Goalie Research: Wins Added

[seventieslord]

In 06/07, I remember reading that Luongo actually had a higher save percentage at ES, PP, and SH than Brodeur, but when you combined the data, because of the discrepancy in times shorthanded Brodeur's save percentage ended up being higher. It'd be interesting to see how the numbers work out if controlled for the number of shots in each scenario.

This is true. Actually, the SH numbers may have been off just because goalies face a very small number of SH shots. But Luongo was equal to or better than Brodeur at ES and vs. the PP but the number of PP shots he faced dragged his sv% down past Brodeur's. It was a real eye-opener to what kind of impact a team can have on a goalie's sv%. In this case they are heliping his numbers greatly by not taking many penalties.

 

[seventieslord]

That is one thing that should really be explored further. All stats, all awards, all numbers are important in determing a goalie's greatness but face it - if we had to eliminate all numbers except one and use just that one going forward as a measure of a goalie's worth, we would use save percentage. Flawed, yes. But it's the best thing we've got. The Hockey Compendium has regular season save percentages for 1954-1967, 1971, 1975, 1976, and then 1983-present. One great thing that could be done with these numbers, is take a look at the team's PIM totals and attempt to approximate how each goalie is faring in equalized circumstances.

Unfortunately, a few assumptions would have to be made, that would have to be accepted as reasonable:

1) That all goalies see a generally uniform decline from their ES sv% to their PP sv%. Some may go down 15 points, others 5 points, but there would have to be a benchmark used.

2) That only a certain percentage of penalty minutes result in power plays for the other team. There are coincidentals which throw it off, and also fights and misconducts which could skew it greatly. But over time, these things can be washed out. This could be approximated by simply looking at all teams in the past decade and comparing their team PIM totals to the number of powerplays they gave their opponents. an "all-time powerplays per PIM" constant, if you will. Naturally, for more recent seasons where PP and SH totals are available, we will not need to approximate. We can simply look at the total powerplays a team faced in the season, and attribute a certain percentage of them to the goalie in question. I'd personally remove a minute per PP goal against so that a PP that resulted in a goal was counted as one minute long - not all of them are but they would average out at that mark over a season.

3) That teams generally see their shots against per minute go up at the same rate when killing a penalty versus playing at even strength. We can use recent post-lockout figures of shots by situation, to approximate this rate pre-lockout.

With these assumptions and existing data, we could take a goalie from 1985, look at his shots against and goals against, and reasonably break them down to goals against and shots against versus the PP and at even-strength. These numbers could then be left as-is, or "adjusted" versus league averages to show what a goalie's sv% would be if he had faced an average amount of powerplays.

Who's up to it?

 

[Hockey Outsider]

I love the fact that you factored in the team's offence into the analysis, as it seems to always be ignored when mentioning a goalies win totals. Your chart shows Brodeur's worth better than just save percentage. Yes, he played behind a great defensive team for most of those years, but it was also a team that usually gave little offensive support. He didn't have much room to make any errors. I'd be interested to see how Grant Fuhr's numbers for Edmonton's dynasty years compare.

Thanks. If there's one insight that this work shows, it's that offense is key. People always talk about a goalie being protected or abandoned by their defense, but that's really only half the story.

My data only goes back to 1988 so I only have one year when Fuhr and Gretzky were on the same team. That year Fuhr won the Vezina trophy, finished 2nd in Hart trophy voting, and won 44.5 games. My formula estimates that he added 0.0 wins. This formula doesn't take into account the dangerous scoring chances that Fuhr faced due to his team's aggressive style, though.

Interesting work. The only suggested improvement would be to do it on a game-by-game basis, but that would take forever to tabulate.

Agreed. Theoretically it makes more sense to determine offensive support, and shots faced, on a game-by-game breakdown for each goalie, but that would require far too much work for me.

What was the highest single-season mark you found?

Interestingly, six of the top ten marks have come from 2006 and 2007, probably due to the fact that starting goalies are given an increasingly larger number of starts each year:

Goalie|Season|Wins Added
Martin Brodeur|2007|+15.1
Roberto Luongo|2007|+14.3
Miikka Kiprusoff|2006|+12.2
Curtis Joseph|1994|+11.5
Dominik Hasek|1997|+11.5
Martin Brodeur|2006|+10.8
Tomas Vokoun|2006|+10.6
Ed Belfour|1993|+10.2
Roberto Luongo|2006|+9.9
Patrick Roy|1989|+8.9

I was surprised to see Hasek here only once, but that's mostly a function of games played. He also has years #11, #19, #24, #58, #59 and #64.

