A Tale of Two Talents: One True and One Not
Some of you may have seen this plot last week in my look at the persistence of shooting percentage. I looked at whether some players exhibited an ability to score more often from a given spot on the ice than other players. This "talent" regressed 75% to the mean, and a high level of skill was really only identifiable in Ilya Kovalchuk and Alex Tanguay:
Compare that to a "legitimate" talent - getting to a good spot on the ice and taking a shot from there. A player's ability to get to his shooting locations regresses just 11% to the mean - in other words, a player controls where he gets opportunities from, and it is an inherent part of his play, unchanged from game-to-game:
Note that I have captured shot location not as distance from the net, but as expected shooting percentage for an average shooter based on shot distance and angle.
So which forwards put themselves into the best and worst average shooting locations - initial shots at even-strength only, no rebounds:
| Best | Exp. Sh% | Worst | Exp. Sh% |
| Crosby | 7.57 | Kotalik | 4.51 |
| Stastny | 7.59 | Rolston | 4.70 |
| Neil | 7.59 | B. Richards | 4.71 |
| Briere | 7.62 | Samuelsson | 4.72 |
| J. Staal | 7.67 | Vrbata | 4.73 |
| H. Sedin | 7.82 | Steen | 4.87 |
| Frolov | 7.93 | Modano | 4.97 |
| S. Koivu | 8.21 | Langenbrunner | 4.98 |
| Holmstrom | 8.55 | Stillman | 5.02 |
| Brunette | 8.74 | Savard | 5.04 |
And my favorite comparison - how often can you do a sports experiment with identical twins?
| Player | Shots | Exp Sh% |
| H. Sedin | 591 | 7.82 |
| D. Sedin | 1052 | 7.03 |
Daniel Sedin produces a ridiculously high shot volume compared to his brother, and suffers only minimally in terms of where he takes his shots.
You may have noticed that the spread between the highest and lowest shooting percentages is quite small - much smaller than the spread in actual shooting percentage, indicating that shot location accounts for just 27% of shooting percentage. If we take all of these results together, we can make the following claim about the components of shooting percentage:
- Ability to get to a location to take a shot = 24%
- Transient ability in getting to that location = 3%
- Shooting talent (ie - ability to exceed average shooter performance) = 19%
- Transient ability to score = 54%
I think I have this right - and it strikes me as somewhat of an unintuitive result. We know that a lot of what appears to be shooting ability is transient - see Henrik Sedin's 4/54 streak to finish the season if you don't believe me. But the way a forward plays - the spots he gets to to take his shots - is barely more important to his shooting percentage than his shooting ability even though it is very consistent even over the course of five seasons. So Tomas Holmstrom's ability to get position in front of the goal isn't much different from Ilya Kovalchuk's ability to pick the top corner. (Of course, we've only considered initial shots here, so Holmstrom's rebound ability is uncaptured.) The bottom line is that a player doesn't even control 50% of his shooting percentage in the long-term.
23 comments
|
0 recs |
Do you like this story?
Comments
Great article. I’ve got something coming up in a similar vein, and my conclusions are close to yours: shot location and shooting ability are of the same magnitude, but one has much more noise than the other. However, I’m not 100% certain of your last statement. How many odd/even games did your analysis represent? Is this really “long-term”? Did you take into account that you split the sample size by half by dividing even/odd?
I’m not trying to nitpick, but this could be one of those important posts that will be linked to for years to come, so I want to make sure we get the conclusions right!
Five seasons of data, so for most of the players in the sample, there were 170+ games in the two buckets. That’s the longest period of time that I can imagine talent being constant for, and perhaps the time frame is actually too long to make that assumption. Like I said, I think I have this right, but I’m open to a different conclusion.
It represents all uncaptured performance related to scoring. If you want to say that it’s ‘making your shots’ or ‘gripping the stick a little tighter’, that’s your prerogative…
The Avs
Let Andrew Brunette go and signed Darcy tucker 2 years ago
/still drives me insane.
The New Improved Avalanche. Now with Real Coaches!
Jibblescribbits: C'mon over and waste some time
by Jibblescribbits on May 24, 2010 12:03 PM EDT reply actions
The 4/54 streak
I don’t quite understand what it means. So Henrik’s expected shooting percentage given shot location is 7.82%, and his 4/54 streak translate toactual shooting percentage at 7.4%. Those are pretty close, no?
Except Henrik shot 20% for most of the season and people were saying it was because he finalyl learned how to score.
