Well there is some correlation. The question is how much. It can't be completely written off. Take my last post and tell me how it's false, or not related.
1992-1993..Professional scouts rate the NHL players
- Thread starter Habsfan18
- Start date
Also, most of the highest shooting percentages come from players who don't take a whole lot of shots.
BraveCanadian
Registered User
- Jun 30, 2010
- 14,705
- 3,573
This is for this last season. I have only included the goalies who played the entire game. Otherwise we would get a bias since goalies who let in the first couple of goals tend to get pulled and have horrendous save percentage.
Shots Against | Goals Against | Frequency | Save%
12 | 1.00 | 4 | 91.67
13 | 2.00 | 3 | 84.62
14 | 2.00 | 7 | 85.71
15 | 2.00 | 5 | 86.67
16 | 1.50 | 12 | 90.63
17 | 2.05 | 20 | 87.94
18 | 2.00 | 16 | 88.89
19 | 2.32 | 28 | 87.78
20 | 2.44 | 36 | 87.78
21 | 2.26 | 38 | 89.22
22 | 2.26 | 73 | 89.73
23 | 2.17 | 88 | 90.56
24 | 2.44 | 106 | 89.82
25 | 2.42 | 120 | 90.30
26 | 2.32 | 139 | 91.09
27 | 2.28 | 120 | 91.57
28 | 2.59 | 141 | 90.75
29 | 2.24 | 145 | 92.27
30 | 2.31 | 121 | 92.31
31 | 2.44 | 128 | 92.14
32 | 2.31 | 108 | 92.80
33 | 2.67 | 113 | 91.90
34 | 2.98 | 107 | 91.23
35 | 2.42 | 104 | 93.08
36 | 2.66 | 87 | 92.62
37 | 2.33 | 66 | 93.69
38 | 2.78 | 55 | 92.68
39 | 2.77 | 61 | 92.90
40 | 2.43 | 44 | 93.92
41 | 2.66 | 53 | 93.51
42 | 2.82 | 22 | 93.29
43 | 2.45 | 20 | 94.30
44 | 2.71 | 17 | 93.85
45 | 2.63 | 19 | 94.15
46 | 2.38 | 8 | 94.84
47 | 3.20 | 10 | 93.19
48 | 3.20 | 5 | 93.33
49 | 4.33 | 3 | 91.16
50 | 2.75 | 4 | 94.50
51 | 1.50 | 2 | 97.06
52 | 7.00 | 1 | 86.54
55 | 3.00 | 1 | 94.55
And this is how it looks graphically where the size of the markers are proportional to the number of occurences.
seventieslord
Student Of The Game
Those are extremely poor sample sizes. Save your smileys for if the resjlts agree with you when alanlyzed properly.
overpass
Registered User
- Jun 7, 2007
- 5,271
- 2,807
Was the original claim of a SOG - SV% relationship about the game level or the season level? Because it's definitely true on the game level. Matnor's numbers prove that without a doubt.
On the other hand, the relationship is much weaker on a season level, and might be non-existent after correcting for arena recording bias.
seventieslord
Student Of The Game
Right. And since season level is what everyone talks about, it makes sense that it should be what is used in any discussion of correlation.
seventieslord
Student Of The Game
Unfortunately, I could not separate the goalies' TOI by situation - I thought NHL.com had that, but it doesn't. So I just used raw sv%. I only used goalies with a reasonable sample size - the top-45 in minutes played.
The correllation between shots against per minute and total sv% is 0.286. Weak.
The correllation between shots against per minute and total sv% is 0.286. Weak.
BraveCanadian
Registered User
- Jun 30, 2010
- 14,705
- 3,573
I think most rules of thumb put .30 as Moderate so there is most likely something there..
More specifically it depends on how much you believe save percentage to be due to chance.
If you believe that the results are prone to chance then the significance is weaker. If you believe that save percentage results are not very prone to chance results then the significance could be stronger.
But I'm getting way too far removed from my stats classes though so..
I don't want to hijack the thread, especially since this has been discussed before but I should just give a few comments.