I was very surprised to see Vokoun on this list. Still, Nashville allowed so many shots that they were expected to win about 48% of their games. In real life, with Vokoun in net, they won 65% of their games. I'm not sure if I had the chance to watch even one Predators game in 2006-- does anybody have some insights about how Vokoun played?

 

[Hockey Outsider]

The Wins Added method seems to give all the credit for a team outperforming their goals for and against to the goalie. I don't think this is an ideal model, as forwards can score timely goals and the defence can tighten up with a lead. However, it's probably better than ignoring the issue entirely, and it's been handled clearly and consistently here.

My biggest criticisms of save percentage and GAA (whether it's adjusted in some way or not) is that it's an average. Therefore it doesn't take into account if all the saves occur in the last minute of the third period with a 2-1 lead, or whether the saves occur during a 5-0 blowout. By looking at wins added, there's probably a higher chance that clutch play is reflected. Your counter-point that goalies get credit for clutch scoring/defense is completely valid, though.

I have one suggestion, if you are interested in adding more complexity in the search for more accuracy. Try adjusting the goalie's save% based on the number of times his team was shorthanded vs the league average. SH Sv% has been considerably lower than EV Sv% over the past few years, and I expect it's always been that way. I'm not sure if historical data is available on EV Sv% vs SH Sv%, but there might be possibilities there in any case.

This would definitely improve the relevance of the numbers I posted. It would be tricky but I'm sure I can make the adjustments if I had the data -- unfortunately I haven't been able to find this in Excel or text format.

I ran the numbers for the NHL from 1968-99 (when they added overtime loss points) and I got an exponent of 2.04 as the best fit - not significantly different from 2. However, the best exponent to use varies with the goals per game. Use the formula (total goals per game)^0.375 to get the best exponent for a given level of scoring.

I used an exponent of 2.0 for simplicity but it will be easy to correct with Excel. I've found that an exponent of around 1.8-1.85 works best, but that was before that addition of three-point games, which undoubtedly mess up the numbers.

I suggest reading "Win Probabilities" by Alan Ryder which has a rigorous analysis of many different win probability formulas.

 

[Bear of Bad News]

1) I believe that we should be looking at how many actual decisions a goalie had, not how many minutes. For example, in the Hasek example, he played 70.3 games based on his minutes, but adding up his decisions gets you 69 games. His wins per game should be based on wins divided by total decisions, not wins per minutes/60.

I use actual decisions when I do similar work (my support-neutral W/L statistics). But I can see a justification for doing it this way.

In that event, I'd recommend that instead of using 60 for the divisor, use the actual average minutes per NHL game in the season in question - otherwise you're going to end up with more decisions than games played at year's end. But it's not that big of a difference, so maybe I wouldn't bother with it at all.

 

[Bear of Bad News]

My biggest criticisms of save percentage and GAA (whether it's adjusted in some way or not) is that it's an average. Therefore it doesn't take into account if all the saves occur in the last minute of the third period with a 2-1 lead, or whether the saves occur during a 5-0 blowout. By looking at wins added, there's probably a higher chance that clutch play is reflected. Your counter-point that goalies get credit for clutch scoring/defense is completely valid, though.

Something I've worked on in the past is a measure of save percentage standard deviation, so that a goaltender who consistently stops 90% of shots game-in and game-out could be distinguised from one who gets shutouts half the time and lets in 20% of shots half the time.

I was never happy with how partial games were handled - although I may have come to a satisfactory resolution, but I haven't had a chance to test it enough (and I don't want to suffer from premature jocularity).

On the other hand, I wasn't sure where to go next from there. If you're on a horribly bad team, you'd want the erratic goaltender because in the games where he's phenomenal, you have a chance to win. If you're on a great team, then you want the consistent goaltender. On a team in the middle, I'm not sure how much of a difference it makes.

 

[Hockey Outsider]

You then do the various calculations and adjustments, reaching the conclusion that Dominik Hasek added 8.8 wins to the team and with an average goalie the team would have had 18 fewer points, finishing 10th and out of the play-offs.

