"Clutch" Goaltenders

[Doctor No]

I'm working on an article about clutch performance of goaltenders, and the way that I typically write articles is to leave data here and there, and then I can't find it, and then I move onto something else without actually writing the article.

So here we go...

One definition that we could use to define goaltending "clutchness" is how they perform in these situations:

• Plain old playoff games
• Games in which they can eliminate an opponent
• Games in which their team can be eliminated
• Mutual elimination games (typically a game seven)

If they perform in a statistically-significant fashion better in the more "key" games, then one might suggest that the goaltender has shown signs of being clutch.

Of course, this is not the only way to define clutchness (perhaps not even the best way), but it's what I'm going with because I became interested with all of the "Jonathan Quick is so clutch when facing elimination" talk this spring.

Click to read more...

 

[Doctor No]

Of course, the goaltender's performance has to be measured against what he could reasonably be expected to do. If you're always the #8 seed in the conference, then you may lose a lot of "games where your team could be eliminated" solely because you're constantly facing great teams.

For each goaltender in a game, I measure their expected number of saves as: (shots faced) multiplied by (one minus the opponent's non-empty-net shooting percentage).

For the opponent's shooting percentage, I use their entire season's data (regular season plus postseason) because otherwise, especially for teams that only play one playoff round, the goaltender I am measuring would determine the metric. I realize that this isn't perfect (since typically save percentages go up in the playoffs overall relative to regular season) and I'm working on an alternative.

For instance, Dan Bouchard played against Boston in the playoffs on April 5, 1983 (this just happens to be the first data point in my set of data). He faced 37 shots on goal, and the Bruins that year had a non-ENG shooting percentage of 13.27%. We would therefore expect Bouchard to make (37)*(1-.1327) = 32.18 saves on the game. He actually made 33 saves in a 4-3 loss. (This actual data point may be found under NHL GAME LOGS on Bouchard's link above). So in this case, although Bouchard lost the game, he actually performed slightly better than would have been expected.

His opponent, Pete Peeters, was expected to make 17.97 saves on 21 shots, and he actually made 18.

Obviously, in a single game this metric may not be particularly fair - I couldn't tell you offhand whether or not Bouchard let in any weak goals among the four that he allowed, or how many "big saves" he made, and my guess is that I'm not alone in that regard. The premise is that if we aggregate many of these results, then the unfairness could wash out.

 

[Doctor No]

Okay, let's play a bit. I have data currently loaded from 1983 through 2014 (am working on 1981 and 1982, but let's leave those aside for now).

The metric I prefer for this process is something that I've got on my site - S+/30. For every thirty shots faced by a goaltender (approximately one game), how many goals do they prevent (beyond an average goaltender)?

This also allows for more apples-to-apples comparisons between eras (for instance, if there were more mutual elimination games in the 1980s than today, we might see that save percentages dropped in those situations if we looked at raw totals).

Remember I said before that save percentages go up in the playoffs? Part of that is because better goaltenders play in the playoffs, and part of that is because systems improve. This shows up here, by situation:

Situation | Actual SV% | Expected SV% | S+/30
No Elimination|0.905|0.896|+0.30
Can Eliminate|0.908|0.896|+0.35
Can be Eliminated|0.904|0.895|+0.26
Mutual Elimination|0.915|0.896|+0.57

(These results form a partition, meaning that I have excluded Mutual Elimination games from the two categories above it)

Conclusion one: yes, save percentages are higher in the playoffs. Even in non-elimination situations, goaltenders do better than "expected" as a matter of course; for every 30 shots faced, they prevent an extra 0.30 goals.

Conclusion two: there's not a lot of evidence that (on the whole) goaltenders step up in situations where they can eliminate, or where they can be eliminated (beyond the 0.30 number).

Conclusion three: in mutual-elimination games, it does appear that goaltenders step up their game on the whole (preventing an average of 0.57 goals beyond expected). The sample I have covers 15,743 minutes of action, or about 262 regulation games.

 

[Doctor No]

Okay, let's have a little bit of fun now, and play with a few goaltenders.

Patrick Roy has a pretty good reputation for clutch play.

Situation | GP | W | L | Actual SV% | Expected SV% | S+/30
All Games|247|151|94|0.918|0.894|+0.73
Can Eliminate|56|30|26|0.918|0.894|+0.70
Can be Eliminated|24|12|12|0.924|0.897|+0.83
Mutual Elimination|13|6|7|0.907|0.897|+0.31

(Unlike the above table, these results form do not form a partition. Mutual Elimination games are included in all three "elimination" categories)

A few conclusions - compared to the numbers above, Roy steps his game up in the playoffs overall. In mutual-elimination games, he doesn't look so hot; this is suggestive of small samples (although there's certainly no evidence to suggest that he's "extra clutch" in game sevens).

