Sunday, October 31, 2010

Top 5 and Top 10 Save Percentages

This is a bit of a data dump post, but I wanted to post the results after compiling the number of top 5 and top 10 save percentage finishes for goalies in the post-expansion era. Note that evidence suggests that team effects were greater in the 1970s and 1980s than in the 1990s and 2000s (see this post for example), which means that it isn't fair to directly compare, say, a goalie who played on a bad team in the 1970s or a goalie who played on a great team in the 1980s with one of today's netminders.

With that in mind here's the list, with the first number being top 5 finishes and the second the number of top tens:

Patrick Roy: 9 / 15
Tony Esposito: 8 / 12
Dominik Hasek: 9 / 11
Billy Smith: 6 / 11
Ken Dryden: 6 / 8
Bernie Parent: 6 / 8
Chico Resch: 5 / 7
John Vanbiesbrouck: 4 / 7
Roberto Luongo: 3 / 7
Rogie Vachon: 1 / 7
Martin Brodeur: 4 / 6
Ed Belfour: 4 / 6
Tom Barrasso: 4 / 6
Kelly Hrudey: 3 / 6
Dan Bouchard: 3 / 6
Andy Moog: 3 / 6
Pete Peeters: 5 / 5
Gump Worsley: 4 / 5
Jacques Plante: 3 / 5
Denis Herron: 3 / 5
Doug Favell: 3 / 5
Reggie Lemelin: 3 / 5
Curtis Joseph: 2 / 5
Cesare Maniago: 2 / 5
Mike Richter: 1 / 5
Don Beaupre: 1 / 5
Tomas Vokoun: 4 / 4
Jeff Hackett: 3 / 4
Mark Fitzpatrick: 3 / 4
Nik Khabibulin: 2 / 4
Don Edwards: 2 / 4
Miikka Kiprusoff: 2 / 4
Glen Hanlon: 2 / 4
Mike Liut: 2 / 4
Sean Burke: 1 / 4
J.S. Giguere: 1 / 4
Bob Essensa: 1 / 4
Ed Giacomin: 1 / 4
Gerry Cheevers: 1 / 4
Grant Fuhr: 1 / 4
Evgeni Nabokov: 0 / 4
Marty Turco: 3 /3
Roman Cechmanek: 3 / 3
Rolie Melanson: 3 / 3
Mike Palmateer: 2 / 3
Bunny Larocque: 2 / 3
Olaf Kolzig: 2 / 3
Ron Hextall: 2 / 3
Jose Theodore: 2 / 3
Guy Hebert: 2 / 3
Daren Puppa: 2 / 3
Manny Fernandez: 2 / 3
Tim Thomas: 2 / 3
Niklas Backstrom: 2 / 3
Bruce Gamble: 1 / 3
Glenn Hall: 1 / 3
Gilles Villemure: 1 / 3
Gilles Meloche: 1 / 3
Chris Osgood: 1 / 3
Felix Potvin: 1 / 3
Ron Tugnutt: 1 / 3
Henrik Lundqvist: 1 / 3
Ryan Miller: 1 / 3
Rick Wamsley: 1 / 3
Manny Legace: 0 / 3

As a postscript, here are the goalies with 500 career games played since expansion who did not make the cut:

Mike Vernon (2/2), Bill Ranford (0/1), Kirk McLean (0/2), Greg Millen (0/1), Jocelyn Thibault (1/1), Ken Wregget (0/1), Arturs Irbe (1/1), Gary Smith (0/0), Tommy Salo (0/0), Roger Crozier (1/2), Dwayne Roloson (2/2)

Tuesday, October 26, 2010

Don Cherry and Swedish Goalies

One of the current goalie hotbeds is the country of Sweden. Six different Swedish netminders made their NHL debuts in either this season or last, to raise to 11 the total number of Swedes who have played in goal since the lockout. Henrik Lundqvist remains the nation's best, but this year Anders Lindback has emerged as a surprise in Nashville, Jonas Gustavsson has played well for a surprising Toronto squad, and Robin Lehner has started to make the case that he might be the future solution to Ottawa's goaltending woes. Despite all this Swedish success, the best may still be yet to come in the person of Jakob Markstrom, considered by many to be the top-rated goalie prospect in the world today.

