Monday, March 2, 2009

Shots Against Effects

I thought of another way to break down shots against and save percentage - by looking at goalies with similar league rankings from year-to-year to see if goalies who faced fewer than average or higher than average shots were likely to have a save percentage that differed from the norm. Even if there is no relationship in an overall sense, there could still be some relationship for goalies with extreme shots against results, whether high or low. After all, when we are comparing say Turco in Dallas to Vokoun in Florida, we are talking about the opposite ends of the spectrum in terms of team prevention.

I took a sample of the top 40 goalies in minutes played for each season from 2000-01 to 2008-09, and then looked at the save percentage stats and rank for all the goalies who ranked #1, #5, #10, #15, #20, #25, #30, #35, and #40 in shots against per 60 minutes each season, where #1 was the fewest and #40 was the most.

Here is the breakdown:

1st in SA: .908 save %, avg rank 21
5th in SA: .908 save %, avg rank 24
10th in SA: .909 save %, avg rank 21
15th in SA: .903 save %, avg rank 27
20th in SA: .898 save %, avg rank 30
25th in SA: .911 save %, avg rank 17
30th in SA: .909 save %, avg rank 19
35th in SA: .915 save %, avg rank 13
40th in SA: .916 save %, avg rank 13

Those results are pretty consistent across the board. The only ones that are really striking are the low results for #15 and #20, which wouldn't be expected by any commonly advanced shots against theory, and the high results for #35 and #40, which could potentially indicate an advantage to facing higher shots against totals.

The most obvious factor that can be influencing this analysis is goalie quality. Since any goalie can finish in any one of these spots, one grouping could be Brodeur, Roy, Belfour and Hasek while the next could be Cloutier, Raycroft, Esche and Kidd. The actual groupings weren't that extreme, but goalie quality does explain the weird results for the goalies who ranked 20th, as Vernon, Turek, Grahame, Conklin, Khabibulin, Tellqvist, Ward and Toskala was easily the weakest group of goalies. The goalies who ranked 15th weren't much better (Thibault, Roy, Storr, Weekes, post-lockout Theodore, Ward, Lalime, Price this year). It appears to be just a fluke of a fairly small sample size that those two ranks attracted lesser talent and led to lower save percentages. On the other hand, both the 35th ranked and 40th ranked goalie groups were strong, with goalies like Burke, pre-lockout Theodore, Roloson, Huet, Lehtonen, and Luongo three times, along with a few lesser lights in Rhodes, Hnilicka, Denis, Hedberg, Anderson and Chris Mason.

Here are the numbers adjusted for the average career save percentage for each goalie in the sample and normalized to the league average of .907. This works pretty well for most of the goalies, with the exception of those who have played just 1 or 2 seasons, like Carey Price, or goalies whose careers began in the 1980s, like Patrick Roy.

#1: .901
#5: .910
#10: .911
#15: .903
#20: .902
#25: .908
#30: .911
#35: .914
#40: .913

This list makes it look like it was very difficult to face the fewest shots in the league, but that is not really the case. The adjustment is somewhat misleading, as the average career save percentage is overstating the strength of the group. Dominik Hasek faced the fewest shots against in the league in both 2006-07 and 2007-08, so his .922 career mark is counted twice by this method although he was well past his prime in those two seasons. This year, current fewest shots against leader Steve Mason counts as .920, since as a rookie this season is all we have to go on. Mason may be a good young goalie, but he is unlikely to sustain that mark over an entire career. If we replace these adjustments with average values, then the #1 group improves to a .908 adjusted save percentage, only slightly lower than the others.

The two highest-shot groups finish 1-2 in the adjusted rankings as well, which does seem to indicate that there may be some advantage at work. However, the main reason for these excellent results are that they included the two best high-workload, high-save percentage seasons in the period (Jose Theodore in 2001-02 and Roberto Luongo in 2003-04). Remove those two years, and both groups end up with an adjusted save percentage of .911, which is in line with everything else.

There might be a slight save percentage advantage to facing high shot totals and possibly even a slight disadvantage to facing low shot totals, but it is probably at most a couple thousandths. Otherwise, over an 8 season sample it did not appear to make much difference at all whether goalies faced fewer shots than average or more shots than average. The results were certainly within the margin of error given the sample size, or could be explained by differences in goalie quality. The evidence continues to mount that there appears to be no significant relationship between save percentage and shots against.

9 comments:

Cornelius Hardenbergh said...

Some teams will give up a bunch of shots, but only low-quality shots, this could lead to the higher rating of people with higher shot totals...especially if the official counting shots wants to pad their stats.

FatMan said...

Great work as usual, CG. In the interest of eliminating noise, though, have you thought of doing something like, instead of the doing it the way you have shown, putting the goalies in larger groups, such as 1-5 in shots faced, 6-10, etc., which could help eliminate the problems of having group with overly strong or weak goaltending in any individual group.

Bruce said...

1st in SA: .908 save %, avg rank 21
5th in SA: .908 save %, avg rank 24
10th in SA: .909 save %, avg rank 21
15th in SA: .903 save %, avg rank 27
20th in SA: .898 save %, avg rank 30
25th in SA: .911 save %, avg rank 17
30th in SA: .909 save %, avg rank 19
35th in SA: .915 save %, avg rank 13
40th in SA: .916 save %, avg rank 13
The evidence continues to mount that there appears to be no significant relationship between save percentage and shots against.


