Saturday, November 1, 2008

Expected Shutouts

The main problem with the adjusted shutout totals I calculated in a recent post is they do not take team factors into account. A goalie playing on a strong defensive team in a higher-scoring era (like, for example, Tony Esposito) would rank a lot higher by the adjusted shutouts metric than they probably deserve.

To deal with this issue, I have developed another way of looking at shutouts. It is based on comparing a goalie's shutout performance to how many shutouts an average goalie would probably have recorded while playing the same minutes and facing the same shots. The expected number of shutouts was based on the goalie's minutes played and the predicted probability of a shutout for an average goalie, which was calculated by raising the league-average save percentage to the power of the number of shots faced per game by the subject goalie. For example, if a goalie faced 30 shots per game when the league average was .900, the expected probability of recording a shutout in any given game would be .900^30, or .042. If the goalie played 4000 minutes that season, that would project to an expected season total of 2.8.

This probability calculation is based on the assumption that the goalie would face exactly the same number of shots every game, and that each shot against is equally likely to be stopped, two assumptions which are obviously not true. However, since it is based on averages, it is likely that the method will come up with a predicted shutout total that is within a reasonable range from the actual result.

To test whether this method approximates reality, I compared the expected results against actual (not including the current season). Since 1983-84, i.e. the seasons Hockey-Reference has save percentage data, the expected shutout percentage has been 4.4% of games, and the actual has been 4.8%. This is pretty close. I expected the actual result to slightly exceed the expected result because goalies do not face the average number of shots every game and it gets exponentially easier to get a shutout whenever you face fewer shots. For example, a goalie who faces 20 shots one game and 40 the next is much more likely to get a shutout in one of those two games than a goalie who faces 30 shots per game twice in a row. This method has no corrections for shot quality, so goalies that face higher than average shot quality have a more difficult time recording shutouts regardless of their shots against totals. With these issues in mind, we need to interpret this statistic by looking at ranges and relative rankings, rather than raw totals only.

To show how this works, let's look at a couple of specific examples. Cristobal Huet has 17 career shutouts, while Dan Cloutier has 15. On the career rankings list, they look pretty similar. But are they? I calculated Cloutier's expected career shutout total to be 22.7, meaning Cloutier is 7.7 shutouts (or 34%) below average. Huet, on the other hand, has just 9.8 expected shutouts, meaning he has outperformed the metric by 7.2 shutouts, or 73%. Even though they have similar career totals, Huet is among the very best and Cloutier is among the very worst when you look at performance relative to expected.

Unfortunately I cannot apply this ranking to every goalie throughout history because we have limited shot data. I therefore had to focus on goalies who played mostly in the save percentage era, and if shot data was missing I filled in the league average shots per game for those seasons. Here are the top 20 goalies who have at least 20,000 career minutes between 1982 and 2008, ranked by ratio of shutouts to expected shutouts:

RankGoalieSOExp SODiff%
1.Dominik Hasek8143.1+37.9+88%
2.Roberto Luongo3920.8+18.2+88%
3.Mike Liut2514.4+10.6+74%
4.Pete Peeters2113.0+8.0+62%
5.Greg Hanlon138.4+4.6+55%
6.Tom Barrasso3825.1+12.9+51%
7.Greg Millen1711.6+5.4+47%
8.Guy Hebert2819.8+8.2+41%
9.Patrick Roy6647.7+18.3+38%
10.Daren Puppa1914.5+4.5+31%
11.Evgeni Nabokov4031.0+9.0+29%
12.Ed Belfour7660.2+15.8+26%
13.J.S. Giguere3024.2+5.8+24%
14.Martin Brodeur9677.6+18.4+24%
15.Patrick Lalime3529.2+5.8+20%
16.Jocelyn Thibault3932.9+6.1+19%
17.Jon Casey1613.5+2.5+19%
18.Andy Moog2823.9+4.1+17%
19.Rick Wamsley1210.3+1.7+17%
20.Curtis Joseph5144.0+7.0+16%

The two usual suspects, Hasek and Luongo, end up at the top. Martin Brodeur's record compared to average is good but not exactly dominant. Brodeur does have a high career "shutouts above average" mark with +18.4, second only to Hasek, edging out Roy's +18.3 with Luongo right behind at +18.2. Hasek crushes the field with almost 38 more shutouts than average. If Hasek played Brodeur's minutes and outperformed average by the same 88% margin, the Dominator would have had 146 shutouts. Just something to keep in mind whenever somebody tries to argue that Brodeur's eventual shutout record means that he was the best.