A look at his mutual-elimination games may be in order:

Game | Result | Score | Shots | Exp Saves | Actual Saves
4/29/1986 vs. Hartford|W|2-1 OT|25|21.6|24
4/29/1991 at Boston|L|1-2|29|25.7|27
5/1/1992 vs. Hartford|W|3-2 2OT|41|36.7|39
4/29/1994 at Boston|L|3-5|31|28.1|26
5/4/1998 vs. Edmonton|L|0-4|17|15.5|13
6/4/1999 at Dallas|L|1-4|25|22.5|21
5/27/2000 at Dallas|L|2-3|29|26.2|26
5/9/2001 vs. Los Angeles|W|5-1|26|23.3|25
6/9/2001 vs. New Jersey|W|3-1|26|23.2|25
4/29/2002 vs. Los Angeles|W|4-0|23|20.8|23
5/15/2002 vs. San Jose|W|1-0|27|24.1|27
5/31/2002 at Detroit|L|0-7|16|14.5|10
4/22/2003 vs. Minnesota|L|2-3 OT|30|27.3|27

I'm not entirely sure what to make of this, although Roy's overall record in game sevens may be beaten to death at this point.

 

[Doctor No]

I'll do Martin Brodeur next, and save some others for later, since he's been on my mind lately. Also, to the extent that save percentage does Brodeur a disservice, at least I'm comparing him to himself (so it should affect him somewhat equally, as long as he plays about half of games at home in all situations).

Situation | GP | W | L | Actual SV% | Expected SV% | S+/30
All Games|205|113|91|0.919|0.904|+0.44
Can Eliminate|39|22|17|0.925|0.902|+0.67
Can be Eliminated|27|14|13|0.919|0.909|+0.31
Mutual Elimination|10|6|4|0.926|0.905|+0.63

(Unlike the above table, these results form do not form a partition. Mutual Elimination games are included in all three "elimination" categories)

Brodeur's pattern is reasonably consistent with the overall data; if anything, he and the Devils may have an ability to get that final knockout punch when an opponent is down and (almost) out.

At the moment, I'm not sure why my game logs for Brodeur total 0.919 where his real total is 0.920 - if I find something, I'll report back (and edit as appropriate).

Here are Marty's mutual-elimination games:

Game | Result | Score | Shots | Exp Saves | Actual Saves
4/29/1994 vs. Buffalo|W|2-1|18|16.0|17
5/27/1994 at NY Rangers|L|1-2 2OT|48|43.0|46
5/4/1999 vs. Pittsburgh|L|2-4|13|11.5|9
5/26/2000 at Philadelphia|W|2-1|27|24.6|26
5/9/2001 vs. Toronto|W|5-1|16|14.4|15
6/9/2001 at Colorado|L|1-3|22|19.6|19
5/23/2003 at Ottawa|W|3-2|26|23.4|24
6/9/2003 vs. Anaheim|W|3-0|24|21.8|24
4/28/2009 vs. Carolina|L|3-4|31|28.4|27
4/26/2012 at Florida|W|3-2 2OT|45|41.5|43

Some famous losses there, but overall pretty decent (again, the aggregate is consistent with the overall pattern).

 

[Doctor No]

Anyhow, thoughts welcome (including suggestions for which goaltenders to report on next).

I'd like to do Hasek and Quick, and certainly Fuhr (although I'd be missing his 1982 efforts for the moment).

Also, this is a plug for pointing out any data errors that you might see (either here or directly in my game logs). Accuracy is important to me, but there's a lot of numbers in my database .

 

[Doctor No]

Dominik Hasek is an interesting case, I think:

Situation | GP | W | L | Actual SV% | Expected SV% | S+/30
All Games|119|65|49|0.925|0.901|+0.71
Can Eliminate|18|13|5|0.937|0.900|+1.10
Can be Eliminated|13|5|8|0.932|0.899|+1.01
Mutual Elimination|3|1|2|0.946|0.896|+1.50

(Unlike the above table, these results form do not form a partition. Mutual Elimination games are included in all three "elimination" categories)

Now obviously sample size plays a consideration here, but Hasek starts out solid and sure looks a lot better when someone can go home at the end of the night.