It was not always the case, however, that the Tre Kronor produced top-level goalies. In fact, there were only a handful of Swedes who played net in the NHL in the 20th century. The entire list is as follows: Tommy Salo, Pelle Lindbergh, Tommy Soderstrom, Hardy Astrom, and Goran Hogosta.

Salo and Lindbergh are likely familiar to most hockey fans. Salo had a lengthy NHL career (526 games played), although he may unfortunately be remembered most for his gaffe against Belarus at the 2002 Olympics. Lindbergh won the 1985 Vezina Trophy, then had his career tragically cut short the following season when he died in a car crash at the age of 26.

The other three had much less memorable careers, although I suspect many Canadian hockey fans have at least a passing familiarity with the name of Hardy Astrom. Astrom played just three years in the NHL, but has achieved a level of infamy as a result of a few stories told by TV personality Don Cherry on CBC's Coach's Corner and on his personal Rock 'em Sock 'em videos.

The way Cherry tells it, Astrom was a bumbling goalie that couldn't stop soft lobs from center ice in practice. It's entirely possible that Astrom let in some stinkers, especially while getting bombarded behind the hapless Colorado Rockies defence. Still, does it make sense that a guy who represented Sweden in the 1976 Canada Cup and two other world championships and attracted enough attention to play pro hockey in North America at a time when Europeans were underrepresented in the NHL was laughably inept?

Sounds to me like this guy might just have been stigmatized by a bad goal or two at the wrong time, a la Tommy Salo. Let's see what the numbers say. Here are the results for all Colorado goalies from 1979-80 to 1980-81. Breaking it into Astrom vs. everyone else, we get the following:

Hardy Astrom: 3.76, .870
All other goalies: 4.20, .854

I'm starting to think Don Cherry may have been exaggerating just a little.

Astrom was pretty unlucky in the goal support department, because his winning percentage lagged most of his teammates. That might have led to a perception that he wasn't a "winner". Yet for all Cherry likes to rag on Astrom, the coach still gave him 49 games in net in 1979-80. Swede jokes may entertain Canadian TV audiences, but by what economists called revealed preference it is unlikely that Cherry really thought his Swedish goalie was truly that horrible, based on the coaching decisions he made with his job on the line.

The numbers suggest Astrom was likely a pretty average netminder, especially when you factor in how bad the Rockies were. I bet Cherry coached quite a few other goalies who were worse than Astrom during his tenure in the NHL, and it seems kind of unfair (but perhaps not surprising, given Cherry's general anti-European sentiment) that he seems to have picked out one Swedish guy all these years later to be the butt of his jokes.

Tuesday, October 19, 2010

Questioning Ed Giacomin

(Like Tyler Dellow, I usually feel like I don't have much to talk about early in the season. Anything involving goalies is going to necessarily be an attempt to find meaning in impossibly small sample sizes. My posts will be mostly focused on historical items until we can be in a better position to guess at how things are shaking out.)

If you were asked to list off the easiest goalie jobs in NHL history, there would be a number of dynasty teams that would come immediately to mind. Right after those powerhouses, I think a high ranking on that list should be reserved for the starting role on an Original Six team shortly after expansion in the late '60s and early '70s. That was a position that was pretty much guaranteed to make you look better than you were.

To nobody's surprise, the teams that were already full of NHL talent dominated the expansion teams in the first few seasons after the league doubled in size. This effect lasted for quite some time as new expansion fodder kept getting tossed into the mix throughout the decade, until the late '70s when the Hawks, Leafs and Wings had fallen back into the pack and a number of expansion teams including the Flyers, Islanders, and Sabres had become legitimate contenders for the Cup.

There were 11 Hall of Fame goalies that were active during the period from 1968 to 1975, but a handful of them were at the end of their careers (Plante, Worsley, Hall, Bower, Sawchuk), and another (Billy Smith) was suffering through the growing pains on an expansion team and had yet to make his mark. That leaves Ken Dryden, Tony Esposito, Gerry Cheevers, Ed Giacomin and Bernie Parent as the goalies who had their peak in the post-expansion period.

Or, to express it in a slightly different but perhaps more meaningful way, Montreal's goalie, Chicago's goalie, Boston's goalie, New York's goalie, and Bernie Parent.