Say what? The guys in positions 1, 5, 10, 15, and 20 each have an average rank outside the top 20, and the guys ranked 25, 30, 35, and 40 all have an average rank inside the top 20, (with 35 and 40 being the only ones inside the top 15), and this is evidence for NO significant relationship?

Interesting conclusion.

The Contrarian Goaltender said...

Bruce:

Did you read the rest of the post? Obviously the raw data suggest that it is easier to face more shots, as I pointed out. I feel most of the discrepancy is explained by goalie quality and the impact of outlier seasons by Theodore and Luongo. I'm certainly open to counterarguments, however.

I think it is possible there is a small effect that makes it more difficult for low shot goalies and less difficult for high shot goalies, but the evidence here does not provide significant proof. I also think that the reason this may be the case is the effect of blocked shots, rather than some fundamental relationship between number of scoring chances and goals against.

"Inside the top 20" means little to me. I don't see why it is significant whether something falls inside an arbitrary boundary defined by a round number. The difference between 21 and 19 is the same as the difference between 23 and 21. Especially when the actual save percentage difference is .001-.002.

Bruce said...

Did you read the rest of the post?

Not well enough. For one thing I think I read the nature of your adjustment for the #1's (Hasek * 2, S.Mason) backwards.

Obviously the raw data suggest that it is easier to face more shots, as I pointed out.

Obviously.

I feel most of the discrepancy is explained by goalie quality and the impact of outlier seasons by Theodore and Luongo. I'm certainly open to counterarguments, however.

How about that those outlier seasons were in part made possible due to the high numbers of shots faced by Jose the Hirsute and BobbiLu. Not sure which is chicken and which is egg, a question which warrants at least a passing mention.

I think it is possible there is a small effect that makes it more difficult for low shot goalies and less difficult for high shot goalies, but the evidence here does not provide significant proof.

Depends on what you mean by "significant" I guess. To me the fact that the five samples of goalies who finished in the Top 20 of fewest shots against ALL finished OUTside the Top 20 in Sv%, and that the four samples of goalies who finished outside the Top 20 of fewest shots against ALL finished INside the Top 20 in Sv% is significant.

[Insert hand-waving analysis here]

I also think that the reason this may be the case is the effect of blocked shots, rather than some fundamental relationship between number of scoring chances and goals against.

As in, a shot blocking team reduces shots against, but the ones that do get through are a little more likely to go in? I can go along with that.

"Inside the top 20" means little to me. I don't see why it is significant whether something falls inside an arbitrary boundary defined by a round number. The difference between 21 and 19 is the same as the difference between 23 and 21.

Right, of course. What's significant is the word "ALL" in my paragraph above. Small differences in some cases, but consistently favouring the high shots against guys when it comes to posting a better Sv%.

Especially when the actual save percentage difference is .001-.002.

We are always talking about tiny difference when it comes to Save Percentages. It seems like they are significant or insignificant depending on whether they work in favour or against whatever case is being made. :)

Don't know how much work is involved, but I do think FatMan has a good suggestion with the five-man groupings. Of course that introduces "arbitrary boundaries" between #5 and 6, 10 and 11, etc. But a denser set of numbers should reduce the effect of those outlier seasons.

Anonymous said...

Take a look at save percentage vs games played.... You always say how can Brodeur be the best goalie of all time if he has never led the league in save percentage... Well, since some goalies are dealing with a smaller sample size, it only makes sense that the variance from the mean will extend farther than a goalie who has played almost twice as many games. I see this in the case every year, whether its Kipper, Luongo, or Brodeur. Always guys with 30 or 40 something games are up at the top...

The Contrarian Goaltender said...

Anonymous: I agree with you on the effects of variance on seasonal save percentage, but to have Dan Ellis types lead the league in save percentage is a fairly recent phenomenon. Look at the history of save percentage leaders (available here). The typical winner is a workhorse starter.

I think this is evidence of the depth of goalie competition out there today. There isn't a Hasek or a Roy that is dominating everyone else, and there are more good goalies than spots to put them in. That is why I keep saying that goalies are mostly interchangeable, and it is usually the rest of the team that determines the result.

Note also the period from 1993-94 to 1998-99, which is why this site is so vocal in advocating that particular goalie as the best all-time.

Anonymous said...

Look at goalies like Turco, Nabakov, Lundqvist and Kipprasoff. All those guys had decent save percentages, if not spectacular ones, when playing around 50 games a year. Every one of them since jumping to a 70+ a year guy has seen dropoffs in save percentage numbers.

The Contrarian Goaltender said...

Lundqvist hasn't dropped off that much in the last few seasons, and Nabokov has posted good save percentages in high workload seasons earlier in his career. I agree that Turco and Kiprusoff have declined, but I don't think that is necessarily related to usage.

Researching the effect of playing a lot of games is on my to-do list. I'm holding off judgment until I see a comprehensive look, because it is too easy to cherry-pick examples one way or the other.