Jun 012011
 

There seems to be some confusion, or lack of clarity, about my post on corsi vs shooting percentage vs shooting rate the other day so let me clear it up in as straight forward a way as I can.

“Hawerchuk” over at BehindTheNetHockey.com writes the following:

“I’m not totally sure what he’s getting at. People use Fenwick because it’s persistent, and PDO because it’s not. Over the course of a single season, observed shooting and save percentage drive results, but they are not persistent.”

Dirk Hoag over at OnTheForecheck.com writes:

“Here’s an example of when NOT to use correlation as a tool in statistical analysis (when the variables in question are linked by definition). David makes a bad blunder here, by looking at scoring leaders, seeing a bunch of high shooting percentages, and concluding that shooting percentage is the true “talent”. The problem is that shooting percentage swings wildly from season to season, whereas shooting rates are much more consistent.”

The great advantage of corsi/fenwick has over goals as an evaluator of talent is the greater sample size associated with it.  The greater the sample size the more confidence we can have in any results we conclude from it and the less chance that ‘luck’ messes things up.  Year over year shooting percentage fluctuates a lot, but that doesn’t necessarily mean that it isn’t a talent or doesn’t have persistence, it could mean that the sample size of one year is too small.  The four year shooting percentage leader board seems to identify all the top offensive players so it can’t be completely random.  So what happens if we increase the sample size?  Here are correlations of fenwick shooting percentages while on ice in 5v5 even strength situations for forwards:

Year(s) vs Year(s) Corrolation
200708 vs 200809 0.249
200809 vs 200910 0.268
200910 vs 201011 0.281
200709 vs 200911 (2yr) 0.497

As you can see, there isn’t a lot of persistence year over year but for 2 years over 2 years we are starting to see some persistence.  Still not to the level of corsi/fenwick, but certainly not non-existant either, and the greater correlation with scoring goals makes fenwick shooting percentage on par with fenwick as a predictor of future goal scoring performance when we have 2 seasons of data as I pointed out in my last post.

For the record, year over year correlation for fenwick for rate is approximately 0.60 depending on years used  and 2 year vs 2 year correlation is 0.66.

But as I pointed out in my previous post, you would probably never use shooting percentage as a predictor because you may as well use goal rate instead which has the same sample size limitations as shooting percentage but also factors in fenwick rate.  Year over year correlation of GF20 (goals for per 20 minutes) is approximately 0.45 depending on years used and the 2 year vs 2 year correlation is 0.619 so GF20 has persistence and has a 100% correlation with itself making it as reliable (or more) a predictor of future goal scoring rates as fenwick rate with just one year of data and a better predictor when using 2 years of data.  Let me repost the pertinent table of correlations:

Year(s) vs Year(s) FenF20 to GF20 GF20 to GF20
200708 vs 200809 0.396 0.386
200809 vs 200910 0.434 0.468
200910 vs 201011 0.516 0.491
Average 0.449 0.448
200709 vs 200911 (2yr) 0.498 0.619
200709 vs 200910 (2yr vs 1yr) 0.479 0.527

The conclusion is, when dealing with less than a years worth of data, fenwick/corsi is probably the better metric to identify talent and predict future performance, but anything greater than a year goals for rate is the better metric and for one years worth of data they are about on par with each other.

Note:  This is only true for forwards.  The same observations are not true about defensemen where we see very little persistence or predictability in any of these metricts, I presume because the majority of them don’t drive offense to any significant degree.

May 302011
 

The general consensus among advanced hockey statistic analyzers and is that corsi/fenwick stats are the best statistic for measuring player and team talent levels.  For those of you who are not aware of corsi and fenwick let me give you a quick definition.  Corsi numbers are the number of shots directed at the goal and include shots, missed shots and blocked shots.  Fenwick numbers are the same except it does not included blocked shots (just shots and missed shots).  I generally look at fenwick and will do that here but fenwick and corsi are very highly correlated to the results would be similar if I used corsi.

The belief by many that support corsi and fenwick is that by looking at fenwick +/- or fenwick ratio (i.e. fenwick for /(fenwick for + fenwick against)) is an indication of which team is controlling the play and the team that controls the play more will, over time, score the most goals and thus win the most games.  There is some good evidence to support this, and controlling the play does go a long way to controlling the score board.  The problem I have with many corsi/fenwick enthusiasts is that they often dismiss the influence that ability to drive or suppress shooting percentage plays in the equation.  Many dismiss it outright, others feel it has so little impact it isn’t worth considering except when considering outliers or special cases.  In this article I am going to take an in depth look at the two and their influence on scoring goals on an individual level.

I have taken that last 4 seasons of 5v5 even strength data and pulled out all the forwards that have played at minimum 2000 minutes of 5v5 ice time over the past 4 seasons.  There were a total 310 forwards matching that criteria and for those players I calculated the fenwick shooting percentage (goals / fenwick for), fenwick for rate (FenF20 – fenwick for per 20 minutes of ice time) and goal scoring rate (gf20 – goal for per 20 minutes ice time) while the player was on the ice. What we find is shooting percentage is more correlated with goal production than fenwick rate.

Shooting % vs GF20 R^2 = 0.8272
FenF20 vs GF20 R^2 = 0.4657
Shooting % vs FenF20 R^2 = 0.1049

As you can see, shooting percentage is much more highly correlated with goal scoring rate than fenwick rate is which would seem to indicate that being able to drive shooting percentage is more important for scoring goals than taking a lot of shots.

Here is a list of the top 20 and bottom 10 players in fenwick shooting percentage and fenwick rate.