Here are Dom's mutual-elimination games:

Game | Result | Score | Shots | Exp Saves | Actual Saves
4/29/1994 at New Jersey|L|1-2|46|41.1|44
5/10/2001 vs. Pittsburgh|L|2-3 OT|28|25.0|25
5/31/2002 vs. Colorado|W|7-0|19|17.3|19

 

[Doctor No]

How about the modern-day clutchmaster, Jonathan Quick:

Situation | GP | W | L | Actual SV% | Expected SV% | S+/30
All Games|76|45|31|0.923|0.913|+0.30
Can Eliminate|18|10|8|0.923|0.913|+0.27
Can be Eliminated|12|9|3|0.937|0.913|+0.70
Mutual Elimination|4|4|0|0.940|0.912|+0.85

(Unlike the above table, these results form do not form a partition. Mutual Elimination games are included in all three "elimination" categories)

Quick (and the Kings) does seem to raise his game when the possibility of cleaning out his locker the following day is a reality. When they have the edge? Not as much.

Here are Jonathan's mutual-elimination games (they will likely look familiar):

Game | Result | Score | Shots | Exp Saves | Actual Saves
5/28/2013 vs. San Jose|W|2-1|26|24.1|25
4/30/2014 at San Jose|W|5-1|40|36.7|39
5/16/2014 at Anaheim|W|6-2|27|24.3|25
6/1/2014 at Chicago|W|5-4 OT|41|37.1|37

 

[Doctor No]

Okay, I'm doing the study, and my favorite goaltender is Kirk McLean. So we're doing Kirk McLean. Plus, it's enlightening:

Situation | GP | W | L | Actual SV% | Expected SV% | S+/30
All Games|68|34|34|0.907|0.886|+0.62
Can Eliminate|10|6|4|0.909|0.887|+0.66
Can be Eliminated|16|11|5|0.923|0.887|+1.06
Mutual Elimination|5|3|2|0.937|0.887|+1.51

(Unlike the above table, these results form do not form a partition. Mutual Elimination games are included in all three "elimination" categories)

Wow - go get 'em, Captain Kirk! When you remember that McLean is customarily considered a run-of-the-mill starter, his overall playoff numbers (compare to the first table) are very good. And when the Canucks can be eliminated, wow. Those save percentages would be considered good today, let alone when the expectation was 88.7%.

Here are Kirk's mutual-elimination games:

Game | Result | Score | Shots | Exp Saves | Actual Saves
4/15/1989 at Calgary|L|3-4 OT|46|40.4|42
4/30/1992 vs. Winnipeg|W|5-0|33|29.6|33
4/30/1994 at Calgary|W|4-3 2OT|49|43.6|46
6/14/1994 at NY Rangers|L|2-3|35|31.4|32
5/19/1995 at St. Louis|W|5-3|44|38.7|41

Spot this game (and this save) on the list above:

 

[Raccoon Jesus]

This is really freaking cool CB.

 

[VanIslander]

Irbe

 

[DougieSmash]

Cam Ward never lost a game #7 in his career - with Red Deer or Carolina. He's pretty darn good at those kind of games.

 

[Doctor No]

This is really freaking cool CB.

Thanks! It's fun having the data available at this point - all of those years in the microfilm room have finally paid off.

I'll do Irbe and Ward (and Lundqvist) next.

 

[Doctor No]

One of my favorites, largely for his hand-repaired equipment, Arturs Irbe:

Situation | GP | W | L | Actual SV% | Expected SV% | S+/30
All Games|51|23|27|0.902|0.897|+0.15
Can Eliminate|7|3|4|0.899|0.895|+0.14
Can be Eliminated|7|3|4|0.910|0.893|+0.48
Mutual Elimination|2|1|1|0.882|0.888|-0.16

(Unlike the above table, these results form do not form a partition. Mutual Elimination games are included in all three "elimination" categories)

First, Irbe's overall save percentage (0.902) isn't preserved in my game logs (0.900) - I'll dig into that when able.

The elimination subsets are pretty small here, so I don't read too much into them.