The Leafs and Red Wings might have been able to put somebody in the Hall of Fame too if they hadn't spread the workload around. Both teams used 14 different goalies in those 8 seasons, and Roy Edwards was the only one who played in over 200 games. The Leafs did have both Plante and Parent on their team at times during this period, which helped both of their resumes although both of them would have become honoured members even without their tenures in Toronto.

If team effects had the potential to create Hall of Famers, then which goalies got lucky and which ones were unlucky? I'd submit Ed Giacomin as probably the worst of the lot. If you look at the Rangers' year-by-year GAAs against Original Six teams compared to expansion teams, you start to get a sense of the lack of balance in the league:

1967-68: 2.76 vs. Original Six, 1.88 vs. Expansion
1968-69: 3.15 vs. Original Six, 1.94 vs. Expansion
1969-70: 2.80 vs. Original Six, 2.14 vs. Expansion
1970-71: 2.73 vs. Original Six, 1.98 vs. Expansion
1971-72: 3.10 vs. Original Six, 2.06 vs. Expansion
1972-73: 3.31 vs. Original Six, 2.35 vs. Expansion
1973-74: 4.26 vs. Original Six, 2.67 vs. Expansion
1974-75: 4.52 vs. Original Six, 3.07 vs. Expansion
Period Averages: 3.14 vs. Original Six, 2.34 vs. Expansion

That's a 34% increase in GAA when playing against a fellow Original Six team. The Rangers also shut out the expansion teams 39 times, compared to just 17 blankings of their older foes.

Giacomin was voted the best goalie in 1966-67, which appears to be an impressive feat given that there were still only five other teams that season. However, that was an unusual year where most of league's netminders were in platoon situations. Roger Crozier was the only other goalie who played in more than 44 games, and considering Crozier played on a much weaker Detroit team it looks like Giacomin was named the best goalie more or less by default.

Giacomin also was voted the top goalie in 1970-71, a year where he finished second in save percentage with an impressive .922. Although this was Giacomin's career year, he still never should have received that honour, not when Jacques Plante posted an incredible .944 save percentage in just 5 fewer games played. In addition, Giacomin's journeyman backup Gilles Villemure had very similar stats (2.30, .919 compared to Giacomin's 2.16, .922), which suggests that the Rangers were making it tough for the opposition to score.

In this period Giacomin got many of the typical edges of a goalie playing for a defensive, disciplined team. New York faced fewer opposing power plays than average every year between 1968 and 1975 except for 1971-72. The Rangers were also routinely among the top teams in shots prevention. Newly released shot data shows that Giacomin's career shots faced per game rate was 28.8, just 0.3 higher than Ken Dryden's career mark. Giacomin also got a ton of starts, which can be interpreted as either a fortunate opportunity or evidence of his durability, depending on your outlook.

Giacomin did not have much longevity, and he only had about 5-6 peak seasons (which with the single exception of 1966-67 came in the post-expansion period). His career save percentage of .902 exactly matched league average for the seasons he played. He was nothing special once he was shipped out from Broadway to the Red Wings. His playoff record was also decidedly mediocre. The main argument for Giacomin seems to be his award recognition. In addition to being twice named the best goalie as mentioned above, Giacomin was voted the second best goalie at season's end an additional three times. Given that he did not rank in the top 5 in the league in save percentage in any of those three campaigns, despite the advantages mentioned above, it seems reasonable to question whether he truly deserved that recognition.

When evaluating players, I think it is a fair argument to claim that someone with a Hart Trophy or an Art Ross Trophy should have an advantage over a similar player without similar award recognition. With goalies I am hesitant to give the same weighting to individual awards. The two reasons for that is that I believe writers are more likely to get the voting wrong with goalies than with forwards, and that there appears to be more of a luck element involved in a single season's worth of goaltending than for a single season by a forward. Giacomin was repeatedly recognized as one of the game's best, which could have been because he was actually one of the best or it could have been because he was an established goalie playing in a large market on one of the league's best defensive teams that profited greatly from pounding on the league's weak sisters.

As for which goalies were unlucky, I think it is extremely unlikely that Rogie Vachon was a worse goalie than Ed Giacomin. Vachon was much younger when he broke into the NHL and had a longer career, he had a better save percentage (.905) over the seasons that Giacomin also played in the NHL, he played well on more than one team, he had a better Hart Trophy voting record and he represented Canada internationally. I think it's a pretty open-and-shut case, and it makes no sense to me that Giacomin is in the Hall of Fame while Vachon is not.