Rank Player FenSh% Player FenF20
1 MARIAN GABORIK 8.07 HENRIK ZETTERBERG 16.7
2 SIDNEY CROSBY 7.83 ALEX OVECHKIN 16.3
3 ALEX TANGUAY 7.64 PAVEL DATSYUK 16.15
4 HENRIK SEDIN 7.63 TOMAS HOLMSTROM 16.05
5 BOBBY RYAN 7.60 NICKLAS BACKSTROM 16.05
6 STEVE DOWNIE 7.58 ERIC STAAL 16.04
7 EVGENI MALKIN 7.57 RYANE CLOWE 15.91
8 DANIEL SEDIN 7.55 ALEXANDER SEMIN 15.85
9 ILYA KOVALCHUK 7.49 SCOTT GOMEZ 15.8
10 NATHAN HORTON 7.44 ZACH PARISE 15.8
11 J.P. DUMONT 7.43 ALEXEI PONIKAROVSKY 15.79
12 JASON SPEZZA 7.39 JOHAN FRANZEN 15.78
13 PAUL STASTNY 7.36 JIRI HUDLER 15.74
14 PAVOL DEMITRA 7.33 DAN CLEARY 15.71
15 DANY HEATLEY 7.30 SIDNEY CROSBY 15.71
16 RYAN MALONE 7.29 JUSTIN WILLIAMS 15.68
17 JONATHAN TOEWS 7.28 CHRIS KUNITZ 15.61
18 THOMAS VANEK 7.24 MIKHAIL GRABOVSKI 15.56
19 SERGEI KOSTITSYN 7.24 JOE PAVELSKI 15.43
20 DREW STAFFORD 7.24 MIKAEL SAMUELSSON 15.39
301 BLAIR BETTS 4.20 CHUCK KOBASEW 11.34
302 ERIC NYSTROM 4.12 TRAVIS MOEN 11.31
303 SAMUEL PAHLSSON 4.10 IAN LAPERRIERE 11.23
304 SHAWN THORNTON 3.99 ERIC NYSTROM 11.21
305 TRAVIS MOEN 3.89 ROB NIEDERMAYER 10.94
306 TODD MARCHANT 3.88 TODD MARCHANT 10.91
307 NATE THOMPSON 3.75 SAMUEL PAHLSSON 10.87
308 FREDRIK SJOSTROM 3.70 JERRED SMITHSON 10.76
309 CRAIG ADAMS 3.52 JAY PANDOLFO 10.74
310 STEPHANE VEILLEUX 3.49 BLAIR BETTS 10.67

For both lists, the players are the top of the list are for the most part considered top offensive players and the players at the bottom of the list are not even close to being considered quality offensive players.  So, it seems that both shooting percentage and fenwick do a reasonable job at identifying offensively talented players.  That said, the FenF20 list includes 7 players (Zetterberg, Datsyuk, Holmstrom, Franzen, Hudler, Cleary and Samuelsson) who have played mostly or fully with the Detroit Red Wings and it seems unlikely to me that 7 of the top 20 offensive players are Red Wing players.  Furthermore, the fenwick list also includes guys like Ponikarovsky, Samuelsson, Hudler, Cleary, Williams, etc. who would probably be considered secondary offensive players at best.  From just this cursory overview it seems to confirm what we saw with the correlations – Shooting Percentage is a better indicator of offensive talent than Fenwick For rates.

It is actually no surprise that the Red Wings dominate the fenwick rate leader board because the Red Wings organizational philosophy is all about puck control.

“It’s funny because our game looks at numbers just like other games,” says Red Wings general manager Ken Holland, “but as much value as we assign to puck possession and how essential it is to winning, we really don’t have a numerical value for it that everyone can agree on. Remember when [A’s general manager] Billy Beane started emphasizing on-base percentage in baseball? It wasn’t just a curious number; it changed the game. It redefined the type of player you wanted on your team. It’s coming in hockey; we just have to figure out how.”

This got the pro-corsi crowd riled up a bit as they said “Umm, yeah, we have that stat and it is called corsi” and were a bit bewildered at why NHL GMs didn’t make that recognition.  But anyway, what the above shows is that an organization that focuses on puck control dominates the corsi for statistic so I guess what that shows is that corsi/fenwick probably is a good measure of puck control.  But, as we have seen, fenwick (i.e. puck control) doesn’t automatically translate into goals scored.  There are no Red Wing players among the top 20 in fenwick shooting percentage and Datsyuk is the only Red Wing player in the top 20 in goals for per 20 minutes so while they take a lot of shots (or at least shot attempts), they aren’t the best at converting them into goals.

For me, and I am sure many others, the above is enough to conclude that shooting percentage matters a lot in scoring goals, but for the staunch corsi supporters they will argue that corsi is more persistent from season to season and thus is a better predictor of future performance.  So which is the better predictor of future performance?  The following table shows the correlation between shooting percentage and fenwick rate with the following seasons goal scoring rate.

Year(s) vs Year(s) FenSh% to GF20 FenF20 to GF20
200708 vs 200809 0.253 0.396
200809 vs 200910 0.327 0.434
200910 vs 201011 0.317 0.516
Average 0.299 0.449
200709 vs 200911 (2yr) 0.479 0.498
200709 vs 200910 (2yr vs 1yr) 0.375 0.479

Note:  For the above season(s) vs season(s) correlation calculations, only players with at least 500 5v5 even strength minutes in each of the four seasons are included.  This way the same players are included in all season(s) vs season(s) correlation calculations.

As you can see, when dealing with a single season of data the correlation with GF20 is much better for fenwick rate than for fenwick shooting percentage.  The gap closes when using 2 seasons as the predictor of a single season and is almost gone when using 2 seasons to predict the following 2 seasons.  It seems that the benefit of using corsi over shooting percentage diminishes to near zero when we have multiple seasons of data and though I haven’t tested it shooting percentage probably has an edge in player evaluation with 3 years of data.

Of course, you would never want to use shooting percentage as a predictor of future goal scoring rate when you could simply use past goal scoring rate as the predictor.  Past goal scoring rate has the same ‘small sample size’ limitations as shooting percentage (both use goals scored as it sample size limitation) but scoring rate combines the prediction benefits of shooting percentage and fenwick rate.  The table below is the same as above but I have added in GF20 as a predictor.