Efficiency is not a problem here. I did not break it down to one-game samples. What I did was to aggregate the save percentage for each number of shots in a game. For instance, a goalie faced 20 shots 36 times with an average save percentage of 87.78.
You are right that it would be important to take special team effects into account. However, as you say, we would expect the bias to go in a downward direction. That is, a goalie that is often on the penalty kill will face more shots in total that are also of higher quality.
What do you mean poor sample sizes?
Yes, my claim was only on a game-by-game basis. It is theoretically possible that the relationship disappears when aggregating. For instance, take Brodeur as an example. It could be the case that on a game-by-game basis he has better save percentage when he faces more shots. On the other hand, he has historically played on a defensive-minded team which may have lowered both the shot quality and the number of shots he faced. So, when comparing him to another goalie he would have both better save percentage and having faced fewer shots. In that way, the aggregate effect could be different.
I tested this using aggregate data (save percentage by season and goalie) from 83/84 to 07/08 for goalies who played at least 40 games and regressed the save percentage on the average number of shots per game faced by each goalie and season. And the result is statistically significant on any conventional significance level. On average, each extra shot per game results in a higher save percentage by almost 0.1 percentage point. That is around half the effect found on the individual level (where an extra shot per game leads to a higher save percentage with around 0.2 percentage point).
To visualize the result I split the goalies into five groups for each season sorted by the number of shots they faced on average. I then aggregated the save percentage for each group and this is the result:
Group|SA/GP|Save%
1 | 25.3 | 89.85%
2 | 27.28 | 89.96%
3 | 28.60 | 90.07%
4 | 30.17 | 90.05%
5 | 32.45 | 90.27%
As can be seen, the difference in save percentage between the fifth of the goalies with the least shots against and the fifth of the goalies with the most shots against are around 0.4 percentage points. So, I'm not argue that it's a big difference but the effect is there.
You are right that arena-recording bias could confound the result. To test that one would have to aggregate game-by-game data after removing the arena effect. Unfortunately, it would require some data work that I'm not sure I have the time for. I might check that later (or you might want to do it
)
I guess it was my fault for using the term correlation. A correlation coefficient (or R-squared for that matter) is only useful if my claim would have been that the number of shots a goalie faces is the dominant determinant of save percentage. Of course that is not my point, a goalie's skill is much more important. That being said, it may still be worth taking into account. Specifically, two important questions need to be adressed. First, is the effect statistically significant (and it is, though we may worry about confounding factors)? Second, is the effect is of significant size? That is, is it important to take into account that a goalie who faces one additional shot have a lower save percentage with 0.1 percentage point? If the answer is yes to both these questions than the fact that the correlation coefficient is small is pointless.
All in all, sorry for derailing the thread. I think this is interesting and is something that is worth studying more. My conclusion is that there is a clear positive correlation on a game-by-game basis and that it looks as if this is also true on a goalie-and-season level though there are a number of unresolved issues that needs to be addressed. I guess my main point is that I sometimes here arguments such that "not only did he have an outstanding save percentage, he had it while facing the most shots in the league" and I think that is a problematic statement.
Efficiency is not a problem here. I did not break it down to one-game samples. What I did was to aggregate the save percentage for each number of shots in a game. For instance, a goalie faced 20 shots 36 times with an average save percentage of 87.78.
You are right that it would be important to take special team effects into account. However, as you say, we would expect the bias to go in a downward direction. That is, a goalie that is often on the penalty kill will face more shots in total that are also of higher quality.
What do you mean poor sample sizes?
Yes, my claim was only on a game-by-game basis. It is theoretically possible that the relationship disappears when aggregating. For instance, take Brodeur as an example. It could be the case that on a game-by-game basis he has better save percentage when he faces more shots. On the other hand, he has historically played on a defensive-minded team which may have lowered both the shot quality and the number of shots he faced. So, when comparing him to another goalie he would have both better save percentage and having faced fewer shots. In that way, the aggregate effect could be different.