In a very unbalanced competitive environment, I think it is right to at least question any goalie who only had elite success for a short period of time in a single team environment. It is possible that they had a short peak and then tailed off, but I think the more probable explanation is that they had team advantages during a certain portion of their career that they did not have at other times.

Tuesday, October 12, 2010

The Best Calendar Year Ever?

Recently two different publications, one by Hockey Night in Canada, the other by The Hockey News, have come out with historical goalie rankings. Both of them made what I consider to be a very serious error in underrating Dominik Hasek. I've been beating the drum for Hasek pretty much since day one in this space, not because I am a personal fan of him or any of his teams, but because it is impossible to get deep into goalie statistics without being impressed by the ridiculous level of dominance Hasek displayed in the 1990s.

So here we go again, another chance for me to throw out numbers supporting the great Czech netminder, trying to properly illustrate the unrighteousness of ranking the Dominator as the 5th best goalie of all-time, as THN does, or the 3rd best goalie since 1967 (a la HNIC).

In the calendar year of 1998, including the regular season, playoffs and Olympics, Dominik Hasek had the following stat line: 54-20-14, 1.75, .943, plus 16 shutouts. He averaged nearly 1 shutout every 5 games, playing mostly on a Buffalo Sabres team that was the worst team in the league at shot prevention in 1997-98 and 5th worst in 1998-99. The Sabres' win threshold over those two seasons was .917, meaning that they needed a very good goalie just to be a .500 team. To further stack the deck against the Dominator, the Sabres took more penalties than average in both seasons as well.

The league average save percentage in the regular season was around .906 in that period. In the 1998 playoffs, all goalies other than Hasek combined for an average of .912. Playoff averages usually rise slightly because teams only play their starting goalies. A difference of .006 suggests that the scoring environment was pretty similar between the regular season and playoffs. In the Olympics, the average save percentage was .904 (that's not including Kazakhstan, which got completely shelled in every game).

I'll take .906 as the league average and assume Hasek faced average shot quality, was not impacted by scorer bias and did not play a role in his team's shot prevention (or at least that the effects of all three ended up netting out to zero). Based on those assumptions, Hasek was about 120 goals above average in a 12 month span. Considering he did it in minutes equivalent to 90.3 full games, Hasek averaged 1.33 goals better than average per game for an entire year, during a time period when the average NHL team scored 2.60 goals per game.

That is why Hasek should be talked about among Gretzky, Orr, Lemieux and Howe when people are discussing the greatest peaks in hockey history. It's certainly not conventional wisdom to put Hasek up in the stratosphere with those legends, but there is a numerical case for it. It's possible that, like Hasek, Gretzky had some crazy calendar year that was better than any of his full seasons (maybe 1983, which included most of his ridiculous 51-game point streak), but his most impressive full season may have been 1984-85. Counting regular season, playoffs and the Canada Cup, the Great One scored 95 goals and 172 assists for 267 points and a +126 rating in 106 games.

That stat line probably looks way more impressive than Hasek's to most hockey fans, but a lot of that is probably because we have more intuitive sense about the level forwards produce at than the level goalies produce at. We know that 267 points is far beyond the curve for forwards, but while we realize that .943 is great, we may not have a sense of exactly how great (adjusted for scoring environment, it would be the equivalent of Patrick Roy in the '93 playoffs, or J.S. Giguere in the '03 postseason, for 90 games in a row). Depending on your assumptions about his ice time, the strength of his teammates, Gretzky's defensive ability and the production of an average forward, it's possible to argue that Hasek contributed more on a per-game basis than even the Great One.

For example, let's assume that Gretzky had average ice time, was average defensively and played with average linemates (two of those are clearly false, but bear with me). Since Hasek was being compared to the average goalie, let's use the average first-line forward as the baseline for Gretzky. In 1984-85 the average first-liner, excluding Gretzky himself, averaged 1.09 points per game, which equates to 116 points in 106 games. As a result, we can conclude that Gretzky scored 151 points above average, 1.42 points per game higher than the average forward. That is just slightly better than Hasek's mark even based on the prior assumptions and giving Gretzky sole credit for all of his points. Take into account the fact that he played on the same line as Jari Kurri and the same team as Paul Coffey, factor in Gretzky's heavy ice time, and maybe he doesn't beat the Dominator after all.