Year(s) vs Year(s) FenSh% to GF20 FenF20 to GF20 GF20 to GF20
200708 vs 200809 0.253 0.396 0.386
200809 vs 200910 0.327 0.434 0.468
200910 vs 201011 0.317 0.516 0.491
Average 0.299 0.449 0.448
200709 vs 200911 (2yr) 0.479 0.498 0.619
200709 vs 200910 (2yr vs 1yr) 0.375 0.479 0.527

The above table tells you everything you need to know.  When looking at single seasons both GF20 and FenF20 perform similarly at predicting next seasons GF20 with fenwick shooting percentage well behind but when we have 2 years of data as the starting point, GF20 is the clear leader.  This means, when we have at least a full seasons worth of data (or approximately 500 minutes ice time), goal scoring rates are as good or better than corsi rates as a predictor of future performance and beyond a years worth of data the benefits increase.  When dealing with less than a full season of data, corsi/fenwick may still be the preferred stat when evaluating offensive performance.

So what about the defensive side of things?

Year(s) vs Year(s) FenA20 to GA20 GA20 to GA20
200708 vs 200809 0.265 0.557
200809 vs 200910 0.030 0.360
200910 vs 201011 0.120 0.470
Average 0.138 0.462
200709 vs 200911 (2yr) -0.037 0.371
200709 vs 200910 (2yr vs 1yr) 0.000 0.316

Defensively, fenwick against rate is very poorly correlated with future goals against rate and it gets worse, to the point of complete uselessness, when we consider more seasons.  Past goals against rate is a far better predictor of future goals against rate.

Where it gets interest is unlike offense correlation drops when you consider more seasons which seems a bit strange.  My guess is the reason we are seeing this is because I am just looking at forwards and defense is more driven by goaltending and defensemen and as more time passes the greater the difference are in goalie and defensemen teammates.  Furthermore, forward ice time is largely driven by offensive ability (and not defensive ability) so many of the quality defensive forwards may be removed from the study because of the 500 minute per season minimum I am using (i.e. the group of players used in this study are biased towards those that aren’t focusing on defense).  Further analysis is necessary to show either of these as true though but the conclusion to draw from the above table that, for forwards at least, goals against rates are by far the better indicator of defensive ability.

In summary, it should be clear that we cannot simply ignore the impact of a players ability to drive or suppress shooting percentage in the individual player performance evaluation and so long as you have a full year of data (or > 500 or more minutes ice time) the preferred stat for individual player performance evaluation should be goal scoring rate.  Corsi/fenwick likely only provide a benefit to individual performance evaluation when dealing with less than a full year of data.

May 172011
 

I wrote an article a few weeks ago about the offensive and defensive contribution (i.e. their HARO+ or HARD+ rating multiplied by ice time) of each position (C, LW, RW, D and G) but I have come to realize that my methodology is incorrect and thus the conclusions are incorrect (at least when looking at league-wide results).  The reason is, in my rating system contribution is evenly distributed among the 5 players on the ice so if I sum up all contributions of all players playing at a particular position I should see each position be given an equal share, and for the most part that is what I saw.  The exception being centers being given more influence and wingers less, this is because players that are listed as being centers often play the wing where as wingers are less often on the ice playing as centers.

The proper method for identifying the contribution a position has on offense and defense is not to sum up their contribution but to look at the variation observed in the players ratings for that position.  Recall that with my ratings a 1.00 is a neutral rating or an indication that the player has no positive or negative effect for that aspect of the game (offense or defense) compared to the expected level of performance when quality of competition and quality of teammates are considered.  Anything less than 1.0 implies a negative impact and anything above 1.0 implies a positive impact.  So, if a position can significantly influence offensive production then we should see a larger variance among centers HARO+ ratings.  The good players at that position will have ratings well above 1.00 and the weaker players well below 1.00.  For positions that do not have a significant impact we should see players at that position have ratings much closer to 1.00 and less variation between the best and worst players.   So, here is what we find.

HARO+ HARO+ HARD+ HARD+ HART+ HART+
Position Average StdDev Average StdDev Average StdDev
C 0.918 0.171 0.994 0.116 0.956 0.091
RW 0.927 0.162 1.001 0.096 0.965 0.084
LW 0.939 0.167 0.993 0.099 0.966 0.086
D 0.894 0.095 0.990 0.101 0.942 0.068
G 0.984 0.080 0.992 0.040

The above uses four year ratings (2007-11) and only forwards and defensemen with at least 2000 minutes of 5v5 even strength ice time and goalies with 3000 minutes were considered.  The resulting group included 122 centers, 85 LW, 103 RW, 194 defensemen and 53 goalies.

On offense, the three forward positions have significantly higher standard deviations (0.162-0.171) than defensemen (0.095) which intuitively makes sense.  It means that forwards have a greater ability to influence offensive production than defensemen which is no surprise.  Defensively the greatest variation in HARD+ occurs for centers with defensemen and wingers more or less the same a step below centers and goalies another step back again.  It is possible centers rank ahead of wingers and defensemen in part because they are the ones who take face offs and thus are a major factor in the team gaining control of the puck.

The other thing that you’ll notice is that for HARO+ the average rating is well below 1.00 for both the forwards and the defense.  This probably indicates that the big minute players are the offensive players which makes the average rating (which is ice-time neutral) well below the ice time weighted average (which in theory should be very close to 1.00).  Lets take a look at how the players rate according to total ice time.