Just to clarify, the main purpose of my post was to come up with a new way to measure goalie performance. Saying that "the '98 Sabres would have fallen to 10th in the conference without Hasek" is more of a illustrative footnote, not the crux of the analysis.

1.) To who did the points go to? You just dropped 18 points from the final standings
BUT those 18 points have to go somewhere for the standings to balance. You do
not seem to have a ready justification for the disappearance of these points.

You are correct. The 9 wins would have to be allocated to the league's other 25, probably in proportion to how many times they played the Sabres. The average is less than 1 point per team, but it is still fair to make the adjustment.

2.) Statistical fairness dictates that your study apply the same analysis to all the
teams and the goalies for the 1997-98 season before your claim that a 10th place
finish is the only result possible.

On the contrary, I am trying to isolate a single variable: Dominik Hasek's performance. I'm not concerned with re-casting the Eastern conference standings had every team had an average goalie. I'm simply trying to illustrate Hasek's worth by showing how much he meant to his team, all other factors being held constant.

 

[Hockey Outsider]

I wonder how Sean Burke would fare using this comparison...

Burke does fairly well and has some excellent years in Hartford and Phoenix. He ranked in the top five in 1988, 1996 and 2002 and had several years in the top ten.

I estimate that Burke added 12.2 wins over the course of his career, which ranks him 18th during this era. To put that into perspective, he's immediately below Nikolai Khabibulin and Guy Hebert, and immediately above Henrik Lundqvist and John Vanbiesbrouck. That sounds "about right" to me-- Khabibulin and Hebert are good comparables. Vanbiesbrouck's numbers are only this low due to the awful final few years of his career.

 

[Hockey Outsider]

Great questions!

HO, I have a couple questions for you regarding your assertion that Brodeur was "average" between 1999 and 2004. I don't want to appear to be a homer Devils fan. It's just that as one, I know more about Brodeur than I do any of the other goalies you have analyzed. So on to the questions:

Before I respond I should mention that my first post probably underrates Brodeur due to his longevity. Even if it's true that a statistically average goalie would have won as many game as Brodeur did from 1999-2004, very few goalies could maintain that level of play over 75 games each year. It might be more fair to compare Brodeur to 60-65 games of an average goalie, and then 10-15 games of a below-average backup. That obviously will increase Brodeur's value, because his consistency/durability should be rewarded.

1) Would a goalie who plays "average" on a great team and has awesome stats (SV% and GAA as well as wins) be measured similarly to a goalie who plays "average" on a bad team and has crappy stats? For instance a team like Detroit has won and competed for the Presidents Trophy consistently since the mid-1990's. Would a goalie on that team need to have great stats (SV% > .910 and/or GAA <2.00) to be considered average and have a wins added of +/- 0? Likewise with a bad team (the LA Kings come to mind but I don't want to assume all their years were bad).

Basically, yes. If an average goalie would be expected to earn 35 wins on a Stanley Cup contender, on the basis of that team's offense and defense, and they actually win 35 or 36 games, then they haven't added much to that team's success.

In 2002 Brodeur went 38-29-6 on a team that was ranked 1st in the league defense and 20th in offense. That same year, Luongo went 16-33-4 on a team that was 27th in offense and 28th in defense. My formula estimates that they were essentially equally good-- Brodeur added about 1 win above and beyond what could be expected by playing on such a strong team, and Luongo met expectations (0 wins added/lost) playing on such a weak team. I'd go as far as saying that had they switched teams, Brodeur's Panthers would still be contenders for the #1 draft pick, and Luongo's Devils would still be contenders for the Presidents Trophy. (Keep in my that this analysis is for the regular season only).

Does this unfairly penalize goalies on good teams, or reward goalies on bad teams? Possibly. Comments are welcome on this.

2) For a goalie like Brodeur, the period that you mention is one in which the team had some of its best offensive years, leading or coming close to leading the league in GF at the beginning of the period. What affect would that have on Brodeur, given that he was fairly consistent in his play over the years?

Since the Devils had great offense during that span, it would increase the expectations on Brodeur. Since his W/L/T record remained fairly constant the formula says "Brodeur's won as much as he did in prior years, except this time he's supported by an excellent offense, so some of the game that he won in prior years were probably won by the team's great offense these years".

3) It appears that your formula focusses on save percentage in comparison to wins when you break down the essential components of the formula. Save percentage has been Brodeur's weakest statistic over almost his entire career when compared to other "elite" goalies. Would this be the greatest contribution to Brodeur being "average" from 1999 to 2004?