I tested this using aggregate data (save percentage by season and goalie) from 83/84 to 07/08 for goalies who played at least 40 games and regressed the save percentage on the average number of shots per game faced by each goalie and season. And the result is statistically significant on any conventional significance level. On average, each extra shot per game results in a higher save percentage by almost 0.1 percentage point. That is around half the effect found on the individual level (where an extra shot per game leads to a higher save percentage with around 0.2 percentage point).
To visualize the result I split the goalies into five groups for each season sorted by the number of shots they faced on average. I then aggregated the save percentage for each group and this is the result:
1 | 25.3 | 89.85%
2 | 27.28 | 89.96%
3 | 28.60 | 90.07%
4 | 30.17 | 90.05%
5 | 32.45 | 90.27%
As can be seen, the difference in save percentage between the fifth of the goalies with the least shots against and the fifth of the goalies with the most shots against are around 0.4 percentage points. So, I'm not argue that it's a big difference but the effect is there.
You are right that arena-recording bias could confound the result. To test that one would have to aggregate game-by-game data after removing the arena effect. Unfortunately, it would require some data work that I'm not sure I have the time for. I might check that later (or you might want to do it
)I guess it was my fault for using the term correlation. A correlation coefficient (or R-squared for that matter) is only useful if my claim would have been that the number of shots a goalie faces is the dominant determinant of save percentage. Of course that is not my point, a goalie's skill is much more important. That being said, it may still be worth taking into account. Specifically, two important questions need to be adressed. First, is the effect statistically significant (and it is, though we may worry about confounding factors)? Second, is the effect is of significant size? That is, is it important to take into account that a goalie who faces one additional shot have a lower save percentage with 0.1 percentage point? If the answer is yes to both these questions than the fact that the correlation coefficient is small is pointless.
All in all, sorry for derailing the thread. I think this is interesting and is something that is worth studying more. My conclusion is that there is a clear positive correlation on a game-by-game basis and that it looks as if this is also true on a goalie-and-season level though there are a number of unresolved issues that needs to be addressed. I guess my main point is that I sometimes here arguments such that "not only did he have an outstanding save percentage, he had it while facing the most shots in the league" and I think that is a problematic statement.
seventieslord
Student Of The Game
I mean, single games.
If using single games to analyze sv% trends works, then why not break it down into periods too? I'm being facetious when I say that, of course. But I obviously have much more confidence in results that are based on greater samples of data.
overpass' point about arena bias is important because these would be "phantom" shots, not goals, added to make a sv% appear higher.
also, using only full games confounds the results, too. A goalie who gets bombed for 5 GA after facing 25 shots in half a game (we've seen it happen, right?) would not show up in your study, but that's a case of a goalie who faces 50 shots "per game" and had an 80% sv%. If you think about it, it's not that likely that a goalie will stay in for a full game of 40+ shots if his sv% is very much below 90%. He'd probably get pulled. So of course you're more likely to capture instances of 0-3 GA in 40+ shot games doing it that way.
Aggregation level has to do with what one wants to analyze (game-by-game effects) or (season-goalie effects) but it has nothing with precision as long as the sample is the same. I know it may sound unintuitive but aggregating or disaggregating does not do anything for statistical efficiency per se.
Yes
That's a good point so I tried using shots against per 60 minutes instead but requiring at least 10 minutes played. In that case the effect is still there but much smaller (0.07 instead of 0.19). Setting at least 30 minutes as the threshold the effect is close to the one above (0.17 instead of 0.19). I haven't thought about it too much, and this way estimates something slightly different, but I think your point may be quite important.
It’s funny, I was going through this THN issue this evening and was about to make a thread before finding this one that I posted over a decade ago!
MarkusKetterer
Shoulda got one game in
I’m curious. When it listed the players per team, was there a cut off? Like say, no more than 25 players (to account for GP and/or injuries to players)?
Ad
Upcoming events
-
-
-
-
-
Game 3 Florida Panthers @ Tampa Bay Lightning - Florida Panthers leads series 2-0Wagers: 1Staked: $100.00Event closes- Updated: