On-ice Shooting percentage - luck or skill?

overpass

Registered User
Jun 7, 2007
5,254
2,736
5 vs 4

Again, using data from 2007-08 through 2010-11.

Using the binomial approximation to a normal distribution, I calculated the z-score for each NHL player's on-ice shooting percentage over this time period, where z-score = (On-ice GF - On-Ice SOGF*LgSH%)/(Standard deviation calculated using binomial approximation)

Here are the players with the highest and lowest z-scores:

Player | GF | SOGF | On-ice shooting % | Z-score
MIKEKNUBLE | 124 | 753 | 16.5% | 3.36
ANDREIMARKOV | 134 | 825 | 16.2% | 3.32
KIMMOTIMONEN | 157 | 1012 | 15.5% | 2.97
NICKLASBACKSTROM | 173 | 1129 | 15.3% | 2.94
MIKERICHARDS | 134 | 854 | 15.7% | 2.89
CLAYTONSTONER | 2 | 3 | 66.7% | 2.85
DANHINOTE | 2 | 3 | 66.7% | 2.85
TJHENSICK | 2 | 3 | 66.7% | 2.85
JONATHANTOEWS | 137 | 883 | 15.5% | 2.78
ALEXANDEROVECHKIN | 125 | 805 | 15.5% | 2.66
JOEJENSEN | 1 | 1 | 100.0% | 2.65
JUSTINFALK | 1 | 1 | 100.0% | 2.65
MATTSMABY | 1 | 1 | 100.0% | 2.65
MICHAELBLUNDEN | 1 | 1 | 100.0% | 2.65
MIKEIGGULDEN | 1 | 1 | 100.0% | 2.65
TIMSESTITO | 1 | 1 | 100.0% | 2.65
SEANHILL | 6 | 19 | 31.6% | 2.53
TIMSTAPLETON | 6 | 19 | 31.6% | 2.53
TIMBRENT | 10 | 39 | 25.6% | 2.50
CHRISTIANEHRHOFF | 133 | 876 | 15.2% | 2.47
TOMASHOLMSTROM | 126 | 826 | 15.3% | 2.46
ANDREWBRUNETTE | 126 | 827 | 15.2% | 2.44
RYANKESLER | 124 | 819 | 15.1% | 2.35
ALEXANDERSEMIN | 135 | 899 | 15.0% | 2.35
PATRICKKANE | 138 | 924 | 14.9% | 2.31
BRYANSMOLINSKI | 7 | 27 | 25.9% | 2.13
DENISGREBESHKOV | 47 | 284 | 16.5% | 2.10
JOFFREYLUPUL | 74 | 476 | 15.5% | 2.06
JARRETSTOLL | 108 | 723 | 14.9% | 2.04
PETRSYKORA | 76 | 493 | 15.4% | 2.01

Player | GF | SOGF | On-ice shooting % | Z-score
FRANCOISBEAUCHEMIN | 39 | 495 | 7.9% | -3.07
NICKBOYNTON | 7 | 138 | 5.1% | -2.62
IANWHITE | 45 | 508 | 8.9% | -2.44
DAVIDBOOTH | 53 | 578 | 9.2% | -2.38
BRETTLEBDA | 10 | 159 | 6.3% | -2.35
WOJTEKWOLSKI | 58 | 618 | 9.4% | -2.30
JANHLAVAC | 4 | 89 | 4.5% | -2.27
DAINIUSZUBRUS | 53 | 568 | 9.3% | -2.24
ANSSISALMELA | 4 | 87 | 4.6% | -2.21
CALO'REILLY | 7 | 120 | 5.8% | -2.19
LUCACAPUTI | 1 | 48 | 2.1% | -2.17
JASONGARRISON | 6 | 108 | 5.6% | -2.17
DEREKMEECH | 7 | 119 | 5.9% | -2.17
DERICKBRASSARD | 41 | 450 | 9.1% | -2.14
BRANDONDUBINSKY | 57 | 596 | 9.6% | -2.12
ARRONASHAM | 2 | 59 | 3.4% | -2.11
SCOTTHANNAN | 3 | 71 | 4.2% | -2.10
TORREYMITCHELL | 3 | 71 | 4.2% | -2.10
JAYBOUWMEESTER | 96 | 941 | 10.2% | -2.07
JOHN ODUYA | 29 | 332 | 8.7% | -2.04
TRAVISZAJAC | 94 | 919 | 10.2% | -2.02
FEDORTYUTIN | 89 | 874 | 10.2% | -2.02
CRAIGCONROY | 12 | 164 | 7.3% | -1.99
BENNFERRIERO | 1 | 42 | 2.4% | -1.97
SHEAWEBER | 89 | 865 | 10.3% | -1.91
MICHAELSANTORELLI | 16 | 200 | 8.0% | -1.90
VALTTERIFILPPULA | 46 | 480 | 9.6% | -1.89
NIKLASHJALMARSSON | 0 | 25 | 0.0% | -1.88
BYRONRITCHIE | 2 | 52 | 3.8% | -1.88
KRISRUSSELL | 60 | 605 | 9.9% | -1.87

