Sep 032012
 

A month and a half ago Eric T at NHLNumbers.com had a good post on quantifying the impact on teammate shooting percentage.  I wanted to take a second look at the relative importance the impact on teammate shooting percentage can have because I disagreed somewhat with Eric’s conclusions.

For a very small number of elite playmakers, the ability to drive shooting percentage can be a major component of their value. For the vast majority of the league, driving possession is a more significant and more reproducible path to success.

It is my belief that it is important to consider impact on shooting percentage for more than a “very small number of elite playmakers” and I’ll attempt to show that now.

The method that Eric used to identify a players impact on shooting percentage is to compare that players line mates shooting percentages with him to their overall shooting percentage.  As noted in the comments the one flaw with this is that their overall shooting percentage is impacted by the player we are trying to evaluate which will end up underestimating the impact.  In the comments Eric re-did the analysis using a true “without you” shooting percentage and the impact of driving teammate shooting percentages was greater than initially expected but he concluded the conclusions didn’t  chance significantly.

Overall average for the top ten is a 1.2% boost (up from 0.9% in story) and 5 goals per year (up from 4.5). I don’t think this changes the conclusions appreciably.

In the minutes that a player is on the ice with one of the very best playmakers in the league, his shooting percentage will be about 1% better. For a player who gets ~150-200 shots per year and plays ~40-60% of his ice time with that top-tier playmaker, that’s less than a one-goal boost. It’s just not that big of a factor.

He also suggested that using the “without you” shooting percentage instead of the “overall shooting percentage” would probably result in “more accurate but less precise” analysis.  This is because a guy like Daniel Sedin would get very few shots when playing apart from Henrik Sedin because they rarely play apart and this small “apart” sample size might be subject to significant small sample size errors.

In this work I have taken a hybrid approach.  I took 5 years of 5v5 zone start adjusted data (2007-08 through 2011-12) and estimated how many goals a players line mates would score based on their without you shooting percentage, except in instances where the player had fewer than 100 without you shots.  In situations where the player had fewer than 100 shots apart the player being evaluated I used the players overall shooting percentage.  Regardless of which one is used, let’s call the shooting percentage used the players expected shooting percentage.   I then calculated an estimated goals scored by line mates by multiplying the expected shooting percentage by the number of shots taken when on the ice with the player being evaluated and summed up the expected goals across all line mates.  This gave me an actual goals scored by a players line mates and an expected goals for by a players line mates and I also know the actual shots taken by a players line mates so I can calculate actual and expected shooting percentages.  The following table shows the top 20 and bottom 20 forwards and their impact on their line mates shooting percentage (only players whose line mates had >1000 shots are considered).

 Player  Pos  LM Goals  LM Shots  LM Sh%  Exp Goals  Exp Sh% Sh%-ExpSh%
HENRIK SEDIN C 240 2012 11.928 177.3 8.811 3.117
DANIEL SEDIN L 174 1551 11.219 129.5 8.35 2.869
SAM GAGNER C 150 1394 10.76 115.4 8.281 2.479
ANDREI KOSTITSYN L 139 1390 10 105.6 7.598 2.402
BOBBY RYAN R 148 1439 10.285 114.4 7.952 2.333
MARIAN GABORIK R 130 1153 11.275 104.3 9.05 2.225
SIDNEY CROSBY C 151 1413 10.686 119.6 8.461 2.225
JASON SPEZZA C 178 1691 10.526 140.6 8.313 2.213
ALEX TANGUAY L 165 1500 11 132.8 8.852 2.148
MIKE RIBEIRO C 177 1668 10.612 143.1 8.579 2.033
NATHAN HORTON R 169 1562 10.819 137.5 8.804 2.015
J.P. DUMONT R 107 1050 10.19 85.9 8.179 2.011
JOE THORNTON C 216 2207 9.787 171.8 7.784 2.003
PAVEL DATSYUK C 196 1836 10.675 160.4 8.738 1.937
ALEX OVECHKIN L 186 1718 10.827 155.0 9.019 1.808
JOFFREY LUPUL R 109 1038 10.501 90.4 8.708 1.793
MARTIN HAVLAT R 134 1304 10.276 110.8 8.499 1.777
NICKLAS BACKSTROM C 211 1880 11.223 178.2 9.48 1.743
THOMAS VANEK L 151 1527 9.889 124.5 8.155 1.734
MARTIN ST._LOUIS R 218 1912 11.402 185.1 9.679 1.723
JOE PAVELSKI C 126 1624 7.759 143.1 8.811 -1.052
RYAN CALLAHAN R 96 1223 7.85 109.3 8.933 -1.083
RYAN O’REILLY C 71 1004 7.072 82.0 8.169 -1.097
DANIEL WINNIK C 82 1286 6.376 96.6 7.513 -1.137
TOMAS HOLMSTROM L 111 1331 8.34 126.9 9.532 -1.192
JEFF HALPERN C 74 1014 7.298 86.5 8.53 -1.232
MARTY REASONER C 84 1217 6.902 99.3 8.162 -1.26
RYAN KESLER C 133 1606 8.281 153.5 9.558 -1.277
SAMUEL PAHLSSON C 77 1211 6.358 92.7 7.654 -1.296
TROY BROUWER R 100 1203 8.313 116.4 9.677 -1.364
MIKE GRIER R 64 1032 6.202 78.2 7.582 -1.38
SEAN BERGENHEIM L 77 1153 6.678 93.4 8.097 -1.419
TRAVIS MOEN L 77 1186 6.492 93.9 7.92 -1.428
SERGEI SAMSONOV L 83 1257 6.603 101.3 8.062 -1.459
JARRET STOLL C 87 1302 6.682 106.9 8.211 -1.529
CHRIS KUNITZ L 147 1808 8.131 174.9 9.671 -1.54
MICHAEL FROLIK R 86 1099 7.825 103.5 9.42 -1.595
MAXIME TALBOT C 85 1202 7.072 104.9 8.724 -1.652
KYLE WELLWOOD C 83 1191 6.969 104.0 8.729 -1.76
MIKE KNUBLE R 109 1477 7.38 144.1 9.759 -2.379
-LM Goals is the actual goals scored by the players line mates while on the ice with the player.  
-LM Shots is the actual shots taken by the players line mates while on the ice with the player.
-LM Sh% is the actual shooting percentage of the players line mates while on the ice with the player.
-ExpGoals is the number of goals the players line mates are expected to score if their shooting percentage while on the ice with the player was equal to their shooting percentage when not on the ice with the player (or if <100 apart shots the line mates overall shooting percentage).
-ExpSh% is the expected shooting percentage of the players line mates (ExpGoals / LM Shots)
Sh%-ExpSh% is the difference between the actual and expected shooting percentage, or how much the player boosts their line mates shooting percentages.

I always say that order is the enemy of randomness and the fact that the top 20 are predominantly populated by offensive oriented players, many known as playmaking centers, and the bottom 20 are significantly populated by 3rd liners and defensive players tells me that the results we are seeing aren’t due to random error or errors in the methodology.

All totaled there were 222 players in the study, 26 boosted their line mates shooting percentage by at least 1.5% and 53 players boosted their line mates shooting percentage by at least 1.0%.  Only 20 players (the 20 listed above) had an observed penalty of at least 1.0% to their line mates shooting percentage and only 86 showed any negative influence whatsoever.  This assymetry and bias towards having more players boost shooting percentage and by larger amounts is due to selection bias.  By only considering players whose line mates take at least 1000 shots while the player is on the ice eliminates all the bad players who were so bad that they didn’t get enough ice time to have 1000 shots taken by their line mates.

I should mention here that we are not taking into account quality of line mates that might impact the results.  Since both the Sedin’s play together their impact on their line mates shooting percentage is really compounded (particularly for Daniel who rarely played without Henrik).  There is probably a very good chance than Daniel Sedin doesn’t have that significant an impact on line mate shooting percentage, but we just can’t separate his abilities from his twin brothers abilities so they both look really good.  But that said, one or some combination of both of the Sedin’s have a very significant impact on their line mates shooting percentages.

The next thing we need to do is attempt to quantify what these boosts in shooting percentages mean in terms of goals.  To do this I assumed each player played 15 minutes of 5v5 ZS adjusted ice time per game over 82 games (i.e. the play a heavy ES work load over a full 82 game season).  The top 25 players and bottom 25 players are listed in the table below.

 Player G/Yr  Player G/Yr
HENRIK SEDIN 15.17 MIKE KNUBLE -9.90
DANIEL SEDIN 11.78 CHRIS KUNITZ -8.13
SIDNEY CROSBY 11.11 KYLE WELLWOOD -7.98
JOE THORNTON 10.36 SERGEI SAMSONOV -7.15
ANDREI KOSTITSYN 9.95 MAXIME TALBOT -6.98
JASON SPEZZA 9.90 MICHAEL FROLIK -6.29
SAM GAGNER 9.82 JARRET STOLL -6.21
BOBBY RYAN 9.75 TROY BROUWER -5.94
ALEX TANGUAY 9.73 TOMAS HOLMSTROM -5.89
J.P. DUMONT 9.34 SEAN BERGENHEIM -5.69
NATHAN HORTON 9.25 MIKE GRIER -5.68
PAVEL DATSYUK 9.23 RYAN KESLER -5.40
NICKLAS BACKSTROM 8.54 TRAVIS MOEN -5.25
MIKE RIBEIRO 8.53 MARTY REASONER -4.96
MARIAN GABORIK 7.78 SAMUEL PAHLSSON -4.88
MARTIN HAVLAT 7.59 RYAN O’REILLY -4.79
THOMAS VANEK 7.58 DANIEL WINNIK -4.73
JONATHAN TOEWS 7.21 JOE PAVELSKI -4.73
MARTIN ST._LOUIS 7.16 JEFF HALPERN -4.56
JOFFREY LUPUL 7.15 PATRICK MARLEAU -4.40
CORY STILLMAN 7.08 CHAD LAROSE -4.26
ALEX OVECHKIN 6.93 RYAN CALLAHAN -4.07
EVGENI MALKIN 6.86 CLARKE MACARTHUR -3.74
STEVE DOWNIE 6.58 JOCHEN HECHT -3.74
BRENDEN MORROW 6.49 SCOTTIE UPSHALL -3.48