That's just a quick-and-dirty method, we could also use a metric that is designed to measure value such as GVT. Gretzky had a 59.1 GVT rating in the regular season and playoffs in 1984-85. Hasek averaged 54.0 GVT in 1997-98 and 1998-99, and added an additional 13.6 in the 1998 playoffs. Including the Olympics and looking just at that calendar year, Hasek probably had a GVT over 70. That means that both in total and on a per-game basis, Hasek's numbers in this period would easily beat not only Gretzky's best season but also the best seasons of Bobby Orr and Mario Lemieux (although, to be fair, GVT typically does rank the top goalies above the top skaters).

Having said all that, Gretzky's peak is extremely impressive because he was able to maintain it for such a long period of time (averaging 203 points per year over a six-season stretch, plus another 31 per year in the playoffs). Hasek was amazing throughout the '90s but I think most would agree that 1998 was his absolute peak, which means that he was probably both playing out of his mind while also having some luck in terms of having the puck hitting him. If you're rating careers then no doubt Gretzky wins, and if you're rating extended primes than there's a good case for the Great One as well. However, for one game, at the absolute height of their respective powers, I'd definitely think twice about it. At the very least I think Hasek down to the next goalie is a bigger gap than Gretzky to Lemieux.

Some people will probably tell you that Gretzky was on a different level than Hasek because the Great One was in a class of his own far ahead of the rest of the league, while Hasek was only just a bit better than Brodeur and Roy. Those people are flat-out wrong. This is what Hasek's main rivals were doing over the same 12 months, see if you think any of them are even close:

Roy: 32-32-6, 2.36, .910, 6 SO
Brodeur: 39-22-11, 2.12, .909, 8 SO
Belfour: 45-18-7, 1.87, .919, 6 SO
Joseph: 41-32-3, 2.38, .914, 9 SO

For an even better expression of the relevant difference, here are the goals against per 30 shots numbers for those four guys and Hasek:

Dominik Hasek: 1.71
Ed Belfour: 2.43
Curtis Joseph: 2.58
Patrick Roy: 2.70
Martin Brodeur: 2.73

Belfour may have been the second-best goalie that year, and he was still dusted by Hasek. Remember that Belfour was playing on the back-to-back President's Trophy-winning Dallas Stars, a terrific defensive team coached by Ken Hitchcock that also took fewer penalties than average. Even if you don't like save percentage or you think that there are other major factors in play like puckhandling, just compare those win/loss records for Belfour on the best team in the league (.875 win threshold) against Hasek on the overmatched Sabres and the underdog Czech Olympic team:

Belfour: 45-18-7, .693
Hasek: 54-20-14, .693

At the very least, I think it's fair to say that the number of players in league history that have had that big of an effect on winning can be counted on one hand.

In my mind it will be a travesty if Hasek is remembered as the third-best goalie of his generation, which is a perception that a lot of media-types are currently doing their best to entrench. In terms of actual performance, the Dominator really does stand alone.

Tuesday, October 5, 2010

SRS

Hockey Reference has a metric called the Simple Rating System (SRS) that is designed to measure team strength in terms of predicted goal differential per game. I haven't yet run the complete numbers to see exactly how accurate it is in terms of predicting playoff results, but I do know from hacking around a bit with the results that teams that have a large edge in SRS over their opponents are much more likely to win their playoff matchups.

It appears that the threshold where victory becomes highly likely for the favoured team is when the SRS gap between competing teams is .3 or more. Since SRS claims to be a measure of a team's per-game goal differential above average, that is the equivalent of a difference in goal differential of about +25, which equates to a difference of about 4 wins over 82 games in terms of "true talent" (i.e. the rating is designed to adjust for schedule strength and to remove some of the effects of variance over the course of a single season, such as a team that wins a lot of shootouts or close games).