I'll finish for now by identifying some potentially overlooked goalies who did well but have played fewer games. I'll set 10,000 minutes as the cutoff here (note that both Kari Lehtonen (+75%) and Marc-Andre Fleury (+45%) narrowly missed the minimum but would have made this list):
GoalieSOExp SODiff%
Allan Bester72.8+4.2+150%
Bob Froese135.7+7.3+128%
Murray Bannerman84.0+4.0+100%
Clint Malarchuk126.9+5.1+74%
Cristobal Huet179.8+7.2+73%
Mathieu Garon149.1+4.9+54%
Roman Cechmanek2516.8+8.2+49%
Miikka Kiprusoff2620.0+6.0+30%
Pat Riggin84.0+4.0+29%
Henrik Lundqvist1713.2+3.8+29%
Jamie Storr1612.9+3.1+24%


Scott said...

Thanks CG. This looks great. The methodology also seems to me to be very reasonable. If you remove seasons for which shot data isn't available, do you end up with any significant variations?

The Contrarian Goaltender said...

Not really, no, it was usually just a season or two of shots missing for the guys I looked at, and the early 1980s had relatively few shutouts and low league-leading shutout totals (3 or 4 per year were usually enough to win).

Scott said...

Thanks CG. Good to know. It would be interesting to compare these guys with long-term goalies who don't make this list. It might give some indication of consistency. If a guy like Hebert has 41% more shutouts then expected but his save percentage is average his median game would suffer.

I'm not sure which is more important in a playoff context, though I suppose it would depend on the team. A poor team would probably want someone that has higher highs while a good team would probably want someone more consistent.

Anonymous said...

i dont really see the point in saying "had hasek played brodeurs minutes" because hasek only retired last season. so if he didnt play brodeur like minutes, then the falt is his own, and obviously there is a reason brodeur was able to do this and hasek wasnt. chances are this same reason has a lot to do with what separates brodeur from hasek and others when comparing their greatness.

ILR said...

Good ol' Roman Cechmanek and Patrick Lalime show up a bit unexpectedly here. If Osgood is the definite league average goalie, Cechmanek and Lalime might be the quintessential subpar stoppers on a great team. Those Flyers and Senators teams at the end of the dreaded Dead Puck Era definitely had a cup run in them (as did St.Louis and Vancouver, also victims of shoddy goaltending on a potent team). It's telling that Brian Boucher got the Flyers further than Dancin' Roman ever did.

Based on his official stats, Cechmanek was a monster. What did he have, 1,8 GAA over the course of three Flyers seasons, or so? As the memories fade and a new breed of stats-junkies grow up, we're bound to see a Cechmanek revival. Yes, I see it now: 'The Most Underrated Player of Early 2000s'

The Contrarian Goaltender said...

i dont really see the point in saying "had hasek played brodeurs minutes" because hasek only retired last season.

The point is that Brodeur's minutes were played behind the New Jersey defence. It is hardly Dominik Hasek's fault that he was drafted by Chicago and not New Jersey. I understand why you would prefer Brodeur's durability to Hasek's, but if you look at the minutes both of them did play, Brodeur's expected shutouts per game rate is 35% higher than Hasek's. If Hasek played on New Jersey, I think there is little doubt that he would already be the all-time shutouts leader.

If minutes played are really enough to separate Brodeur from Hasek, then Glenn Hall should be considered the greatest goalie to ever play hockey. But I have never heard anyone make the case that Hall's durability makes him better than Plante and Sawchuk. The people who do make that case tend to refer to other things, such as for example Hall's 7 First Team All-Star selections. You know, measures of skill rather than measures of how much the team is willing to pay their backup goalies.

The Contrarian Goaltender said...