There is considerably less variation in 5 vs 4 shooting percentage than in 5 vs 5 shooting percentage. We can see that, unlike in 5 vs 5 lists of top and bottom performers, some are obviously present simply because of bad luck or good luck in a very small sample.

The standard deviation of the z-scores is 1.00. This implies that the results are entirely the result of random variation! If I cut out all players who were on the ice for less than 200 shots, to focus on players who had more significant samples (making sure to recalibrate the mean and the standard deviation) I get a standard deviation of the z-scores of 1.06. That's still a very low amount of skill showing up relative to luck.

Lets look at this another way. How did the group of players who received the most power play time compare to other groups?

Players who were on the ice for:

1000+ shots: 13.2 on-ice SH%
800-1000 shots: 13.0 on-ice SH%
600-800 shots: 12.7 on-ice SH%
400-600 shots: 12.2 on-ice SH%
200-400 shots: 12.1 on-ice SH%
0-200 shots: 11.1 on-ice SH%
League average: 12.4 on-ice SH%

Now causality probably runs in both directions here to some degree. Players who are selected to play the most on the PP have the best SH% skill - but also players who are on the ice for high shooting percentages, possibly because of random variation, may be asked to play more on the power play.

Here's the group of players who were on the ice for 1000+ PP shots for:

Player | SOGF | On-ice SH%
MARTINST. LOUIS | 1292 | 12.4%
NICKLASLIDSTROM | 1241 | 14.0%
DANBOYLE | 1235 | 13.3%
ILYAKOVALCHUK | 1216 | 12.1%
MIKEGREEN | 1214 | 14.1%
JOETHORNTON | 1199 | 13.4%
EVGENIMALKIN | 1178 | 13.5%
BRADRICHARDS | 1171 | 12.6%
TOMASKABERLE | 1150 | 11.7%
DIONPHANEUF | 1141 | 12.0%
NICKLASBACKSTROM | 1129 | 15.3%
JAROMEIGINLA | 1111 | 12.7%
DANYHEATLEY | 1109 | 13.4%
HENRIKZETTERBERG | 1106 | 12.3%
PATRICKMARLEAU | 1074 | 13.5%
BRIANRAFALSKI | 1074 | 13.8%
SIDNEYCROSBY | 1072 | 13.2%
CHRISPRONGER | 1071 | 13.5%
VINCENTLECAVALIER | 1069 | 11.8%
RYANGETZLAF | 1069 | 13.6%
HENRIKSEDIN | 1069 | 13.7%
ERICSTAAL | 1054 | 13.5%
COREYPERRY | 1043 | 13.5%
LUBOMIRVISNOVSKY | 1040 | 13.5%
ANZEKOPITAR | 1033 | 11.8%
MIKERIBEIRO | 1033 | 12.1%
DANIELSEDIN | 1027 | 13.3%
TOBIASENSTROM | 1019 | 12.4%
PAVELDATSYUK | 1017 | 14.3%
KIMMOTIMONEN | 1012 | 15.5%

I think that looks more like a group of the most skilled, best PP players in the league, rather than a group of players who got lucky with their shooting percentage and got more playing time as a result. But keep in mind that the numbers posted above may be affected by selection bias to some degree.

In any case, it appears that 5 vs 4 on-ice shooting percentage includes less skill and more luck than 5 vs 5 on-ice shooting percentage - possibly because the set of players who play 5 vs 4 are selected for their skill. Keep in mind that this is not a universal truth for hockey - only for the set of NHL players who played on the PP from 2007-08 to 2010-11.
 

seventieslord

Student Of The Game
Mar 16, 2006
36,080
7,132
Regina, SK
This is a really simple example, but look at the list of the highest shooting percentages in NHL history:

http://www.hockey-reference.com/leaders/shot_pct_season.html

The list is littered with players who played on the same line as all-time great playmakers. Mario Lemieux's famous leaches (Rob Brown and Warren Young) are right near the top. Charlie Simmer was Marcel Dionne's triggerman.