Centers

Ice time HARO+ HARD+ HART+
>4000 min. 1.042 0.968 1.005
3000-3999 0.906 0.988 0.947
2000-2999 0.864 1.015 0.940
1000-1999 0.784 1.025 0.905

Left Wing

Ice time HARO+ HARD+ HART+
>4000 min. 1.089 0.939 1.014
3000-3999 0.987 0.990 0.989
2000-2999 0.824 1.015 0.920
1000-1999 0.760 1.036 0.899

Right Wing

Ice time HARO+ HARD+ HART+
>4000 min. 1.071 0.963 1.018
3000-3999 0.953 1.003 0.979
2000-2999 0.871 1.008 0.940
1000-1999 0.775 1.047 0.911

For the three forward positions it is clear that the top offensive players get the most playing time while players who get less playing time are slightly better defensive players.  This isn’t really a big surprise as the majority of a team’s offense comes from their top line(s).  The question is, how much does coaching/playing style influence the results.  By that I mean, would first line forwards be better defensively if they were on the third line and asked to play a defensive role as opposed to being on the first line and being asked to and expected to produce offense?  I suspect for most players the answer would be yes.  I suspect the reverse (third/fourth line guys having better offensive ratings if given first line roles) is also true, but probably to a lesser extent.

Defense

Ice time HARO+ HARD+ HART+
>5000 min. 0.923 0.988 0.955
4000-4999 0.919 0.997 0.958
3000-3999 0.871 0.998 0.934
2000-2999 0.874 0.974 0.923
1000-1999 0.864 1.025 0.944

For defensemen the best offensive defensemen still get the most ice time, though the variation is much less than seen with the forwards.  Defensive ability seems to have very little variation across ice times until you get to the lower minute players who appear to be more defensive specialists.

Goalie

Ice time HARD+
>10000 min. 1.040
>8000 min. 1.028
>6000 min. 1.012
>4000 min. 0.992
>2000 min. 0.984

As one would expect, the best goalies are given the most time in goal.  There were 9 goalies with greater than 10,000 minutes of 5v5 ice time and all had ratings over 1.00 except Tomas Vokoun whose rating was 0.978.  According to my rating system, Vokoun is a pretty ordinary goalie which means he is likely one of the more over rated goalies in the NHL because some (or most) consider him elite.  It’ll be interesting to see where he ends up this summer as a UFA and how that team performs next year.  Could Vokoun be another goalie failure in Philadelphia?  Could happen.

Apr 182011
 

By all accounts, Corey Perry had an exceptional season in 2010-11 and this is particularly true down the stretch when he flew by Steven Stamkos for the lead in goals scored and pushed himself into serious contention from the Hart Trophy as the leagues most valuable player.  There is no doubt that Perry’s production level surpassed anything he had previously done in his career, but was he truly more valuable to the Ducks than in previous seasons?  Let’s look at the numbers.

Season GP Goals Assists Points +/- PPG PPA PP Points
2010-11 82 50 48 98 9 14 17 31
2009-10 82 27 49 76 0 6 17 23
2008-09 78 32 40 72 10 10 14 24
2007-08 70 29 25 54 12 11 6 17

Based on the raw stats, he has been better in 2010-11 in terms of goal scoring and fairly consistent in terms of collecting assists but despite his increase in goals and points, his +/- hasn’t increased significantly.

Let’s look a little deeper into Perry’s even strength 5v5 statistics.

Season GF20 GA20 GF% TMGF20 TMGA20 TMGF% OppGF20 OppGA20 OppGF%
2010-11 0.928 0.882 0.513 0.876 0.843 0.510 0.774 0.745 0.509
2009-10 1.047 0.828 0.558 0.694 0.807 0.463 0.776 0.759 0.505
2008-09 1.113 0.754 0.596 0.712 0.775 0.479 0.756 0.751 0.501
2007-08 1.003 0.683 0.595 0.674 0.551 0.550 0.724 0.725 0.500

(source:  http://stats.hockeyanalysis.com/showplayer.php?pid=2)

For those unfamiliar with my terminology, GF20 is Perry’s goals for by team while on the ice per 20 minutes of ice time, GA20 is the same for goals against and GF% is GF20/(GF20+GA20) and represents what percentage of all goals scored while he was on the ice were scored by his team.  The TM stats are the same but for his team mates when they are not playing with Perry and the Opp stats are the same but for Perry’s opponents when they are not playing against Perry.

Now, the first observation you may make is that Perry’s GF20 was lower in 2010-11 than in any of the previous season so while Perry produced more offense (goals in particular) in 2010-11 individually, the team produced somewhat less when Perry was on the ice.  In other words, Perry’s goal/point production may have come at the cost of his line mates goal/point production.  The same thing is true defensively.  More goals were scored against Perry while Perry was on the ice than in any previous season.

Now, looking at team mate production when his teammates are not on the ice with Perry we find that they produce slightly fewer goals per 20 minutes (0.876 without Perry vs 0.928 with) but also give up slightly fewer goals too (0.843 without Perry, 0.882 with).  What is interesting though is Perry’s line mates this season appear to be better offensive players than in prior seasons as their 2010-11 GF20 was 0.876 vs 0.694 in 2009-10 though they also had a slightly higher GA20 in 2010-11 as well.  So from these numbers it seems that overall Perry played with significantly better offensive players in 2010-11 than in prior years and slightly worse defensive players in 2010-11 than in prior years.

As for quality of opposition, the offensive production of Perry’s opponents in 2010-11 was about the same as in 2009-10 while defensively they were slightly better.

So, in summary we can state that when Perry was on the ice in 5v5 even strength situations the Ducks produced less in 2010-11 than they did in 2009-10 and gave up more goals in 2011-10 than they did in 2009-10.  Furthermore, overall his line mates appear to have been significantly better offensive players in 2010-11 than in 2009-10 and only slightly worse defensive players while his opposition appears to be similarly skilled offensively and marginally less skilled defensively.