By the way, I think any stat junkie worth his salt would name Roberto Luongo the most underrated goalie of the early 2000s. Lalime is certainly a good example of a subpar goalie on a great team. But Cechmanek is an interesting case, because as you say his statistics were very impressive. Some of it has to be because of easier than average shot quality, but Cechmanek far outperformed the other goalies he played with in Philadelphia which makes me think he was doing something right.

I think Roman Cechmanek was a good goalie, at least for a few seasons. His problem was that he got stigmatized because he was prone to letting in soft goals. I don't think what kind of goals a goalie allows should matter, just the number of goals he allows, but that nevertheless affects a goalie's evaluation and reputation. The playoff failures also didn't help, but in my opinion they weren't entirely Cechmanek's fault. Everyone seems to forget how many goals the team scored when they evaluate goalies on playoff success. Just look at the goal support:

Boucher '00: 2.44 GFA, 2.03 GAA
Cechmanek: 2.04 GFA, 2.33 GAA
Esche '04: 2.78 GFA, 2.32 GAA

No wonder Boucher and Esche went farther than Cechmanek. Lalime was actually a similar case, in that his teammates were also big playoff underachievers and scored at a much lower rate in the postseason. Whenever you talk about wins and losses, it is always a team effort.

Bruce said...

I think Roman Cechmanek was a good goalie, at least for a few seasons.

So do I. Sucker used to break Canadian hearts (mine included) at the Worlds year after year, for one thing. He's played in 8 Worlds so far. Fwiw, he and Hasek both played over 300 games in the Czech league; Cechmanek's GAA was 2.22, Hasek's 2.78. Different teams, different eras, but suffice to say the Czech maniac was good.

At the NHL level the guy posted three consecutive Sv% over .920 which is no mean feat no matter what team you're own. His luck ran dry in the playoffs, he let in the odd soft goal which really tended to hurt given the utter lack of goal support that you referenced. In 2002 against Lalime's Senators, the Flyers scored 2 goals in a 5-game series, 1 of them in OT in the one game Cechmanek managed to win, 1-0.

Lalime played like a man possessed in that series, he's always been hot and cold but when he was hot he was mighty tough. Unfortunately he couldn't beat Toronto when it mattered most.

ilr said...

Yeah, it's quite possible that Cechmanek got a rawer deal than he deserved due to just simply bad luck when it counted the most. The small sample size of playoff hockey tends to do that sometimes. To Cechmanek's and Lalime's credit, they actually do show up on this list whereas the goalies of those pinnacle St.Louis & Canucks teams are nonexistent (Turek) or actually harmful (Fuhr, Cloutier).

Overall, the shutout metric makes for a comprehensive list that is well in line with the other analyzes on this site, although some statistical anomalies such as for example Ken Wregget's statistically great but shutoutless(!) 94-95 season makes him seem worse than he objectively probably should be. Also some guys (Storr, Casey, Barrasso) stick out like sore thumbs on the 'good' list. Of those, Barrasso has a high-ish exp SO value, so his greatness is unlikely to be just blind luck. The other two I'm going to count as flukes until further proof. There's probably a case to be made for Casey but I can't imagine what kind of voodoo would be required to pimp up Jamie Storr.

The Contrarian Goaltender said...

I can't imagine what kind of voodoo would be required to pimp up Jamie Storr.

I'm not sure much voodoo is required. I think Storr's biggest failing was more not fulfilling expectations as a high first round draft pick than performing poorly. His stat line is actually fairly decent for a goalie who never established himself as an NHL starter.

Career save percentage .908, .002 better than league average during his career. Storr outperformed his expected shutout results and recorded shutouts at a much higher rate than his teammates. Between 1997-98 and 2002-03, Storr led the Kings in save percentage 4 times in 6 years, and Los Angeles was 79-73-19 (.518) with Storr in net and 141-141-39 (.500) with everyone else, so hindsight suggests the Kings maybe should have played him more often.

As a highly touted junior goalie and a high draft pick, there is no doubt that Storr was a talented goalie, and raw talent tends to show itself more in things like shutouts. I am not really that familiar with Storr because of his short career mostly on the West Coast, but I'm guessing he was likely held back from developing into a top starter because of other factors, such as technique or consistency. I think if Jamie Storr was drafted by a top defensive team with a good goalie coach, he might be a starting goalie in the NHL today.