Yes and no.

"No" because I never use any goalie's save percentage in my formula. All my calculations are done without any reference to the individual goalie's save percentage. I just compare their actual to expected wins.

"Yes" because I use the league average save percentage when calculating expected wins (which is then compared to actual wins). Therefore, any flaws in save percentage would affect my expected wins.

4) The New Jersey Devils / CAA arena have been notorious over the years of having low shot totals for and against during games. Given my question in number 3, would this have had a significant affect on Brodeur's rating during his career had his shots against totals increased in numbers to the average (thus inflating his SV%), while not affecting the total games he won during the year?

See above. If save percentage is biased against Brodeur due to playing in a low-shot environment, then that's not directly held against him since I never use his individual save percentage.

However if save percentage is biased at a more fundamental level against all goalies league-wide, then that would affect the expected wins calculation. That would affect every goalie.

 

[Hockey Outsider]

1) I believe that we should be looking at how many actual decisions a goalie had, not how many minutes. For example, in the Hasek example, he played 70.3 games based on his minutes, but adding up his decisions gets you 69 games. His wins per game should be based on wins divided by total decisions, not wins per minutes/60.

This probably makes more sense since I'm comparing expected and actual wins (decisions). I doubt this makes a big difference but I can make the adjustments.

2) Ties should count. They're not wins, but they're being treated as being just as bad as losses, if I understand this correctly. A goalie on a team with low offensive numbers and low goals against is going to find themselves at a disadvantage in this formula because they are more likely to have more ties. These ties are being counted as "a game that they didn't win" for the purposes of this exercise.

I'm counting wins as wins + 50% of ties, OTLs & SOLs. Two ties equal a win. (The downside is that the goalie's records are now W/L instead of W/L/T, so it makes them a bit less comparable to the real-life numbers).

On the Brodeur boosters side, however, they can always say he makes a lot of poke checks which don't show up as shots on goal. They can also say that his puckhandling skills result in fewer shots against. This is probably true but it is also difficult to quantify. I would attempt to guess exactly how many instances of puckhandling resulted in a shot being prevented, per game, then multiply that by games played, and look at his save percentage over the years and determine how many of those prevented shots would have been allowed had he not prevented them. It wouldn't be exact, though. And the kicker is really in the original number used, the number of actual shots he is preventing on a nightly basis. You would also need to subtract the number of times an average goalie does this per game so as to not over-reward Brodeur, because if he's preventing 7 shots per game and an average goalie prevents 4, he's really preventing 3 more than an average goalie, not 7. The rest of the formula would write itself. This is a metric that would obviously help Brodeur but I question how significant it could really be in attempting to prove anything significant, i.e. that he is better than Roy or Hasek, arguments that get blown out of the water any time a detailed analysis is done.

I wonder if there's any way to quantify this objectively.

A very rough calculation suggests that Brodeur would need to prevent about 4 shots per game through puckhandling and pokechecking, relative to the league average, to get up to Hasek-level save percentage. Brodeur is, in my opinion, an average poke-checker so there's no help there. I know he handles the puck perhaps 15 times per game... how many of those would have result in shots against?

Johnny Bower would probably benefit from this more than any goalie-- not only did he lead the league in save percentage 5 times, he also has a reputation of being one of the best poke-checkers in history.

 

[Big#D]

Thanks for the answers HO.

By the way in Q#3, I mentioned talked about save percentage because you used shots against (unless I'm reading it wrong you mean the number of shots the goalie in question had to stop) and the league SV%. If the goalie had a lower number of shots against due to miscalculation, it would affect the caluclation of Wins Added. But I think 70sLord answered the question as to what possible affect it might have had on said goaltender.

 

[Canadiens1958]

Moving This Along

Just to clarify, the main purpose of my post was to come up with a new way to measure goalie performance. Saying that "the '98 Sabres would have fallen to 10th in the conference without Hasek" is more of a illustrative footnote, not the crux of the analysis.



You are correct. The 9 wins would have to be allocated to the league's other 25, probably in proportion to how many times they played the Sabres. The average is less than 1 point per team, but it is still fair to make the adjustment.



On the contrary, I am trying to isolate a single variable: Dominik Hasek's performance. I'm not concerned with re-casting the Eastern conference standings had every team had an average goalie. I'm simply trying to illustrate Hasek's worth by showing how much he meant to his team, all other factors being held constant.