So, what does this all mean?  Here are Perry’s offensive and defensive ratings:

Season HARO+ HARD+ HART+
2010-11 1.164 0.852 1.008
2009-10 1.300 0.917 1.109
2008-09 1.391 0.953 1.172
2007-08 1.325 0.979 1.152

With all things considered, despite scoring 50 goals this past season, one could make an argument that 2010-11 was well below his performance during the three previous seasons.  It seems that his improved individual numbers may have come at the cost of his team mates and that made him less valuable to the Ducks overall.

Apr 152011
 

Before I get into the main subject of this post let me first point out that I have updated stats.hockeyanalysis.com to include all 1, 2, 3 and 4 year player ratings that can be calculated using the last 4 years of NHL data.  For more information on my player ratings read this.

I generate offense, defense and overall ratings for each and every player in the NHL and I wanted to get an idea of how much each position contributes to the performance of the team.  To accomplish this I multiplied each players offensive and defensive ratings (HARO+, HARD+) by their ice time (5v5 ratings and ice time used) and summed them up by position and then compared the positions total to the overall total.  I did this using the ratings calculated for the past 4 seasons combined as well as for each of the past 4 individual seasons.  This is the result I came up with :

Offense:

Season(s) Center RW LW D D
2007-11 24.64% 18.04% 17.14% 20.09% 20.09%
2007-08 26.91% 16.22% 16.47% 20.20% 20.20%
2008-09 25.23% 18.01% 16.66% 20.05% 20.05%
2009-10 23.93% 18.47% 17.49% 20.06% 20.06%
2010-11 25.13% 18.02% 16.76% 20.04% 20.04%

Defense:

Season(s) Center RW LW D D G
2007-11 20.67% 15.08% 14.27% 16.72% 16.72% 16.55%
2007-08 22.46% 13.83% 13.81% 16.75% 16.75% 16.39%
2008-09 21.06% 15.49% 13.76% 16.67% 16.67% 16.35%
2009-10 19.98% 15.46% 14.79% 16.73% 16.73% 16.30%
2010-11 21.35% 15.08% 14.21% 16.51% 16.51% 16.35%

Average of Offense + Defense:

Season(s) Center RW LW D D G
2007-11 22.65% 16.56% 15.71% 18.40% 18.40% 8.28%
2007-08 24.69% 15.03% 15.14% 18.48% 18.48% 8.19%
2008-09 23.14% 16.75% 15.21% 18.36% 18.36% 8.17%
2009-10 21.95% 16.96% 16.14% 18.39% 18.39% 8.15%
2010-11 23.24% 16.55% 15.48% 18.27% 18.27% 8.17%

Note:  I split the defense contribution over 2 positions.

Now, the first thing I noticed with these numbers is how surprisingly consistent they are from season to season, especially for defense and goaltending.  Up front players frequently shift from center to wing and from left wing to right wing so that may account for some of the (still relatively small) seasonal fluctuations.  Maybe I shouldn’t be surprised at this consistency but it does give me some confidence in my rating system that it is consistent across seasons as well as with multiple season ratings.

The second thing that caught my attention was the importance of defensive contribution to the offense.  Approximately 40% of offensive production can be attributed to the two defensemen on the ice and the defensemen are more important than the wingers. Part of this is simply that defensemen get more ice time than forwards since there are only 3 defense pairs versus 4 forward lines.  The other part is probably that they play an integral part of collecting rebounds and transitioning the team from defense to offense so they may have greater influence in the percentage of time played in the offensive zone.

Of the three forward positions, the center position is clearly the most important but we probably figured that.  Face offs might be a contributing factor but also we might just find that the most talented players end up playing center.  Right wings are slightly more important than left wings but the difference is not substantial.

Next I wondered what this data would mean to what teams should allocate for salaries.  For a 60 million payroll the average salary for position should work out to the following:

Pos Salary (Million$)
Center 13.6
RW 9.9
LW 9.4
D 11.0
D 11.0
G 5.0

Of course elite players skew the team payroll structure a fair bit.  As a LW earning over $9.5M Alexander Ovechkin is eating up the entire Capitals allotment for LWs and Crosby, Malkin and Staal are way over budget for the Penguins but you have to work around the talent you have.  A couple months ago Behind the Net Hockey Blog had a post outlining the salary allocated to players by position (split between forwards, defense, and goaltending).  Forwards were allocated 59.1% of a teams payroll, defense 32.2% and goaltending 8.7% over the past 4 seasons which compares to 54.9%, 36.8% and 8.3% for my ratings.  That would mean that forwards are overpaid (relative to their contribution) by about 4.1%, defense under paid by 4.6% and goalies over paid by about 0.4%.

For interest sake I decided to take a look at the Vancouver Canucks performance distribution since they have a fairly well balanced team and are a serious cup contender.  Here is what I found:

2007-11 2010-11
Position Offense Defense Average Offense Defense Average
Center 23.44% 19.96% 21.70% 21.04% 17.15% 19.10%
RW 11.44% 9.88% 10.66% 9.97% 10.34% 10.15%
LW 25.14% 21.88% 23.51% 31.12% 25.11% 28.11%
D 19.99% 17.21% 18.60% 18.94% 15.92% 17.43%
D 19.99% 17.21% 18.60% 18.94% 15.92% 17.43%
G 0.00% 13.86% 6.93% 0.00% 15.55% 7.77%

(Note:  The above is calculated using the current roster using the ratings and ice time over the past season or four seasons regardless of whether that ice time was with the Canucks.  This is an evaluation of the team ending the 2010-11 season with the Canucks, not the Canucks team performance over past seasons.  Also four season ratings should give a better player evaluation than single season ratings due to the larger sample size so I would consider them closer to true value.)

The Canucks are definitely a team driven by a group of quality left wingers or at least players listed as playing LW such as D. Sedin, Burrows, Raymond, Torres but I suspect some get shifted to RW from time to time.  Also, as good as Luongo is the quality and depth of the team in front of him reduces his relative contribution to his team to below average levels.  In the future I’ll take a look at some other teams as it’ll be interesting to see how goalie contribution changes from good teams with subpar  goalies (Detroit maybe) to bad teams with good goalies (Florida – Vokoun!! Though my ratings don’t value him as highly as many others do).