Here are Arturs' mutual-elimination games:

Game | Result | Score | Shots | Exp Saves | Actual Saves
4/30/1994 at Detroit|W|3-2|38|26.5|28
5/14/1994 at Toronto|L|2-4|21|18.8|17

 

[Doctor No]

Cam Ward looks pretty good when battling back:

Situation | GP | W | L | Actual SV% | Expected SV% | S+/30
All Games|41|23|18|0.917|0.902|+0.46
Can Eliminate|12|6|6|0.904|0.900|+0.11
Can be Eliminated|6|5|1|0.935|0.904|+0.95
Mutual Elimination|4|4|0|0.932|0.902|+0.90

(Unlike the above table, these results form do not form a partition. Mutual Elimination games are included in all three "elimination" categories)

The "Can Be Eliminated" and the "Mutual Elimination" row are basically the same (and both look really good); that's mainly because the four mutual-elimination games are the majority of the "can be eliminated" set.

Related, the mutual elimination games are also one-third of the "can eliminate" set (where Ward looks terrible) - so imagine how bad the stats there look when the mutual elimination games are removed.

Here are Cam's mutual-elimination games (as noted above, he's undefeated):

Game | Result | Score | Shots | Exp Saves | Actual Saves
6/1/2006 vs. Buffalo|W|4-2|24|21.4|22
6/19/2006 vs. Edmonton|W|3-1|23|20.7|22
4/28/2009 at New Jersey|W|4-3|35|32.1|32
5/14/2009 at Boston|W|3-2 OT|36|32.3|34

 

[Doctor No]

Last but not least (for tonight), Henrik Lundqvist just might be the "king" of helping his team get back into series:

Situation | GP | W | L | Actual SV% | Expected SV% | S+/30
All Games|92|43|48|0.926|0.909|+0.49
Can Eliminate|14|8|5|0.912|0.909|+0.08
Can be Eliminated|20|12|8|0.958|0.910|+1.44
Mutual Elimination|6|5|1|0.965|0.908|+1.71

(Unlike the above table, these results form do not form a partition. Mutual Elimination games are included in all three "elimination" categories)

(Yes, Lundqvist's overall save percentage from my logs does not match his career totals either).

When the Rangers are down, Lundqvist appears to not be out - the "can be eliminated" results are borderline statistically significant; he goes from +0.49 G+/30 (all playoff games) to +1.44 G+/30 (in games where Lundqvist can be eliminated).

This includes the +1.71 G+/30 in mutual elimination games, listed here:

Game | Result | Score | Shots | Exp Saves | Actual Saves
4/28/2009 at Washington|L|1-2|24|21.8|22
4/26/2012 vs. Ottawa|W|2-1|27|24.7|26
5/12/2012 vs. Washington|W|2-1|23|20.9|22
5/13/2013 at Washington|W|5-0|35|31.6|35
4/30/2014 vs. Philadelphia|W|2-1|27|24.6|26
5/13/2014 at Pittsburgh|W|2-1|36|32.7|35

 

[DTMAboutHeart]

I know this shouldn't have to be said but, it's really hard to draw any conclusive conclusions when dealing with such tiny sample sizes.

Still a cool exercise though

 

[West]

Couple years ago I did something where I took players from Top Career points and just did the point totals from there best 8yrs (career prime). The idea was to weed out players like Mike Gartner who had exceptionally long careers but would never be considered best player in the league and do a reasonable comparision between players like Lemieux and Gretzky or Orr and Coffey.

Does limiting players to there prime/best years noticeably affect the data?

 

[Doctor No]

Agreed on sample size - one of my goals was (is) to see if any players' samples were statistically significant even with the smaller set of data. Spoiler: while some are on the border, none are obviously over the threshold.

I like the idea of looking at subsets of players in their primes, although that exacerbates the above issue.

One alternative theory I've heard a lot is that older goaltenders may lose some of their overall ability, but still "know how to win" in clutch games.

I think that the best way to look at both of these ("prime goaltenders" and "old goaltenders") would be by looking at age cohorts in aggregate - I'll post that a bit down the line.

 

[yanko6]

Do Tim Howard.

 

[VanIslander]

Olaf "Godzilla" Kolzig!

 

[charlie1]

This is great! Thanks for sharing.

I'm surprised that the results are not statistically significant. Are they still non-significant if you bin all the elimination games together and ask whether the goaltender is better in elimination vs. non-elimination? I'm just thinking that the sub-categories really cut down on the sample size so if you just make two categories (elimination and non-elimination) you might find that the results are significant. Maybe you have already tried this? Also, what are you using as your significance test? Z-score with binomial distribution for the variance?

 

[stahl]

How about Ed Belfour?

 

[Luongofan558*]

Please,Luongo is the most clutch goalie.

 

[Patchey*]

Is there a chart for least clutch?