Thank you for taking the time to explain your position. Basically in the quoted section above there are three paragraphs.

First paragraph we part company after the first sentence. To this point I applaud your efforts. Beyond the first sentence I do not agree and will use your second paragraph with the data I posted previously to illustrate my point.

Second paragraph - If I correctly understand your data over an 82 game schedule
Dominik Hasek projects to 39.5 wins.

Dominik Hasek 72GP 33W 23L 13T
Steve Shields 16GP 3W 6L 4T

Over 82 games means that it would be reasonable to transfer the wins that Steve Shields had to Dominik Hasek yielding 36W. The remaining 3.5 W would have to come from the loses or ties credited to Shields. Your actual swing in points is not or would not be as great as your example shows. Basically the 4T = 2W leaving 1.5W that could be considered as exclusive to Dominik Hasek. You definitely would not have to distribute as many points over as many teams as you claim.

Third paragraph - basically there is nothing in your work that justifies crediting Dominik Hasek with the 8.8 added wins. The true number seems to be closer to 1.5W.

 

[Bear of Bad News]

I didn't hear back from you a few months ago, so I was afraid that you'd tabled things.

So it turns out that I'm a doofus (not that most people around here didn't know that already). I checked my sent PMs and couldn't find the most recent response (which I thought I had sent in early May). Mea culpa.

 

[Bear of Bad News]

Second paragraph - If I correctly understand your data over an 82 game schedule
Dominik Hasek projects to 39.5 wins.

Dominik Hasek 72GP 33W 23L 13T
Steve Shields 16GP 3W 6L 4T

Over 82 games means that it would be reasonable to transfer the wins that Steve Shields had to Dominik Hasek yielding 36W. The remaining 3.5 W would have to come from the loses or ties credited to Shields. Your actual swing in points is not or would not be as great as your example shows. Basically the 4T = 2W leaving 1.5W that could be considered as exclusive to Dominik Hasek. You definitely would not have to distribute as many points over as many teams as you claim.

Third paragraph - basically there is nothing in your work that justifies crediting Dominik Hasek with the 8.8 added wins. The true number seems to be closer to 1.5W.

Your argument seems to be that Steve Shields doesn't have enough losses or ties to pass along to Hasek as wins. But that's not what this discussion is about. First of all, Hasek already has the 8.8 added victories, so I'm not sure why you're trying to add more victories to his total. Second of all, Hasek doesn't play NHL games against Steve Shields (at least not in the year in question), so why would Hasek's improved performance result in decisions being taken from Shields?

 

[Canadiens1958]

Simple

Your argument seems to be that Steve Shields doesn't have enough losses or ties to pass along to Hasek as wins. But that's not what this discussion is about. First of all, Hasek already has the 8.8 added victories, so I'm not sure why you're trying to add more victories to his total. Second of all, Hasek doesn't play NHL games against Steve Shields (at least not in the year in question), so why would Hasek's improved performance result in decisions being taken from Shields?

The projection was over 82 games. Now Hasek was not going to play for another team was he? No he would play for the Sabres. So if Hasek played all 82 games then Steve Shields would obviously not play any. Over 82 games Hasek's win total projects to 39.5.

In the games where Hasek was credited with a W/L/T he actually had 33W. Shields had 3W 6L 4T. Do you think that Hasek would have lost the 3 games Shields won?
So it is fairly reasonable to add those three wins to Hasek yielding 36. No adjustments are required for the league. Do you think the points from the 4T in games Shields played would be loses if Hasek played - at worst they would stay ties.
4T = 2W yielding the equivalent of 38 W for Hasek. We are left with the 6L generated when Shields played. The projection for Hasek playing 82 games is 39.5 W
so we are the equivalent of 1.5 W short which have to be accounted from the 6 L.Minimal adjustments.

Hasek may have made the difference of 1.5 W not the 8.8 wins claimed.

 

[Bear of Bad News]

The cause for your misunderstanding is now clear. Please go back and re-read the original post.

 

[Hockey Outsider]

So it turns out that I'm a doofus (not that most people around here didn't know that already). I checked my sent PMs and couldn't find the most recent response (which I thought I had sent in early May). Mea culpa.

No worries. Been very busy lately finishing up some professional exams so I put this research off for a bit.