(I also mixed data here: 4/54 does not include missed shots and includes PP; his 7.82% is EV only, includes missed shots and excludes rebounds.)
Got it. Definitely I don’t agree with the argument that Henrik Sedin is a very good goal scorer just from watching him play all these years. His playmaking and passing, however, is probably near the top of the league. But I don’t think we have data on passing other than assists, which is not a very accurate measure. It would be great if not long shot locations are known, but also the player who passed to puck to those locations, or whether the shooter carried the puck there or the opponent team’s mistakes resulted in the shots.
Does it change the results much if you include rebounds? In the case of Holmstrom in particular, getting into a position to pot rebound goals is as much as part of his game as driving to the net.
I’m curious as to how you get from this:
This “talent” regressed 75% to the mean, and a high level of skill was really only identifiable in Ilya Kovalchuk and Alex Tanguay
to this:
But the way a forward plays – the spots he gets to to take his shots – is barely more important to his shooting percentage than his shooting ability
It just seems odd. It seems especially odd to me that 19% of shooting percentage is shooting talent and yet we can only identify two players with a high level of talent. Shouldn’t we see more players who vary from the mean consistently if this is the case? Further, what would the results look like if Kovalchuk had decided to play out his career in Russia? Does that 19% drop like a rock?
I read this post three times at work and I’m concluding…Kovalchuk really is in a class of his own either by uncaptured performance or by incredibly luck? Maybe that?
Devils in my heart! Devils in my mind! Devils in my eyes! Devils until I die!
In Lou We Trust - The New Jersey Devils SBN Blog
by John Fischer on May 24, 2010 5:06 PM EDT up reply actions
Kovalchuk doesn’t change the R^2 at all.
Essentially, as Vic and Likens and many others have shown, shot location (aka quality) occupies a very small slice of goal scoring. So the non-location based part is the amount by which you over- or under-perform the locations you shoot from. We know that’s huge – so many things are wrapped up in it – goaltending, shooting talent, defensive positioning, screens, rushes, whether the shot came off a turnover. A small part of it is shooting talent…But a small part of a big part is as big as a big part of a small part. That’s the only way I can think of it.
I’m surprised, obviously.
I guess the surprising thing to me is that we’re unable to identify examples of players with shooting talent. If it really is 19% of shooting percentage, that seems like a big chunk, big enough that we should be able to find more examples of clearly above average players. And yet… it seems like we just plain old can’t. Do you think the primary cause of that is sample size, or is it something else entirely?
by Scott Reynolds on May 24, 2010 7:05 PM EDT up reply actions
I think you can’t discount the population. There were 215 players in my sample over five seasons. So we’re saying that there’s not a lot of guys who stand out in terms of shooting talent – from a given spot on the ice – among regular top-6 forwards. Everybody’s trying to put the puck in the net – do we really expect to see a lot of difference there? You’d think if players were bad shooters, they wouldn’t be allowed to take 200+ EV shots a year anymore.
For shot location, on the other hand, among regular shooters, we have a mix of guys who shoot from the wing and guys who mix it up in front of the net, which is a deterministic and controllable difference.
I thought your sample was 500 or more shots since the lockout? That’s more like 100 or more EV non-rebound shots per season and most guys don’t get half of their EV shots on rebounds. That said, you’re right that we are capturing data from only the best shooters in the league over this time period. It would be interesting to include a “chaff” group in the analysis where we take all of the data from any given team and combine it into one “player” for all of the guys who don’t meet the qualification requirements. So for the Oilers, take all of the guys who don’t qualify and combine their results to get “Oiler replacement.” Do the same for each club and see what we get. That might give us a baseline for shooting talent.
by Scott Reynolds on May 25, 2010 9:51 AM EDT up reply actions
Interesting idea. I’m sure I will re-visit this topic in the future.
I think there were a lot of players in the sample who had 200+ EV shots in a single season. Or at least 150. Anyways, my point was just that these guys are allowed to drive the shots on goal for some period of time, so you’d assume they weren’t in the George Parros realm.
very interesting stuff
I would not have guessed Andrew Brunette with 600 mulligans.
I’d really love to see a similar analysis with rebounds included. Also, is it possible to see the full list (as opposed to just top and bottom 10)?
a question i had...
i’d be curious about that worst and best shooting percentage list and the rebounds they generate. i understand that we’re just talking about finding shooting location as a talent and not trying to optimize anything, but i wonder if the worst shooters list demonstrably generates more rebound goals in the 2-5 seconds after their ES shots than the best shooters list.

by 