Mar 182011
 

The guys over at Behind the Net have initiated a ‘prove shot quality exists’ competition and in response to that Rob Vollman took a quick and dirty look at shooting percentage suppression.  As I showed the other day, Rob’s logic was a little off.

Rob started off by identifying a number of players with high on ice save percentages over the past 3 seasons.  Some of these guys included low minute players mostly playing on the fourth line against other fourth line caliber players, but there were a handful of players who played relative significant number of minutes and still put up good on ice save percentages.  Let me remind you of a few names that Rob identified:  forwards Marco Sturm, Manny Malhotra, Tyler Kennedy, Travis Moen, Taylor Pyatt, Michael Ryder, defensemen Kent Huskins, Sean O’Donnell, Mike Weaver, Mark Stuart.  I’ll get back to these guys later but I’ll claim that Rob dismissed some of them prematurely by claiming they played against weak competition.

As you may or may not know I have developed offensive and defensive ratings for every player and these can be found at http://stats.hockeyanalysis.com/ Furthermore, I have created these using goals for/against as well as shots for/against, fenwick for/against, and corsi for/against.  For clarification, fenwick is shots + missed shots while Corsi is shots + missed shots + blocked shots.  For this study I decided to use fenwick instead of shots because I had the data handy and I was too lazy to get the shot data in the right format but there shouldn’t be a significant difference (the two are very highly correlated).

Continue reading »

Mar 152011
 

I thought this debate had been fully hashed out already but apparently some people still don’t believe that the game score has an impact on shooting percentage (and shot quality).  The following table shows the shooting percentages by game score over the past 3 seasons (2007-08 to 2009-10) during even strength situations where neither goalie is pulled for any reason (including delayed penalty situations).

Situation Shots Goals SH% Prob<= Prob>
Down2+ 23650 1852 7.83 0.3794 0.6206
Down1 30447 2356 7.74 0.1696 0.8304
Tied 60753 4427 7.29 0.0000 1.0000
Up1 26842 2288 8.52 0.9999 0.0001
Up2+ 19351 1779 9.19 1.0000 0.0000
Overall 161043 12702 7.89 0.5024 0.4976

The Situation, Shots, Goals, and SH% columns are self explanatory.  As you can see, shooting percentage is at its lowest in game tied situations, increases slightly for teams that are trailing and increases significantly for teams that are leading.

The second last column titled Prob<= show the probability (according to a binomial distribution) that that number of goals or fewer would be scored on that number of shots if the expected shooting percentage was 7.89%, the same as the overall 5v5 shooting percentage.  The last column titled Prob> is simply 1-Prob<= and shows the probability of getting more than that number of goals on that number of shots.  So, in down 2+ goal situations, there is a 37.94% chance of their being 1852 or fewer goals scored on 23650 shots which indicates that the down2+ shooting percentage isn’t different from the 5v5 mean at any reasonable confidence level.  The same conclusion can be drawn about down1 situations.  But, the shooting percentages in game tied, up1 and up2+ situations are statistically different at an extremely high confidence level.  Essentially there is zero chance that game tied, up1, or up2+ situations have the same natural shooting percentages as game overall 5v5 situations.  In no way can luck be the sole reason for these differences.

So, does this conclusively tell us that shot quality exists and varies according to game score?  It probably does, but I can’t say it is conclusive as it could mean that teams that trail a lot have bad goaltending (the reason they are trailing) and this results in the team leading having an inflated shooting percentage.  So, what if we looked at shots against a particular team.  Let’s say, for example, against the NY Rangers.  Here is what that looks like.

Situation Shots Goals SH% Prob<= Prob>
Overall 5159 386 7.48 0.5135 0.4865
Up1 843 73 8.66 0.9116 0.0884
Up2+ 485 46 9.48 0.9571 0.0429
Leading 1328 119 8.96 0.9800 0.0200
Tied 2004 138 6.89 0.1658 0.8342

I chose the Rangers because they use predominantly one goalie and that goalie is generally speaking a quality goalie.  As you can see, the confidence levels aren’t quite as strong as league wide mostly because of the smaller sample size but if we combine the up1 and up2+ categories we can say that shot quality against the Rangers when the opposing team is leading is statistically different than shooting percentage against the Rangers overall.

If you are interested in seeing what happens with a team that has had chronically bad goaltending, here is the same table for the Maple Leafs.  We see the same sort of things.

Situation Shots Goals SH% Prob<= Prob>
Overall 5309 491 9.25 0.5120 0.4880
Up1 938 94 10.02 0.8098 0.1902
Up2+ 906 100 11.04 0.9698 0.0302
Leading 1844 194 10.52 0.9712 0.0288
Tied 1985 149 7.51 0.0034 0.9966

So what have we learned.

  1. Shooting percentages vary according to game score.
  2. Those shooting percentage differences can’t be attributed to luck.
  3. Those shooting percentage differences can’t be attributed to goaltending.

That means, it must be the quality of the shots that varies across game scores.  In short, we can conclude that when teams get down in a game they open up and take more chances offensively which in turn gives up higher quality shots against which makes perfect sense to me.

When we combine this with my previous post on the Washington Capitals shooting percentage last season, it is probably safe to assume that shot quality exists and we can’t safely assume that all shots can be treated equal in all situations.

Jan 092011
 

The Los Angeles Kings have signed Jack Johnson to a 7 year contract extension which will pay him $3.5 million the first 3 seasons and $5 million the final four seasons with a cap hit that works out to a cap hit of $4.36M per season.  So the question is, is it a good deal for the Kings?  I am not sure it is.

First, let me start off by saying that I really don’t watch the Kings that much so I haven’t seen Jack Johnson play all that much.  My comments here are based purely on a statistical analysis.  For some of you that makes these opinions objective, for others it probably means you think I am out to lunch, how can you fairly evaluate someone without having watched him a lot.  So be it.