Something I've worked on in the past is a measure of save percentage standard deviation, so that a goaltender who consistently stops 90% of shots game-in and game-out could be distinguised from one who gets shutouts half the time and lets in 20% of shots half the time...

This is a bit of a tangent, but the NHL already rewards inconsistency in the sense that shutouts are given an inordinate amount of praise. Even ignoring the fact that it's easier to get a shutout on a good defense team, if two goalies had a similar GAA and save percentage, but one had a lot of shutouts, it's very likely that to balance out the shutouts, that goalie also had a lot of bad games. Goalie A might give his team a great chance to win due to 7 shutouts... but is his team really better off if he has a lot of, say, 5- or 6-goal games to offset those shutouts?

On the other hand, I wasn't sure where to go next from there. If you're on a horribly bad team, you'd want the erratic goaltender because in the games where he's phenomenal, you have a chance to win. If you're on a great team, then you want the consistent goaltender. On a team in the middle, I'm not sure how much of a difference it makes.

Have you heard of basketball research Dean Oliver? He's published a lot of articles arguing the same point -- it pays to be inconsistent when you're the underdog, and it pays to be consistent when you're the favourite. (Example article). Theoretically a goalie's performance rating, under any metric, should be adjusted to show whether his (in)consistenty helped or hurt his team. I wouldn't even know where to begin in terms of quantifying this, however.

 

[dcinroc]

Have you heard of basketball research Dean Oliver? He's published a lot of articles arguing the same point -- it pays to be inconsistent when you're the underdog, and it pays to be consistent when you're the favourite. (Example article). Theoretically a goalie's performance rating, under any metric, should be adjusted to show whether his (in)consistenty helped or hurt his team. I wouldn't even know where to begin in terms of quantifying this, however.

Ken Dryden made a similar point in his book "The Game" when he talked about good "bad team" goalies (such as Mike Palmateer) that could steal games from better teams with fantastic play, but who could also give up soft goals in a rout. Those soft goals just aren't noticed as much when you play for a bad team.

 

[Njdevilsfireonice30]

Very interesting stuff, I have to take a longer look when I get home later...

It still stands, though - no matter what board you go on, what time of day it is, how good or bad the Devils were the year before, someone is shredding Brodeur... he may be the worst goalie of all time, who is on the verge of breaking the all-time wins and shutouts records solely because Scott Stevens has made 30 saves a game for him for the past 12 years to some, but I still like the guy.

 

[Bear of Bad News]

This is a bit of a tangent, but the NHL already rewards inconsistency in the sense that shutouts are given an inordinate amount of praise. Even ignoring the fact that it's easier to get a shutout on a good defense team, if two goalies had a similar GAA and save percentage, but one had a lot of shutouts, it's very likely that to balance out the shutouts, that goalie also had a lot of bad games. Goalie A might give his team a great chance to win due to 7 shutouts... but is his team really better off if he has a lot of, say, 5- or 6-goal games to offset those shutouts?

A good point - my guess is that you'd have to look at it on a game-by-game basis, something I hope to be able to do by the end of the summer at least.

Have you heard of basketball research Dean Oliver? He's published a lot of articles arguing the same point -- it pays to be inconsistent when you're the underdog, and it pays to be consistent when you're the favourite. (Example article). Theoretically a goalie's performance rating, under any metric, should be adjusted to show whether his (in)consistenty helped or hurt his team. I wouldn't even know where to begin in terms of quantifying this, however.

I've never read Oliver's work (now that my Sonics are gone, the NBA is dead to me), but I will check his stuff out - I imagine that due to the flow of the game, there's quite a bit of overlap between what will work for hockey and what will work for basketball. Except of course that I like goaltending, and that's illegal in basketball.

I'm not doing anything particularly interesting this summer, although the stuff I'm doing will hopefully make it much easier for interested people to look at most things in much more detail. Basically, I'm translating all of my website (statistics and game logs) into MS-Access so that I do analytics easier. (Believe it or not, right now my stats and logs are in the exact format you see on my web page and nothing more. Makes it hard to do anything with it, but when I started it way back when, I never thought I would do anything with it).

So eventually I'll be able to pass around things like 40-save performances by goaltenders, consecutive win streaks, playoff performances by round, and then start doing more complex things such as minor league transformations and whatnot. It's boring right now (very boring), but I'm hoping that it will be worth it.