I wouldn’t necessarily consider the above numbers insignificant.  All those players in the top 25 list plus 5 others boost their line mates goal scoring by 6 or more goals per season solely by boosting their line mates shooting percentages.  This means there are 30 players who contribute an extra win to their team every season solely by boosting their line mates shooting percentages.  The standard deviation among all 222 players is 4.18.

To put this into perspective, I compared a forwards ability to boost line mates fenwick for rate per 20 minutes of ice time (FF20).  Sidney Crosby had the greatest ability to boost line mate FF20 over the past 5 seasons.  His line mates FF20 when not with Crosby was 12.92 but it was 15.83 with Crosby.  This equates to an extra 179 fenwick for events per season (assuming 15 minutes of ice time over 82 games) while Crosby is on the ice (note that this includes Crosby’s individual fenwick for events so it isn’t just Crosby’s ability to boost his line mates fenwick for events).  Applying the league average fenwick shooting percentage of 6.29% to those 179 events we see that Crosby’s ability to boost on-ice FF20 results in an extra 11.26 goals per season.  The standard deviation among all players is 3.59, or slightly less than the standard deviation for impact on goals due to influence on line mate shooting percentage.  Although there is greater margin for error in the shooting percentage analysis, it is theoretically possibly that a players ability to boost his line mates shooting percentage has a greater impact on goal scoring than the players ability to drive fenwick for events.

Let’s look at this in yet another way.  If we take (LM Goals – ExpGoals)/LM Goals we will find the percentage of goals that the players line mates scored that were due to the players ability to boost their shooting percentages.  The top and bottom 25 look like this.

 Player  LM Goals  Exp Goals %G due to Sh% Influence
HENRIK SEDIN 240 177.269 26.1%
DANIEL SEDIN 174 129.512 25.6%
ANDREI KOSTITSYN 139 105.61 24.0%
SAM GAGNER 150 115.439 23.0%
BOBBY RYAN 148 114.434 22.7%
JASON SPEZZA 178 140.568 21.0%
SIDNEY CROSBY 151 119.559 20.8%
JOE THORNTON 216 171.803 20.5%
J.P. DUMONT 107 85.878 19.7%
MARIAN GABORIK 130 104.341 19.7%
ALEX TANGUAY 165 132.78 19.5%
MIKE RIBEIRO 177 143.092 19.2%
NATHAN HORTON 169 137.52 18.6%
PAVEL DATSYUK 196 160.437 18.1%
THOMAS VANEK 151 124.534 17.5%
CORY STILLMAN 102 84.301 17.4%
MARTIN HAVLAT 134 110.829 17.3%
JOFFREY LUPUL 109 90.386 17.1%
MIKKO KOIVU 129 107.39 16.8%
ALEX OVECHKIN 186 154.955 16.7%
FRANS NIELSEN 97 81.722 15.8%
NICKLAS BACKSTROM 211 178.224 15.5%
BRENDEN MORROW 133 112.624 15.3%
LOUI ERIKSSON 170 143.991 15.3%
MATTHEW LOMBARDI 103 87.285 15.3%
CLARKE MACARTHUR 117 128.858 -10.1%
SHAWN HORCOFF 105 116.296 -10.8%
PATRICK MARLEAU 151 168.988 -11.9%
SCOTTIE UPSHALL 74 82.866 -12.0%
JOE PAVELSKI 126 143.091 -13.6%
CHAD LAROSE 102 115.857 -13.6%
RYAN CALLAHAN 96 109.252 -13.8%
TOMAS HOLMSTROM 111 126.873 -14.3%
RYAN KESLER 133 153.497 -15.4%
RYAN O’REILLY 71 82.019 -15.5%
TROY BROUWER 100 116.417 -16.4%
JEFF HALPERN 74 86.489 -16.9%
DANIEL WINNIK 82 96.614 -17.8%
MARTY REASONER 84 99.328 -18.2%
CHRIS KUNITZ 147 174.858 -19.0%
SAMUEL PAHLSSON 77 92.69 -20.4%
MICHAEL FROLIK 86 103.529 -20.4%
SEAN BERGENHEIM 77 93.363 -21.3%
TRAVIS MOEN 77 93.927 -22.0%
SERGEI SAMSONOV 83 101.341 -22.1%
MIKE GRIER 64 78.245 -22.3%
JARRET STOLL 87 106.912 -22.9%
MAXIME TALBOT 85 104.859 -23.4%
KYLE WELLWOOD 83 103.962 -25.3%
MIKE KNUBLE 109 144.147 -32.2%