So, lets start off with the good.  Over the past couple of seasons he has significantly improved his offensive output, especially on the PP.  In 2007-08 he had 3g, 11pts in 74 games.  In 2008-09 he had 6g, 11pts in just 41 games.  Last season was a bit of a breakout year for him as he posted 8g, 36pts in 80 games and this season he has taken that up another level with 4g, 31pts in just 41 games.  That said, the majority of his point production increase this season has been on the power play where he has 3g and 21 points or 68% of his points vs 36% one year ago.  Of course, his PP ice time has risen from 2:48 a game to 4:02 a game so that was a factor.  His PP performance so far seems to be coming at the expense of Drew Doughty who has seen his PP points drop significantly this season from last.    He had 23 even strength points last season and is on pace for 20 this season so even strength there is no real improvement.

Now for the bad, or should I say ugly.  It can be shown that statistically he has been and still is one of the worst defensive defensemen in the NHL.  Of the 110 NHL defensemen who have had 200+ minutes in 5v5 game tied situations, Johnson ranks 109th in my HARD+ rating system with a 0.588 score (a 1.00 score would be an average defenseman) which evaluates a players defensive performance while taking into consideration the quality of both his teammates and the opposition he plays against.  This isn’t anything new.  Of the 92 defensemen who played 400+ 5v5 game tied minutes last season Jack Johnson finished dead last in my HARD+ rating.  If you are one of those people who prefer to use fenwick/corsi, Jack Johnson finished 86th of 92 in my FenHARD+ rating last season.

If you don’t fully understand my HARD+ rating systems that’s OK, you can take a look here to see all Kings defensemen sorted by FenF% (Fenwick For / (Fenwick For + Fenwick Against) and you will see that last season he was dead last among Kings defensemen with 50+ minutes 5v5 game tied.  To to be fair, he is a bit better so far this season in FenF% but that isn’t the case in GF% (goals for / (goals for + goals against)).  It seems the coaches are questioning his defensive responsibility as well as his short handed ice time has been cut from 1:35 a game last year to 1:07 a game this year.

No matter how you look at the numbers, Jack Johnson has probably been  somewhere between bad and dreadful defensively thus far in his career and while he looks to be developing into a good, or maybe very good, offensive defenseman, particularly on the PP, one has to wonder if making a 7 year big $$ commitment to him was a wise decision.  It probably isn’t unusual for defensemen to improve their defensive skills as they age but Johnson has a long way to go to even become an average defensive defenseman.  It was a risky signing in my opinion that the Kings may regret down the road.  It’s a lot to pay for a PP specialist, especially when you already have Doughty, a much better player in all aspects of the game including probably the PP, already on your roster.

Dec 162010
 

In the Hockey Statistical Analysis world Tomas Vokoun is an interesting case study because depending on how he gets evaluated he either shows up as an very good goalie or in some cases a true elite goalie in the NHL.  Most ways we evaluate goalies has to do with save percentages.  We either look at overall save percentage or even strength save percentage or even even strength game tied save percentage.  Under all of these scenarios Vokoun excels to various degrees.  A recent Behind the Net Hockey Blog post asked several hockey statistic analysts to discuss “elite goalies” and Tomas Vokoun’s name came up frequently.  What is dumbfounding to me is Vokoun’s record because his won-loss record (79-80-25) is notably worse over the past 3 seasons than his backups (32-22-8).  That can’t be a sign of an elite goalie, even if his backups have been relatively good (i.e. Craig Anderson).  One may postulate it is due to facing tougher competition as backup goalies often get the to play against weaker teams or one may postulate it is just due to bad luck.  Or maybe, he just isn’t a great goalie.

Since shots totals and shooting/save percentage is often affected by game score I’ll focus on 5v5 even strength game tied statistics to balance everything out.  Over the last 3 seasons (2007-08 to 2009-10) there are 35 goalies with 1500 or more 5v5 game tied minutes.  Of these goalies, Tomas Vokoun ranks 8th in 5v5 game tied save percentage which may not be elite, but still very good.  Jonas Hiller tops the list with a .942 save % with Vokoun at .933 and Chris Osgood trails the list with a .906 save %.  So, Vokoun looks pretty good.

But, Tomas Vokoun ranks just 23rd in goals against average which isn’t great and probably average at best.  Those who are in love with fenwick numbers will note that Vokoun has the second highest fenwick against of any goalies with 1500+ 5v5 tied minutes and he gives up so many goals because Florida gives up so many shots and scoring chances.  Of course, I believe that not all shots against are equal and shot totals can be influenced by style of play as much as talent.  If you don’t believe style of play affects shot totals and scoring chances, ask yourself why there are score effects on shot/corsi totals?  The answer is depending on the score, teams play differently.  But teams play differently when the score is tied as well.  Some teams play a defense first style, even when game is tied, and others play a more wide open offensive style.  Florida, without any true elite offensive stars, probably plays more of a defensive game which would naturally lead to more shots against, but not necessarily more quality scoring chances against.

So yes, Florida gives up a lot of shots, but how good is Tomas Vokoun’s competition really.  He does play in the weakest division in the NHL and yet he can’t produce a good won-loss record.  Just looking at Vokoun’s opposition, his opponents rank dead last in goals for per 20 minutes so compared to other goalies he is playing against relatively weak opponents offensively.  His oppositions GF% (goals for / goals for + against) is also fourth worst so overall so he plays against very weak opposition in terms of scoring goals and stopping goals.  For those who prefer Fenwick, his opposition has a FF% (fenwick for / fenwick for + against) of .499, good for 27th among the 35 goalies.  So his opposition isn’t good and his performance in goals against average isn’t good either.  That isn’t a good combination if you want to be considered an elite level goalie.