 

[Bear of Bad News]

It still stands, though - no matter what board you go on, what time of day it is, how good or bad the Devils were the year before, someone is shredding Brodeur... he may be the worst goalie of all time, who is on the verge of breaking the all-time wins and shutouts records solely because Scott Stevens has made 30 saves a game for him for the past 12 years to some, but I still like the guy.

I don't really see where anyone is shredding Brodeur in this thread. Could you elaborate?

 

[ContrarianGoaltender]

Interestingly, six of the top ten marks have come from 2006 and 2007, probably due to the fact that starting goalies are given an increasingly larger number of starts each year:

I think this explanation is unconvincing. Goalie games played numbers were more or less the same pre- and post-lockout. In fact, the average starting goalie played more games in both 2002-03 and 2003-04 than 2005-06.

Surely this effect is due to the introduction of the shootout loser point, which skews the average winning percentage from .500 to something like .550? Did you use the same exponent for league-average Pythagorean projections for post-lockout as you did pre-lockout? If so, then it is not surprising that the '06 and '07 seasons had high numbers, and they should probably be corrected for that.

Take the supposedly best season as an example, Brodeur in 2006-07. Brodeur's record was 48-23-7, so you would credit him with 51.5 wins. Except he was involved in 16 shootouts, so his record after 65 minutes or less would be just 38-23-23, the equivalent of 49.5 wins for pre-lockout goalies.

Also, I don't have the exact numbers but I guess you projected New Jersey to be slightly outscored with an average goalie in net. The result is that the Pythagorean formula predicts a record a bit under .500, but any goalie playing in the shootout era would probably have a .530 or .540 record based on those expected goal numbers because of shootout losses. So factor that in, remove the additional bonus shootout wins, and Brodeur's season is the in +9 or +10 range, which is still good but not the best.

Otherwise nicely done. This type of analysis shows the strong impact of team strength on goalie stats like wins, and also how rare and valuable truly elite goalies like Roy and Hasek really were.

 

[Hockey Outsider]

Surely this effect is due to the introduction of the shootout loser point, which skews the average winning percentage from .500 to something like .550? Did you use the same exponent for league-average Pythagorean projections for post-lockout as you did pre-lockout? If so, then it is not surprising that the '06 and '07 seasons had high numbers, and they should probably be corrected for that.

Good catch. I completely forgot about the effect of OTLs and SOLs -- obviously that would overrate goalies from 2000-present (when the OTL point was started) and especially from 2006-present (when the SOL point was awarded).

Let's suppose that the actual points percentage in 2007 was 55%. Perhaps it would make sense to divide each goalie's actual wins by 55%, and then compare it to expected wins? (Alternatively I could compare their actuals wins to 55%/50% = 110% of projected wins, but that would give the same result).

I would expect the difference to be very minimal from 2000-2004, but it could be significant to 2006-2008. Thanks for pointing out the error, that's the only way to improve projects like this.

 

[overpass]

Good catch. I completely forgot about the effect of OTLs and SOLs -- obviously that would overrate goalies from 2000-present (when the OTL point was started) and especially from 2006-present (when the SOL point was awarded).

Let's suppose that the actual points percentage in 2007 was 55%. Perhaps it would make sense to divide each goalie's actual wins by 55%, and then compare it to expected wins? (Alternatively I could compare their actuals wins to 55%/50% = 110% of projected wins, but that would give the same result).

I would expect the difference to be very minimal from 2000-2004, but it could be significant to 2006-2008. Thanks for pointing out the error, that's the only way to improve projects like this.

I think your proposed method looks good for OTL points and SOL points.

The extra point for a shootout win is another matter. If you want to have a strict measure of regular season value, the most correct method would be to measure the goalie's shootout contribution, right? With your method, I suppose that would be to measure the goals scored by his team in the shootout, find the performance of the average goalie in the shootout, and compare actual shootout record to expected shootout record. I don't know if the Pythagorean formula applies in the shootout.

Personally, I don't mind the stat leaving out shootout contributions, as they have no value come playoff time. However, as your stat that measures value and is tied to wins, it seems like the shootout should be included, as it leads to wins.

 

[Bear of Bad News]

That's how I think shootout win expectation to be partitioned, and how I've seen it done in the past (by Mr. Ryder or Mr. Fyffe I believe).

It's actually kind of nice for us analytical types, because the shootout is much easier to quantify in terms to win expectation due to its one-against-one nature.