PP sucks will also have a higher shooting percentage over time, due to a higher proportion of their shots being during the PP than most players.
 

AlienWorkShop

No, Ben! No!
Oct 30, 2004
3,455
328
In any case, it appears that 5 vs 4 on-ice shooting percentage includes less skill and more luck than 5 vs 5 on-ice shooting percentage - possibly because the set of players who play 5 vs 4 are selected for their skill. Keep in mind that this is not a universal truth for hockey - only for the set of NHL players who played on the PP from 2007-08 to 2010-11.
Nice work, thanks for running the numbers!

Just a thought, perhaps PP SH% has a greater dependence on the goalie? I don't know if it's ever been tested, but it's a common refrain that goalies are your best penalty killer, so that would suggest less dependence on your linemates' skill and your own playmaking ability, and more on the goaltender's skill for your on-ice SH%.

Have the numbers for 4v5 SV% been run for goaltenders? That could answer that question. The blog you posted was done for even strength.
 

Czech Your Math

I am lizard king
Jan 25, 2006
5,169
303
bohemia
What if one limited the sample to players with at least N shots for on ice while on PP, then compared only those players to each other in shooting% for on ice 5v5 and on PP? I think there's still some sort of bias coming through, but that's just a hunch, very well could be wrong. There's so many players with only mediocre skill 5v5 and PP is limited mostly to those with superior skill.
 
Mar 31, 2005
1,675
10
East Coast
I tried a different (simpler) approach. I ran a regression to see what 'explains' a shooting percentage.

My independent variables were shots, average distance, TOI (are they good), GP (do we have a sample size), and shooting percentage last season. Basically, if shooting percentage was luck then last year should have no effect on this year. I used players in 4 seasons, 2007-2011, that played more than 80 total games in the two consecutive seasons.

Turns out last years success is significant, there was a positive and statistically significant correlation. Distance was very significant, (an extra 5 feet to your average decreases your expected shooting percentage by 1%) I'm thinking of adjusting everyone's shooting percentage for distance and re-running it, but I won't get around to it this weekend.

Feel free to pick the the procedure apart, I had some difficulty decided what to include/exclude.

Parameter Standard
Variable Label DF Estimate Error t Value Pr > |t|

Intercept Intercept 1 8.24130 0.44404 18.56 <.0001
SH_current 1 -0.00486 0.00169 -2.87 0.0042
DIST_current DIST 1 -0.20319 0.00991 -20.51 <.0001
TOI_current TOI/60 1 0.28239 0.03429 8.23 <.0001
GP_current GP 1 0.01383 0.00513 2.70 0.0071
SPCT_lag SPCT 1 0.15235 0.02299 6.63 <.0001
 

Czech Your Math

I am lizard king
Jan 25, 2006
5,169
303
bohemia
I tried a different (simpler) approach. I ran a regression to see what 'explains' a shooting percentage.

My independent variables were shots, average distance, TOI (are they good), GP (do we have a sample size), and shooting percentage last season. Basically, if shooting percentage was luck then last year should have no effect on this year. I used players in 4 seasons, 2007-2011, that played more than 80 total games in the two consecutive seasons.

Turns out last years success is significant, there was a positive and statistically significant correlation. Distance was very significant, (an extra 5 feet to your average decreases your expected shooting percentage by 1%) I'm thinking of adjusting everyone's shooting percentage for distance and re-running it, but I won't get around to it this weekend.

Feel free to pick the the procedure apart, I had some difficulty decided what to include/exclude.

Parameter Standard
Variable Label DF Estimate Error t Value Pr > |t|

Intercept Intercept 1 8.24130 0.44404 18.56 <.0001
SH_current 1 -0.00486 0.00169 -2.87 0.0042
DIST_current DIST 1 -0.20319 0.00991 -20.51 <.0001
TOI_current TOI/60 1 0.28239 0.03429 8.23 <.0001
GP_current GP 1 0.01383 0.00513 2.70 0.0071
SPCT_lag SPCT 1 0.15235 0.02299 6.63 <.0001

It looks like GP & shots are not particularly significant. I don't remember what the thresholds are, but they may not really be adding that much to your model (check the R^2 or whatever stats with & w/o those variables). TOI appears significant, but this may be a flaw in the model, as its more likely Sh% causes TOI than TOI causes Sh%. I would remove TOI for that reason, possibly GP & shots, and maybe add Shots/60 to see how significant that is.
 