How about a direct comparison with his backups.  In 2007-08 his goals against average per 20 minutes was significantly worse than Craig Anderson’s (0.949 for Vokoun, 0.538 for Anderson) while Anderson’s opponents had a slightly better goals for per 20 minutes (0.678 vs 0.671).  In 2008-09 Vokoun had a much better season giving up 0.697 goals per 20 minutes compared to Anderson’s 0.896 though Anderson played against slightly better offensive competition.  In 2009-10 Vokoun had a much better goals against than Clemmensen (0.621 vs 1.058) but played against weaker competition as well (OppGF20 of .714 vs 0.743 for Clemmensen’s opponents).  Generally speaking Tomas Vokoun had a very weak 2007-08 season but much better 2008-09 and 2009-10 seasons even though he always seemed to play against weaker offensive opponents.

In terms of my Hockey Analysis Ratings, Tomas Vokoun ranked 16th out of 35 goalies in 2007-10 HARD and 18th in 2007-10 HARD+ rankings.  Middle of the pack.  The seasonal breakdown positioned him 35th of 38 in HARD+ for goalies with 500+ minutes in 2007-08, 19th of 35 in 2008-09, and 6th of 37 in 2009-10.  So far this season he is closer to the bottom again.

Is Tomas Vokoun an elite goalie, or even great goalie?  Probably not.  He just posts good save percentages because his team gives up a lot of shots, but not necessarily quality scoring chances, and he plays against weak offensive competition.

Dec 032010
 

(Updated to include 3 seasons of data as I now realize that more luck data was available)

The other day there was a post on the Behind the Net Blog which used betting odds to estimate how lucky a team was during the 2009-10 season.  In many ways it is quite an ingenious way to evaluate a teams luck and I recommend those who have not read it go take a look.  Last night I was watching, sadly, the Leafs-Oilers game and thinking about luck in a hockey game and whether a team has any control over the luck they experience.   It got me thinking, does a team which controls the flow of the play mean that team is more likely to have more ‘good luck’ stuff happen to them than ‘bad luck’ stuff.

I defined luck as being how many standard deviations their actual point totals were from their expected point totals as defined in the document referenced in the Behind the Net blog post and in an updated document with 4 years of data.  I have only included 3 seasons in this analysis since I have only been working with 3 seasons of data recently and I was too lazy to go back and calculate a fourth season right now.

The most used stat to indicate how well a team controls the play is corsi or fenwick percentage which is basically the number of shots a team directs at the goal divided by the number of shots that they and their opponents teams directed at the goal.  I’ll be using Fenwick % here which includes shots and missed shots but not blocked shots.  So how does Fenwick % correlate with luck?

The correlation is fairly low but a correlation exists.  Maybe good teams can generate their own luck.  Here is a table of a teams luck and fenwick% for 2009-10.

Team Luck Fen%
Chicago Blackhawks 0.777 0.578
Detroit Red Wings 0.395 0.541
Boston Bruins -0.534 0.536
Pittsburgh Penguins -0.156 0.530
Toronto Maple Leafs -1.282 0.528
New Jersey Devils 0.459 0.522
St. Louis Blues 0.186 0.519
Phoenix Coyotes 2.092 0.515
Nashville Predators 1.225 0.514
Calgary Flames -0.590 0.513
Washington Capitals 1.883 0.512
San Jose Sharks 1.020 0.512
Philadelphia Flyers -1.157 0.511
Ottawa Senators 0.083 0.508
Los Angeles Kings 1.040 0.498
Buffalo Sabres 0.302 0.496
Atlanta Thrashers -0.347 0.496
New York Rangers -0.753 0.495
Vancouver Canucks 0.471 0.495
Carolina Hurricanes -0.555 0.491
New York Islanders -0.201 0.490
Columbus Blue Jackets -0.855 0.488
Dallas Stars -0.212 0.480
Anaheim Ducks -0.087 0.467
Tampa Bay Lightning -0.604 0.466
Florida Panthers -0.726 0.465
Montreal Canadiens 0.052 0.464
Minnesota Wild -0.486 0.459
Colorado Avalanche 0.599 0.449
Edmonton Oilers -1.993 0.446

When I was looking through the table something caught my attention.  Of the bottom 15 teams in Fenwick%, only four teams had positive luck.  These were Buffalo, Vancouver, Montreal and Colorado.  Generally speaking, these four teams had good to very good goaltending.  Of the top 15 teams in Fenwick%, only five teams had negative luck.  These were Boston, Pittsburgh, Toronto, Calgary and Philadelphia.  Boston and Calgary had good to very good goaltending (especially once Boston switched mostly to Rask) but Philadelphia, Pittsburgh and Toronto had mediocre to poor goaltending.  That got me to wondering whether goaltending correlated with luck at all so I took a look at the correlation between 5v5 game tied shooting and save percentages with luck.

Like fenwick%, there is an indication of a small correlation between shooting percentage and luck and there is a bit more of a correlation with save percentage.  Next I looked at combining all three factors.  Initially I was going to look at combining all three through some sort of average but then decided to look at goals for percentage instead (goals for divided by goals for plus goals against) since that basically encompasses everything anyway and we find that combined we get a relatively strong correlation with luck.

Now we are getting into correlation that might actually mean something, but what does it all mean?  To be honest, I am not sure.  Regardless of what ‘skill’ we look at there does seem to be a small positive correlation between how good a team is and how good their luck is (as calculated from the betting lines).  Does this mean that a bad team and especially a team with bad goaltending opens itself up to more bad luck than good teams or teams with good goaltending, or does it mean that luck manifests itself mostly in bad goals against or does it simply mean that the people who bet on hockey games trend towards betting the underdog which would push their expected winning percentage up and good teams expected winning percentage down which would result in a poor estimation of luck?  I am not sure how you determine what the exact cause of the correlation is but if it is the latter I have a word of advice, always bet the favourite.