

One thing in the meantime that I figured I'd do would be to look at the careers of some of the era's better goaltenders to see what we can make of their usage patterns.

I'll start with Martin Brodeur since he's easier (all with the same team, so less things to check in the database, and consistently plays most of the Devils' games).

Season | Brodeur | Other Goaltender (SoS) | Diff
1991-92|-0.47| Chris Terreri 0.11|-0.58
1993-94|-0.08| Chris Terreri 0.00|-0.08
1994-95|-0.12| Chris Terreri -0.03|-0.09
1995-96|+0.01| Corey Schwab 0.06|-0.05
1996-97|-0.08| Mike Dunham 0.07|-0.15
1997-98|-0.07| Mike Dunham -0.15|+0.08
1998-99|-0.06| Chris Terreri 0.00|-0.06
1999-00|-0.20| Chris Terreri 0.10|-0.30
2000-01|-0.15| Chris Terreri -0.25|+0.10
2001-02|-0.08| John Vanbiesbrouck 0.03|-0.11
2002-03|-0.14| Corey Schwab -0.23|+0.09
2003-04|-0.03| Scott Clemmensen -0.92|+0.89
2005-06|-0.14| Scott Clemmensen -0.20|+0.06
2006-07|-0.19| Scott Clemmensen -0.18|-0.01
2007-08|-0.14| Kevin Weekes -0.02|-0.12
2008-09|-0.06| Scott Clemmensen 0.02|-0.08
2009-10|-0.05| Yann Danis -0.33|+0.28
2010-11|-0.03| Johan Hedberg 0.05|-0.08
2011-12|-0.05| Johan Hedberg 0.02|-0.07
2012-13|-0.06| Johan Hedberg 0.05|-0.11

Now remember that backup goaltenders typically have a higher strength of schedule than starters, because when are backups more likely to enter a game midway? (I'd like to look at this only for games started by goaltenders at some point to try and measure the effect).

With that said, other than Brodeur's rookie year (and since his games were largely consecutive, that's more a schedule function than anything deliberate) and perhaps 1999-2000, I don't see anything here.

Brodeur typically played the vast majority of Devils' games, and it doesn't look like there was any attempt to put the backup in for an easy opponent. The fact that his strength of schedule is typically negative is because the Devils are typically an above-average franchise (who doesn't have to play themselves).

(By the way, if you want to see what a terrible schedule looks like, click on Scott Clemmensen and look at who he played in 2003-04 - and I'll remind you that Pittsburgh and Washington were not the class of the league that they are today; there's a reason they got Crosby and Ovechkin in the drafts. Clemmer got the Pens twice, the Caps once, and rounded things out with a game against Atlanta - who was actually decent).

Doctor No · Nov 14, 2013

hatterson said:
A couple ideas to normalize opponent save percentage across eras.

1.) Look at difference from average. Similar to wins above replacement or other such similar stats you could easily calculate the difference between a goaltenders actual performance and the expected performance of a perfectly average goaltender in the same scenario. Although this doesn't adjusted the expected save percentage it does give you an easier way to compare goalie performances. If expected was .880 and actual was .890 that would be roughly equivalent of expected being .910 and actual being .920

2.) Adjust based on league average save percentage. Calculating difference from league average and then adjusting that to an arbitrary era average should let you see the difference between what a goalie in 1986 was expected to do and a goalie in 2012 was expected.

Still working my way through these (I wanted to let them percolate a bit in my brain)...

I like both of these ideas, and I think that I prefer the second one. So for instance, if the league-average save percentage in a season was 0.880, and a goaltender's opponent-weighted expected save percentage, I could report the schedule difference as "+0.010". What do you think about that?

My plan is to replace the current metric with this on my pages - since, as you can see here (for instance):

http://hockeygoalies.org/bio/cechmanek.html

The regular season statistics / postseason statistics tables are already getting a bit wide. :laugh:

Doctor No · Nov 14, 2013

In our "great goaltenders of the modern era" journey, let's continue with Patrick Roy:

Season | Roy | Other Goaltender (SoS) | Diff
1985-86|+0.02| Doug Soetaert 0.11|-0.09
1986-87|+0.00| Brian Hayward -0.08|+0.08
1987-88|-0.09| Brian Hayward -0.04|-0.05
1988-89|-0.13| Brian Hayward -0.06|-0.07
1989-90|-0.18| Brian Hayward +0.04|-0.22
1990-91|-0.08| Andre Racicot -0.21|+0.13
1991-92|-0.14| Rollie Melanson +0.02|-0.16
1992-93|+0.03| Andre Racicot -0.48|+0.51
1993-94|-0.03| Andre Racicot -0.06|+0.03
1994-95|-0.01| Ron Tugnutt +0.13|-0.14
1995-96 (M)|-0.01| Jocelyn Thibault -0.20|+0.19
1995-96 (C)|-0.25| Stephane Fiset -0.09|-0.16
1996-97|-0.07| Craig Billington -0.17|+0.10
1997-98|-0.09| Craig Billington -0.01|-0.08
1998-99|-0.11| Craig Billington -0.09|-0.02
1999-00|+0.03| Marc Denis +0.04|-0.01
2000-01|+0.03| David Aebischer -0.23|+0.26
2001-02|+0.04| David Aebischer +0.01|+0.03
2002-03|+0.04| David Aebischer -0.03|+0.07

The 1995-96 seasons are mainly noise due to the trade, but I include them anyhow (even though Roy and Thibault never existed together on the Canadiens, for instance). I did leave off his 1984-85 "season" (one period of action against Winnipeg).