In total there were 27 players which had >15% of all their line mates goals attributed to their ability to boost their line mates shooting percentage and there were 59 players who had >10% of their line mates goals attributed to their ability to boost their line mates shooting percentage.  I don’t consider this insignificant.

One final thing to look at is how a players assist rates compare to their ability to boost line mate shooting percentages.

That is actually a fairly strong relationship between a players assist rate and the players ability to boost line mates shooting percentages.  It really isn’t unexpected, but it is good to see.  Playmaking is a valuable skill in generating offense.

 

 

  4 Responses to “Relative Importance of a Players Impact on Teammate Shooting Percentage”

  1.  

    I think this overstates things because you haven’t regressed the results to the mean at all. Over ~1000-2000 shots, simple variance will leave us with ~1-1.5% uncertainty on the shooting percentage. That variance will tend to broaden the distribution from the true talents, making the extremes look more extreme than their long-run performance will be and overstating the importance of this effect.

    •  

      I will grant you that, the effects listed above quite possibly are over emphasized. But with that said, I am not sure standard methods for accounting for uncertainty necessarily apply. There is a lot more going on in hockey than goes on in ‘coin flipping’ so I am not certain uncertainty can be accounted for in the same way. For example, Mike Knuble has the largest negative impact. We also know that Knuble has one of the highest shooting percentage because he takes shots from high percentage zones (basically he stands in front of the net chipping in rebounds). So, when Knuble is on the ice he is taking away these high percentage shots away from his line mates. When his line mates are on the ice with someone other than Knuble, these high percentage shots may be available for them to take. This means some chunk of Knuble’s negative influence is systematic and not subject to uncertainty. Or, put in a different way, because of Knuble’s style of play, the mean Knuble’s line mates shooting percentages would ‘regress to’ when on the ice with him is different than another players teammates mean shooting percentage that we should ‘regress to’.

      •  

        Erm, don’t get me wrong — I’m not arguing that Knuble and Crosby are equals in this regard.

        Just that any finite measurement will be broadened relative to the underlying talent, and the extent of that broadening depends on the variability of the thing being measured and the sample size — and that 1000-2000 shots is a decent but not great sample size for a shooting percentage.

        Phil Kessel has a 10.8% career shooting percentage after 1532 shots; would you be shocked if a magic probing device told us that his true talent is that of a 10.2% shooter (meaning he ran hot by eight goals over the course of his career to date)?

        Perhaps the existence of such a device would shock you, but I don’t think its output would, because you recognize that ~1000-2000 shots gives us a good but not great sense of how good a shooter someone is. The noise on the measurement in this post is even larger, because the with-you portion of the analysis has ~1000-2000 shots and there is additional uncertainty on the without-you analysis.

        You could run some split half correlations to figure out the reliability and estimate the regression to the mean to figure out whether accounting for that would reduce the measured effect by 25% or 75%, but there’s no doubt it will reduce it.

        •  

          For sure and I did grant you that my numbers in the original article are most likely over emphasized. The point I was making is I am not certain using a typical coin flipping model will in any way accurately estimate this because there are so many other factors involved.

          As for Phil Kessel’s shooting percentage his true-talent shooting percentage may well be 10.2%. Might be 11.2% too. Of course, if he played with Sidney Crosby it may be 11.8%, or if he learned to play more like Mike Knuble it might be 12.8% or higher. For that matter, zone starts may very well affect his shooting percentage and if he started getting more defensive and neutral zone faceoffs his expected shooting percentage may in fact rise. The point is, it is really difficult to determine how lucky, or unlucky, Phil Kessel has been because we have no clue what his expected level based on his talent level and the situations he has played in.

          As for the without-you uncertainty, that may not be an issue because while any particular line mates without-you number may be full of uncertainty, across the numerous line mates any particular player has had the uncertainties are likely to start to cancel each other out and the overall uncertainty in the without-you stats probably approaches something fairly small (especially those that have changed teams and have had a lot of line mates) relative to any with-you uncertainty. The total number of shots across all of all the without-you stats I suspect would be well into the 10’s of thousands (particularly for any player who has changed teams a few times).

Sorry, the comment form is closed at this time.