Mar 31, 2005
1,675
10
East Coast
It looks like GP & shots are not particularly significant. I don't remember what the thresholds are, but they may not really be adding that much to your model (check the R^2 or whatever stats with & w/o those variables). TOI appears significant, but this may be a flaw in the model, as its more likely Sh% causes TOI than TOI causes Sh%. I would remove TOI for that reason, possibly GP & shots, and maybe add Shots/60 to see how significant that is.

I think a t-score of ~3 is pretty rock solid, so they are right on the cusp. I did already drop observations without a significant amount of games, so games can go.

TOI was simply a 'value' factor (plus it was available), and might be a little too correlated with some thing like shots. Shots/60 seems to roll everything up in one metric. Thanks for the input, I'll try these next week and repost. IIRC the R2 was 43%, so there's certainly room for improvement.
 

Czech Your Math

I am lizard king
Jan 25, 2006
5,169
303
bohemia
I think a t-score of ~3 is pretty rock solid, so they are right on the cusp. I did already drop observations without a significant amount of games, so games can go.

TOI was simply a 'value' factor (plus it was available), and might be a little too correlated with some thing like shots. Shots/60 seems to roll everything up in one metric. Thanks for the input, I'll try these next week and repost. IIRC the R2 was 43%, so there's certainly room for improvement.

TOI is def. correlated to shots. If you're using TOI/60 vs. Sh/60, I would still expect a correlation, but less so. I wouldn't say Sh/60 rolls everything into one metric, but if there's a correlation, it should at least be in the right direction. On a team level, the correlation is neutral or slightly negative, but not very significant.
 
Mar 31, 2005
1,675
10
East Coast
I've found some inconsistencies with NHL.com data and the behindthenet data I have been using, with update when I understand what is going on.
 
Last edited:

Czech Your Math

I am lizard king
Jan 25, 2006
5,169
303
bohemia
I've found some inconsistencies with NHL.com data and the behindthenet data I have been using, with update when I understand what is going on.

I would at least calculate your own Sh%s if you are using BTN's data. Others have said SF is actually saves, not shots. Either way, their calculations for Sh% appear to be incorrect.

I emailed them about it a few days ago, and of course they have not replied.
 

overpass

Registered User
Jun 7, 2007
5,254
2,736
What if one limited the sample to players with at least N shots for on ice while on PP, then compared only those players to each other in shooting% for on ice 5v5 and on PP? I think there's still some sort of bias coming through, but that's just a hunch, very well could be wrong. There's so many players with only mediocre skill 5v5 and PP is limited mostly to those with superior skill.

Here are the 5v5 on-ice shooting numbers by # of PP shots.

On-ice PP shots | On-ice EV GF | On-ice EV SOGF | On-ice EV SH%
1000+ | 7329 | 75440 | 9.7%
800-999 | 10056 | 112726 | 8.9%
600-799 | 13530 | 153290 | 8.8%
400-599 | 15838 | 188636 | 8.4%
200-399 | 12269 | 148519 | 8.3%
0-199 | 32414 | 424143 | 7.6%
Average | 91436 | 1102754 | 8.3%

So if you are looking for an expected on-ice shooting percentage for 5v5, it's worth looking at how much the player plays on the PP. If he very rarely plays on the PP, 7.6% is a better estimate of his expected on-ice SH% at 5v5 than the league average of 8.3 SH%. And if he is a star player who plays heavy PP minutes, 9.7% might be a better estimate of his expected on-ice 5v5 SH%.
 

oilerbear

Registered User
Jun 2, 2008
3,168
199
The answer to the question, is of course, both. But in what degree?


I'll see if I can get the 2011-12 data to test the predictions.

Hockey is attacks, Counter attack, and muti phase poscession.
That is a whole different look at how a puck is directed at the net.

The shot is a binary point in time Stat.
of all the pucks directed at the net, they are the ones to get to the scoring area.

Shooting % is a weak measure of the success of shots.

your study ignores so many variables.

1. Type of Shot
2. Location of Shot
3. Distance of Shot
Alan Ryder Did a study of the Success rate of Shots. Showing curves for each shot type relative to distance. There was another study done that looked at shots by location with no measure of type.
points 1-3 are the variables related to the release of a puck. the shooter can control to some extent.

4. Path of Shot
Clear, Screened, Deflected, Tiped.
Foxes Glowing Puck technology allows for the tracking of 3D puck path. (As well as all puck movement before directing to net)

5. Elevation of Goalie (position)
6. Movement of Goalie side to side and limbs.
CBS gametracker shows a basic elevation of goalie showing puck location relative to the net.
Like fox trackers strike zone. A more complex elevation is needed.

points 4-6 allso greatly affect success rate.