If we accept the tenet that backups will have a slightly higher strength of schedule solely because they're more likely to spell a starter against a tough opponent, then my main takeaway here is that the Avalanche coaches were protective of a young David Aebischer. Craig Billington looks to have been treated like a typical backup (not overly protected).

The 1992-93 season is already documented above, but it looks like Andre Racicot was similarly protected (Melanson barely eked out Racicot on the list above in 1991-92, but Racicot's 1991-92 SoS was -0.26).

Roy/Fiset in 1995-96 strikes me as somewhat interesting, but being here first-hand to witness the season, my guess is that Colorado's early-season schedule was tougher than after Roy joined. Fiset actually played quite a bit for a superstar's backup that year (and I find that interesting).

Brian Hayward served a valuable role as Roy matured into a top goaltender (similar to Chris Terreri's role above).

hatterson · Nov 14, 2013

Chalupa Batman said:
Still working my way through these (I wanted to let them percolate a bit in my brain)...

I like both of these ideas, and I think that I prefer the second one. So for instance, if the league-average save percentage in a season was 0.880, and a goaltender's opponent-weighted expected save percentage, I could report the schedule difference as "+0.010". What do you think about that?

I like the concept, but I fear it would over represent 80s goaltenders (or others during an era of overall low save percentages)

+0.010 represents a significantly larger step in difficulty if league average is .930 than if it is .880

Doctor No · Nov 14, 2013

hatterson said:
I like the concept, but I fear it would over represent 80s goaltenders (or others during an era of overall low save percentages)

+0.010 represents a significantly larger step in difficulty if league average is .930 than if it is .880

Really good point. How about (1-opp. wtd sv%)/(1-league sv%) - 1?

Then, if the league were at 0.880 and the goaltender were at 0.870, we would get 8.3% (so the goaltender's job was 8.3% tougher than league average).

If the league were at 0.930 and the goaltender were at 0.920, we would get 14.3% (so the goaltender's job was 14.3% tougher than league average).

Does that hold water? Stated differently, when the league-wide error rate is higher, it takes more of a difference (in absolute terms) to be meaningful.

hatterson · Nov 14, 2013

Chalupa Batman said:
Really good point. How about (1-opp. wtd sv%)/(1-league sv%) - 1?

Then, if the league were at 0.880 and the goaltender were at 0.870, we would get 8.3% (so the goaltender's job was 8.3% tougher than league average).

If the league were at 0.930 and the goaltender were at 0.920, we would get 14.3% (so the goaltender's job was 14.3% tougher than league average).

Does that hold water? Stated differently, when the league-wide error rate is higher, it takes more of a difference (in absolute terms) to be meaningful.

I *think* that works, at least it seems to pass the sniff test.

If the average save percentage is .950 and a goalie's OpS is .900 his job is theoretically twice as hard as a goalie in a completely average situation. Same with .900 vs .800.

If the average save percentage is .900 and a goalie's OpS is .950 his job is 50% easier? If it's .900 and a goalie's OpS is .925 his job is 25% easier?

Although is it right so say it's 25% easier, or that it's 75% of the difficulty? Sorry, it's late and I'm doped up on cold medication :laugh:

Doctor No · Nov 24, 2013

I've finally made this change (took me long enough).

Easier to see now if and when the two separate metrics correlate (they seem to for the most part).

The first (SoS) measures the average overall strength of a goaltender's opponents, while the second (OS%+) measures the average relative shooting percentage of a goaltender's opponents. So they wouldn't necessarily correlate, but intuition would suggest that good overall teams would be more likely to have good shooting percentages.

A couple places for those who want to sniff test:
Some popular player bio:
http://www.hockeygoalies.org/bio/lundqvist.html
http://www.hockeygoalies.org/bio/roy.html
http://www.hockeygoalies.org/bio/hasek.html
http://www.hockeygoalies.org/bio/brodeurm.html

And the Florida Panthers' goaltending history:
http://www.hockeygoalies.org/bio/nhl/florida.html

Doctor No · Nov 24, 2013

Just poking around with it, I think that I like it.

NHL Goaltenders and Strength of Opponents

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Student Of The Game

Registered User

Effortless sexy.

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Registered User

Ad

Ad

Ad