Even the best goalies in recent times built their playoff records largely on series victories where their team had an SRS rating that was .3 or more ahead of the opposition. Here are playoff series win/loss records for the consensus top 4 goalies of the last two decades split by the strength of their teammates and opponents:

Patrick Roy:
On teams >.3 better than the opposition: 21-2
On teams >.3 worse than the opposition: 2-3
Playoff series against all other opponents: 10-8

Martin Brodeur:
On teams >.3 better than the opposition: 15-2
On teams >.3 worse than the opposition: 0-1
Playoff series against all other opponents: 5-9

Ed Belfour:
On teams >.3 better than the opposition: 12-2
On teams >.3 worse than the opposition: 1-4
Playoff series against all other opposition: 6-6

Dominik Hasek:
On teams >.3 better than the opposition: 5-0
On teams >.3 worse than the opposition: 0-3
Playoff series against all other opposition: 7-4

Combined results:
On teams >.3 better than the opposition: 53-6
On teams >.3 worse than the opposition: 3-11
Playoff series against all other opposition: 28-27

Playing on a much superior team, these goalies combined to win 90% of the time. When playing on a much weaker team, they only managed to win 20% of the time. In all other series it was essentially a coin flip.

It should be noted that the SRS rating is based on regular season performance that includes the contribution of the goaltender themselves. Keep in mind that in some of these situations, the goalie's own play was largely responsible for lifting their team ahead of the opposition. For example, some of the teams had lower win thresholds than their opponents yet higher SRS ratings, which would be an indicator of quality goaltending.

Also, just because the goalie is playing on the better team does not mean they weren't playing well, or were only overseeing 8-0 blowouts. It is important to avoid the common bias in favour of goalies on great teams, but we shouldn't overcompensate by never giving any credit to any goalie with decent teammates. Upsets can easily happen in seven game playoff series. A good team with a good goalie in net becomes tough to beat, as even when the team is off its game the goaltender can sometimes come to its rescue.

As a point of comparison, Chris Osgood was 12-5 in series where his team had a significant advantage. I'm not saying Osgood was to blame for all of those losses, but poor goaltending, or perhaps an average goalie who goes on an unlucky streak, can sometimes sink even a very good team. The high winning percentage in the mismatched series above does seem to indicate strong goaltending, although certainly the rest of the team played a huge role as well.

As an footnote, here are the numbers for Curtis Joseph:
On teams >.3 better than the opposition: 2-3
On teams >.3 worse than the opposition: 2-6
Playoff series against all other opposition: 7-4

In the latter two categories, Joseph's record stacks up against any of the other guys. The main difference was that he did not have nearly as many chances to pound weaker opponents as Roy, Brodeur and Belfour. Cujo was also quite unlucky in terms of goal support, as the Red Wings' failure to score on Giguere in '03 and Kiprusoff in '04 ended up being the main reason for Detroit's downfall in two of his three losses against weaker opposition.

Saturday, October 2, 2010

Some Help from the Official Scorer?

Pekka Rinne, career:
at Home: 42-17-5, 2.21, .924, 29.12 SA/60
on Road: 20-15-4, 2.87, .897, 27.84 SA/60

Those splits certainly make one at least a bit skeptical about whether Rinne's save percentage is being accurately reported in his home rink in the Music City, and whether we can therefore rely on those numbers to accurately reflect his performance. Nashville has been one of the places suspected of a shot recording bias (for more info see Tom Awad or JLikens), which increases the probability that there may be overcounting going on in Rinne's favour.

Unfortunately, this case is not entirely clear cut, because former Predators goalie Dan Ellis doesn't have the same extreme splits. In fact, Ellis actually has a higher career save percentage on the road despite playing all but one game of his career for the same Predators team as Rinne:

Dan Ellis, career:
at Home: 18-16-5, 2.59, .910, 28.82 SA/60
on Road: 32-26-5, 2.68, .913, 30.78 SA/60

Backup goalies can sometimes have skewed numbers because of relief appearances, but even after taking out the 12 games where Ellis came in off the bench, the pattern persists:

at Home: 18-15-4, 2.57, .913, 29.58 SA/60
on Road: 30-22-4, 2.65, .916, 31.62 SA/60

This is a fairly small sample for both Rinne and Ellis, and I expect that some of the effect is simple random variance. It is likely that the .027 difference between Rinne's home and road performance has been a bit of a statistical quirk, as has his good fortune to face almost 4 fewer shots against per game than Dan Ellis while on the road, but I'm still slightly suspicious of the official scorer in Nashville. It will be interesting to see if similar results continue this season with Ellis in Tampa and Rinne expected to once again be the main man in the crease in Nashville.