This is data that is availible to establish a true value to a shot.

A saved puck directed at the opositions goalie from inside your own blue line is of equal shot value as a tip in save were a goalie leaped across the crease to snag the puck glove hand.

That is not right!

I want to know that of a players shots total
1. the % of each type
2. locations of his types
To establish an Expected Success rate.
3. % of failed path
4. # of passes from High % shooting positions.
5. elevation targets of the shots.

Give me Eberles low shot count High % close to the net shots that yeild a high Shooting % and Goal /pocession with limited opposition counter attack ability.

over a high count shooter from many locations that yeilds a low goals
/ pocession count and high counter attack rate.

That is observed and not full data supported which individual shot history would supply.
 

overpass

Registered User
Jun 7, 2007
5,254
2,736
I want to know that of a players shots total
1. the % of each type
2. locations of his types
To establish an Expected Success rate.
3. % of failed path
4. # of passes from High % shooting positions.
5. elevation targets of the shots.

Give me Eberles low shot count High % close to the net shots that yeild a high Shooting % and Goal /pocession with limited opposition counter attack ability.

over a high count shooter from many locations that yeilds a low goals
/ pocession count and high counter attack rate.

That is observed and not full data supported which individual shot history would supply.

Sounds good to me! I suggest you:

1. Acquire the necessary data for every player in the league.
2. Find the frequency and success rate of the different types of shots.
3. Examine the observed distribution and derive the skill distribution for each stat.

I don't have the necessary data so I won't do it.
 

overpass

Registered User
Jun 7, 2007
5,254
2,736
Breaking down on-ice shooting percentage by position

Defencemen
On-ice PP shots | On-ice PP GF | On-ice PP SOGF | On-ice PP SH% | On-ice EV GF | On-ice EV SOGF | On-ice EV SH%
1000+ | 781 | 5981 | 13.1% | 1239 | 13974 | 8.9%
800-999 | 1305 | 10002 | 13.0% | 2162 | 24528 | 8.8%
600-799 | 2280 | 17848 | 12.8% | 4359 | 51286 | 8.5%
400-599 | 2122 | 17192 | 12.3% | 5284 | 65258 | 8.1%
200-399 | 1385 | 11534 | 12.0% | 5315 | 65122 | 8.2%
0-199 | 1331 | 12131 | 11.0% | 16326 | 201037 | 8.1%
Average | 9204 | 74688 | 12.3% | 34685 | 421205 | 8.2%

Forwards
On-ice PP shots | On-ice PP GF | On-ice PP SOGF | On-ice PP SH% | On-ice EV GF | On-ice EV SOGF | On-ice EV SH%
1000+ | 3605 | 27287 | 13.2% | 6090 | 61466 | 9.9%
800-999 | 4499 | 34535 | 13.0% | 7894 | 88198 | 9.0%
600-799 | 4529 | 35688 | 12.7% | 9171 | 102004 | 9.0%
400-599 | 4811 | 39674 | 12.1% | 10554 | 123378 | 8.6%
200-399 | 2427 | 20069 | 12.1% | 6954 | 83397 | 8.3%
0-199 | 2098 | 18806 | 11.2% | 16088 | 223106 | 7.2%
Average | 21969 | 176059 | 12.5% | 56751 | 681549 | 8.3%

As in previous posts, the data is from the 2007-08 season through the 2010-11 season. I've broken players into bins based on the number of PP shots they were on the ice for over this time period, as a rough estimate of skill.

It appears that the 5v5 spread in shooting talent is considerably wider among forwards than among defencemen. The forward groups range from a 9.9 on-ice SH% at the most skilled to a 7.2 on-ice SH% at the least skilled. The comparable range for defenders is 8.9 SH% to 8.1 SH%.
 

seventieslord

Student Of The Game
Mar 16, 2006
36,080
7,132
Regina, SK
5. Elevation of Goalie (position)
6. Movement of Goalie side to side and limbs.

Those are things that the goalie has control over. Why would we use them in an analysis of the quality of shots sent his way?

As an example, if he was "down and out" thanks to making an initial save, and then the puck squirted out to the point, where a usually harmless shot was directed to the net, and it scored, a normal shot quality metric would say that a relatively easy shot was successful. One that was adjusted for the goalie's position/location/etc would say it was a very difficult shot. Certain types of goalies would be "creating" more difficult shots for themselves by being seemingly out of position often. What would that tell us about shot quality?
 

Ad

Upcoming events

Ad

Ad

-->